Sustaining high-quality respirometry measurements on animals for estimating metabolic rate and fuel use is challenging. I present a general discussion and selected methods for automated measurements spanning over >4 mo with little need for adjustments or maintenance. 1) Lack of compensation for respiratory volume change may cause 6% error in the results on a fasting animal. The Haldane transformation provides the simplest calculation method for both O2 and CO2 measurements. 2) Use of Nafion tube dryers configured as countercurrent moisture exchangers provides maintenance-free drying of gases, with typical outlet dew points at −25 to −38°C and no CO2 adsorption. The accuracy is improved by low dead space, making it feasible to pass gases though the same dryer. 3) A software algorithm employing a triple interpolation technique allows corrections for automated calibrations of O2 and CO2 with both zero-reference and span gas. 4) Burning known amounts of 100% ethanol provides total system verification of both O2 consumption and respiratory quotient. A calculation method to supply instant CO2 calibration gas from an alcohol burn is presented. 5) Automatic flow switching triggered by low/high O2 thresholds improves accuracy of measurements and safety for the animals experiencing large ranges of O2 consumption; this is a special requirement for recording metabolism in small hibernating mammals.
- oxygen consumption
- indirect calorimetry
- software algorithm
- metabolic rate
- gas drying
while open-flow respirometry has been employed in animal physiology for more than 150 years to produce estimates of metabolic rate and metabolic fuel use (19), obtaining high quality measurements is still a challenge and especially if long-term measurements without regular attendance is desired. Computerized technology helps; however, practical knowledge on how to employ it correctly, together with physical configuration and calibration of systems, has often resided with traditions specific to each laboratory and inherited by its students. Some more general tutorials have lately become available (15, 16, 30). Long-term measurements extending for multiple weeks or months, e.g., in hibernation research, monitoring in animal facilities or the field, is even more challenging due to instrument drift and changes in environmental conditions and can require significant time for maintenance and calibrating the system. I present here a general discussion on open-flow respirometry and methods for automated measurements spanning over days to months with reduced need for adjustments and other maintenance.
Principles and Calculations
An open-flow system includes holding the animals within a metabolic chamber or similar use of a breathing mask. Air inside the chamber is depleted of O2 and enriched with CO2 by the respiration of the subject and is continuously replenished by pushing or pulling gas through the chamber inlets and outlets. The rate of gas exchange by the animal is calculated according to the following general equations: (1) (2) where V̇o2 is O2 consumption (l/min); V̇co2 is CO2 production (l/min); V̇i is incurrent flow rate; V̇e is excurrent flow rate (l/min); FiO2 and FiCO2 are the fraction of O2 and CO2, respectively, in incurrent air; and FeO2 and FeCO2 are the fraction of O2 and CO2, respectively, in excurrent gas. Similar equations apply to any other gas exchanged by the subject. Usually it is only practical to measure one of V̇i or V̇e with adequate accuracy. Respiratory quotient (RQ) = V̇co2/V̇o2, and, unless RQ = 1, we cannot make the approximation V̇i = V̇e (Fig. 1). Haldane (9) pointed out that there can be considerable errors if we do not correct for the respiratory volume change. For example, an animal with RQ = 0.704, typical for a fasting animal, will incorrectly be determined to have an RQ of 0.75 and an error in V̇o2 of −6.5%. This seems counterintuitive, as the relative difference between V̇i and V̇e is typically so small that it would be difficult to measure. However, the products in Eq. 1 are large compared with the resulting difference between them, and this amplifies the effects. The error is much less in V̇co2 where FiCO2 is very low with air as background gas.
There are two principle approaches to correct for respiratory volume change. The first, the Haldane transformation (9), validated by Luft et al. (17), Wagner et al. (27), and Wilmore and Costill (29), is based on the assumption that there is a fraction of gases that passes through the metabolic chamber without being altered by the animal's respiration. This fraction, labeled as “N2” in Fig. 1, may also include other inert gases that are not exchanged. Its incurrent and excurrent rate will be the same: (3) where FiN2 and FeN2 are the fraction of nitrogen and other gases that are not exchanged in incurrent and excurrent gas, respectively. This can be solved for the flow rate that is not measured.
If V̇e is measured, (4a) If V̇i is measured, (4b) Then we can express FiN2 and FeN2 as a function of the fractions of gases that are exchanged, in this case only O2 and CO2: (5a) (5b) Thus, depending on whether excurrent or incurrent flow is measured, we can express: (6a) (6b) The above equations assume that no water vapor is present (all gas rates are at stpd conditions); it can be expanded to include fractions of incurrent and excurrent water vapor, FiH2O and FeH2O, respectively, and other gases that may be exchanged, e.g., methane in ruminants, FiCH4 and FeCH4: (7a) (7b) V̇i and V̇e are now stp flow rates, not stpd rates as in Eqs. 6a and 6b. While we could substitute any of these (30) into Eqs. 1 and 2, it is more efficient to calculate the flow rate numerically in a computer program and apply the numerical result to Eqs. 1 and 2 or similar equations for exchange of other gases. As an example of how this can be implemented, Fig. 2 shows the listing of a principal Pascal code segment from the author's own data acquisition program [a modified version of LabGraph (24)]. Each nitrogen fraction is calculated in a de-accumulation loop after initiation to unity, and another loop calculates the gas rates. The calculations are similar for all gases, with the exception that a sign factor is applied to express results as positive for either gas produced or consumed by the subject. If any gas is absorbed before measurement, either incurrent or excurrent, the fraction of that gas will be zero (the Boolean flags GasInPresentFlag[Gas] or GasPresentFlag[Gas] set to false), and thus not included in the de-accumulation loop. If CO2 and H2O are absorbed from incurrent and excurrent air, the equation will become similar to Depocas and Hart (5), Eqs. 8 and 10, respectively, expressed in fractions instead of partial pressures; if CO2 is only absorbed from excurrent air, the equation will become similar to Hill's (10) Eq. 6.
A special case is where excurrent CO2 is absorbed from a subsample and not before measurement of excurrent flow. This may be the most practical way to absorb CO2 in a large-animal setup with high flow rate. Withers' (31) Eq. 4b describes this situation; however, this equation requires CO2 to be absorbed from incurrent air too. Instead, V̇e can be corrected to the CO2-free value: (8) where Fe′O2 and V̇e′ are fraction of O2 and flow of excurrent air after absorption of CO2, respectively, both of which can then be used in the Haldane calculation. This correction is very insensitive to RQ, only 0.06% for a change in RQ from 0.7 to 0.85; assuming a fixed RQ of 0.85 will give satisfactory results under normal conditions. Thus the Haldane-based algorithm in Fig. 2 is very flexible and can be adapted to a number of conditions, provided that proper flow corrections are performed.
A second approach to correct for respiratory volume change is using as starting point that (9) Substituted into the general Eq. 1, it becomes Depocas and Hart (5) Eq. 3. It is assumed that no other gases than O2 and CO2 are exchanged. Thus, even if N2 fractions are not part of the derivation, it would have the same problems as the Haldane transformation, if there were any exchange of N2 or other gases that are not measured. In the case where only V̇o2 is measured because only an O2 analyzer is available, and we do not absorb CO2 but RQ is known or assumed, this second approach becomes useful as V̇co2 = RQ·V̇o2. If V̇e is measured, it results in Wither's (31) Eq. 3a: (10a) If V̇i is measured, it results in Wither's (31) Eq. 1d after converting to dry gas: (10b) Interestingly, if results from Eqs. 10a or 10b are converted to energy units according to Kleiber (12), and an intermediate RQ is assumed, most of the errors resulting from deviation of actual RQ from the assumed one will be canceled out (13).
If V̇co2 is also measured, it can be described by Withers' (31) Eq. 3b: (11) With the approximation that V̇co2 = V̇e·(FeCO2 − FiCO2), one can substitute and arrive at an equation that contains expressions with O2 depletion and CO2 enrichment, (12) Similar equations have been developed for other conditions of flow measurements; a complete set based on the second approach can be found on the Warthog Analysis X software documentation, as developed by Chappel (3) and repeated by Lighton (15). While the errors with air as background gas are negligible compared with the Haldane transformation, some care must be taken with unusual gas mixtures.
Because of the ease of processing equations and corrections in computer programs, the author prefers the mathematically correct Haldane transformation, which will allow any gas mixtures. The exception is where there is no inert gas present, e.g., incurrent gas is a mixture of O2 and CO2 or only pure O2. Those conditions becomes problematic with the second approach too; Eq. 11 and the corresponding equation where V̇i is measured will either fail or produce very large errors with 100% O2 as background gas. Equations 10a and 10b will also be very sensitive to incorrectly assumed RQ under those conditions; assuming an RQ of 1 when RQ is 0.7 will now result in an error of −30% in V̇o2.
While metabolic chambers scale to the volume of the subject and thereby the body mass (BM), metabolic rate of mammals and birds scales approximately to BM0.75. Thus to obtain similar differential gas concentrations, the chamber for a large subject will inherently need a much lower flow rate relative to the chamber volume and thereby a longer time constant than for a small subject. (A mask is sometimes preferred in very large animals.) Several different flow arrangements are possible though the metabolic chamber. The configurations of a push (positive pressure) system where V̇e is measured and flowmeter is downstream, or a pull (negative pressure) system where V̇i is measured and the flowmeter is upstream should be avoided if possible, as no leaks in the metabolic chambers can be tolerated without causing errors. This leaves us the following.
1) A push system where V̇i is measured; flowmeter is upstream. Chamber leaks can be allowed; however, the sample from the outlet of the chamber is then to be regarded as a subsample from the chamber. Thus the air must be completely mixed within the chamber, usually with one or more fans for the gas sample to be representative.
2) A pull system where V̇e is measured; flowmeter is downstream. This is the flow arrangement preferred by the author for most situations and also applicable to use of a face mask. Chamber leaks can be allowed, provided that supply to the chamber is from surrounding air without absorbing CO2 from the incurrent airstream, and that flow rate is high enough to prevent any loss of respiratory gas from the chamber. Thus speed of air though any inlet to the chamber, whether through a designated inlet or through a leak, must exceed the rate of a diffusion front out of any of those openings. Pressure inside the chamber must always be lower than outside pressure. A consideration in this respect is barometric pressure (Patm) changes due to breathing; the chamber is, in effect, a whole body plethysmograph (6). These respiratory pressure changes can be measured with a sensitive differential pressure transducer and compared with pressure differences with and without flow through the chamber to evaluate integrity of the seals. A long flow path into the chamber can create a buffer zone that may prevent respiratory gas from escaping. Another cause of leaks can be convective flow driven by heat loss from a subject in a cold environment. Because of this chimney effect, special attention should be paid to sealing near the top of the chamber. With proper design of a pull system, all respiratory gas will eventually be caught and passed through the flowmeter; thus even without mixing of chamber air, gas rates will be correct when integrated over time. The lack of mixing may create artifacts in the time courses, if the subject moves within the chamber and changes proximity to the outlet, and also with sudden mixing of previously unmixed respiratory air. Distribution of outlets at multiple points throughout the chamber may reduce artifacts. Convective flow, as noted above, may also contribute to mixing. If feasible, a mixing fan should be used to minimize artifacts.
Response Time Correction
Due to the time needed to exchange the volume of air in the metabolic chamber, the change in fraction of a gas in the excurrent airstream is in a non-steady state and only gradually approaches equilibrium. With complete mixing, the volume of the respirometry chamber will work the same way as the capacitor in an electrical RC filter, the changes in fractions of a gas out of the chamber described by an exponential function of time. The Z-transform can then be used to calculate back to the initial instantaneous excurrent gas fraction (2). A two-phase correction to be used with dual compartments in the sampling system was suggested by Frappell et. al. (8). Lighton and Halsey (16) provide a further discussion of issues with response time correction. Instead of correcting the excurrent gas fractions, applying response time correction to calculated O2 works well, as transient effects of RQ changes on O2 are very small. Response time correction amplifies any noise in the signal. As in the authors' own data acquisition program, noise can be reduced by applying a sliding average over a few measurements before response time correction of V̇o2. An interesting variant of this is using a polynomial line-fitting method to smooth the signal before response time correction (20). Artifacts from incomplete mixing of chamber gas may also be strongly amplified, if response time correction is attempted. For limited measurement periods, the author has successfully applied it to animals that remain immobile. Because of difficulties with correct determination of the effective chamber volume, it is easy to overcompensate, with resulting over- and undershoots. It is suggested that both uncorrected and corrected traces should be presented in publications to offer readers the possibility to evaluate the validity of the correction. Perhaps a conservative approach that undercompensates may be better than one that overcompensates and creates spike artifacts.
Water Vapor and Sample Drying
The Haldane transformation can include water vapor fractions, and no further corrections are then needed to do “wet” measurements, provided that both FiH2O and FeH2O are known and V̇i or V̇e can be correctly determined under those conditions (21). Water vapor pressure (Ph2o), usually derived by applying proper equations to relative humidity (RH) and temperature data, or obtained directly from an instrument, is converted to fractions by Fh2o = Ph2o/Patm. An alternative is to correct both V̇i, V̇e, and each gas fraction to dry conditions (20). However, typical accuracy of RH sensors is ∼2–3% RH and exhibit an additional 1–2% hysteresis. At 25°C and 50% RH, an error of only 1% RH and 0.1°C in measurement of gas temperature can translate to 1.8% error in V̇o2. Calibration of Ph2o difference of dry and humid reference gas to the corresponding O2 deflections can improve this accuracy (20). Careful attention would also have to be paid to the material of tubing and filters. Opposed to Teflon-based filters, the glass fiber-based filters used in many popular gas analyzers will adsorb H2O and may cause apparent drift before they have equilibrated. Tygon tubing also adsorbs H2O. The humidity meter is an additional instrument to keep calibrated, which adds complexity when performing long-term measurements. Also, if the infrared CO2 analyzer is not of a dual-wavelength type that subtracts influence of H2O vapor, the CO2 reading may be affected by the H2O vapor content of the sample. Alternatively, the gas can be dried before measurements.
Drying of the main airstream before flow measurement may be impractical with large subjects and high flow rates, where high capacity drying canisters would be needed; thus drying a subsample may be a necessary. This causes the need to correct the flow measurement from wet to dry conditions (30): Where V̇e″ and V̇i″ are the wet flow rates when dry gas concentrations are measured.
[One additional source of inaccuracy when using a standard mass flowmeter or controller is that, due to the higher heat capacity, the H2O vapor component of the gas contributes 1.14 times the actual flow of that fraction (26). This can be corrected by multiplying FeH2O or FiH2O with this factor before subtraction in the above flow corrections; while the correction is small, it can be important at high humidity.] The errors resulting from measuring and correcting a humid flow to dry flow is much less than when analyzing and correcting wet gases to dry gases; in the extreme case where it is assumed that RH is 50% at 25°C in the excurrent airstream and the actual humidity is 0% or 100%, the maximum error in V̇e would be ±1.6%. Thus it would be adequate to estimate humidity within ∼10%, which is well within long-term stability of commercially available humidity sensors. Ignoring the correction in an experiment at low temperature would cause very small errors; for instance a dew point of 0°C would only cause 0.6% error in V̇e.
Gas drying represents a problem in itself. A number of common desiccants adsorb CO2 to varying degrees, which will interfere with both the CO2 and O2 measurement. Figure 3 shows the O2 and CO2 response out of a drying canister to a step change in incoming concentrations of O2 and CO2. The new Drierite affected both CO2 and O2. Desiccants that adsorb CO2 should only be used in combination with CO2 absorption before O2 measurements. Partially depleted regenerated Drierite had less CO2 adsorption than new. However, it still took 4 min to 99% response, which is about 10 times that reported by White et. al (28) for recharged Drierite with the same flow rate and canister size. The 99.8% response required for accurate calibrations at 10 ppm resolution (assuming a 0.5% span) was reached after 5 min. The differences could possibly be caused by batch-to-batch variation or initial hydration state.
Of the desiccants tested for effectiveness (Fig. 4), only Mg(ClO4)2 and CaCl2 passed CO2 through without any adsorption. While CaCl2 can be regenerated a limited number of times by heating to 225°C, it is one of the least effective desiccants and transforms into a corrosive salt solution near depletion. Mg(ClO4)2 provides a very low Ph2o; however, it presents an explosion hazard if contaminated by combustible matter (4). It fuses to a very solid mass when depleted that is very difficult to remove. Using large quantities of it in long-term measurements could be both a problem of cost and safe disposal. Molecular sieve 3Å provided the best and most stable drying until sudden depletion. It is inert, releases the smallest amount of dust, and can be regenerated indefinitely by heating. It should only be used when CO2 is absorbed from the sample air. None of these desiccants had enough capacity for a 30-ml canister at 200 ml/min flow rate to last for 1 day; a 100-ml canister would be needed to provide some margins. Equilibration time after a gas switch (e.g., between sample and reference gas) would then be too long for it to be feasible to pass all gas through the same canister. If a drying canister is placed on each gas channel, gradual depletion of the desiccants at different rates may cause differential errors in the gas readings.
Alternative Gas Drying: Nafion Tube Dryers
Nafion is a Teflon material that contains strong acid sulfur groups that bind to polarized molecules like water vapor and will let nonpolarized molecules like O2 and CO2 through unaffected. Water vapor is chemically transported through the Nafion material, if there is a difference in Ph2o on the two sides. In the form of one or more parallel tubes (Perma Pure, Toms River, NJ), it can be arranged as a countercurrent moisture exchanger (Fig. 5). After filtering, subsample air is pulled through the Nafion tubing to the gas analyzers at low flow rate (typically 100 ml/min), after which pressure is dropped with a precision needle valve or mass flow controller before it is pulled at reduced pressure in countercurrent direction though a common sheath on the outside of the Nafion tubing to the pump (e.g., NMP830 BLDC, KNF Neuberger, Trenton, NJ). The pressure drop past the needle valve causes a reduction in Ph2o and drives a multiplicator effect. The gas passing though the analyzers will become progressively dryer until equilibrium is reached, with an indefinitely stable low Ph2o. The low dead space of the Nafion tubing (typically 1–12 ml) makes it feasible to pass all gas through the same dryer opposed to the large desiccant canisters that have to be connected in each sample and reference gas stream. While already assembled dryers with one or 50–100 tubes (PD-50T-24-PP and PD-50T-12-PP tested by the author) are available from the only supplier of Nafion tubing material (Permapure), an example of a minimal dryer constructed by the author from three 0.060-in. tubes (TT-060, 0.052-in. inner diameter and 0.063-in. outer diameter) of 30-in. lengths is shown in Fig. 6. These were inserted into an adjusted length of 0.25-in. inner diameter Bevaline tubing with a T-piece glued into each end with marine epoxy. The tubing should not be too dry when assembling, as it may expand and crumble if it gets moist. The dryer can, before gluing, be curled into a loop that could fit into an instrument and has a material cost of ∼10% of the cost of a 50-tube 24-in. dryer. (If higher flow rates are required, more Nafion tubes could be added, and dimensions of the outer sheath and the T-pieces increased.) Figure 7 shows a test of this 30-in.-long dryer at a flow rate of 100 ml/min. The nominal inlet air was kept at a dew point of 20°C. Most of the dryer was submerged in the water bath, controlling the inlet dew point, except the first 10–15 cm, which was exposed to room air at ∼24°C to avoid condensation. The pressure in the reflux of the dryer was ∼0.25 atm. It took 10–15 h for the dryer to equilibrate. The final equilibrium dew point reached was −26.3°C, corresponding to Ph2o of 0.056 kPa. This will only decrease O2 concentration by 0.011% and can easily be compensated by baseline correction. A similar test on a 24-in. PD-50 series dryer took >40 h to completely equilibrate, and final Ph2o reached was 0.023 kPa, corresponding to a dew point of about −35°C. A 12-in. PD-50 series dryer showed intermediate response. All of these dryers could, in an emergency, be used after 3–5 h, if proper baselining technique is used; however, it should be planned to let them run overnight before use when starting from a humid state. Figure 8 shows the response of the 3-tube 30-in. Nafion dryer to alternation from inlet air with 20°C dew point to completely dry air for 3 h and back again. It took 2 h before a change in outlet humidity was observed. Thus the dryer, in addition to drying the gas, has a high buffer capacity against extreme changes in inlet humidity. It is concluded that gas switches will not affect outlet humidity in a way that cannot be compensated by hourly baseline corrections. The 50-tube dryers provide even better buffering.
The total capacity of the Nafion dryer to transport water is enhanced at high temperature. On the other hand, if equilibrium is reached toward the end of the dryer, it is the residual Ph2o of the Nafion membrane that will eventually determine the Ph2o at the outlet, and low temperature gives lower residual Ph2o (14, 22). This effect is not buffered and occurs almost immediately. Figure 9 shows the outlet Ph2o of the dryer at different temperatures (most of dryer submerged in water bath, except ∼15 cm at inlet). From this, it can be calculated that a short-term temperature fluctuation of 3°C, which may be a typical experimental room condition, would cause a fluctuation in the O2 reading of ∼20 ppm. This is <5% of the temperature effect on typical fuel cell sensors. Thus only in cases in which a temperature-controlled O2 analyzer is used to measure very small differences in O2 concentration would the temperature effects on a Nafion dryer have to be considered. This could be remedied with good temperature buffering of the dryer (reducing short-term effects) or temperature control of the dryer outlet.
Nafion dryers used as countercurrent moisture exchangers will provide long-term drying of gases with a minimum amount of maintenance, provided that analyzer flow rate is dimensioned to the capacity of the dryer (e.g., 100 ml/min for the 3-tube dryer). It is important to avoid contamination of the dryer by filtering at <2 μm on the dryer inlet. Gas that reacts with the Nafion membrane, such as ammonia, may damage the dryers, and ammonia scrubbers should be used where there is a chance that it may be released (i.e., from protein breakdown in bedding material, some domestic animal applications).
Drying capacity is a compromise between having a long and efficient dryer and causing too high pressure drop across the drier. During a practical test with the 30-in. three-tube dryer, where flow was controlled at 100 ml/min with either a precision needle valve or several different brands of mass flow controllers (STEC 4400 with piezo valve, Tylan FC-260 with heat expansion valve, Sierra Instruments 840 and Precision Flow Devices with solenoid valve, pump KNF MNP830 BLDC running from a regulated power supply), pressure fluctuations recorded over a 10-min period would only have contributed to 2–4 ppm variation in recorded O2 concentration. Total pressure drop was only 0.16 kPa across the dryer plus a Whatman 1-μm PTFE disk. Thus the flow resistance of this dryer would normally have insignificant effect on the accuracy of the O2 measurement. Long-term changes in analyzer pressure can be compensated with proper baselining.
There are hundreds of ways to connect components of a respirometry system, and the optimal configuration depends on the particular application and what equipment is available. Some considerations are to avoid pressure fluctuations in the measuring cell of an oxygen analyzer, assessment of sensitivity to leaks in different parts of the system, the expected duration of experiments, the range of V̇o2 and V̇co2 expected, and what degree of flexibility and portability is needed. The following shows four different systems to illustrate some issues.
Figure 10A shows a very basic system for recording from a single animal with subsampling of analyzer flow. If experiments are of short duration, then reference air could be sampled before and after the experiment without an animal in the chamber. Longer experiments (typically more than 1–2 h) require reference air to be sampled though a dedicated flow path. A separate pump for reference air/zero gas will be needed. In the figure, reference air is manually connected, but also could be switched with solenoid valves. The connection point is here shown upstream from the Y-piece, where excessive flow is vented. At high flow rates, it may be more practical to move the connection point downstream from the Y-piece. An issue is that it can get challenging to keep a stable flow through the analyzers when switching gas sources. Also, responses to gas switches will be slow; the 30-ml CaCl2 canister in Fig. 3 would require 2.75 min to equilibrate within 10 ppm at 200-ml flow rate. If the canister volume is increased to 100 ml for a drying capacity of 1 day, equilibration time would increase to 9 min.
If an experiment lasts for months, it may be necessary to record from multiple animals at a time. Figure 10B shows an example of a system that was built from existing components and used to measure metabolic rate of hibernating bears (25). Because of the need to switch between animals every 5 min, the large drying canisters (that had several days' capacity) had to be moved to each subsample stream. The system was not ideal, as differences in rate of depletion of the drying canisters could cause inaccuracies in the measurements if the CaCl2 desiccant was not changed often enough. Also, as solenoids only were available for switching between the two metabolic chambers and reference air (FiO2, in this case outside air, also used as baseline for CO2 span), span gas had to be connected manually and was limited to two span calibrations per day.
Due to the large amount of maintenance work in long-term studies with 6 mo of continuous recording, that system was further developed with introduction of Nafion dryers and full automation. Figure 11 shows the final system that was completely rebuilt with new equipment. Only one-half of the system is shown for clarity; the actual system consists of two parallel systems, each measuring from two different animals at a time with 50% time coverage (minus transition time) for a total of four animals. All parts shown are duplicated, except that flow paths for reference/zero gas and span gas were shared up to the point of the dual multiplexer, as only a total of eight flow control channels were available (Flowbar-8 flow monitor and RM-8 gas flow multiplexer version 3, Sable Systems International, Henderson, NV), of which two were used by the analyzers. A flow of 400 ml/min was kept through the shared flow paths vs. a subsample flow rate 200 ml/min through the animal sample paths. The presence of dedicated subsample pumps helps to keep subsample flow stable, even if the main flow (1–50 l/min) is changed. In smaller animals at low metabolic rate, the sample pumps and flowmeters may be omitted, and subsample pumps and flowmeters used instead. Subsample flow is filtered, monitored with additional flowmeters on all channels, and passed into the computer-controlled gas flow multiplexer. The latter was replumbed so that two switching solenoids are independently connected, and they were rewired to conditionally activate with two multiplexer channels.1 After passing through an optional humidity sensor (see Water Vapor and Sample Drying about correction of flow rate for water vapor), samples are referenced to atmospheric condition by venting one-half of the flow through a 10-ml syringe barrel or a Y-shaped vent and then filtered again before passing through the Nafion dryer. The internal filters of the gas analyzers (one dual-channel Oxzilla2 with channels used independently and two CA-10As, Sable Systems International) were removed as they add response time and tend to have some water vapor adsorption, thus interfering with the function of the Nafion dryer. Analyzer flow (100 ml/min) is monitored with a mass flowmeter and controlled with a precision needle valve. Both of these were located in the flow monitor (Flowbar-8), which was replumbed to connect the needle valve after the flow sensor and remove the internal filters for those two channels (see footnote 1 above). The needle valve decreases pressure to ∼0.25 atm in the reflux path. Both the analyzer pump providing the vacuum and the sample pumps are similar, except that all subsample pumps have an additional option for electronic speed control (KNF MNP830 BLDC 4-wire option).
For cost saving, both subsample flow monitoring and analyzer flow monitoring can be performed with inexpensive rotameters with flow control valves. The addition of the Nafion dryer to this configuration compared with Fig. 10B only required the extra cost of the dryer and a precision needle valve. It saved the cost of the drying canisters. The simple system in Fig. 10A could likewise have been improved by adding a Nafion dryer, a pump, and a needle valve, and would additionally have solved the problem with keeping analyzer flow constant. While the system in Fig. 11 is technically a pull system, major parts of the system have positive pressure and is thus insensitive to small leaks; the sensitive parts are between the main mass flowmeter and the subsample pump and between the pressure reference point before the dryer and the analyzers. Leaks in the dryer reflux circuit could decrease dryer performance and should also be avoided. Systematic leak testing of all parts of a system with a pressure sensor after initial assembly can save a lot of work in error seeking at a later stage.
Only the metabolic chambers, sample pumps, and main mass flowmeters in Fig. 11 need to be scaled to animal size; the rest of the system within the dotted line can be kept unchanged with animal sizes from mouse to elephants and could possibly be built into an integrated portable unit in the future, reduced in size. More typically, a respirometry system requires a whole rack or ∼2–3 m bench top, with addition for the data acquisition components and computer.
For very small animals (typically arthropods), a push system as in Fig. 12 may be more optimal. A dual gas flow multiplexer is used both on the upstream and downstream side to allow use of a single mass flow controller for all active chambers. An alternate supply provides flow though the chamber when not connected. Any inaccuracy in flow when the chamber is inactive will be quickly equilibrated, as the chamber volume is only 10 ml. To match any background diffusion through tubing, etc., a reference chamber or equivalent is used. Another, but more expensive, alternative to the dual-multiplexer solution would be a manifold and supply for each chamber through it own mass flow controller on the upstream side. The connection for the span gas is located after the multiplexer due to the much higher concentration of CO2 in the span gas (compressed air) than sample gas. Otherwise, it could bias one branch of the multiplexer if CO2 sticks to the tubing walls or remains in tubing after a channel switch. The atmospheric pressure reference point is placed after the analyzer, which will minimize pressure fluctuations in the analyzer. This is particularly important if an O2 analyzer is added and the O2 extraction is very low. Careful control of analyzer flow relative to sample flow is necessary to avoid drawing in moist air in the reflux path of the dryer. The system could also have been designed with a pressure reference point before the filter into the Nafion dryer, as in Fig. 10. A smaller Nafion dryer (based on a single 24-in. TT-60 tube) is adequate with the lower flow rate of 25 ml/min used in this setup. As the supply gas to the chambers is dry after CO2 removal, some organisms may require the gas to be humidified.
Whether to use a gas flow multiplexer to expand the number of animals studied at a time at reduced equipment costs or just using it for calibrations depends on what data are of interest and the response time of the chamber. In the setup for hibernating bears in Figs. 10 and 11, the time constant of the metabolic chamber was so long that multiplexing did not cause distortion of the response curve of measured V̇o2. When a similar system was applied to 1-kg arctic ground squirrels, the switching every 5 min with <50% time coverage resulted in some jagged artifacts in the shape of the curve. Shorter switch intervals would be problematic, as the fuel cell analyzer and multiplexer needs 2 min to equilibrate after every switch. Thus each animal's data were only obtained for 3 min every 10-min period. It still allows reliable measurement of stable minimum/basal metabolism. More than two animals on each analyzer group would be problematic. In the small-animal setup in Fig. 12, a switch every 5 min would just barely allow equilibration of the analyzer and hide the shape of the metabolic response curve. Instead, it was set to do a full measurement period of 20 min for each animal before a change to the next experimental condition. Thus the number of animals was only affecting the time of the total experiment and the need to maintain the animals in a healthy state in the chamber for the total duration of the experiment.
The traditional method of correcting a system for drift is to pass background gas and calibration gases through the analyzers at regular intervals and then adjust analyzer controls accordingly. A similar approach could be used for corrections in software, using the last known calibration data to adjust data forward from that point. Data will only be correct directly after a calibration. Subsequently, analyzer drift may cause a sloping curve of an otherwise stable signal until the next adjustment, where there will be an apparent step change. Whether this approach is accurate enough depends on the amount of drift between calibrations and required precision. For instance, if an analyzer only drifts by 10 ppm over a 1-h interval while measuring a 0.5% O2 extraction or CO2 enrichment, this simplified approach may be adequate. In practical situations, drift is often much larger, and, in an automated computerized system, the software can instead interpolate between calibrations. Commercially available software can provide various means for doing baseline corrections by interpolation for the background gas; for instance, a splining interpolation technique was recently described for very frequent baselining intervals in combination with channel consolidation (20). However, for longer calibration intervals providing less data, the more conservative approach of linear interpolation may be more suitable. In long-term measurements also, span drift, particularly on the CO2 measurements, is likely to occur. The nominal span is the distance between the nominal span gas value and the nominal zero gas value (typically CO2-free air). These two types of calibrations cannot take place at the same time. Thus, unless the system is virtually free of zero drift, it would be mathematically inaccurate to calculate measured span as the recorded span gas calibration value minus the last zero calibration value, as the system could have drifted between the two calibrations. Instead, the surrounding zero calibrations can be interpolated to time of span gas calibrations to calculate a span correction. Span corrections can then be interpolated to the time of a measurement. To apply the span correction to a measurement, the surrounding zero gas calibrations interpolated to time of measurement must be subtracted first and nominal zero value added after span correction. The following describes in more detail how this triple interpolation technique can be implemented, as exemplified by the authors own data acquisition software.
During measurements, the program records multiplexer state for subjects and calibration gas switches in a status variable (one for each analyzer group). Then a calibration search unit in the software buffers valid calibration values and corresponding times. If data are recorded into multiple files, the search also includes files recorded before and after the currently searched file, as those calibrations may affect corrections in the present file. The search skips calibration measurements during a specified equilibration time right after a gas switch and then averages valid calibration data for the rest of the calibration period. (Similar equilibration times apply when selecting valid measurements when switching from gas to animal measurements or with multiplexed animal switches.)
For each recorded data point of excurrent chamber gas, the nearest buffered span calibrations and corresponding times on each side are selected. Then the closest zero gas reading before and after each of those two span calibrations are interpolated to the time of the span gas reading. Each measured span is then calculated as the difference between span gas reading and the interpolated zero gas reading. The span correction at each span gas reading is calculated as nominal span divided by measured span. These span corrections are then interpolated to the time of the chamber measurement. The closest zero gas readings are also interpolated to the time of the chamber measurement, subtracted from the measured chamber gas value, the result multiplied with the span correction, and then the nominal zero gas value finally added. If a stable source of outside air is used as incurrent air to the chamber, this reference air can be used as “zero” gas. Thus reference calibrations will then serve both as zero gas and reference gas, and the reference values will be corrected to the nominal reference air value. (The latter is similar to baseline drift correction available in commercial software.) If a dedicated zero gas is used, the measured reference air values will be corrected for span and zero drift the same way as the excurrent chamber gas, and the corrected reference air value will reflect true variations in concentrations of gases in incurrent air to the chamber. All of these corrections will be mathematically precise, as long as instrument drift is linear; intervals between calibrations need to be adapted to the instrument drift pattern so that drift can be approximated by linear interpolation. If data are not followed by calibration(s), the last valid calibration points are used instead of interpolations.
The program code carrying out all of the multiplexer control, calibration corrections, and respirometry calculations, including initiation code, optional response time corrections, and lag correction, support for multiple animals and multiple analyzer groups, etc. is extensive. For verification of this code, a number of simulated data sets were designed in Excel. First, an initial set was constructed that did not include simulated analyzer drift and resulted in a linear increase or decrease in V̇o2 and V̇co2 with time and a constant RQ. Then sinoid type drift of span, zero, and reference air was mathematically introduced into all of these data with a period close to, but not exactly, 1 day and phase shifted between the different gases. Simulated calibration values (approximately once per hour in a pattern shifted between analyzer groups) were subjected to the same simulated drift. Figure 13 shows this data set for one channel without corrections (FiO2 is assumed constant), with baseline correction only (reference air value is used to correct all data so that reference air stays at the nominal value), and with full correction of zero, reference, and span. The system is able to correct for drift of a magnitude considerably higher than the measured signal. The most sensitive signal, RQ, shows only slight deviations from the nominal value of 0.6667. The small discrepancy at the end of the data set is due to inclusion of a test for lag correction (featuring separate correction for delays from the multiplexer to the analyzer, and delays upstream from the multiplexer); thus the correct data for the last calibrations would be past the end of the last file of the data set. The effectiveness of these linear interpolation techniques depends on the drift pattern and proper calibration intervals. A major factor influencing instrument drift is temperature environment of the instruments. Changes often happen slowly with weather changes or in a diurnal pattern. As indicated in Fig. 13, extreme drift that could be caused by diurnal temperature fluctuations can be effectively corrected with hourly calibrations. Likewise, drift caused by inadequate or missing Patm compensation would usually be correctable due to the slow nature of those changes. More problematic are the quicker oscillations sometimes caused by room thermostats. In those cases, it might help to buffer the instrument by adding insulation that dampens the oscillations. In an outdoor location, shielding from intermittent sun exposure would be important.
Figure 14 shows these correction methods applied to a 5-mo record in a hibernating American black bear using the respirometry setup in Fig. 11. (The experiment was part of a study approved by the Institutional Animal Care and Use Committee at University of Alaska Fairbanks.) The analyzer controls were only adjusted two times during the entire period. Daily mean RQ is shown instead of 5-min averages, as short-term variability in respiration appears to cause larger fluctuations in RQ.
Total System Verification
Due to the complexity of both the hardware and data processing in a respirometry system, total system checks by simulating metabolism of an animal in a way that can be measured by other means should be performed whenever practical and before and after the study period, as errors could gradually appear over time and bias results. In long-term studies, permanent long-term drift in components, e.g., flowmeters, becomes even more serious when it has more time to accumulate; thus total system checks should be considered mandatory whenever feasible. A system can be tested by infusing a known flow rate of nitrogen into an empty chamber (1, 7). This requires accurate measurement of nitrogen flow, and flow measurements are one of the methods that often are subject to errors. A better alternative is to simulate an animal's metabolism by burning known amounts of a 100% pure alcohol like ethanol, methanol, or propanol, with a small clean-burning lamp or a gas like high-purity propane or butane (5, 15, 18, 20, 30), as this only requires an accurate scale. This has the added advantage that it provides a check for both V̇o2 and V̇co2; RQ can be derived from the reaction equations and is 0.6667 for all three alcohols, 0.6000 for propane and 0.6154 for butane. The amount of O2 used is 1.4596 l/g ethanol, 1.0493 l/g methanol, 1.6784 l/g propanol, 2.5415 l/g propane, and 2.5066 l/g butane. Alcohol lamps can be constructed from one or more 16-gauge cutoff needles with a wick thread and an additional 25-gauge needle (to equalize pressure) inserted into a silicone stopper in a small flask of low height. Wicks are adjusted to give a clean burning blue flame, typically with the top of the wick 1 mm below the top of the needle (only tested with ethanol). A lamp can be inserted into the chamber immediately after lighting, and recording lasts until the lamp is burnt out and washout completed, followed by a full calibration. The area under the V̇o2 curve of the burn is then compared with the amount calculated from the weight loss of the lamp. While there are small errors associated with loss of burn gas before the chamber is closed, this is small and usually negligible if the lamp is quickly lit without letting breath or burned gas from a lighting tool enter the chamber. Evaporation of remaining alcohol after the lamp has burnt out could be a problem with long washout time of large chambers. However, if ambient temperature is low, this effect is also quite small (e.g., when the system in Fig. 11 was new and flowmeters recently calibrated, all four chambers gave results within 1% of the expected). Alternatively, if weight of the lamp or gas burner can be determined continuously, the expected V̇o2 could be determined instantaneously and compared with recorded V̇o2 (15, 18). Constructing clean burning lamps simulating very small or very large animals can be problematic. An alternative for large animals is to burn steel wool and determine the increase in mass of the steel wool due to oxidation (32). While the method avoids problems with potential background evaporation, it does not provide any check of the RQ.
If calibration gases available from commercial suppliers are not analyzed to high enough precision, an independent determination of the CO2 content is called for. Provided that one has a correctly calibrated or corrected O2 analyzer, repeated alcohol burns as described above can be used with an iterative method. For each burn, the nominal CO2 value input into the correction equations of the software (assuming span corrections are available) is adjusted repeatedly until calculated RQ in the plateau phase of the burn is 0.6667. About six burns are usually needed for an accurate average. An ethanol burn can also be used directly as an instantaneous calibration gas, if one has a correctly calibrated or corrected O2 analyzer. By using a small separate respirometry chamber with an alcohol flame supplied with air from the same source as the chambers with the subjects, the CO2 fraction of the dried burn gas drawn from this chamber can be calculated as: (13) Use of this equation is not restricted to manual adjustment of a CO2 analyzer before an experiment starts. Regular CO2 span corrections (as described under calibration corrections) can be performed with nominal CO2 values derived from this equation during an experiment, if FiO2 and FeO2 are subjected to corrections before application of the equation. Difficulties with keeping an ethanol flame burning in experiments lasting for several days could possibly be mitigated by substituting the lamp for a small gas burner. Multiple calibration points at different CO2 concentrations can be achieved by diluting a burn gas to varying degrees with reference air, allowing check of the linearity of a CO2 analyzer over an extended range (limited upwards by O2 extractions too high to sustain a clean burning lamp).
Large Dynamic Range of Measurements
Respirometry in deep hibernators (for instance, arctic ground squirrels) presents the challenge of measuring a 200-fold range of metabolic rates between minimum rates during torpor and peak metabolic rate during arousals (11). Gas analyzers do not provide sensitivity and stability enough to provide accurate measurements over this range, unless flow rate is adjusted to reflect metabolic rate changes. A low flow rate used to measure torpor metabolic rate may cause hypoxic conditions when the animal arouses. Thus the animal needs careful monitoring for signs of arousals. Automatic regulation of flow rate to provide nearly constant O2 differential may give the best accuracy in RQ. The chamber gas concentrations may, however, never reach equilibrium, and it is difficult to use a single pump to provide large enough range of flow rates. Instead, automatic flow switching, triggered by differential O2 thresholds between the sample and the subsample flowmeters and pumps (Fig. 15) was added to a respirometry system similar to the one in Fig. 11. The system takes into account shifts in baselines before testing for the O2 switching thresholds. The switching is effected by three-way low-resistance valves (model BL-1, Sable Systems International) mounted upstream from the main sample pump, but after the branching for subsampling. When metabolism is low, the valve is switched so that this pump will draw false air, and not from the metabolic chamber. This will cause flow rate to be below a low threshold defined in the software for using the subsample flowmeter readings in the calculations. When O2 extraction reaches the high threshold during arousal, the pump is reconnected, and flow values are again taken from the main mass flowmeter. The thresholds must be specified carefully, allowing adequate hysteresis. Artifacts are seen in the switching phase. During rising V̇o2, the transition can be made smoother by averaging flow rates over an interval adapted to the response time of the metabolic chamber.
Chamber Supply Gas Differing From Normal Atmospheric Conditions
Problems with diffusion though chamber walls are usually quite negligible with typical respirometry chamber materials like Plexiglas and differences in inside and outside gas concentrations of typically around 0.5%. However, with larger concentration differences of >10% typical in hypoxia exposures, or very small differences between incurrent and excurrent concentrations relative to the gradient across the chamber walls, the diffusion through chamber walls and tubing needs to be considered. One approach is using nondiffusible material like glass or stainless steel. Another is to draw reference gas through an identical reference chamber, as shown for the arthropod setup in Fig. 12. For short experiments, reference gas samples can also be drawn through the measurement chamber before and after an experiment, thus compensating for the diffusive background flux; however, this will not work correctly if the supply gas changes during the experiment. An alternative approach is to measure background diffusive conductance of the system with an empty chamber and then compensate for flux due to background diffusion in the calculations. In a system in which background diffusion can be approximated by Fick's law (23), the total system conductance, Go2, l·h−1·kPa−1, can be calculated as the background diffusion of O2 out of the chamber when empty, V̇o2empty, divided by the Po2 difference across the chamber walls: (14) where FaO2 is ambient O2 fraction outside the chamber. Then the background diffusion during measurements, V̇o2backgr (the variable GasBackGrChambFlux [Gas] in the program code of Fig. 2) can be calculated as: (15) and subtracted from measured V̇o2. The conductance, Go2, only needs to be measured once, and the equation should work with switches in supply gas to the chamber. Equilibration times of the reference sampling path vs. the measurement chamber will need to be considered in the transition phase. Further considerations need to be taken in avoiding leaks in the chamber.
In some applications, a complete seal is not possible due to the need to run wiring from a tethered animal out through the chamber. A pure push system is not feasible, as a mixing fan cannot be powered in the rotating chamber. An example how this can be solved is shown in Fig. 16A. Here gas was supplied with positive pressure to the cylindrical metabolic chamber though a smaller fixed chamber riding on top of it, with tethers passed out with tight clearance. Reference air is drawn from this chamber. The main metabolic chamber is ventilated with negative pressure through a sealing swivel at the bottom, allowing the chamber to be used on a rotating table. A similar principle could be applied to a mask system, with gas supplied with positive pressure into an additional inflated sleeve, from which reference air and gas supply to the mask could be drawn from (Fig. 16B).
Respirometry systems require careful planning and considerations for both the physics and experimental conditions with respect to the subjects. The simplest systems are by experience not the most trouble-free ones, as complexity can add both flexibility and better control over the experimental situation. This may require modification of standard components. The major part of the work associated with a system for long-term respirometry may be in the construction and verification of the system; once up and running, it is possible to achieve months of reliable operation with minimal care and monitoring. Data processing methods shown here can both increase accuracy and feasibility of performing long-term studies with a reasonable work load.
This study was supported by US Army Medical Research and Materiel Command award nos. W81XWH-06-1-0121 and W81XWH-09-2-0134, and American Heart Association fellowship no. 0020626Z.
No conflicts of interest, financial or otherwise, are declared by the author(s).
Author contributions: Ø.T. conception and design of research; Ø.T. performed experiments; Ø.T. analyzed data; Ø.T. interpreted results of experiments; Ø.T. prepared figures; Ø.T. drafted manuscript; Ø.T. edited and revised manuscript; Ø.T. approved final version of manuscript.
I thank Brian M. Barnes for support and encouragement of this project and for commenting on the manuscript, and Brian T. Rasley for help with construction of the metabolic chamber in Fig. 16A.
↵1 A document containing detailed instructions on modifying and verifying modifications of Sable Systems's Flowbar-8 and RM-8 gas flow multiplexer verson 3 for zero, reference, and span gas calibrations can be provided on request by the author.
- Copyright © 2013 the American Physiological Society