In humans, significant progress has been made to link spatial changes in electroencephalographic (EEG) spectral density, connectivity strength, and phase-amplitude modulation to neurological, physiological, and psychological correlates. In contrast, standard rodent EEG techniques employ only few electrodes, which results in poor spatial resolution. Recently, a technique was developed to overcome this limitation in mice. This technique was based on a polyimide-based microelectrode (PBM) array applied on the mouse skull, maintaining a significant number of electrodes with consistent contact, electrode impedance, and mechanical stability. The present study built on this technique by extending it to rats. Therefore, a similar PBM array, but adapted to rats, was designed and fabricated. In addition, this array was connected to a wireless EEG headstage, allowing recording in untethered, freely moving rats. The advantage of a high-density array relies on the assumption that the signal recorded from the different electrodes is generated from distinct sources, i.e., not volume-conducted. Therefore, the utility and validity of the array were evaluated by determining the level of synchrony between channels due to true synchrony or volume conduction during basal vigilance states and following a subanesthetic dose of ketamine. Although the PBM array allowed recording with high signal quality, under both drug and drug-free conditions, high synchronization existed due to volume conduction between the electrodes even in the higher spectral frequency range. Discrimination existed only between frontally and centrally/distally grouped electrode pairs. Therefore, caution should be used in interpreting spatial data obtained from high-density PBM arrays in rodents.
- high-density EEG
coordinated, synchronized neural interactions within and across different specialized brain areas are responsible for the brain's cognitive and executive functions (Uhlhaas and Singer 2006). Electroencephalography (EEG) is a powerful method to study this neural coordination. Compared with spatial imaging techniques such as functional magnetic resonance imaging and positron emission tomography, EEG has a high temporal resolution and therefore can be used to investigate synchronization across functional networks in the millisecond timescale. Because of novel mathematical methods developed over recent years, the process to understand how this coordination can be achieved progressed significantly by linking spatial EEG changes in spectral density, connectivity strength, and phase-amplitude modulation to (ab)normal neurological, physiological, and psychological correlates. In addition to its technical advantage in terms of temporal resolution, EEG is a powerful translational technique. It is well preserved across species. Correlates of all well-established oscillations in humans are also present in other species such as rodents and cats, in which they have the same rhythmic nature and, most likely, similar underlying mechanisms (Buzsáki and Watson 2012). However, standard rodent EEG techniques employ only few electrodes, resulting in poor spatial resolution due to the restricted number of conventional screw electrodes in relation to drilling holes, the small skull surface, and screw electrode size. This limits the (back)translation of fundamental mechanisms underlying neural coordination between preclinical and clinical findings.
Recently, a technique was developed in mice that could overcome the limitation of (back)translation of fundamental mechanisms underlying neural coordination between preclinical and clinical findings (Choi et al. 2010). This technique was based on a polyimide-based microelectrode (PBM) array applied on the mouse skull covering frontal, central, parietal, and upper parietal brain regions. The technique maintained a significant number of electrodes for spatial spectral mapping and phase dynamics with consistent contact, equal impedance, and mechanical stability. Importantly, the technique is surgically faster to apply and less invasive (reduced surgery time due to minor drilling and defining stereotaxic coordinates) and reaches the minimal number of electrodes corresponding to clinical human EEG recording setups. Therefore, this technique may provide a useful animal-human interface for studying fundamental mechanisms underlying (abnormal) neural coordination. However, the technique is currently limited to mice. Depending on the indication, often rats are more suitable than mice, i.e., in more (complex) behavioral tasks/studies, and new transgenic knockout rats are becoming widely available.
In the present study, we built on the technique developed for mice by extending the method to rats. Therefore, a similar PBM array, but adapted to the rat skull, was designed and fabricated in a way that it could be used in combination with a wireless EEG headstage [W32; Triangle Biosystems International (TBSI), Durham, NC]. This headstage allowed for the recording up to 31 electrodes simultaneously with wireless acquisition and continuous battery charging, combining the study of untethered and free (long-term) home cage behavior with the potential to be applied to complex behavioral tasks such as operant conditioning and mazes. The validity and utility of the approach was evaluated during different vigilance states and well-established pharmacological manipulation, i.e., after the administration of a subanesthetic dose of ketamine, which is known to initially increase EEG gamma power and to change brain connectivity (Anver et al. 2011; Páleníček et al. 2010; Pinault 2008; Sohal et al. 2009). We evaluated the quality of the signals and whether they represented true physiological activity. In addition, importantly, to identify the potential chance of volume conduction errors [e.g., only in human EEG, already more than 95% of a scalp-recorded potential at a certain location is generated by sources within 6 cm (Nunez et el. 1997)], the level of synchrony between channels due to true synchrony or volume conduction was determined. Therefore, spectral analysis and synchrony measures such as the phase-locking value (PLV) (Lachaux et al. 1999) and the debiased weighted phase-lag index (dbWPLI) (Vinck et al. 2011), were used. Both the PLV and dbWPLI estimate whether the phase difference between channels varies over time series. However, in contrast to the PLV, the dbWPLI controls for zero-lag differences, and therefore, for synchrony due to volume conduction. Therefore, estimating both PLV and dbWPLI provides a way to determine whether synchrony between channels is true or due to volume conduction.
MATERIALS AND METHODS
Animal care and experimentation were performed in compliance with protocols submitted to and approved by the local ethics committee of Stockholm North, Sweden.
Fabrication and characterization of PBM array.
The PBM array was designed to cover the exposed skull of an adult male rat (250–350 g) with electrode positions based on a neuroanatomical atlas (Paxinos and Watson 1986). The array had a structure with branches with electrode contact points and a bregma reference branch for easy positioning on the exposed skull (Fig. 1A). The electrode coordinates (mm anterior/posterior from bregma and lateral from midline, respectively) were as follows: 1) +9.5/±1.0 (nasal bone; the left electrode and right electrode served as ground and reference, respectively); 2) +6.7/±1.2 (olfactory bulb, OB); 3) +5.0/±1.8 (prefrontal cortex, PFC); 4) +2.2/±1.5 and ±3 (premotor cortex, PMC); 5) −2.1/±1.5 and ±3.0 (primary somatosensory cortex, SI); 6) −4.2/±1.5, ±3, and ±4.6 (parietal cortex, PC); 7) −6.7/±1.5 and ±3.0 (visual cortex, VC); and 8) −10.8/±1.0 and 2.5 (cerebellum, Cb).
The arrays were fabricated as previously described in detail (Choi et al. 2010). In short, 300-nm-thick platinum electrode contacts (diameter 0.5 mm), connection lines, and connector-matching interconnection pads were deposited on a polyimide substrate (thickness 5 μm), shaped using a lift-off photolithography process, and covered by a second 5-μm polyimide layer. Through selective reactive ion etching, the electrode contacts and the connector-matching interconnection pads were exposed, leaving a contact gap of 4.7 μm between the skull and electrode contact. A male A9707 Omnetics connector mating with the female wireless EEG headstage A9409 Omnetics connector was fixed with conductive glue to the connector-matching interconnection pads. In addition, two wires were connected to the male connector for dorsal neck muscle electromyography (EMG). Finally, the electrode contacts were platinized to improve the electrical properties (Poppendieck et al. 2014). Altogether, the PBM array allowed recording of 26 active EEG and 2 active EMG channels grounded and referenced to the nasal bone.
For initial characterization, the electrode contacts of each fabricated array were electrochemically evaluated by impedance spectroscopy as previously described in detail (Choi et al. 2010). For all the fabricated arrays, the impedance range was 2,000–4,000 Ω between 10 and 10,000 Hz and decreased in a similar way with increasing frequency in all electrode contacts of all used arrays (Fig. 1B). This is a common behavior for a general electrode-electrolyte interface (Choi et al. 2010).
For final characterization and before the move to chronic implantation, the signals between the PBM array and conventional screw electrodes were compared during acute anesthetized preparation. After premedication with 5 mg/kg carprofen (Norocarp; Norbrook Laboratories, Newry, Northern Ireland) and 0.05 mg/kg buprenorphine (Temgesic; RB Pharmaceuticals, Slough, UK), anesthesia was induced in Sprague-Dawley rats (SD; n = 2, 8 wk old, Ntac:SD; Taconic, Ejby, Denmark) with 3–5% isoflurane (Attane; Scanvet Animal Health, Parola, Finland) and maintained with 2–3% isoflurane such that the hind paw withdrawal and righting reflex were absent. Body temperature was continuously monitored and maintained at 37–38°C. The rats were fixed in a stereotaxic frame (Stoelting, Wooddale, IL). The scalp was shaved and disinfected. Following median incision and exposal of the skull by microclamps, lidocaine (Xylocaine; AstraZeneca, Sodertalje, Sweden) was applied before removal of the periosteum from the skull. Next, bleedings were stopped and the skull was thoroughly cleaned and dried. Next, in one rat, screw electrodes with 1-mm-diameter tip (MCS1x2; Angthos, Lidingö, Sweden) were epidurally implanted on 12 positions similarly to the array including 10 active electrodes (bilateral OB, PFC, PMC, SI, PC, and VC) and 1 reference (right) and 1 ground electrode (left) in the nasal bone. These electrodes were wired to an electrode interface board with a mounted A9707 Omnetics connector (part no. 010-0053-10; TBSI) mating with the female A9409 Omnetics connector of the wireless EEG headstage (W32; TBSI). In the other rat, the cleaned and dried skull was gently moisturized with saline before placement of the microelectrode array, because these wet conditions help the branches to be pulled to the skull, allowing a better adherence of the array to the skull (Choi et al. 2010). The array was then carefully aligned so that the bregma marker point of the array met the skull bregma and the vertical array midline met the skull midline. To avoid contact between the electrodes, any redundant moisture was removed, leaving saline only in the contact gaps between the skull and each electrode contact. Next, the wireless EEG headstage was connected to the electrode interface board or the PBM array. For 5 min, signals were continuously acquired with single-ended input using a sampling frequency of 2,000 Hz and a bandpass filter between 0.1 and 200 Hz and were saved in binary .nex format (NeuroWare; TBSI). With both the conventional screws and the array, the EEG showed a typical burst suppression pattern of inhalational anesthetics such as isoflurane. Since the signal level of the array was in a similar range as that of conventional epidurally implanted screw electrodes (Fig. 1D), it was decided to finally characterize the utility of the PBM array in chronic implanted animals.
Chronic PBM array implantation.
For chronic PBM array implantation, adult male SD rats (n = 4, 8-wk-old, Ntac:SD; Taconic) were used. We used a slightly different method of implantation than the one described in detail by Choi et al. (2010). First, although we carefully followed their method, the dental cement in fluid form leaked under the branches of the PBM array, preventing the electrode contact with the skull, which could consequently reduce/block the reception of signals. Second, we designed an approach to allow recovery of the array and its reuse between animals. To this aim, an anesthetic and surgical approach similar to that described above in detail was followed for initial exposure and cleaning of the skull. Next, the array was carefully aligned on a moisturized skull as described above (Fig. 1C5). Subsequently, following desiccation, a presurgically prepared cap of transparent hardened dental cement (Fig. 1C3) was placed on top and secured on the skull by screws (Fig. 1C6). Two pieces of plastic were then embedded in the cap by using dental cement in such a way that all screws were kept accessible (Fig. 1C7). Once hardened and secured, the interconnection pad with the Omnetics connector presurgically secured on a plate of plastic (Fig. 1C4) was turned over and subsequently fixed to the cap via the embedded frontal and distal pieces of plastic using screws (Fig. 1C8). At the distal end, a plastic cover was set in place to prevent postsurgical damage of the interconnection pads and the EMG leads (Fig. 1C8). The EMG leads were inserted unilaterally in the dorsal neck muscle with a distance of ∼2–3 mm in between and were sutured in place. Finally, the skin was sutured, only leaving access to the connector matching with the EEG headstage. Postoperative analgesia with 5 mg/kg carprofen (once a day) and 0.05 mg/kg buprenorphine (twice a day) was provided for at least 2 days following surgery. Initial pilot studies indicated that the followed approach allowed robust and stable fixation of the PBM array and construction on the rat's head for at least a month. In addition, postexperimental observations revealed that the removed periosteum grew back postsurgically over the array below the cap and further contributed to the consistent contact and stability. However, reuse of the array between animals is not recommended, because pilot experiments revealed that signal quality could dramatically decrease when the array is applied in a new animal.
Following a recovery period of 14 days, EEG/EMG was recorded in the chronic implanted animals for 1 h to obtain sufficient epochs of each vigilance state of interest such as active waking (aW; i.e., exploratory behavior), non-rapid eye movement sleep (nREM), and rapid eye movement sleep (REM). Subsequently, for the drug experiments, all animals were treated with vehicle (saline) and 30 mg/kg ketamine (1 ml/kg ip Ketaminol; Intervet International, Boxmeer, The Netherlands) in a random design with at least 48 h of washout. All experiments were performed between 1200 and 1700 (lights off 1800, 12:12-h light-dark regime) while the animals were in their home cage.
Before the start of all recording sessions, the animals were equipped with the wireless headstage through their head-mounted connector. Signals were continuously acquired with single-ended input using a sampling frequency of 2,000 Hz and a bandpass filter between 0.1 and 200 Hz and were saved in binary .nex format (NeuroWare; TBSI).
Preprocessing, spectral analysis, and phase synchrony analysis.
All preprocessing and analysis steps were performed with custom-written (P. J. Stienen) MATLAB scripts (version R2014a; The MathWorks, Natick, MA) based on functions from the Fieldtrip toolbox (Oostenveld et al. 2011). The raw continuously recorded data were imported into MATLAB. The DC offset was removed by subtracting the average signal amplitude over the whole recording for each individual channel. Each individual EEG channel was then referenced to the reference channel. The EMG channels were referenced to each other to obtain a bipolar EMG recording. The signals were subsequently imported into EEGLAB (Delorme and Makeig 2004) to visualize the data. For the drug(-free) experiments, visual classification of vigilance states and artifacts was performed in 4-s epochs as described previously (Scheffzük et al. 2011): 1) regular theta oscillations (4–8 Hz) and the level of EMG activity (aW > REM), 2) the amount of high-amplitude–low-frequency delta activity (0.5–4 Hz) in the frontal area [nREM > passive waking (pW), aW, and REM], 3) the amount of regular theta oscillations in the PC area (REM and aW > pW and nREM), and 4) EMG activity (REM < nREM < pW < aW).
For spectral/phase synchrony analysis of the drug-free experiments, 30 epochs from all the aW, nREM, and REM epochs without artifacts were selected randomly for each animal. For the drug experiments, the first 30 epochs without artifacts were selected (mixed aW and pW) between drug administration and 30 min after drug administration. Spectral analysis was performed on each epoch using Welch's periodogram (4-s Hanning window, 0.25-Hz resolution, no overlap), and spectrograms were subsequently averaged for each vigilance state separately. Phase synchrony analysis was performed by estimating the PLV (Lachaux et al. 1999) and the dbWPLI (Vinck et al. 2011) over the 30 epochs for each vigilance state. Both the PLV and dbWPLI estimate whether the phase difference between channels varies little across epochs. Both measures range between 0 (no synchrony) to 1 (maximum synchrony). However, in contrast to the PLV, the dbWPLI controls for zero-lag differences, and therefore, for spuriously inflated synchrony due to volume-conducted common source activity. In addition, as previously described (Phillips et al. 2014), we used the dbWPLI instead of the direct estimator of the WPLI. The direct estimator of the WPLI is heavily biased by sample size (Vinck et al. 2011), whereas the dbWPLI sample size bias is negligible for even small sample sizes of 20–30 trials (Vinck et al. 2011) such as in the current study. Estimating both PLV and dbWPLI provides a way to determine whether synchrony between channels is true (high PLV/dbWPLI) or false due to volume conduction (high PLV/low dbWPLI).
For each vigilance state of each animal in the drug-free experiments, we tested whether the PLV and dbWPLI significantly exceeded zero (i.e., significant synchrony) by computing jackknife estimates of the standard error of the PLV or dbWPLI (Phillips et al. 2014). These standard errors were subsequently used to calculate z scores and accompanying P values with respect to the normal distribution. An α value of 0.05 corrected for multiple comparisons (325 channel combinations with 26 EEG channels) was applied. PLV/dbWPLI values of channel combinations that showed no significant synchrony were replaced by zero. Finally, these corrected matrices were averaged over all animals. For each animal in the drug experiments, the PLV/dbWPLI estimates of the drug conditions were normalized to their own vehicle conditions. Finally, these matrices were averaged over all animals (n = 3).
Waveforms and spectral analysis during vigilance states.
Waveform and spectral content were vigilance state dependent (Fig. 2). Theta activity (4–8 Hz) was present during aW and REM sleep. During both aW and REM, theta activity was most prominent and of similar level bilaterally above the SI (electrodes 5l/5r and 6l/6r), PC (electrodes 7l/7r to 9l/9r), VC (electrodes 10l/10r and 11l/11r), and Cb (electrodes 12l/12r). In contrast, the most frontal electrodes above OB (electrodes 1l/1r), PFC (electrodes 2l/2r), PMC (electrodes 3l/3r), and Cb (electrodes 13r/13l) showed less theta activity (Fig. 2, A and C). During aW, there was a clear low-gamma activity band (30–60 Hz), which was less prominent during REM (Fig. 2, A and C). nREM was characterized by high activity in the lower frequency range (<10 Hz) with a less prominent distinction between delta (0.5–4 Hz) and theta range compared with aW and REM, but with clear spindle activity (12–16 Hz), which was absent during aW and REM (Fig. 2B).
Synchrony between electrodes during vigilance states.
During aW, the PLV in the theta range reached almost 1 (maximum synchrony) between all electrode pairs (Fig. 2A). Also during nREM and REM, the PLV in the delta and theta range, respectively, reached almost 1 between all electrode pairs, except between the frontally grouped electrodes above the OB, PFC, and PMC (electrodes 1l/1r to 4l/4r) and centrally/distally grouped electrodes above the SI, PC, VC, and Cb (electrodes 5l/5r to 13l/13r); PLV was near 0 (minimal synchrony; Fig. 2, B and C). The same was true for the higher frequency ranges (Fig. 2; aW/REM, low gamma; nREM, spindles).
In contrast to the PLV, the dbWPLI values between nearly all electrode pairs were near 0 (from no to minimal synchrony) for the lower (delta and theta) as well as the higher (spindles and low gamma) frequency ranges for each vigilance state (aW, nREM, REM) evaluated (Fig. 2). Although the PLV decreased with interelectrode distance in aW, nREM, and REM in the frequency ranges studied (delta, theta, spindles, and low gamma), in all cases the PLV was relatively high (>0.5) whereas the dbWPLI was low (<0.3), even in the 5- to 10-mm interelectrode distance (Fig. 3).
Synchrony between electrodes following ketamine challenge.
Ketamine increased low gamma power in a similar way in the electrodes above the SI (electrodes 5l/5r and 6l/6r), PC (electrodes 7l/7r to 9l/9r), VC (10l/10r and 11l/11r), and Cb (electrodes 12l/12r), whereas there was no effect in the most frontal electrodes above the OB (electrodes 1l/1r), PFC (electrodes 2l/2r), PMC (electrodes 3/3r and 4l/4r), and the most outer electrodes above the Cb (electrodes 13l/13r) (Fig. 4, left). Ketamine increased the PLV in the low gamma range between the frontally grouped electrodes above the OB (electrodes 1l/1r), PFC (electrodes 2l/2r), and PMC (electrodes 3l/3r and 4l/4r) and centrally/distally grouped electrodes above the SI (electrodes 5l/5r and 6l/6r), PC (electrodes 7l/7r to 9l/9r),VC (10l/10r and 11l/11r), and Cb (electrodes 12l/12r and 13l) (Fig. 4, right). In contrast to the PLV, there were no clear and robust effects of ketamine on the dbWPLI between the different electrode pairs (Fig. 4, right).
In the present study, a method to measure EEG with high spatial resolution was designed and evaluated with a PBM array applied on the rat skull built on a technique that was previously developed for mice (Choi et al. 2010). The principle findings are, first, that the PBM array allowed recording with high signal quality similar to conventional epidurally implanted screw electrodes. Second, although some discrimination existed between frontally and centrally/distally grouped electrode pairs, in both low- and high(er)-frequency bands during the different vigilance and drug states, the PLV reached values close to 1 between almost all electrode pairs, whereas the dbWPLI was close to 0. Given that the dbWPLI, rather than the PLV, controls for synchrony due to volume conduction (Vinck et al. 2011), the high PLV combined with the low dbWPLI indicates that the observed synchrony between the electrode pairs is due to volume-conducted common source activity, rather than true synchrony. Therefore, one should be precautious when interpreting spatial data resulting from PBM arrays applied to the rat skull.
Despite the presence of volume conduction, the waveforms and spectral content were vigilance state dependent. This supports that the signals obtained by this method were from physiological origin. Regular theta oscillations dominated during aW and REM, whereas delta activity dominated during nREM. In addition, during aW, there was a clear low-gamma activity band, which was less prominent during REM. This is in accordance with previous findings (Scheffzük et al. 2011) as well as the present finding that ketamine increased gamma power (Anver et al. 2011; Pinault, 2008; Sohal et al. 2009). Together, the present findings showed that true physiological brain activity can be obtained using a microelectrode array applied on the rat skull.
The increase in the low-gamma PLV between the frontal and central/distal electrode pairs following ketamine can be explained by signal power changes, rather than increase of true synchronization. In contrast to the dbWPLI, the PLV is sensitive to noise. Evidence exists that increases in signal power improve the signal-to-noise ratio and increase the PLV, while there is no change in true locking (Muthukumaraswamy and Singh 2011; Vinck et al. 2011). Therefore, together with the lack of robust ketamine effects on the dbWPLI, which controls for noise influences (Vinck et al. 2011), the PLV increase between the frontal and middle/distal electrode pairs following ketamine administration does not represent a true increase of synchrony due to ketamine.
Importantly, the present finding of low spatial discrimination, especially in the higher frequencies because of volume conduction, suggests one should be precautious when interpreting data using the current microelectrode array applied on the rodent skull in more advanced experimental procedures. Lower frequencies synchronize activity over longer distances, whereas higher frequencies mediate within short distances and are organized by local (neo)cortical mechanisms (Csicsvari et al. 2003; Nir et al. 2011; Sirota et al. 2003, 2008; Steriade 2001). In particular, rodent neocortex-derived theta rhythms are not from neocortical origin but are volume conducted from the hippocampus-septal nuclei-entorhinal cortex network (Buzsáki 2002). Therefore, the present finding of the high level of synchronization between electrode pairs in the lower frequencies was expected. However, in contrast to spindles, and to some extent, delta (Nir et al. 2011) and gamma (Steriade 2001; Csicsvari et al. 2003; Sirota et al. 2003, 2008), expected to be more spatially localized, a relatively high level of synchronization due to volume conduction in these frequencies was observed. Roughly, one could discriminate between the frontally grouped electrode and the centrally/distally grouped electrode pairs. Therefore, one should be precautious when interpreting data using EEG PBM arrays in rats. Moreover, given the present findings in rats, it would be recommended to mathematically evaluate findings resulting from similar mouse microelectrode arrays (Bergstrom et al. 2013; Choi et al. 2010; Lee et al. 2013) in a way similar to the present study, i.e., using both PLV and dbWPLI measures to investigate whether synchronization between electrodes is true or due to volume conduction. Taking into account the present high level of synchronization between electrodes due to volume conduction in rats and the mouse brain being smaller than that of rats, one should expect even a higher chance of synchronization due to volume conduction in mice over rats.
Despite the low spatial resolution due to volume conduction with the use of the high-density PBM array, the present data do not necessarily imply that a PBM array applied on the rodent skull cannot be practiced. We suggest reducing the number of electrodes, for example, to frontal, central, and distal pairs, thereby keeping the advantages of the technique in terms of electrode contact, equal impedance, and reduced surgical invasiveness. In addition, other implantation strategies such as addition of mineral oil after placement of the microelectrode array (Cuellar et al. 2009) to further seal the implant may be beneficial in preventing volume conduction artifacts.
In conclusion, despite the high-quality recording with the high-density PBM array applied to the skull in the present study, one should be precautious when interpreting its derived spatial data because of the global presence of spuriously inflated synchrony due to volume conduction over the skull. The number of electrodes of the array should be reduced while keeping the advantages of the array in terms of electrode contact, equal impedance, and reduced surgical invasiveness.
While the research described in this manuscript was performed, P.J.S., M.V. and E.Å. had employment at AstraZeneca, a for-profit company engaged in the discovery, development, manufacture, and marketing of proprietary therapeutics. They do not consider that this creates any conflict of interest with the subject-matter of this paper. W.P. and K.P.H. have no conflict of interest.
P.J.S., W.P., K.P.H., and E.Å. conception and design of research; P.J.S., W.P., and K.P.H. performed experiments; P.J.S. and M.V. analyzed data; P.J.S., M.V., and E.Å. interpreted results of experiments; P.J.S. prepared figures; P.J.S. drafted manuscript; P.J.S., M.V., W.P., K.P.H., and E.Å. edited and revised manuscript; P.J.S., M.V., W.P., K.P.H., and E.Å. approved final version of manuscript.
- Copyright © 2016 the American Physiological Society