Abstract
Many have believed that oxygen (O2) crosses red blood cell (RBC) membranes by dissolving in lipids that offer no resistance to diffusion. However, using stopped-flow (SF) analyses of hemoglobin (Hb) absorbance spectra during O2 off-loading from mouse RBCs, we now report that most O2 traverses membrane-protein channels. Two agents excluded from the RBC interior markedly slow O2 off-loading: p-chloromercuribenzenesulfonate (pCMBS) reduces inferred membrane O2 permeability (PMembrane) by ∼82%, and 4,4’-diisothiocyanatostilbene-2,2’-disulfonate (DIDS), by ∼56%. Because neither likely produces these effects via membrane lipids, we examined RBCs from mice genetically deficient in aquaporin-1 (AQP1), the Rh complex (i.e., rhesus proteins RhAG + mRh), or both. The double knockout (dKO) reduces PMembrane by ∼55%, and pCMBS+dKO, by ∼91%. Proteomic analyses of RBC membranes, flow cytometry, hematology, and mathematical simulations rule out explanations involving other membrane proteins, RBC geometry, or extracellular unconvected fluid (EUF). By identifying the first two O2 channels and pointing to the existence of other O2 channel(s), all of which could be subject to physiological regulation and pharmacological intervention, our work represents a paradigm shift for O2 handling.
Results & Discussion
Central to human life is the inhalation of O2, its diffusion across the alveolar wall to pulmonary capillary blood plasma, O2 diffusion across the RBC membrane and transfer to Hb (O2 + Hb[O2]3 → Hb[O2]4), the distribution of RBCs throughout the body, and the offloading of this O2 from Hb to support metabolism. A century ago, Krogh’s mathematical analysis of O2 diffusion—from Hb, through the intracellular fluid (ICF), across the RBC membrane, to the bulk extracellular fluid (bECF) of the capillary blood plasma, and then across several membranes to the ICF of O2-metabolizing cells1—required the implicit simplifying assumption2 that the membranes offer no resistance to O2 diffusion (i.e., RMembrane = 0; see Supplementary Discussion). Although some argued otherwise3,4, Krogh’s assumption became engrained in the literature, and most have assumed that total resistance to O2 diffusion out of RBCs (RTotal = RICF + RMembrane + REUF) comprises exclusively RICF and REUF5–7.
Calling the above prevailing view into question were the discoveries of the first membranes with no detectable CO2 permeability8 and the first membrane protein (i.e., AQP19,10) with an additional role as a CO2 channel.11 pCMBS, which reacts with the –SH of Cys-189 in the extracellular vestibule of the AQP1 monomeric pore and thereby inhibits H2O movement through AQP112, also blocks a component of CO2 traffic through AQP1.13 Moreover, various AQPs exhibit CO2 vs. NH3 selectivity.14,15 Other work showed that rhesus (Rh) proteins also can conduct CO2.16 These observations indicated that gas transport involves more than diffusion through lipids. Work on human RBCs genetically deficient in AQP117,18 or RhAG18,19 showed that these two channels account for >90% of CO2 permeability of RBCs, and that DIDS, which binds to and can react with the – of Lys residues, blocks a component of CO2 traffic through both AQP1 and the Rh complex. Although neither pCMBS nor DIDS is specific, both are valuable tools because they do not interact with membrane lipids.20,21 Thus, we hypothesized that, like CO2, O2 crosses RBC membranes mainly via channels, and that pCMBS or DIDS would block these hypothetical O2 channels.
In the present study, we use SF absorbance spectroscopy to monitor Hb deoxygenation as we mix oxygenated RBCs with a solution containing the O2 scavenger sodium dithionite (NDT; final concentration, 25 mM). The 3D graphs in Figure 1a–b show examples of time courses of Hb absorbance spectra of RBCs—either untreated (panel a) or pCMBS-treated (panel b)—from wild-type (WT) mice. Figure 1c shows the time courses of relative HbO2 saturation (HbSat). An SF assay22 based on the release of carbonic anhydrase (CA) from RBC cytosol to bECF (Methods/Hemolysis) quantitates hemolysis within the SF chamber, and ImageStream flow cytometry (Methods/ImageStream) measures the frequency and diameter of spherocytes (Supplementary Table 8). We find that a 15-min incubation with 1 mM pCMBS increases hemolysis from ∼5.2% to ∼8.2% (Extended Data Figure 1a), and increases the prevalence of spherocytes from ∼1.4% to ∼8.7% (Extended Data Figure 1b). A novel protocol (Extended Data Figure 2) accounts for the effects of extracellular Hb on the Hb-deoxygenation rate constant , and mathematical simulations (below) account for the presence of spherocytes (Methods/Accommodation for Spherocytes). We conclude that that pCMBS decreases by ∼61% (Figure 1d).
To determine whether pCMBS accesses RBC cytosol, we used CA as a sentinel. Although incubating hemolysate with pCMBS for 15 minutes markedly reduces lysate CA activity, treating intact RBCs with pCMBS has no effect (Figure 1e). Thus, pCMBS does not significantly cross RBC membranes.
Figure 2a-b show representative time courses of Hb absorbance spectra of untreated and DIDS-treated RBCs from WT mice, and Figure 2c shows time courses of relative HbSat. For DIDS-treated RBCs, we used a variant of the aforementioned approach to correct raw values for hemolysis (Extended Data Figure 3). A 60-min incubation with 200 μM DIDS increases mean hemolysis from ∼5.2% to ∼13.4% (Extended Data Figure 1a), and increases spherocytosis to ∼41% (Extended Data Figure 1b). After accounting for hemolysis and spherocytosis, we conclude that DIDS reduces by ∼31% (Figure 2d). Although the degree of DIDS-induced spherocytosis is cause for pause, we will see below that protocols with substantially less spherocytosis reveal even greater decreases in .
To determine whether DIDS accesses RBC cytosol, we used Hb as a sentinel (DIDS has little effect on CA). Figure 2e shows that, although incubating hemolysate with DIDS substantially reduces lysate , treating intact RBCs with DIDS does not. Thus, DIDS does not significantly cross RBC membranes.
Because pCMBS and DIDS each decrease by substantial amounts, without crossing the membrane (Figure 1 and Figure 2) or interacting with membrane lipid,20,21 they must block O2-conducting proteins. We began our search for such proteins by examining RBCs from mice genetically deficient in AQP1 or RhAG, which have established roles as major CO2 channels in human RBCs. Figure 3a shows representative experiments comparing of RBCs from WT mice with those from Aqp1–/–, RHag–/–, or dKO mice. Our observation that the individual and double knockouts do not significantly affect hemolysis (Extended Data Figure 1c) is consistent with the report that deletion of mRh and RHag do not increase RBC fragility23. Spherocytosis of RBCs from dKO mice is low (Extended Data Figure 1d). The summary in Figure 3d shows that KOs decrease hemolysis-/shape-corrected values by ∼9% for Aqp1–/–, ∼17% for RHag–/–, and ∼30% for the dKO. Compared to RBCs from WT mice, those from dKO mice reveal substantially less drug-induced spherocytosis (Extended Data Figure 1d), presumably reflecting the absence of two drug targets. Figure 3e reveal that the combination of dKO+pCMBS decreases by 78%, and dKO+DIDS, by 53%. These results show that AQP1, RhAG, and at least one other protein make substantial contributions to RBC O2 permeability.
Because the deletion of one protein could alter the expression of others, we assessed effects of the knockouts on RBC membrane-protein levels by purifying proteins from RBC ghosts, and quantitating proteins by mass spectrometry (label-free LC/MS/MS). We detected 211 plasma-membrane–associated (PMA) proteins. Figure 4a and b show that Aqp1 or RHag deletions produce the expected elimination of the cognate proteins from RBC membranes, as well as mRh, which forms heterotrimers with RhAG24 and falls pari passu with RhAG. The deletions do not significantly affect levels of any of the other 47 PMA proteins with the greatest inferred abundance (Extended Data Figure 4a). Of the 27 proteins in the RBC-ghost preparation exhibiting a significant change with at least 1 genetic deletion (Extended Data Figure 4b), the most abundant is <1% as abundant as AE1. Thus, secondary changes in membrane proteins other than AQP1 and RhAG/mRh are unlikely to account for the observed decreases in .
Because altered RBC geometry, by altering intracellular diffusion distances, could change , we performed hematological analyses on fresh blood from all genotypes (Supplementary Table 2). We observe no significant sex- or genotype-related differences (Extended Data Figure 5a-c) except—in RHag–/– and dKO mice—slightly higher mean corpuscular volume (MCV; see Figure 4c), slightly lower mean corpuscular hemoglobin concentration (MCHC; see Figure 4d), and slightly greater RBC distribution width. Review of blood smears reveals unremarkable RBC morphology, with no differences among genotypes for either control cells (Extended Data Figure 6) or drug-treated cells (Supplementary Data Figure 7). Previous authors noted normal RBC morphology for RHag–/– mice23. DIC microscopy, micrographs (Figure 4e, Extended Data Figure 7) and videos of live, tumbling RBCs (Supplementary Videos) confirm that RBCs of all genotypes are normal biconcave disks. DIC studies (Supplementary Data Figure 8) show that the misshapen cells identified/quantitated by ImageStream are, in fact, spherocytes. Forward light scattering during flow cytometry (Extended Data Figure 8) is consistent with the slightly greater MCVs in knockouts. ImageStream flow cytometry reveals that mean maximal diameters of RBCs+precursors are slightly less in knockouts than WTs (Figure 4f and Supplementary Tables 5 – 6), which implies, together with increased MCVs, that knockout cells are slightly thicker.
To assess effects of small changes in RBC dimensions and [Hb], we performed mathematical simulations using a reaction-diffusion model (see Methods and Extended Data Figure 9a) populated with measured parameter values and known constants. We assume a 1-μm EUF thickness (ℒEUF), which—given our [NTD]o of 25 mM—is probably an overestimate (Supplementary Discussion). For untreated WT RBCs, our model yields a simulated value within ∼1% of experimental values. In simulating for KO mice, we use our measured genotype-specific values of mean corpuscular volume, mean corpuscular Hb, and major diameter, and find that gene deletions would either increase (Aqp1) or produce decreases (RHag, dKO) far smaller than those observed (Extended Data Figure 9b).
Recalling that RTotal = RICF + RMembrane + REUF, what do the present data tell us about the contribution of RMembrane and O2 channels to RTotal? RICF is doubtlessly important, as emphasized recently by Richardson and colleagues25. However, their low [NDT]o of 1 mM (vs. our value of 25 mM) likely leads to a large ℒEUF and thus an overestimated RICF. 5,6,26 Thus, in our simulations, we have assumed that the intracellular diffusion constant for O2 is the arithmetic mean of the generally accepted value and the Richardson value (Supplementary Methods/Parameter values). Our simulations argue that RMembrane—even for WT RBCs without inhibitors—is ∼33% of RTotal (Supplementary Discussion). To the extent that our ℒEUF estimate is low, the RMembrane percentage is even higher. Thus, even for control RBCs from WT mice, the membrane offers substantial resistance to O2 diffusion. However, RTotal and RMembrane are as low as they are only because of O2 channels. Figure 4g and Extended Data Figure 9c-d summarize the predicted dependence of on PMembrane. Thus, the 30% decrease in caused by dKO corresponds to a ∼55% decrease in PMembrane, so that RMembrane now represents nearly half of RTotal. The 78% decrease in caused by dKO+pCMBS corresponds to a ∼91% decrease in PMembrane, so that RMembrane now represents ∼83% of RTotal. It is not clear how drugs confined to the outside of the membrane or deletion of membrane proteins could produce the observed decreases of in any way other than decreasing O2 egress through channels. Moreover, an analysis of 8 other key parameters (e.g., ℒEUF, cell thickness) shows that no reasonable change in any could explain our data (Extended Data Figure 10).
In summary, two channel proteins—AQP1 and the Rh complex—contribute ∼55% of the RBCs O2 permeability. Other proteins blocked by pCMBS likely contribute at least another 35%. Together with earlier CO2 studies, the present work shows that, of the O2 and CO2 traffic across RBC membranes, >90% occurs through channels. Our work raises questions that should trigger new research: What is the molecular mechanism of O2 diffusion through the channels? We suggest that the most likely routes are the hydrophobic central pores at the middle of the AQP1 tetramers and Rh trimers, or other pathways between monomers. Which other membrane proteins contribute to PMembrane in RBCs? And to what extent do AQPs, Rh proteins, and other membrane proteins contribute to O2 permeability in other cell types? An important implication of O2 channels is that they provide a mechanism for cellular regulation of O2 permeability. They also offer the future physician the possibility of raising O2 permeability (e.g., to enhance performance or wound healing) or lowering O2 permeability (e.g., to treat oxygen toxicity) by altering the number of or intrinsic permeability of channels.
End notes
Supplementary Information is available in the online version of the paper.
Author Contributions
P.Z., R.R.G., R.O., G. G., F.J.M. & W.F.B. designed the study; R.R.G., performed the initial stopped-flow experiments; P.Z., A.I.S., A.B.W., S.T., R.O., & F.J.M. performed experiments and collected data; P.Z., A.I.S., A.B.W., D.E.H., H.J.M., R.O., F.J.M., & W.F.B. analyzed data; P.Z., R.O., F.J.M., & W.F.B. wrote the manuscript.
Author Information
The authors declare no competing financial interests. Readers are welcome to comment on the online version of the paper. Correspondence and requests for materials should be addressed to W.F.B. (walter.boron{at}case.edu), P.Z. (pan.zhao2{at}case.edu) or R.R.G. (geyerro1976{at}gmail.com).
Methods
Mice
WT C57BL/6 mice (originally obtained as Aqp1+/– mice on a C57BL/6 background, a generous gift of Alan Verkman) were backcrossed for >20 generations by our laboratory to define our lab-standard C57BL/6 strain. We backcrossed Aqp1–/– mice, with a targeted disruption of the Aqp1 gene27, into our WT mice for >18 generations, and confirmed genotypes by real-time PCR (performed by TransnetYX, Inc.). We backcrossed RHag–/– mice (provided by Jean-Pierre Cartron as RHag+/– mice on a C57/BL6 background), with a targeted disruption of the RHag gene23, into our WT mice for at least 7 generations, and confirmed genotypes of RHag–/– mice were confirmed by standard PCR (Supplementary Methods). From Aqp1–/– and RHag–/– mice, we generated dKO (i.e., Aqp1–/–RHag–/–) mice, and confirmed genotypes as noted above. Mice were allowed access to food and water ad libitum. In experiments, we used mice of both sexes (8 to 19 weeks old). Not shown are data showing that no changes in kinetic or hematological parameters occur up to 6 months of age for all genotypes. All animal procedures were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) at Case Western Reserve University.
Physiological solutions
Supplementary Table 12 shows the compositions of all solutions used in this study. We made pH measurements—either at room temperature (RT) or at 10°C—using a portable pH meter (model A121 Orion Star, Thermo Fischer Scientific (TFS), Waltham, MA) fitted with a pH electrode (Ross Sure-Flow combination pH Electrode, TFS). Beakers containing pH calibration buffers (pH at 6, 8 and 10; TFS), physiological solutions to be titrated, and the pH electrode in its storage solution (Beckman Coulter, Inc. Brea, CA) were equilibrated at either RT or 10°C, as appropriate. For work at 10°C, we used a refrigerated, constant-temperature, shaker water bath (model RWB 3220, TFS) and submersible Telesystem magnetic stirrers (TFS). We adjusted pH with NaOH (5 M) or HCl (5 M). Osmolality was measured using a vapor pressure osmometer (Vapro 5520; Wescor, Inc., Logan, UT) and, as necessary, adjusted upward by addition of NaCl.
Inhibitors
Stock solutions of 2 mM 4-(chloromercuri)benzenesulfonic acid sodium salt (pCMBS; catalog no. C367750; Toronto Research Chemicals, Toronto, ON, Canada) and 2 mM 4,4’-diisothiocya-natostilbene-2,2’-disulfonate (DIDS; catalog no. 1226913; Invitrogen, Eugene, OR) were freshly prepared by dissolving directly into oxygenated solutions, preventing exposure to light. In some experiments, RBC samples were pre-incubated with 1 mM pCMBS for 15 min (Figure 1) or with 200 μM DIDS for 60 min (Figure 2 and Extended Data Figure 3a, right bar). For assessing the effect of millisecond exposures to DIDS on the Hb absorbance spectrum (Extended Data Figure 3a, stippled bar), we added 200 μM DIDS to the deoxygenated solution.
Preparation of RBCs
We collected fresh blood from WT or KO mice for seven assays: SF, proteomics/mass spectrometry, hematology, blood smears, still microphotography, microvideography, and flow cytometry. We used the cardiac-puncture method28 for some SF studies, and we used the submandibular-bleed method29 for the remaining SF and all other studies. For cardiac puncture, a 1-ml syringe (Becton, Dickinson and Co., Franklin Lakes, New Jersey, USA) and attached 23-gauge PrecisionGlide needle (Becton, Dickinson and Co.) was used. For bleeds of the submandibular vein, a 3-, 4-, or 5-mm point-length sterile animal lancet (MEDIpoint, Inc., Mineola, NY) was used, and blood samples (∼250 μl) were taken from the same mouse no sooner than 72 h from a previous sample on the contralateral side.
We collected blood samples for mass spectrometry into 0.6-ml microcentrifuge tubes that we had previously rinsed with 0.1% sodium heparin (H4784, Sigma-Aldrich, St. Louis, MO), and processed the blood as described below. We collected blood for hematology studies into special micropipettes (see below), and for blood smears into 2 ml K2E K2EDTA VACUETTE® tubes (Monroe, NC 28110, USA) (see below). For the other 4 assays (SF, still microphotography, microvideography, flow cytometry), we collected blood into heparinized (see above) microcentrifuge tubes, which we centrifuged in a Microfuge 16 Microcentrifuge (Beckman Coulter, Inc.) at 600 × g for 10 min. We aspirated and discarded the resulting supernatant and buffy coat. To remove residual extracellular Hb, we resuspended the pelleted RBCs in our oxygenated solution (Supplementary Table 12) to a hematocrit (Hct) of 5% to 10%, centrifuged at 600 × g for 5 min. After 4 such washes, RBCs were resuspended in oxygenated solution to a final Hct of 25% to 30%, and maintained on ice (for up to ∼6 h) for experiments. At this point, the RBCs were directed to studies of SF (see below), still microphotography and microvideography (see below), or flow-cytometry (see below).
For SF studies, we computed [Hb] in a suspension of intact RBCs from the Hb absorbance measured on a UV/Vis spectrophotometer (model 730 Life Science, Beckman Coulter, Inc.), using an equation described previously22. As necessary, we produced an RBC lysate by osmotic lysis of 20 μl of freshly prepared, packed mouse RBCs in a 1:8 dilution of Milli-Q H2O (Milli-Q® Integral Water Purification System, EMD Millipore Corporation, Billerica, MA), followed by centrifugation at 15000 × g for 5 min in a Microfuge 16 Microcentrifuge, at RT. We transferred the supernatant to a clean 1.5-ml tube for a spectro-scopic determination of free [Hb], as described above22 for intact RBCs. For all stopped-flow experiments—on intact RBCs or lysates or mixtures thereof—we use a final [Hb] of 2.5 μM in the SF reaction cell.
Quantitation of hemolysis in SF reaction cell
We have developed a novel assay22, based on carbonic-anhydrase activity, for determining RBC hemolysis in the SF reaction cell in a time domain similar to that of our O2-efflux experiments. We use a SX-20 stopped-flow apparatus (Applied Photophysics, Leatherhead, UK; fluorescence spectroscopy mode) to mix solutions A and B (Supplementary Table 12) in the SF reaction cell, and create an out-of-equilibrium (OOE) state that re-equilibrates according to the reaction , catalyzed by CA released from hemolyzed RBCs. Solution A contains HEPES/pH 7.03 + RBCs or hemolysate. Solution B contains ∼1% CO2/44 mM + the fluorescent pH-sensitive dye pyranine (H348, Invitrogen), which reports extracellular pH (pHo). We excite the dye at 460 nm or 415 nm, while monitoring total fluorescence emission using a 488-nm long-pass filter. Immediately upon mixing of A and B, the solution in the SF reaction cell (10°C) has the composition ∼0.5% , which is out of equilibrium. The pHo rises exponentially (rate constant, kΔpH, computed as described previously22) to ∼7.50.
We calculate the actual percent hemolysis (Act%H) of ostensibly 100% intact RBCs (i.e., apparent percent hemolysis, App%H, is 0%) during a SF experiment, as described previously22, using a procedure that requires three rate constants: kuncat, kRBC,Lysate, and kRBC,OstInt. (1) kuncat is the uncatalyzed rate constant (without CA). (2) kRBC,Lysate is kΔpH in the presence of fully lysed RBCs. And (3) kRBC,OstInt is kΔpH in the presence of ostensibly 100% intact RBCs. From these, we compute kcat,min = kRBC,OstInt – kuncat and kcat,max = kRBC,Lysate – kuncat. Finally:
Stopped-flow absorbance spectroscopy and calculation of
We used the SF apparatus (absorbance spectroscopy mode) to determine the rate constant of hemoglobin deoxygenation. We load RBCs ([Hb] = 5 μM, Hct ≅ 0.3%) in the oxygenated solution (Supplementary Table 12) into one syringe, and load the deoxygenated solution containing the O2 scavenger NDT (50 mM) in the other syringe. We record absorbance (A) in the SF reaction cell (10°C) over the portion of the Hb spectrum with highest Signal/Noise (wavelength [λ]: 410–450 nm, at 5 nm intervals), sampling at 10-ms intervals for 4000 ms. For each new set of loaded samples, we began by performing ≥6 “shots” to ensure that the new solutions are loaded into the SF reaction cell, and then sequentially acquired Aλ vs. time during 9 separate shots at λ = 410, 415 … 450 nm. We compiled such data set into 3-D graphs (e.g., Aλ vs. λ vs. time plots in Figure 1a). Then, as described in Supplementary Methods, we use absorbances at 6 wavelengths (i.e., 410, 415, 425, 430, 435, 440 nm)—sufficiently distant from isosbestic wavelengths—to compute Hb saturation at each time point (e.g., Figure 1c) beginning at t = 100 ms, and finally computed the quasi-rate constant (e.g., Figure 1d). We accept a data set only if each of the 9 records of Aλ vs. time approaches an asymptote. We accept 1 to 5 such data sets for each set of conditions, and then average these values to obtain the value used in further analyses.
Correction of for hemolysis
Because Hb released into free solution deoxygenates faster than Hb inside RBCs, we correct raw values, based on Act%H, determined as described above. For each mouse, we generate simulated hemolysis samples for apparent hemolysis levels of 0% (i.e., ostensibly 100% intact RBCs), 2.5%, 5 %, 10% and 25% by mixing ostensibly 100% intact RBCs (100%, 97.5%, 95%, 90%, and 75%) with hemolysate (0%, 2.5%, 5%, 10%, and 25%)—keeping total [Hb] constant at 5 μM. We determine , as described above, for each simulated hemolysis sample, and—using an algebraic translation of the x-axis—convert a plot vs. apparent % hemolysis (e.g., blue elements in Extended Data Figure 1 and Extended Data Figure 3b) to a plot of vs. actual % hemolysis (green elements). Extrapolating back to 0% actual hemolysis yields an estimate of the actual of 1 mouse. We used this approach for RBCs treated with pCMBS or no drug.
For RBCs treated with DIDS, we modified the above approach because DIDS decreases the of free Hb, so that the actual slope of the vs. Act%H regression line (Slope DIDS,Act) is less than the control value (Slope Ctrl,Act). Two factors are at work. (1) During the 1-h pretreatment with DIDS, the drug produces a substantial reduction in (compare hatched and open bars in Extended Data Figure 3a) for the minute fraction of Hb already released22 (∼0.4%) from RBCs at this stage. (2) During the actual SF experiment, DIDS produces a small reduction in (compare stippled and open bars in Extended Data Figure 3a) for a larger fraction of Hb newly released in the SF reaction cell. Thus, in a DIDS experiment, the actual slope of the regression line becomes: Extended Data Figure 3b shows the hemolysis correction for RBCs—from 1 mouse—studied under control conditions (i.e., no DIDS). Here, Slope Ctrl,Act is 6.51 s−1/Act%H. For RBCs from the same mouse, but treated ×1 h with 200 μM DIDS, the Act%H was 12.7%. Inserting these and the other values for this mouse into the above equation yields: Thus, knowing the measured in ostensibly 100% intact RBCs , we can estimate the actual for DIDS-treated RBCs as: As summarized in Extended Data Figure 3b, for this particular mouse:
Accommodation for spherocytes
As described in Supplementary Methods, we generated preparations of RBCs from WT and dKO mice, and then treated these with no drug, pCMBS or DIDS as in experiments. Here, however, after treatment ±drug, we performed hematology (Supplementary Table 7, Supplementary Table 9), blood-smears (Supplementary Data Figure 7), microphotography (Supplementary Data Figure 8), and ImageStream (Supplementary Table 8, Supplementary Data Figure 9) assays. For each of six conditions ([WT vs. dKO] × [control vs. pCMBS vs. DIDS]), ImageStream flow cytometry (see Methods/ImageStream) revealed spherocyte abundance and diameter, from which mathematical simulations yielded spherocyte (Supplementary Methods). Linear combinations of HbSat vs. time for normal RBCs and spherocytes allowed us to reconstruct the of normal RBCs, from which we computed PMembrane. See Methods/Accommodation for Spherocytes for an overview of this reconstruction procedure.
RBC-ghosts preparation and Proteomic analysis
Blood samples (0.2 to 0.5 ml) were collected into heparinized microcentrifuge tubes from 3 mice for each genotype, and then placed the tubes on ice, pending immediate processing into ghosts. We generated erythrocyte ghosts as previously described30, with the following changes to buffer composition (Supplementary Table 12): RBC lysis and post-lysis wash buffers contained 5 mM Tris-HCl/pH 8.0 and complete Protease Inhibitor Cocktail tablets (cPIC; Roche, 04693116001), as opposed to sodium-phosphate buffer/pH 7.5 and PMSF and pepstatin A—a modification required to obtain good mass-spectrometry data. Ghosts were then flash-frozen and held at −80 °C prior to mass-spectrometry analysis.
Mass spectrometry experiments were performed by the CWRU Center for Proteomics and Bioinformatics. Briefly, RBC ghost samples were lysed with 2% SDS and cPIC, using pulse sonification. SDS was removed by filter-aided sample preparation31 detergent cleanup, and total protein concentration was determined by the Bio-Rad Protein assay kit (Bio-Rad, Hercules CA). A 10-µg sample was then digested with LysC/Trypsin, and 300 ng of the resulting product was analyzed via 4-hr LC/MS/MS. Data were processed and quantified using Elucidator, and analyzed statistically using one-way ANOVA. See Supplementary Methods for details.
Automated hematological analyses
We collected fresh blood into a 20-μl plastic Boule MPA Micro pipettes (EDTA-K2, Boule Medical AB, Stockholm, Sweden), which we inserted into a Heska HemaTrue® Veterinary Hematology Analyzer (Heska corporation, Loveland, CO) according to the manufacturer’s protocol. See Supplementary Methods/Inhibitor Studies for details on preparation of RBCs for automated hematological analyses in inhibitor studies.
Blood smears
We collected fresh blood from 3 mice of each genotype into 2-ml K2E K2EDTA VACUETTE® tubes (Monroe, NC 28110, USA). Blood smears were prepared using microscope slides (Fisher scientific, Pittsburgh, USA), and were stained using Wright’s stain on a Sysmex SP-10 autostainer (Kobe, Japan). Air-dried smears were placed in staining cassettes, and staining performed per manufacturer’s specifications. Blood smears images (1000× magnification) were taken on an Olympus BH-2 microscope (Tokyo, Japan) with a DP73 digital camera attachment (Olympus) and visualized with cellSens software (Olympus). See Supplementary Methods/Inhibitor studies for details on preparation of RBCs for blood smears in inhibitor studies.
Still microphotography and microvideography of living RBCs
We performed experiments (still or video) on fresh blood collected as described above from one mouse of each of four genotypes, all on the same day. We repeated this on three different days, for a total of 3 mice of each genotype, for both still and video studies (i.e., a total of 3+3 mice/genotype). RBCs at an initial Hct of 25% to 30% (see above) were suspended 1:10 in oxygenated solution (Supplementary Table 12), for a final Hct of 0.5% to 1%, and stored on ice for 30 – 120 min before imaging. A droplet containing suspended RBCs from one mouse was placed on a glass coverslip that served as the bottom of a recording chamber. The chamber was then mounted on Olympus IX-81 inverted microscope equipped for differential interference contrast (DIC) studies, using either of two oil immersion objectives (60× objective, NA 1.42 for still micrographs or 40× objective, NA 1.35 for microvideos) with a 1.5× magnification selector. The light was detected with an intensified EM-CCD camera (C9100-13, Hamamatsu Corporation, Bridgewater, NJ) with 512 × 512 pixels, and data acquired using SlideBook 5.0 software (Intelligent Imaging Innovation, Denver, CO) for the Hamamatsu camera. We recorded still micrographs or microvideos (1 frame per 5 sec) of the RBC droplet as RBCs fell freely through the plane of focus, toward the coverslip surface. See Supplementary Methods/Inhibitor studies for details on preparation of RBCs for blood smears in inhibitor studies.
Flow cytometry
On one day, fresh blood was collected (see above) from 2 mice of each genotype. Experiments were repeated on a second day, for a total of 4 mice per genotype. To permit gating of viable RBC precursors, 100-μl samples of cells were diluted to 1% Hct in oxygenated solution containing 1 μM Calcein Violet (CV; viability marker; ex/em 405/450 nm; TFS, C3099), 0.1 μg/ml Thiazol Orange (TO; to stain RNA; ex/em 490/530 nm; Sigma-Aldrich 390062), and 5 μM DRAQ-5 (to stain DNA; ex/em 647/683 nm; TFS, 62252), and then incubated for 20 min at RT in the dark. The dye-loaded RBC samples were then washed ×3 in 1 ml oxygenated solution, centrifuged at 600×g for 5 min between washes. Experiments were performed at either 0.06% Hct (∼2 million cells/ml) for light scattering on the LSRII or 1 % Hct for imaging on the ImageStream. Stained cells were maintained on ice for up to ∼2 h on ice for experiments performed that day.
Forward-scatter intensity area (FSC-A) is a measure of cell size; side-scatter intensity area (SSC-A) is a measure of size, granularity, surface projections, and (for asymmetric cells) orientation32. FSC-A, SSC-A, as well as the fluorescence of CV, TO, and DRAQ-5 were measured with an LSRII Flow cytometer (BD Biosciences; San Jose, CA). Data were gated on FSC-A vs. FSC-W (width), enabling separation of RBCs from very small events and aggregates (Supplementary Data Figure 3a). CV-positive cells (>99.9%) were analyzed for TO and DRAQ-5 fluorescence, yielding gated populations (Supplementary Data Figure 3b) of TO-negative/DRAQ-5–negative cells (>96%) and TO-Positive/DRAQ-5–positive cells (∼3%). TO-negative/DRAQ-5–negative cells were then analyzed for light-scatter characteristics in WT, Aqp1–/–, RHag–/–, and dKO mice (Extended Data Figure 8). To evaluate FSC (size) of RBC populations quantitatively, gates were set on the center 90% of “dim” or negative cells in CV vs. DRAQ5, TO vs. CV, and TO vs. DRAQ5 plots. These gates were combined (Boolean AND) with a center 90% gate on an FSC vs. SSC plot. See legend of Extended Data Figure 8 for additional details on analysis.
ImageStream flow cytometry (Amnis, EMD Millipore) analyzes individual cells, in flow, by bright-field and multiple fluorescence parameters. Data are collected as images—two-dimensional spatial grids, in which a third dimension of intensity is captured for each pixel. RBCs are loaded with CV (for viability), TO (for RNA), and DRAQ-5 (for DNA) as described above. Gating schemes (Supplementary Data Figure 4) were established to allow size analysis of individual RBCs (Supplementary Data Figure 5) and RBC precursor types separately from one another. See Supplementary Methods/Inhibitor studies for details on preparation of RBCs for ImageStream analyses in inhibitor studies. Supplementary Data Figure 9 shows images of abnormally shaped cells, and the legend of that figure describes how we identified them using ImageStream. DIC still microphotography (Supplementary Data Figure 8) and DIC microvideography (not shown) show that these misshapen cells are spheres.
Mathematical modeling
We used a reaction-diffusion model—based on one originally developed for CO2 fluxes33—of O2 efflux from an RBC that we modeled as a sphere (Supplementary Methods: Model formulation), as has been done by others previously (Supplementary Methods: Sphere). The diameter of the sphere equals RBC thickness, which we computed from measured hematological and morphological parameters (Supplementary Methods: Thickness). As illustrated in Extended Data Figure 9a, the spherical RBC, a membrane encompassing ICF, is surrounded by a thin layer of EUF, which is in turn surrounded by an infinite reservoir of bECF. Reactions among O2, Hb and HbO2 occur only in the ICF. We model the reaction term as the simple one-step reversible reaction combined with a variation (Supplementary Methods: VRC) of the “variable rate coefficient” model proposed by Moll34, which guarantees that the hemoglobin-O2 saturation curve is sigmoidal. All solutes diffuse within the cytosol and bECF. O2 is the only solute that can move through the plasma membrane. Extended Data Figure 10b and c show examples of fluxes through selected reaction and diffusion events in two simulated HbO2 desaturation time courses. SI Methods contains details on the computational model, parameter values, and the simulation of the time course of deoxygenation of HbO2.
Statistical Analysis
We report results as mean ± s.e.m. In each figure legend, we report which statistical tests we performed, from among the following, to generate unadjusted p-values: (1) paired two-tailed t-test, (2) unpaired two-tailed Welch’s t-test35 or (3) one-way analysis of variance (ANOVA). For comparisons of 2 means, we performed #1. For comparisons among >2 means, we performed #2 or #3 and then, to control for type I errors across multiple means comparisons, we applied the Holm-Bonferroni correction, setting the familywise error rate (FWER) to α = 0.05. Briefly, we order the unadjusted p-values for all 𝒩 comparisons in each dataset from lowest to highest. For the first test, we compare the lowest unadjusted p-value to the first adjusted α value, α/𝒩. If the null hypothesis is rejected, then we compare the second-lowest p-value to the second adjusted α value, α/(𝒩–1), and so on. If, at any point, the unadjusted p-value is ≥ the adjusted α, the null hypothesis is accepted and all subsequent hypotheses in the test group are considered null.
Data availability
The data supporting the findings of this study are available within the paper and its Supplementary Information files. Any further relevant data are available from the corresponding author upon reasonable request.
Extended Data
Acknowledgements
We thank Jean-Pierre Cartron for the gift of the RHag–/– mice and for helpful discussions. We thank Thomas Radford for organizing the husbandry of the mouse colonies; Gerald Babcock for his role as laboratory manager; James W. Jacobberger and Philip G. Woost of the CWRU Flow Cytometry and Imaging Microscopy Core (FCIMC) for their assistance with flow cytometry; Daniela Schlatzer of the CWRU Center for Proteomics and Bioinformatics for their assistance with mass spectrometry. This work was supported by Office of Naval Research (ONR) grant N00014-11-1-0889, N00014-14-1-0716, and N00014-15-1-2060; a Multidisciplinary University Research Initiative (MURI) grant N00014-16-1-2535 from the DoD, NIH grant multi-scale modeling grant 5U01GM111251 (to WFB). R.O. and the modeling were supported in part by NIH grant K01-DK107787. R.R.G was supported by a fellowship grant from the ONR (N00014-12-1-0326). The authors gratefully acknowledge Daniela Calvetti and Erkki Somersalo for having developed the engine of an earlier version of the CO2/pH reaction-diffusion model of an oocyte, which in part served as the starting point for the RBC model.
Footnotes
Title Page: Added Gerolf Gros as an author and added affiliation Page 1: through the to the bulk extracellular... is fixed to to the bulk extracellular... Page 2 in paragraph 1: more that diffusion through... is fixed to more than diffusion through... Page 9: Reprints and permissions information is available at www.nature.com/reprints is deleted in author information. End Notes: Removed Gerolf Gros from acknowledgements and added Gerolf Gros author contribution.