Impaired O₂ unloading from stored blood results in diffusion-limited O₂ release at tissues: evidence from human kidneys

The assertion that O₂ exchange is not diffusion-limited has guided treatments for circulatory shock. Recommendations for increasing D_O2 include positive inotropes (to increase Q), transfusion (to raise Hb content), and ventilation (to raise S_A,O2)⁴ but not improvements to RBC gas-handling kinetics. Perfusion-limited delivery is embodied in early goal-directed therapy (EGDT),^5-8 the protocolized management of circulatory shock secondary to sepsis. In terms of physiological regulation, perfusion-limited O₂ delivery is favorable because the principal factor restricting O₂ supply to cells is arterial flow, which tissues control through vessel resistance.⁹ For instance, if the respiratory demand for O₂ is not met, D_O2 can increase by raising Q. However, this linearity does not apply under a diffusion-limited scenario because increasing Q also shortens the capillary transit time, which reduces fractional O₂ unloading and offsets the effort to raise D_O2. Thus, the kinetics of RBC O₂-handling are critical in determining the transition between perfusion- and diffusion-limited transport, but rarely measured. A commonly reported surrogate is Hb-O₂ affinity, but this steady-state variable cannot predict kinetics. For instance, 2,3-diphosphoglycerate (2,3-DPG) depletion stabilizes HbO₂, but whether this slows O₂ release to the diffusion-limited threshold is unclear. Indeed, a left-shifted O₂-saturation curve could still release the amount of O₂ instructed by respiration by attaining a lower P_O2 at the venous end, provided there is sufficient time for gas equilibration.

To measure O₂-unloading from RBCs, we developed single-cell oxygen saturation imaging. This revealed kinetics that are considerably slower than previous estimates¹⁰ based largely on measurements from Hb solutions.^11-15 The additional delay is attributable to restricted gas diffusion across the macromolecule-packed cytoplasm. Consequently, changes in RBC shape can profoundly affect gas exchange. In storage, RBCs undergo metabolic and morphological changes^16,17 that result in a time- and donor-dependent attrition of O₂-unloading kinetics, reversible upon biochemical rejuvenation,¹⁸ but whether this reaches the threshold for diffusion limitation is unclear. Multiple human trials and animal studies have suggested no negative impact of storage duration on transfusion outcomes,^19-22 although RBC O₂-handling kinetics were not recorded and, instead, randomization used storage duration, a metric that is not an accurate surrogate of RBC quality.²³ Moreover, those trials recruited participants with stable anemia, rather than patients requiring massive transfusion for whom the imperative is to restore O₂ delivery to tissues immediately and where diffusion-limited exchange would be problematic. Because transfused RBCs undergo rejuvenation upon recirculation in a physiological milieu, evidence for an effect of storage lesion should be based on recording responses shortly after transfusion.

To seek evidence for diffusion-limited O₂ delivery with stored blood, we studied ex vivo organs perfused entirely with stored RBCs and recorded tissue responses in real time, such as respiration and oxygenation. Red cell concentrates were obtained from the UK’s National Health Service Blood and Transplant (NHSBT) in accordance with standard storage practice. The kidney is an appropriate organ for these experiments because arterial blood flow determines filtration rate, hence tubular transport and O₂ demand, and any discrepancy with arterial O₂ supply, as defined by Equation 1, would indicate diffusion-limited O₂ exchange (Figure 1A). These measurements were made during prolonged normothermic machine perfusion of the kidney^24,25 conducted as part of a phase 1 trial testing the safety and feasibility of normothermic machine perfusion of the kidney prior to deceased-donor renal transplantation. In a second set of experiments, kidneys deemed unsuitable for transplantation were perfused alternately with RBCs of slow (stored) or fast (rejuvenated) O₂-handling properties to establish causality with tissue readouts. Because renal P_O2 gradients are relatively small owing to modest O₂ extraction (∼10%-15%), any evidence for diffusion-limited delivery would likely apply to organs with greater fractional extraction, such as the heart. Compared with many other organs, kidneys show less variation in vascular function and are more readily available because transplants are common.

Our results indicate that renal respiratory rates were lower in kidneys perfused with RBCs of slow O₂-unloading kinetics. Consistent with a diffusion-limited process, a positive correlation between renal respiration and O₂ delivery is obtained once the rate of O₂ release from RBCs is considered. Switching to biochemically rejuvenated blood during perfusion promptly increased the kidney’s O₂ diffusive capacity and cortical oxygenation. Our findings challenge the paradigm that O₂ exchange at capillaries is invariably rapid, highlight the need to consider RBC gas-handling properties, and justify efforts to optimize blood storage regimes or rejuvenation protocols.

Normothermic Kidney Perfusion Phase 1 (NPK1) was a phase 1 regulated device trial (ISRCTN13292277; ethical approval was obtained from Greater Manchester South Research Ethics Committee REC 20/NW/0442) investigating the safety and feasibility of prolonged-duration normothermic machine perfusion of the kidney prior to transplant; the observational component of our study drew on samples and data from this trial. Additional ethical approval was obtained for experimental work with kidneys unsuitable for transplantation (REC 21/PR/1546). All participants in the clinical trial, blood donors, and donor families (for deceased-donor research organs) gave written informed consent. Organ retrieval procedures were conducted in accordance with normal clinical practice, and kidneys were transported to Oxford using conventional ice-box storage. Conventional back-table surgery was performed before attachment of the kidney to the investigational device and perfusion at 37°C with RBC-based perfusate before transplantation (Figure 1B). All units of blood used for perfusion were produced in accordance with NHSBT procedures, with cold-chain maintained until addition to the perfusion machine. During clinical kidney perfusions, blood gases were measured continuously (Terumo CDI-500) with additional arterial and venous point measurements made using an external blood gas analyzer (ABL-FLEX 90). The supply of O₂ and air was automated to maintain arterial P_O2. Arterial pressure was automatically regulated (90 mm Hg for the first 24 perfusions and 75 mm Hg thereafter), as was temperature (37°C). For rejuvenation experiments, RBC units toward the end of their clinical shelf-life were split, and 1 part was treated with a solution of pyruvate, inosine, phosphate, and adenine (PIPA; marketed as Rejuvesol, Zimmer Biomet). Alternating perfusion between standard-stored blood and rejuvenated blood used 2 perfusion machines with separate blood loops feeding to a common dialysis system.

The kinetics of oxygen unloading from RBCs were studied using a modification of a previously published method.^10,26 

v′_R,O2 varied from 1.9 to 5.5 mL/min O₂, with median of 3.11 mL/min O₂ (Figure 2E). The differences in v′_R,O2 were not due to variation in kidney mass (Pearson ρ = 0.10; P = .24).

The kinetics of O₂-unloading from RBCs were measured using single-cell oxygen saturation imaging under superfusion in a microfluidic chamber (Figure 3A) that allows rapid exchange between a normoxic and anoxic solution (Figure 3B). Owing to the O₂-dependent shift in Hb absorbance, the ratio of CellTracker DeepRed and Calcein Green fluorescence describes the O₂ saturation of superfused RBCs. O₂ unloading was triggered by exposing RBCs to a 10-second period of anoxia (Figure 3C).

To obtain reference values for the O₂-unloading time constant, τ, freshly drawn venous blood was analyzed from 32 registered blood donors. Figure 3D shows their blood group and age. Mean corpuscular volume (MCV, median, 90 fL) and mean corpuscular Hb concentration (MCHC, median, 333 g/L) were normal. The median, was 0.975 seconds, consistent with previous recordings,^10,18 and the change in fluorescence ratio (O₂-carrying capacity, κ) was 0.499 (Figure 3E).

Next, measurements were performed on stored bloods allocated for kidney perfusions. Storage duration varied from 4 to 30 days, with a median of 12.5 days (Figure 3F). RBCs spanned a range of kinetic attrition, including RBCs with substantially slower O₂-unloading compared with reference blood (Figure 3G). Critically, the functional impairment incurred during storage did not correlate strongly with storage duration,¹⁸ which indicates that storage duration is not a good surrogate for changes in RBC functional quality. This strengthens the case for correlating kidney outcomes with a direct measure of O₂ release from RBCs. Venous blood from the kidney recipient was also analyzed for RBC O₂-unloading and suggested a largely normal range with 2 outliers (Figure 3H).

RBCs circulating on the perfusion machine may become rejuvenated upon contact with the organ; thus, to relate renal respiration with RBC gas handling, small samples of blood were obtained during perfusions for measurements of τ and κ (Figure 4A-B). There was an overall trend for RBCs to reduce the rate of O₂ release and then undergo partial recovery. During this time, arterial pH also showed time-dependence, gradually recovering from mildly acidic to physiological levels. Thus, correlations between RBC O₂-handling properties and renal respiration used data once pH had stabilized, labeled here as “end point” (Figure 4C,D). Perfusion was determined to have a significant accelerating effect on O₂ release but, no significant effect on capacity (Figure 4E).

Figure 4F plots calculated v′_RBC,O2 in stored blood before and during kidney perfusion, illustrating the considerable range in terms of ability to release O₂. If gas exchange were diffusion limited at capillaries, then release from RBCs at the lower range of v′_RBC,O2 is most likely to become rate-limiting for renal respiration (v′_R,O2).

Unlike the perfusion-limited definition, the diffusion-limited variant produced a strong correlation with renal respiration (Pearson ρ = +0.58; P < .0001; Figure 5D). The difference between these alternative models relates to the ability of RBC to release a flow of O₂ driven by a respiratory demand. Under perfusion-limited delivery, renal respiration extracts the required amount of O₂ owing to rapid O₂ exchange between tissues and capillaries. Consequently, v′_RBC,O2 becomes irrelevant, because full equilibration is imminent during capillary transit, and small variations in v′_RBC,O2 would not affect tissue oxygenation. In a diffusion-limited scenario, the kidneys can only extract a fraction of the demand set by filtration and tubular transport because RBCs transiting capillaries will not have enough time to equilibrate with tissue P_O2. Consequently, renal respiration will be lower than expected for any given blood flow, explaining the inverse relationship, as presented in Figure 5B. The emergence of a strong positive correlation after introducing a factor describing RBC gas-handling (Figure 5D) indicates that gas exchange at capillaries is, indeed, diffusion-limited.

The correlation between RBC O₂-handling properties and renal respiration is observational, but does not necessarily imply causality. Moreover, the kidneys represent a heterogenous pool of organs, which may affect correlations. Additional evidence for diffusion-limited O₂ release was sought by altering the O₂-handling properties of RBCs during perfusion of the same kidney. Two units of blood (ABO-matched and of equal storage duration within each pair, between 35 and 42 days) were pooled and split. One part was rejuvenated biochemically with a solution containing PIPA previously shown to improve oxygen-unloading kinetics,¹⁸ whereas the other received sham treatment (i.e. with standard saline, adenine, glucose, and mannitol containing additive solution). PIPA may not reverse all aspects of storage lesion, but its efficacy in restoring O₂ unloading is robust and directly relevant to our study. The component of PIPA with the greatest effect on accelerating O₂ handling is likely inosine, because omitting this substance abrogated rejuvenation efficacy (supplemental Figure 2A). Rejuvenated RBCs maintained faster O₂-unloading rates even when treatment was performed 2 weeks before measurements, giving a wide window of effectiveness (supplemental Figure 2B).

Kidney perfusion alternated between standard and PIPA-rejuvenated blood using a twinned dialyzed circuit that keeps all noncellular perfusate parameters and the respiring tissue constant, leaving RBC O₂-handling properties as the sole independent variable (supplemental Figure 3). Rejuvenated RBCs had faster O₂ unloading, without affecting carrying capacity (Figure 6A,C). The effect of perfusion with rejuvenated RBCs was quantified in terms of O₂ diffusion capacity, calculated as the ratio of respiratory rate v′_R,O2 to the P_O2 gradient between capillaries and tissue. Capillary P_O2 was approximated from arterial PO₂, whereas organ-averaged tissue P_O2 was approximated from urine P_O2 measured in the renal pelvis. Blood flow and urine production were constant during the experiment. If rejuvenation of O₂-unloading kinetics had a meaningful impact on gas exchange at capillaries (i.e., the system was diffusion-limited), then the O₂ diffusion capacity would increase during perfusions with rejuvenated blood, and reverse upon return to standard-stored blood perfusion. Faster O₂ diffusivity between RBCs and renal mitochondria would allow a smaller P_O2 gradient (i.e., higher tissue P_O2) to drive an adequate O₂ flux that meets respiratory demand.

Figure 6B shows the time courses of P_O2 in the renal artery, cortex, and urine during this alternating perfusion protocol performed on 3 kidneys. Rejuvenation increased P_O2 in the cortex and urine, consistent with faster O₂ unloading having a favorable effect on overall diffusive capacity. This finding is in line with a diffusion-limited system. Notably, cortical P_O2 increased by 60% after the transition from standard to rejuvenated blood (Figure 6D). The diffusion capacity for O₂ increased significantly during perfusions with rejuvenated RBC, indicating that the rate of O₂ unloading from RBCs can meaningfully affect O₂ transport from capillaries to respiring cells. Because the microvasculature and energetic demand from tubular transport are constant during the experiment, the increase in diffusion capacity must relate to a change in RBCs, consistent with diffusion-limited transport.

The results of our study highlight the circumstances that can lead to diffusion-limited O₂ exchange at systemic capillaries, and challenge the canonical view that this is invariably perfusion-limited. With an O₂-unloading time constant near 1 second in freshly drawn blood¹⁰ and slowing further during storage,¹⁸ it is possible that capillary transit times may not allow P_O2 equilibration.^27,28 Using human kidneys perfused with stored blood, we show that the attrition in RBC O₂-handling properties limits O₂ extraction driven by renal respiration. This relationship is unlikely to be affected by hypoxic signaling (eg, erythropoietin, EPO, production) because the duration of the perfusion periods in our interventional experiments precludes transcriptional responses. Arterial blood flow drives O₂ demand by setting tubular transport and also supplies O₂ for respiration; thus, a linear relationship between arterial O₂ delivery and respiration is expected under perfusion-limited transport. However, we observed no correlation between arterial O₂ delivery and renal respiration, unless the former included a factor describing O₂ release from RBCs, a defining feature of diffusion-limited exchange. Compelling evidence for diffusion-limited exchange was obtained by alternating the O₂-handling properties of RBCs during kidney perfusion. Here, the only factor being varied was the O₂-unloading time constant from RBCs, from slow, in standard-stored blood, to fast in its rejuvenated counterpart. Rejuvenation of stored blood hastened O₂ unloading and increased the O₂ diffusive capacity across the kidney, which allowed a smaller P_O2 gradient to drive the necessary flow of O₂ for respiration. Strikingly, a simple biochemical intervention performed on blood was able to increase cortical oxygenation by 60%. We conclude that O₂ release in the kidney becomes a diffusion-limited process at least in the case of RBCs manifesting a kinetic dysfunction, such as that incurred in storage. The emergence of diffusion limitation in an organ, such as the kidney, characterized by a relatively small degree of O₂ extraction, makes it plausible that other organs, with higher extraction rates, also show such properties. Our results are consistent with an earlier report in rats receiving isovolumic transfusion of stored blood,²⁹ which found that blood samples stored for 5 to 6 weeks produced a greater decrease in renal microvasculature P_O2 than units stored for 3 days, but the effect of prolonged storage could be reduced by rejuvenation. All other factors being equal, a bigger drop in P_O2 is consistent with a greater diffusive barrier to the flow of O₂ between the blood and mitochondria. If RBCs are the independent variable, the most likely explanation is slower release in a diffusion-limited system.

Our findings have implications for our fundamental understanding of gas exchange. Contemporary models of blood transport should consider the possibility that at least under certain conditions, capillary P_O2 may not reach equilibrium with interstitial P_O2, leading to a mismatch between demand and supply. If kinetically-compromised RBCs transiting capillaries have insufficient time to equilibrate P_O2, then the cardiovascular responses that are normally effective in restoring oxygenation under a perfusion-limited scenario may not produce the desired outcome. Faster blood flow will further reduce transit time, and any increase in arterial delivery would be offset by incomplete fractional unloading of RBCs. Consequently, a diffusion-limited process is more likely to incur an oxygen debt, and the only way of improving delivery would be to address the rate-limiting process underpinning slow gas exchange in capillaries, that is, restoring RBC O₂-handling kinetics. In the case of pulmonary capillaries, diffusion-limited exchange may cause arterial hypoxemia, particularly under strenuous exercise. Indeed, there are reports of arterial hypoxemia in elite female athletes because of incomplete O₂ uptake by RBCs,³⁰ and any exacerbation of RBC kinetic properties would make this scenario more likely.

Diffusion-limited O₂ delivery with stored blood is predicted to reduce the maximal respiratory rate in exercise (v′_O2^max). Attempts to test this have been made for elite cyclists receiving autologous transfusion with stored blood.³¹ At least 1 study³² showed that long-stored blood is less effective in improving exercise endurance compared with short-stored blood, which is consistent with our findings. Some studies have shown that v′_O2^max invariably increases with transfusion,³¹ which could be considered as evidence against diffusion limitation. However, there are a number of factors to consider before reaching such a conclusion. Firstly, donor age and fitness may affect vulnerability to storage lesion, and it is plausible that elite cyclists undergo a less pronounced kinetic attrition in storage, compared with the general population. Because O₂-handling kinetics of RBCs were not measured in these studies, the significance of donor profile is unknown. Secondly, autologous transfusions will increase total Hb, which should increase O₂ delivery even if most of the transfused RBCs are kinetically inferior. This is because RBC O₂-handling shows considerable variation, and at least some cells will retain favorable properties, allowing v′_O2^max to rise, even if by a small but significant amount. Thus, both long- and short-stored blood are likely to increase v′_O2^max, when total Hb is not controlled for. Finally, stored RBCs upon recirculation in a physiological milieu undergo a gradual restoration of key surrogates of storage lesion, including 2,3-DPG.³³ If v′_O2^max measurements are taken after in vivo rejuvenation is complete, the consequences of diffusion-limited exchange will not be apparent. Diffusion-limited O₂ delivery is most likely to emerge when the body attempts to increase tissue oxygenation by increasing blood flow under circumstances in which O₂-unloading from RBCs is impaired. Because cyclists are highly trained, their physiology may be well-adapted to handle an increase in blood flow within the limits of perfusion-limited delivery. A different situation will occur in patients who are critically ill, for whom attempts to increase blood flow could lead to diffusion-limited exchange after transfusion with stored RBCs. This may explain why there is no apparent benefit of liberal transfusion strategies over more restricted approaches.^34,35 

Although we have demonstrated diffusion-limited O₂ delivery during perfusion with RBCs compromised by the storage lesion, we have not shown its physiological consequence on the kidney, such as on signaling through hypoxia-inducible factor (HIF) and EPO production. In this study, it was critical to minimize bleeding to achieve technical success, which precluded taking repeated biopsies to examine tissue HIF induction. In our experiments, we sought to avoid possible on-circuit rejuvenation of RBC O₂-handling, which would be a confounding factor; therefore, our perfusions were shorter than the timeframe for HIF-evoked transcriptional responses. However, HIF stabilization and EPO production are aspects that should be investigated in future studies.

Evidence for diffusion-limited O₂ delivery has implications for clinical management of shock.⁴ With diffusion-limited gas exchange, tissue oxygenation will not respond to inotropes because these would abbreviate transit time and reduce fractional release. Transfusions that aim to raise [Hb] should consider the O₂-unloading kinetics of the blood product and the speed with which a treatment effect is clinically required. In some circumstances (eg, top-up transfusions and no critical ischemia), kinetically-impaired transfused RBCs may gradually recover after transfusion and provide the intended benefit within hours. As a result, diffusion-limited O₂ transfer may not be apparent when considering clinically-relevant end points. However, when a treatment effect is required immediately due to critical ischemia or a high ratio of transfused product to native circulating volume (eg, ex vivo organ perfusion, major hemorrhage, large-volume pediatric transfusion for cardio-pulmonary bypass, or repeated blood transfusions), slower O₂ release could offset the increase in O₂ content and not improve tissue oxygenation. EGDT uses various hemodynamic variables to guide the administration of fluids, packed red cells, inotropes, and vasopressors to increase blood flow, increase [Hb], and reduce the oxygen extraction ratio. Although early studies were promising,⁵ a meta-analysis of 3 pivotal randomized, controlled trials indicated little benefit.^36,37 Systematic review and meta-analysis of studies specifically considering transfusion threshold have shown no benefit from transfusion.³⁸ Consequently, the International Guidelines for Management of Sepsis and Septic Shock recommend a restrictive transfusion policy.³⁹ It is plausible that a reason for the failure of EGDT to show more clinical benefit is that O₂ release from RBCs becomes diffusion-limited. Thus, the failure of transfusion to improve outcomes could relate to the kinetic attrition of stored RBCs, particularly when O₂ extraction is high. Beyond transfusion medicine, diseases that reduce oxygen release from RBCs, such as hereditary spherocytosis, may result in severely diffusion-limited O₂ delivery to tissues.

Our findings highlight the importance of monitoring the kinetic quality of stored blood. We previously showed that the rate of kinetic attrition varies among donor units.¹⁸ Storage duration itself is not an accurate proxy of kinetic quality and should be avoided when attempting to match units of stored blood to recipients, or when randomizing units for clinical trials studying transfusion efficacy. We propose that RBC quality should be determined using direct functional assays such as that presented here. When deemed necessary, blood should be rejuvenated biochemically to ensure that recipient outcomes are not compromised by problems associated with diffusion-limited gas exchange. Kinetic monitoring and biochemical rejuvenation may also extend the storage window for packed RBCs to help mitigate shortages of rarer blood types. Rejuvenation before blood product use is labor intensive, and we estimate a burden of 3 to 4 hours that may limit its use in settings, such as major hemorrhage. Although there is capacity to improve the formulation using τ as a performance readout, and the possibility of rejuvenating blood before issue, future studies should evaluate alternative storage regimes (eg, hypoxic storage²⁶), the addition of phosphate and guanosine, or adjustments in pH. This may be particularly important in circumstances where the transfusion-load is a substantial fraction of the recipient’s blood volume or when blood is given to alleviate critical ischemia.

It would be important to expand our study to other organs, such as the heart, brain, and skeletal muscle, where O₂ delivery must be as high as possible. Future trials should investigate how different storage protocols or rejuvenation techniques improve outcomes and test methods for measuring RBC kinetics so that donations can be matched to the most appropriate recipient.

Aspects of this project (specifically normothermic kidney perfusion phase 1, a phase 1 trial of prolonged duration normothermic kidney perfusion before transplant) were funded by the National Institute for Health and Care Research (NIHR) under its Invention for Innovation program (grant number NIHR200022, Chief investigator: P.F.). Research was selected by the European Research Council (ERC) and funded by UK Research and Innovation (UKRI; ERC and MSCA Other Guarantee Calls), EP/X021548/1 (P.S.). Funding support was received from Corpus Christi College, Oxford, and the John Fell Fund (0010677). T.K., S.B., and P.K. received funding from the Norwegian Financial Mechanism GRIEG-1 grant managed by the National Science Centre, Poland (2019/34/H/NZ6/00699). This project also received grant support from the Oxford Transplant Foundation (R.D.), and support in-kind (perfusion machines and disposable sets) from OrganOx Ltd.

The views expressed are those of the authors and not necessarily those of the NIHR or the Department of Health and Social Care.

Contribution: P.S. wrote the first draft of the manuscript, and all authors contributed to revisions and approved the final version; R.D. was the trial manager and lead researcher for NKP1; P.F. was the chief investigator; R.D., J.H., S.K., A.W., R.P., C.C., and P.F. were the trial management group, local investigators, designed the trial, and have access to the primary or identifiable trial data; R.D. conducted all clinical perfusions, assisted by J.F.; for single-cell saturation imaging and data analysis the diffusion-limited oxygen transfer hypothesis was developed by P.S., R.D., and J.R.; R.D. provided deidentified perfusate samples from NKP1; RBC kinetic analysis was developed and conducted by P.S. and J.R., respectively; P.S., R.D., and J.R. performed data analysis; R.D., J.R., and P.S. developed the hypothesis-testing experiments for rejuvenation experiments; R.D. designed the twinned-circuit perfusion setup and performed the perfusions assisted by J.F. and J.R.; J.R. performed blood rejuvenation; R.D., J.R. and P.S. performed data analysis and interpretation of results; T.K., S.B., and P.K. provided research materials (microfluidics); and D.V., J.B., M.E., and C.C. engineered the perfusion machines and provided technical support.

Conflict-of-interest disclosure: P.F. and C.C. are founders, shareholders, and directors of OrganOx Ltd, and serve, respectively, as its Chief Medical and Technical Officers for which they receive consultancy fees. R.D., S.K., and J.F. have received consultancy fees from OrganOx Ltd. The remaining authors declare no competing financial interests.

Correspondence: Pawel Swietach, Department of Physiology, Anatomy and Genetics, Parks Rd, Oxford OX1 3PT, United Kingdom; email: pawel.swietach@dpag.ox.ac.uk.