Advantages of a Centrifugal Pump in the Development of a Swine Model of an Arti cial Placenta

Alex J. Charest-Pekeski University of Toronto Steven K.S. Cho University of Toronto Tanroop Aujla University of Toronto Liqun Sun The Hospital for Sick Children Alejandro A. Floh The Hospital for Sick Children Mark J. McVey The Hospital for Sick Children Ayman Sheta The Hospital for Sick Children Marvin Estrada The Hospital for Sick Children Lynn Crawford-Lean The Hospital for Sick Children Celeste Foreman The Hospital for Sick Children Dariusz Mroczek The Hospital for Sick Children Jaques Belik The Hospital for Sick Children Brahmdeep S. Saini University of Toronto Jessie Mei Lim University of Toronto Olivia J. Moir The Hospital for Sick Children Fu-Tsuen Lee University of Toronto Megan Quinn University of South Australia Jack R.T. Darby University of South Australia Mike Seed The Hospital for Sick Children Janna L. Morrison University of South Australia Christoph Haller (  christoph.haller@sickkids.ca ) The Hospital for Sick Children


Introduction
In resource rich nations, extreme prematurity, de ned as delivery prior to 28 weeks' gestation, remains the leading cause of childhood mortality and morbidity 1,2 . Survival rates decline with decreasing gestational age (GA), with only 6% surviving at 22 weeks GA compared to >90% surviving at 28 weeks GA 3 . In Canada alone, the economic burden of caring for extremely preterm infants approaches $600 million nationally per annum 4 . Unfortunately, despite advances in medical technology and neonatal intensive care, improvements in the outcomes of children born extremely preterm over the last 15 years have been limited 5 .
At the biological limit of viability (22)(23)(24)(25) weeks GA), preterm infants commence pulmonary gas exchange during the late-canalicular and early saccular stages of lung development. Exposure to the positive pressure mechanical ventilation and high partial pressure of inspired oxygen required to achieve adequate gas exchange results in cessation of alveolarization and pulmonary microvascular injury, which is associated with high rates of chronic lung disease and pulmonary hypertension 6 . Furthermore, approximately 50% of children born at the threshold of viability exhibit neurological disabilities at 30 months corrected age, with half of these cases being classi ed as severe 7 . Despite major advances in the outcomes of prenatal infants resulting from the widespread administration of prenatal steroids and exogenous surfactant, recent improvements in the morbidity and mortality have been more incremental, emphasizing the need for new innovative approaches to supporting the fragile physiology and development of extremely preterm infants [1][2][3] .
An arti cial placenta (AP) represents a novel approach that aims to maintain the innate fetal circulation while promoting normal prenatal development. Gas exchange is achieved with a low-resistance hollow-ber membrane oxygenator connected to the fetus via the umbilical vasculature, while incubating the fetus in a uid lled environment. The rst attempt to support previable human infants was reported in the late 1950s 8 ; with subsequent progress made using animal models. These early experiments were complicated by the development of heart failure and infections and concurrent improvements in conventional neonatal intensive care led to diminished enthusiasm for AP technology 8, 9,10 . However, over recent years, several research teams have demonstrated the feasibility of supporting preterm goat and sheep fetuses with a variety of pumpless and pumped arteriovenous and venovenous AP systems using different approaches to establishing vascular access 8,9,[11][12][13][14][15][16][17][18][19][20][21][22][23][24] . Two research teams have successfully demonstrated physiologic fetal sheep hemodynamics with normal organ maturation and minimal injury for periods of up to one month on pumpless arteriovenous AP systems [17][18][19]23 .
Despite these advances, AP is yet to be proven in a realistic animal model of the extremely preterm human infant. At a comparable stage of lung development to a previable human fetus, the preterm sheep is approximately twice the weight 25,26 . Fetal size determines important anatomical considerations for establishing an extracorporeal membrane oxygenation (ECMO) system such as blood vessel diameter and blood pressure. In addition, fetal sheep possess two umbilical arteries (UA) and two umbilical veins (UV). This may allow a more stable transition to the AP circuit using one pair of UA and UV while maintaining native gas exchange via the placenta using the other pair of umbilical vessels, thus minimizing the cessation of oxygenation from the mother 27,28 . By contrast, human and pig fetuses usually have one UV and two UAs, which may represent a more technically challenging approach to the initiation of AP support.
We have previously proposed 27 that Yucatan miniature pigs delivered at approximately 95 days gestation represent a realistic model of human infants born at the lower limit of viability in terms of their body weight and stage of lung development 25 . We recently demonstrated the feasibility of cannulating the umbilical vessels of preterm minipigs and transferring them to an AP system consisting of a warm, uid environment and a pumpless ECMO circuit comprising a commercial neonatal oxygenator 27 . However, our experiments were characterized by subphysiologic circuit ows, tachycardia, and the development of hydrops. Echocardiography further revealed evidence of diminished right ventricular function, which we attributed to excessive ventricular afterload resulting from the large priming volume of the circuit and small umbilical cannulas 27 . We hypothesized that the addition of a pump might improve the hemodynamics of our animals on the circuit by reducing afterload through the generation of negative pressure downstream of the umbilical arterial cannulas, thereby improving venous return and cardiac output. Herein, we sought to investigate the hemodynamics of a pumped AP system, comparing our ndings to those obtained using the pumpless circuit 27 .

Methods
Experimental groups.
The results presented in this report are comprised from fetal pigs studied in utero and while maintained using a pumped AP circuit and of our previously published ndings using a pumpless AP system 27 . Fetal Large White Landrace pigs were studied in utero at 98, 106, and 112 days GA (n = 24 ) and Yucatan miniature pigs at 107 ± 3 days (n = 16), providing reference physiologic data regarding heart rate (HR), blood pressure, blood gases, electrolytes, lactate and glucose concentrations, and UV ow 27 . Fetal Yucatan miniature pigs (n =13) were maintained using an umbilical arteriovenous AP circuit consisting of a centrifugal pump and oxygenator and using a pumpless AP circuit (n = 12) 27 . We included AP subjects that survived on the system for a minimum of 3 hours.
Animals and approvals.  27 . Yucatan pigs were housed in pairs for at least two weeks prior to surgery to increase socialization between the animals and allow for acclimation to human handling and to their new environment. Sows were provided with ad libitum food and water, and environmental enrichments as per SickKids standard operating procedures.
Protocol for in utero studies.
Large White Landrace Cross gilts (n = 6; 98 ± 7 days GA; term = 115 days) were anaesthetised with an intramuscular injection (I.M.) of 20 mg/kg ketamine and inhalation of iso urane. Gilts were intubated and general anaesthesia was maintained using iso urane with 2 L/min O 2 and 4 L/min medical air. Gilts were positioned on the operating table on their backs, an incision was made along the abdomen, the uterus was incised, and a fetal head was exposed. Fetal pigs (n = 24) were cannulated via the UV, and venous blood was sampled for partial pressure of oxygen (PO 2 ), partial pressure of carbon dioxide (PCO 2 ), oxygen saturation (SO 2 ), pH, hemoglobin (Hb), bicarbonate (HCO 3 -), base excess (BE), sodium (Na + ), potassium (K + ), and calcium (Ca 2+ ) as previously described 27,28 . In a subset of fetuses (n = 21, 105 ± 7 days GA), the carotid artery (CA) was instrumented and fetal BP, and HR were measured and continuously recorded in LabChart 8 Pro (ADInstruments Inc., Colorado Springs, USA) 27 . Following in utero experiments gilts and their fetal pigs were humanely euthanized with an intravenous overdose of sodium pentobarbital (Virbac, New South Wales, Australia).
Surgical protocol for pumped AP studies.
Pregnant Yucatan pigs (n = 19; 101 ± 4 days GA; term = 115 days) were anaesthetized with an I.M. injection of 10 mg/kg ketamine hydrochloride, 0.20 mg/kg acepromazine, and 0.015 mg/kg atropine sulfate (CDMV Inc., Saint-Hyacinthe, Canada), with maintenance of general anaesthesia with inhalation of 2-3% iso urane (Fresenius Kabi Canada, Toronto, Canada). To prevent aortocaval compression, anaesthetized sows were positioned on the operating table in the left lateral position. Umbilical blood ow was measured in utero in 4 Yucatan sows in Toronto using cine phase contrast MRI as described in our previous publication. 27 A lower antero-lateral laparotomy was performed for a caesarean section. Following the delivery of all fetal pigs, sows were humanely euthanized with 106 mg/kg Euthanyl (CDMV Inc., Saint-Hyacinthe, Canada).
Fetal surgical procedures.
In a subset of fetuses (n = 11/13 successful pumped AP experiments), a small incision was made along the right side of the neck to expose the jugular vein (JV) and CA 28, 55 . A size-matched custom-made PVC tubing was then inserted in the JV and CA for monitoring of central venous pressure (CVP) and fetal mean arterial pressure (MAP), respectively. The tubing was stabilized using silk sutures, and the neck incision sutured closed. Fetuses were delivered to minimize excessive torsion and stretching of the umbilical cord and were positioned on the maternal abdomen and subsequently weighed. Fetal pigs were then anaesthetized using an I.M. injection of 1 mg/kg rocuronium bromide (Sandoz Inc., Mississauga, Canada), and 5 mg/kg ketamine hydrochloride (CDMV Inc., Saint-Hyacinthe, Canada). Fetal normothermia was maintained by continuously bathing the fetus and the umbilical cord in warmed normal saline. The umbilical cord was treated with a topical application of 100 mg/kg papaverine hydrochloride (Sandoz Inc., Boucherville, Canada).
In all AP studies, umbilical catheters were then placed in both UAs and the UV (12-gauge and 10-gauge custom-made cannulas, respectively), and secured into place using silk sutures (Ethicon Inc., New Jersey, USA) at the insertion of the cannula within the vessel. In smaller fetuses, UA cannulas were downsized to 14 GA angiocaths (Becton Dickinson Canada Inc., Mississauga, Canada). Cannulas were then connected to the AP circuit described below, via modi ed ¼" perfusion adapters (Medtronic of Canada Ltd., Brampton, Canada) to facilitate the initiation of UV ows. Re-positioning of the umbilical catheters was performed as needed to assist in the establishment of ow.
AP circuit design.
The arteriovenous AP circuit consisted of a commercially available centrifugal pump, connected to a low-resistance membrane oxygenator, and modi ed commercially available intravenous umbilical arterial and venous cannulas via ¼" internal diameter x 1/16" wall thickness, 3/16" x 1/16" TYGON PVC tubing with P.h.i.s.i.o coating (LivaNova PLC., London, England). In two experiments, a commercially available cardiopulmonary bypass oxygenator was used (Table 2); however, this was subsequently changed to another commercially available ECMO oxygenator (n = 11; Table 2). Total priming volumes for these neonatal membrane oxygenators were 40 and 65 ml, respectively. Circuits were then initially primed with plasmalyte (Baxter Inc., Mississauga, Canada) and then replaced with heparinized maternal blood, which was continuously recirculated to prevent clotting within the circuit and warm the blood prior to connection to the fetus. UA blood passed through the oxygenator inlet, exiting through the outlet before returning to the heart via the UV ( Figure 1). The sweep gas supplied the oxygenator with a mixture of medical air and oxygen and was titrated to achieve normal physiologic levels of UV PO 2 of 35-40 mmHg and a PCO 2 of 50-60 mmHg 29 based on in utero fetal pig data.

Fluid incubation
Following cannulation, fetal pigs were enclosed in a custom-made silicone infused thermoplastic polyurethane lm 'biobag'. Approximately 2-4 litres of lactated ringer (LR; Baxter Inc., Mississauga, Canada) crystalloid solution was warmed to 39 ± 1°C in a large uid reservoir and circulated via the in ow and out ow ports of the Biobag every 10 hours. The Biobag was equipped with a temperature port that facilitated monitoring of uid temperature using a temperature probe (ADInstruments Inc., Colorado Springs, USA). The biobag was covered to prevent transmission of light to the fetus to better simulate uterine conditions and maintained at an appropriate temperature using a contact heat pad underneath the biobag and an overhead heater.
Fetal pig maintenance on the AP circuit.
Following cannulation and transition to the AP, fetal pigs received a maintenance infusion of 6 µg/kg/h prostaglandin E1 (P zer Canada Inc., Kirkland, Canada) with 100 units/kg/h heparin (Fresenius Kabi Canada, Toronto, Canada) to maintain patency of the ductus arteriosus, and prevent clotting in the AP circuit 19 . Dextrose (418 mg/kg/h; P zer Canada Inc., Kirkland, Canada) was delivered via the circuit for the rst 8 hours of AP support, which was then exchanged for total parenteral nutrition solution with the goal of supporting fetal energy requirements and maintaining fetal glucose concentrations of >100 mg/dL. Calcium chloride (80 mg/kg; Omega Laboratories Ltd., Montreal, Canada) and heparin (100 units/kg) were dosed empirically to achieve a target Ca ++ concentration of >1.4 mmol/L, and an activated clotting time (ACT) of 250-300 seconds, respectively. In early experiments (n = 7/13 successful experiments), 6 mg/kg/h of papaverine was administered I.V. to prevent vasospasm of the umbilical vessels and augment circuit ows. In a subset of experiments (n = 3/13), this was subsequently changed to a maintenance infusion of 30 µg/kg/h milrinone lactate (Aurobindo Pharma, Hyderabad, India). Reconstituted hydrocortisone (8 mg/kg/day, P zer Canada Inc., Kirkland, Canada) and a broad-spectrum empirical antibiotic (piperacillin/tazobactam; 300 mg/kg/day, Sandoz Inc., Boucherville, Canada) was administered I.V., every 6 hours, and 8 hours, respectively. Whole maternal blood was transfused (10 ml/kg) to replete circulating blood volume following circuit phlebotomy. In addition, rocuronium bromide (1 mg/kg) and ketamine hydrochloride (5 mg/kg) were given for excessive fetal movements or perceived fetal agitation. Albumin (25%; CSL Behring Canada Inc., Ottawa, Canada) and furosemide (0.5 mg/kg; P zer Canada Inc., Kirkland, Canada) were given intermittently during longer experiments to address perivascular edema and hydrops fetalis.
Fetal CA, and umbilical arterial and venous blood gases including PO 2 , PCO 2 , pH, Hb, HCO 3 -, SO 2 , BE, Na + , K + , lactate, glucose and ACT were sampled every 1-3 hours and analyzed using a handheld blood analyzer (Abbott Point of Care Inc., Nepean, Canada). Fetal oxygen delivery (DO 2 ) and consumption (VO 2 ) were calculated based on the combination of UV and UA oxygen carrying capacity and indexed UV ow indexed to fetal weight (measured at surgery).
UV ow was continuously measured using a HXL 3/16" tubing ow probe (Transonic) and CVP, fetal BP, pre-pump circuit pressure, pre-oxygenator circuit pressure, and post-oxygenator circuit pressure were measured using Deltran uid lled blood pressure transducers (ADInstruments Inc., Colorado Springs, USA). Data were sampled at 1000 Hz, digitized, and continuously recorded using LabChart Pro 8 (ADInstruments Inc., Colorado Springs, USA). At the end of each study, the data was extracted in consecutive 30-second intervals and analyzed in Excel (Microsoft Corporation, Washington, USA).
Quanti cation of UV blood ow in utero using MRI: 3-D Volumetry and cine phase-contrast.
In utero fetal weight and UV ow were measured in sixteen fetuses from 4 pregnant Yucatan pigs (n = 4; gestational age 107 ± 3 days GA) as previously described using a 3 Tesla magnetic resonance imaging (MRI) system 27 .
Statistical analysis.
Comparisons of anthropometric data between pumpless and pumped AP circuits were analyzed using a two-way ANOVA with Bonferroni correction for multiple comparisons. Comparison of the duration of AP support between pumpless and pumped circuits was analyzed using a Mann-Whitney U test. Changes in temperature, HR, and UV ow between pumpless and pumped AP circuits over the rst 3 hours of support were analyzed using a mixed-effect model, with a Bonferroni correction for multiple comparisons. Differences in mean indexed and absolute UV ow and HR between in utero, pumpless and pumped AP circuits were compared using a repeated measures one-way ANOVA with a Bonferroni correction for multiple comparisons. Analysis of fetal HR and indexed UV ow were performed using a linear regression. Comparisons of mean blood gases, electrolytes, lactate, and glucose concentrations between animals studied in utero, or on pumpless and pumped AP circuits were analyzed using a repeated measure one-way ANOVA, and a Kruskal-Wallis test (when appropriate) with Bonferroni correction for multiple comparisons. UV ow, HR, fetal BP, CVP, and temperature are presented in 5-minute averages. Time post-cannulation versus UV ow is presented in 1 hour averages and analyzed using a one-way ANOVA with Bonferroni correction for multiple comparisons. Differences in MAP between fetal pigs studied in utero and those maintained using a pumped AP were analyzed using a mixed-effect model with Bonferroni correction for multiple comparisons. Indexed UV ow versus oxygen extraction was analysed using a linear regression. *P < 0.05 was considered statistically signi cant.
All statistical analyses were performed using Prism 9 (GraphPad, San Diego, USA). Data are presented as mean ± standard deviation (SD), unless otherwise indicated.

Results
The addition of a centrifugal pump improves successful transition to AP and support times.
Nineteen sows and 89 fetal pigs were used for pumped AP experiments (Table 1). A total of 32 fetal pigs were cannulated and transitioned to the AP. Types of perioperative complications preventing us from maintaining AP support were similar to those experienced in the pumpless AP circuit group and included the inability to establish circuit ows upon connection to the circuit (n = 11), accidental decannulation (n = 4), persistent vasospasm of the umbilical cord (n = 3), and excessive spiralling of the umbilical cord, which prevented cannulation or adequate cannula position (n = 1; Table 1). We successfully maintained 13 fetal pigs at an initial GA of 102 ± 4 days (range = 93-107 days) with a body weight of 616 ± 139 g (range = 390-820g) on a pumped AP circuit for 46.4 ± 46.8 hours (range = 3.4-177.8 hours). This represented a signi cant improvement in the duration of support over the 12 fetal pigs (98 ± 4 days GA; 743 ± 350 g; 11 ± 13 hours) maintained using a pumpless AP circuit (P = 0.009; Table 1 and Figure 2). There was also a marked improvement in the rate of successful AP runs per litter using a pumped AP circuit compared to the pumpless AP circuit (62% pumped AP vs. 28% pumpless AP; Table 1). Successfully supported fetuses were similar in weight and GA in both pumped and pumpless groups (616 ± 139g vs. 743 ± 350g, P = 0.41; 102 ± 4 days vs. 98 ± 4 days, P = 0.24). On the pumped and pumpless AP circuits, fetal body weights were similar between successfully and unsuccessfully supported animals (616 ± 139g vs. 523 ± 145g, P = 0.47; 743 ± 350g vs. 644 ± 184g, P = 0.53). However, unsuccessfully supported fetuses or fetuses that we did not attempt to cannulate were signi cantly smaller in the pumped group compared to the pumpless group (523 ± 145g vs. 644 ± 184g, P = 0.04; 557 ± 124g vs. 892 ± 362g, P < 0.0001). Fetal weight at the time of termination of the AP study was signi cantly higher compared to pre-cannulation weight. Fetal weight gain correlated strongly with length of AP support (r = 0.89), with weight at the end of the experiment exceeding the expected growth trajectory of these fetuses.
Individual data on supported fetuses, support times, and reasons for termination are shown in Table 2.  Pumped AP circuits achieve higher umbilical venous ow.
Animals on a pumped AP circuit maintained a physiologic core body temperature throughout the experiment. The temperature was lower in fetal pigs supported on a pumped system than pigs supported on a pumpless system (P < 0.0001, Figure 3A). UV ow rates of fetuses supported by the pumped AP circuit were comparable to in utero controls, while UV ow on the pumpless AP system was subphysiologic (in utero; 108 ± 24 ml/min vs. pumped; 87 ± 28 ml/min vs. pumpless; 70 ± 18 ml/min, P = 0.001; in utero; 173 ± 45 ml/min/kg vs. pumped; 143 ± 40 ml/min/kg vs. pumpless; 97 ± 39 ml/min/kg, P = 0.005; Figure 3C, E and F). Absolute UV ow measured within the rst 3 hours post-cannulation was positively correlated to HR in pumped AP circuits (r 2 = 0.44; P < 0.0001; Figure 3G), but negatively correlated in fetuses maintained on a pumpless circuit (r 2 = 0.45; P < 0.0001; Figure 3G), with differences between the slopes being extremely signi cant (P < 0.0001). Figures 5A-M demonstrate the changes in UV ow, HR, temperature, MAP, and CVP over time for the 13 fetal pigs successfully maintained using a pumped AP circuit. Upon initiation of AP support, we observed supraphysiologic UV ow, despite a low pump rate setting ( Figure 5A-N). In longer trial runs, this period of supraphysiologic UV ow persisted for approximately 6 hours, then decreasing to subphysiologic ow, plateauing at ~12 hours of support ( Figure 5O). In several experiments (n = 8/13) excessive spasmodic fetal movements resulted in a complete cessation of UV ow and consequently led to the termination of the experiment ( Table 2). In four studies, development of a thrombosis in one of the UAs in the rst 12 hours of support was associated with a signi cant reduction in UV ow.
Fetal pigs were tachycardic for a large proportion of the AP studies ( Figure 3B and D; and Figures 5A-M). Both pumped (205 ± 28 bpm; P < 0.0001; Figure 3D) and pumpless (206 ± 38 bpm; P < 0.0001; Figure 3D) fetal pig groups were similarly tachycardic throughout the experiment compared to in utero controls (130 ± 10 bpm; Figure 3D). Compared to in utero control animals, fetal pigs were hypertensive on the pumped AP system (P = 0.013; Figure 6). Approximately 20 minutes after initiation of a milrinone infusion, there was a marked increase in UV ow, followed by a steady decline over a 3 hour period ( Figure 7A). MAP and HR decreased following the start of the milrinone infusion, which persisted for ~1 hour before returning to a steady state ( Figure 7B and C.).
Upon initiation of AP support, fetal oxygen delivery was physiologic 29,30 . We observed a decline in oxygen delivery and increase in oxygen consumption over time ( Figure 8A). UV ow and oxygen extraction fraction correlated negatively (r 2 = 0.45; P = 0.0003; Figure 8B).

Discussion
To our knowledge, this is the rst study to examine the hemodynamics and blood gas status of preterm pigs supported using an AP system incorporating a small centrifugal pump. We previously showed the feasibility of supporting our animal model using a pumpless AP circuit connected to the fetal circulation via the umbilical arteries and vein 27 ; however, our experiments were characterized by signi cant hemodynamic decompensation within hours. We attributed this to afterload imbalance, elicited by supraphysiologic circuit resistance and further exacerbated by impaired umbilical venous return, which we surmised resulted in higher sympathetic tone, further exacerbating the increase in afterload 27 .
Although other research groups have highlighted potential limitations of pumped ECMO circuits, including an increased risk of pump-induced hemolysis 17 , afterload imbalance, myocardial strain, and impaired autoregulation of UV blood ow 31 , we hypothesized that the addition of a small pump might reduce right ventricular afterload and sustain adequate UV ow, thereby improving survival on the AP. The present study examined differences in hemodynamics, blood gases, electrolytes, biochemistry, and survival of fetal pigs supported on a pumped versus pumpless AP system 27 . The increase in AP support to 46.4 ± 46.8 hours represented a signi cant increase in survival over fetal pigs supported without a pump possibly due to more reliable and physiologic UV ow compared to a pumpless system. Despite the pump support, following a period of supraphysiologic UV ow, the animals continued to experience a slower but consistent decline in UV ow and the development of signs of circulatory deterioration. We have yet to establish an AP system capable of meeting the long-term physiologic needs of animals that are similar in size to the extremely preterm human infants that are the proposed target of this approach.
In a recent report, Partridge et al. 17 demonstrated AP support of fetal sheep weighing 1-2 kg using a pumpless system for up to 28 days without the use of vasopressors. Using a similar AP circuit, Usada et al. 18 reported survival of fetal sheep for up to one week. To investigate the potential utility of AP systems for extremely preterm infants born at 22-25 weeks GA, two research groups cannulated sheep born at 85-95 days GA weighing ~0.5-0.8 kg. Although cannulation was technically feasible, one study reported the development of hydrops fetalis following 5-8 days of support 32 , whereas the other required the use of aggressive pharmacological interventions in order to achieve hemodynamic stability 19 . Although fetal sheep delivered at 85-95 days GA share similar body weight to human fetuses delivered at the biological limit of viability, they are developmentally immature and their pulmonary development is analogous to a human fetus delivered at 18 weeks GA 10,26,33,34 . By contrast, fetal minipigs delivered at 93-107 days GA weighing 0.39-0.82 kg are in their canalicular and saccular stages of lung development and comparable to preterm human fetuses delivered at 22-28 weeks GA. Thus, fetal pigs delivered at 95 days GA would therefore be equivalent in body weight and lung maturity to human fetuses born at 22-25 weeks GA, and may represent a more appropriate model for the development of an AP system 27,33,34 .
Upon initiation of AP support, we observed supraphysiologic UV ow in all pumped AP experiments. Umbilical vein blood ow was higher in fetuses supported with a pumped circuit than in fetuses supported using a pumpless system. Our experiments revealed a positive correlation between fetal HR and UV ow on the pumped AP, with persistent tachycardia seen with both systems. Contrary to our speculations regarding supraphysiologic resistance in the pumpless circuit 27 , we propose that the negative pressure generated by the centrifugal pump may have diminished right ventricular afterload and minimized resistance to ow across the UA cannulas, resulting in higher than normal ow rates. The association we observed between fetal HR and UV ow in animals supported on the pumped circuit may be attributable to the Frank-Starling mechanism and Anrep effect 35 , whereby increased ventricular preload enhances myocardial stretch and tension, resulting in increased contractility. Although this relationship holds true for slight increases in end-diastolic lling pressures, the fetal heart has little preload reserve 36-39 .
Thus, further augmentation of cardiac output and contractility are also driven by the Bowditch effect 40 through increases in fetal HR 41 . In contrast to the physiology seen in a pumpless circuit, we observed a shift of placental-to-systemic blood ow in favor of the AP, effectively volume loading the fetal heart. Therefore, we speculate that these supraphysiologic circuit ows lead to increased preload, tachycardia, sympathetic nervous system activation, and likely higher cardiac output within the rst hours of AP support.
After ~24 hours of AP support, we observed marked reductions in UV ow compared to the high ows seen at the start of AP support and in many experiments, UV ows then remained subphysiologic. The drop in UV blood ow and persistent tachycardia could be explained by increased sympathetic nervous activation and continuous peripheral vasoconstriction. Although the introduction of milrinone improved UV ow and reduced BP, its inotropic and systemic vasodilatory effects appeared to be short-lived, indicating the presence of overwhelming perturbation of fetal cardiac loading conditions. However, our ndings suggest that phosphodiesterase inhibitors may have a positive impact on the circulation of fetal animals supported on the AP 19,42 . Despite similar SO 2 , blood gases showed higher PO 2 , lower PCO 2 , as well as a more alkalotic pH in fetuses on the pumped AP compared to in utero controls. We were able to manage O 2 uptake and CO 2 elimination better on the pumped circuit than pumpless.
However, the gas exchange capacity of the oxygenator used in both groups easily outperforms the native placenta, resulting in supraphysiologic UV PO 2 . High oxygen tension is a known contributor to umbilical cord and ductal constriction and may have negatively affected our experiments 43,44 . Gradual umbilical vessel spasm or constriction at the umbilicus, where the cord passes through the abdominal wall, could explain the gradual decline in UV ow we observed. Mean blood pressure in the UV is 6-8 mmHg in near-term sheep and human fetuses 29,30 . In our pumped AP studies, we observed pre-and post-oxygenator pressures of 25 mmHg and 18 mmHg, respectively (Supplementary Figure 1.). Although we did not measure blood pressure in the UV, it is likely that the pumped circuit exposed the umbilical vein to elevated blood pressure, even if there was some degree of pressure difference across the venous cannula. Berman et al. 45 advanced balloon catheters into the common UV of fetal sheep and demonstrated marked reductions in placental blood ow with increasing UV pressures. This was attributed to a decreased pressure gradient across the umbilical circulation, resulting in diminished umbilical ows. We hypothesize that a similar phenomenon may be occurring in the early phase of our AP experiments, with UV constriction induced by the high pressures and shear stress resulting from the elevated post-pump pressures.
We observed a steady widening of the veno-arterial saturation gradient across the umbilical circulation during our AP runs that was negatively correlated with UV ow, indicating higher oxygen extraction at the tissues in the setting of a net reduction in UV ow and oxygen delivery. One factor that might additionally contribute to the progressive reduction in oxygen delivery was the low hemoglobin concentration we observed in animals supported on the pumped system. The large priming volume of the circuit and oxygenator contributes to fetal anemia because the circuit is primed with maternal blood. During pregnancy, maternal blood volume increases by ~40% and there is a progressive decline in hemoglobin concentrations 46-48 . Conversely, fetal hemoglobin increases with GA to sustain fetal tissue oxygenation in the setting of declining PO 2 29,49,50 .
Thus, the mixing of maternal and fetal blood in the AP circuit contributes to hemodilution of fetal blood, while ongoing blood sampling further depletes fetal red blood cells. To circumvent this problem in future experiments, it may be possible to hemoconcentrate the circuit prime.
Furthermore, pump-induced hemolysis may have also contributed to fetal anemia in these experiments. Other groups have recommended the administration of daily doses of erythropoietin to minimize the need for successive maternal donor blood transfusions 17 . Miniaturization of the oxygenator and pump head could also help to reduce fetal anemia by reducing circuit volume.

Study limitations.
There are several limitations of the present study. Firstly, we have not yet characterized the etiology of the variation in UV ows on the pumped AP circuit. We speculate that a combination of UV constriction and supraphysiologic UV pressures, as well as supraphysiologic oxygen tension may be contributing to progressively diminished UV ows. However, this hypothesis should be tested with direct measurements of UV pressure.
Similarly, future experiments would be strengthened by the simultaneous measurement of umbilical artery pressure. We speculate that adrenergic drive may be contributing to tachycardia and peripheral vasoconstriction on the AP that would be enhanced by the measurement of circulating catecholamines. Secondly, although the addition of a centrifugal pump to the AP improved fetal pig hemodynamics, we periodically encountered air-entrainment into the circuit, as well as cavitation of fetal blood at excessively high negative pressures, resulting in diminished circuit ows. Although we did not detect pump-induced hemolysis, this could be a limiting factor during long-term AP experiments and could explain the fetal anemia on the circuit. Finally, systemic in ammation may have also contributed to the hemodynamic instability and cardiovascular decompensation we observed in pumped AP fetal pigs as centrifugal pumps have been reported to induce low-level systemic in ammation 51 . The analysis of in ammatory cytokines may be helpful to delineate the role of in ammation in fetal pig circulatory physiology on the AP.
Future directions.
Recent developments in AP technology have enabled sustained extrauterine fetal life on a pumpless arteriovenous ECMO circuit. Of note, these experiments were conducted using customized hollow-ber membrane oxygenators that are unavailable for commercial use. Our initial approach in developing a pumpless AP system involved cannulas with the largest luminal diameter to wall thickness ratio possible to minimize circuit resistance. Our failure to demonstrate hemodynamic stability using a pumpless circuit encouraged our team to explore the use of mechanical support in the extremely preterm fetal pig. While the addition of a pump to the circuit improved survival, we observed supraphysiologic circuit ows, evidence of adrenergic drive, and high cardiac output at the start of support, followed by diminished UV ows with persistent tachycardia and hypertension in keeping with diminished cardiac output. This pattern may indicate a progressive increase in circuit resistance resulting from constriction in the cord vessels. An alternative approach to normalizing UV pressures and minimizing the pressure difference across the UV cannula would be to introduce a pressure drop in the circuit through restriction of the circuit or cannula lumen. According to Hagan-Poiseuille's law, reducing the calibre of the venous side of the circuit would increase circuit resistance and limit the supraphysiologic circuit blood ow seen at the onset of our AP experiments.
Miniaturization of the oxygenator and pump head may also help to achieve hemodynamic stability in our animal model. In sheep, placental blood volume at ~116 days GA (term = 145 days) is ~60-80 ml/kg, which represents ~30-40 ml for a 0.5 kg fetus 52,53 . The AP circuit used in the present study has a priming volume that is nearly double the expected placental blood volume for a fetal sheep delivered at 116 days gestation, which likely accounts for the increased afterload we seemed to induce in our pumpless AP experiments, as well as contributing to fetal anemia.
Reductions in the oxygenator membrane surface area could also reduce the need for excessive heparin administration. Thus, modi cations to the circuit through miniaturization of the oxygenator and centrifugal pump may be helpful for establishing an AP system capable of supporting the long-term physiological requirements of extremely preterm fetal pigs.

Conclusion
Over the past decade, important advances have been achieved in the eld of arti cial womb/AP technology. In experiments using a miniature pig model of the AP in which commercial neonatal ECMO oxygenators were connected the fetal circulation via the umbilical cord we demonstrated a marked improvement in survival, UV ow, and oxygen delivery in circuits that incorporated a small centrifugal pump compared with animals maintained on a pumpless circuit. However, despite the addition of the pump, we observed a progressive diminution of UV ow with persistent tachycardia and hypertension, which we attribute to preload imbalance, increased sympathetic tone, and UV hypertension. Thus, despite observing a clear short-term bene t with the addition of a centrifugal pump in supporting the fetal pig hemodynamics, we conclude that further modi cations to the AP circuit are needed. A reduction to the size of the UV cannula could represent an alternative approach for mediating physiologic circuit ows and venous pressures. Given the limitations of neonatal intensive care therapies in preventing iatrogenic organ injury and neonatal death, we remain hopeful that the AP could provide reductions in the mortality and morbidity associated with preterm birth by maintaining a fetal circulation while allowing for normal growth and development of fetal organ systems.

Declarations
Competing interests The authors have no competing interests to disclose. Blood gases, electrolytes, lactate, and glucose concentrations in fetal pigs studied in utero versus fetal pigs supported using a pumped and pumpless AP circuit 27 . In utero fetal pig data represented by open triangles (n = 24), fetal pigs supported on a pumped AP circuit as black circles (n = 13), and fetuses supported on a pumpless AP circuit as open circles (n = 9). If two P-values are presented, the rst P-values represents the pumped AP circuit, and the second P-value represents the pumpless AP circuit.  MAP in fetal pigs studied in utero and maintained using a pumped AP circuit. Fetal pigs studied in utero are represented in open triangles (n = 21) and fetal pigs maintained using a pumped AP as black circles (n = 6).

Figure 7
Effects of milrinone lactate infusion on fetal pig hemodynamics while maintained using a pumped AP circuit. Maintenance infusion of milrinone lactate was used in successful AP experiments (n = 5) for hemodynamic support. 'Milrinone infusion start' indicates the time point at which milrinone lactate IV infusion began. Data are presented as 5-minute epochs and expressed as mean ± SEM.