Insulin Increases Central Aortic Stiffness in Response To Hyperglycemia in Healthy Humans: A Randomized Four-Way Crossover Study

Introduction: Increasing arterial stiffness is a feature of vascular aging that is accelerated by conditions that enhance cardiovascular risk, including diabetes mellitus. Multiple studies demonstrate divergence of carotid-femoral pulse wave velocity and augmentation index in persons with diabetes mellitus, though mechanisms responsible for this are unclear. Materials and Methods: We tested the effect of acutely and independently increasing plasma glucose, plasma insulin, or both on hemodynamic function and markers of arterial stiffness (including carotid-femoral pulse wave velocity, augmentation index, forward and backward wave reection amplitude, and wave reection magnitude) in a four-arm, randomized study of healthy young adults. Results: Carotid-femoral pulse wave velocity increased only during hyperglycemic-hyperinsulinemia (+0.36 m/s; p=0.032), while other markers of arterial stiffness did not change (all p>0.05). Heart rate (+3.62 bpm; p=0.009), mean arterial pressure (+4.14 mmHg; p=0.033), central diastolic blood pressure (+4.16 mmHg; p=0.038), and peripheral diastolic blood pressure (+4.09 mmHg; p=0.044) also signicantly increased during hyperglycemic-hyperinsulinemia. Conclusions: We conclude that the acute combination of moderate hyperglycemia and hyperinsulinemia preferentially stiffens central elastic arteries. This effect may be due to increased sympathetic activity. (ClinicalTrials.gov . Data from numerous studies demonstrate that increased arterial stiffness (assessed by carotid-femoral pulse wave velocity (cfPWV) 5-11 , augmentation index (AIx) 12, 13 , reection magnitude (RM) 14, 15 , or backward wave amplitude (Pb) 16 ) predicts cardiovascular events and/or mortality and may explain some of the notable residual cardiovascular risk associated with even well-controlled hypertension 17


Introduction
Arterial stiffness develops from dynamic interactions between structural and cellular elements of the vessel wall in uenced by both hemodynamic forces and extrinsic factors (including hormones, salt, and glucose) 1 . Arterial stiffness increases with vascular aging due to gradual loss of arterial elasticity 2 and is accelerated by conditions that increase cardiovascular risk, including diabetes mellitus (DM) 3 . Notably, arterial stiffness often antedates and is itself a strong risk factor for many cardiovascular diseases (e.g., heart failure, myocardial infarction, etc.) 3,4 . Data from numerous studies demonstrate that increased arterial stiffness (assessed by carotid-femoral pulse wave velocity (cfPWV) [5][6][7][8][9][10][11] , augmentation index (AIx) 12,13 , re ection magnitude (RM) 14,15 , or backward wave amplitude (Pb) 16 ) predicts cardiovascular events and/or mortality and may explain some of the notable residual cardiovascular risk associated with even well-controlled hypertension 17 .
Arterial stiffness increases with deteriorating glucose tolerance status 18 and is associated with insulin resistance 19 ; however, the independent and combined roles of acute hyperglycemia and/or hyperinsulinemia on arterial stiffness are not well-de ned 20 . Moreover, prior studies have identi ed divergence of arterial stiffness markers (speci cally increased cfPWV and reduced AIx) in persons with DM [21][22][23] , though mechanisms responsible for this are also unclear. A recent editorial by Currie and Delles encouraged investigation of healthy cohorts to better-understand mechanisms contributing to alterations in arterial stiffness 24 . As arterial stiffness is increased in even adolescents with DM 25 , healthy young adults without vascular dysfunction comprise the ideal population in which to investigate metabolic mechanisms linked to acute changes in arterial stiffness. With these points in mind, we sought to quantify the independent effects of elevated concentrations of insulin, glucose, and both on arterial stiffness in healthy humans. Green eld et al. previously showed that AIx decreases in response to the magnitude of postprandial glucose and insulin responses after low-and high-carbohydrate meal ingestion 26 . To isolate the effects of insulin and glucose from those of incretins and the autonomic changes that occur with oral glucose or meal ingestion, we used intravenous glucose and insulin infusions with co-administration of octreotide (OCT) to block endogenous insulin secretion. We then measured cfPWV, AIx, RM, forward wave amplitude (Pf), Pb, and systemic hemodynamic function during euglycemia, hyperglycemia, euglycemic-hyperinsulinemia, and hyperglycemic-hyperinsulinemia.

Recruitment and Study Population
Recruitment for this study was achieved by public advertisement. Healthy young adults met inclusion criteria if they were ≥18 and ≤35 years old, had normal body mass index (18-25 kg/m 2 ), and had a screening fasting plasma glucose <100 mg/dL and a screening blood pressure <140/90 mmHg. Subjects were excluded if they were current smokers or quit smoking <5 years ago, had a rst-degree relative with type 2 DM, were taking vasoactive medications (e.g., anti-hypertensives, diuretics, statins, etc.), were pregnant (i.e., positive pregnancy test) or nursing, had history of allergy or prior adverse reaction to octreotide, or signi cant morbidity that could, in the investigator's opinion, affect outcome measures or subject safety.

Clinical Assessment and Initial Screening
All screening visits and infusion studies were conducted at the University of Virginia (UVA) Clinical Research Unit (CRU). Each subject gave written informed consent at their initial visit prior to being carefully screened to verify inclusion/exclusion criteria. Screening included a detailed medical history and physical examination along with fasting measures of complete blood count, comprehensive metabolic panel, lipid panel, plasma glucose, and serum pregnancy test.

Study Design
We followed the Consolidated Standards of Reporting Trials (CONSORT) guidelines 27 to analyze and report this randomized trial. Randomization of study sequence was conducted by study personnel using a 1:1:1:1 starting allocation from a computer-generated sequence program 28 . The protocol was designed to assess: (1) Protocol A followed by Protocol B (or vice-versa), with subsequent crossover to Protocols C and D (in randomized order); or (2) Protocol C followed by Protocol D (or vice-versa), with subsequent crossover to Protocols A and B (in randomized order). Subjects underwent four infusion protocols ( Figure  1) to test the discrete effects of euglycemia, hyperglycemia, euglycemic-hyperinsulinemia, and hyperglycemic-hyperinsulinemia on arterial stiffness. All protocols were approved by the UVA Institutional Review Board (#19948), with each protocol being performed ≥2 but ≤4 weeks apart for individual subjects to allow a washout period between studies. Within each protocol, we measured cfPWV, AIx, Pf, Pb, RM, systolic blood pressure, diastolic blood pressure, pulse pressure, mean arterial pressure, and heart rate immediately before (i.e., baseline) and at the end of the infusion period ( Figure 1). All vascular assessments for this study were measured per expert recommendations by the same trained operator 29 . After randomization, study personnel were blinded to subject and protocol when evaluating outcome measures. Study participants were instructed to avoid alcohol, exercise, and caffeine for 24 hours and to fast overnight prior to admission to the CRU. We placed intravenous catheters in the right wrist for blood sampling and in the right antecubital fossa for administration of insulin, glucose, and octreotide (OCT).
Studies then began with simultaneous infusion of regular insulin and OCT to maintain plasma insulin near basal levels. We did not replace glucagon or growth hormone, as there is currently no evidence that acutely suppressing basal levels of either hormone affects vascular function.
Protocol A (Euglycemia): A 90-minute saline infusion was initiated, with baseline vascular function measurements obtained during the nal 30 minutes ( Figure 1A). Then, OCT (30 ng/kg/min) with basal insulin replacement (0.15 mU/min/kg) was infused for 240 minutes. Blood glucose was sampled every 10 minutes and plasma insulin every 30 minutes. Euglycemia (EU) was maintained by a variable-rate glucose infusion using the negative feedback principle 30 . We repeated vascular measurements over the nal 30 minutes of OCT infusion.
Protocol B (Isolated Hyperglycemia): Octreotide with basal insulin replacement was continuously infused for 90 minutes while euglycemia was maintained. Baseline vascular measurements were obtained over the nal 30 minutes ( Figure 1B). Then, a primed, variable-rate 20% dextrose infusion began to acutely raise and subsequently maintain blood glucose at ~200 mg/dL using the hyperglycemic clamp method 30 . Blood glucose was sampled every 5 minutes and plasma insulin every 30 minutes, with repeat vascular measurements obtained over the nal 30 minutes of hyperglycemia.
Protocol C (Euglycemic-Hyperinsulinemia): Euglycemia was maintained throughout this protocol by a variable-rate 20% dextrose infusion using the negative feedback principle 30 . Baseline arterial stiffness measurements were obtained during the nal 30 minutes of an OCT (30 ng/kg/min) plus basal insulin (0.15 mU/min/kg) infusion ( Figure 1C). Then, hyperinsulinemia was initiated with a primed (2 mU/kg/min x 10 min), continuous (1 mU/kg/min x 110 min) infusion and OCT continued for 120 minutes. Blood glucose was sampled every 5 minutes and plasma insulin every 30 minutes, with repeat arterial stiffness and hemodynamic measurements obtained during the nal 30 minutes of the insulin clamp.
Protocol D (Hyperglycemic-Hyperinsulinemia): As in Protocol C, a variable-rate 20% dextrose infusion maintained euglycemia while OCT (30 ng/kg/min) and basal insulin (0.15 mU/min/kg) were simultaneously infused for the rst 90 minutes of this study ( Figure 1D). Then, a primed, variable-rate 20% dextrose infusion began to acutely raise and subsequently maintain blood glucose at ~200 mg/dL. Blood glucose was then sampled every 5 minutes and plasma insulin every 30 minutes, with baseline arterial stiffness measurements obtained over the nal 30 minutes of the 120-minute hyperglycemic period ( Figure 1B). Subsequently, hyperinsulinemia was initiated with a primed (2 mU/kg/min x 10 min), continuous (1 mU/kg/min x 110 min) infusion with OCT and hyperglycemia maintained for 120 minutes. Blood glucose was sampled every 5 minutes with plasma insulin every 30 minutes, and repeat arterial stiffness and hemodynamic measurements were again obtained during the nal 30 minutes of the insulin clamp.
Hemodynamic Function: Clinical hemodynamic assessments were obtained at two time points during each protocol ( Figure 1). Blood pressure, pulse pressure, mean arterial pressure, and heart rate were measured and/or calculated with a SphygmoCor® tonometer (ATCOR USA; Napierville, IL).
Carotid-Femoral Pulse Wave Velocity (cfPWV): To assess central aortic stiffness, cfPWV was measured with a SphygmoCor tonometer. To minimize the effects of sympathetic activity on cfPWV measurements, participants laid supine in a temperature-controlled room for at least 15 minutes prior to measurement.
We measured the distance from the suprasternal notch to the carotid pulse and from the suprasternal notch to the femoral pulse on the same side. For each cfPWV measure, 10 seconds of carotid and 10 seconds of femoral arterial waveforms were recorded. cfPWV measures were made in duplicate and the mean value was reported. cfPWV intraobserver reliability was also assessed by having the operator record three serial cfPWV measurements on the same subject over a 4-hour period. The coe cient of variation for cfPWV was 3.63%, indicating good intraobserver reliability 31 . Of note, the cfPWV data in this manuscript were included in a separate report examining macro-and microvascular functional responses to the two insulin clamp protocols 32 .
Augmentation Index (AIx): To assess muscular conduit arterial stiffness, we measured AIx noninvasively with a SphygmoCor tonometer. AIx measurements were obtained at the radial artery by the same trained operator with participants lying in the supine position in a temperature-controlled room for at least 15 minutes prior to measurement. AIx was calculated as the difference of the amplitude of the late systolic peak to the early systolic peak divided by the pulse pressure and expressed as a percentage. AIx values were determined for each pulse over a 30 second period, and a mean value was calculated by the device for each patient and corrected for a heart rate of 75 beats per minute.
Wave Separation Analyses (Pf, Pb, and RM): We derived the central aortic pressure waveform by recording radial artery pressure waves through aplanation tonometry using a SphygmoCor tonometer.
Brie y, after 20 sequential waveforms were acquired and an ensemble was averaged, a validated general transfer function was used to synthesize the central aortic pressure wave noninvasively. Wave separation analysis then decomposed pressure and ow waveforms into their forward (Pf) and backward (Pb) components 29 . Re ection magnitude (RM) was calculated as the ratio of the amplitudes of the backward/forward pressure waves 33 34 project le repository hosted at UVA. The datasets generated and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Statistical Analyses
Sample Size: A sample size calculation using the Cohen's d effect size from a previous study of cfPWV changes during euglycemic-hyperinsulinemia 35 in metabolic syndrome subjects indicated that ≥8 participants in each protocol would have ≥80% power to detect meaningful cfPWV changes in the current study.
Outcomes: The primary outcome for each protocol was change in cfPWV. Secondary outcomes for each protocol included changes in AI, Pf, Pb, RM, systolic blood pressure, diastolic blood pressure, mean arterial pressure, pulse pressure, and plasma insulin.
Descriptive Summarization: Patient demographics were summarized using common descriptive statistics. The arithmetic mean and standard error of mean, standard deviation, median, and interquartile range were used to summarize continuous scaled outcome measures.
Data Transformation and Summarization: Given that the data distributions for some outcome measures were initially skewed, we rescaled the pre-and post-intervention outcome measures for each protocol to the natural logarithmic scale (i.e., log e ). The analytical outcome data were then derived by subtracting the log e transformed pre-intervention outcome measures from the log e transformed post-intervention outcome variable measures. For all outcome measures, the point estimate for the mean pre-intervention outcome measures, the point estimate for the mean pre-to post-intervention outcome measure change, and the point estimate for the difference between the mean of the pre-intervention outcome measures and the mean pre-to post-intervention outcome measure change for each protocol were converted via natural logarithmic antilog transformation (i.e., e x ) to a geometric mean ratio scale.
Outcome Measure Analyses: The pre-to post-intervention outcome variable change for each protocol was compared by linear mixed model (LMM) analysis of covariance. Signi cance was set at α= 0.05 (twotailed test). All statistical analyses were performed with SAS Studio 3.8 (SAS; Cary, NC). LMM Speci cation: The analytical outcome data of each protocol served as the dependent variable measurements of the LMM. An indicator variable to identify the protocol (i.e., A, B, C, or D) served as one of the LMM independent variables, and the log e transformed pre-intervention outcome measurements of each protocol served as a second LMM independent variable. Note that the log e transformed preintervention measurements were included as part of the LMM so that the between-admission comparison of the mean pre-to post-intervention change in the outcome measure could be standardized to a common reference pre-intervention measurement value. Hypothesis testing: Primary hypotheses under the null tested whether the mean within-protocol change in each outcome measure was equal to zero.

Results
Baseline Subject Characteristics and Demographics Table 1 details baseline demographics of the 19 total study participants. All had normal BMI, blood pressure, and fasting plasma glucose. Notably, 13 subjects completed Protocol A, 10 subjects Protocol B, 14 subjects Protocol C, and 12 subjects Protocol D. Nine subjects completed all four protocols. Three subjects voluntarily withdrew after completing only Protocols C and D, two after completing only Protocol C, and two after completing only Protocol A. One subject moved out of the area after completing Protocols A and B, one experienced scheduling di culty after completing only Protocols A and D, and one completed Protocol D but then experienced abdominal cramping (due to octreotide) that led to study termination during their second admission. An electrode wire malfunction resulted in no post-intervention assessment of cfPWV for one subject in Protocol B. Figure 2 shows the time course for mean plasma glucose (upper panel), mean glucose infusion rate (middle panel), and mean plasma insulin levels (lower panel) throughout each protocol. Plasma glucose levels rose signi cantly from baseline within Protocols B and D, and plasma insulin concentrations rose signi cantly from baseline within Protocols C and D. These increases did not differ between respective protocol pairs. Arterial Stiffness Table 2 shows the summary statistics for pre-and post-intervention measures of arterial stiffness within each protocol. cfPWV did not change during euglycemia, hyperglycemia, or euglycemic-hyperinsulinemia (all p>0.05), but signi cantly increased after hyperinsulinemia was added to hyperglycemia (+0.36 m/s; p=0.032). AIx signi cantly decreased during euglycemia (Protocol A), but did not signi cantly change within any other protocol. There were no signi cant changes in Pf, Pb, or RM within any protocol. Table 3 details the within-protocol changes for all hemodynamic parameters. Heart rate (+3.62 bpm; p=0.009), mean arterial pressure (+4.14 mmHg; p=0.033), central diastolic blood pressure (+4.16 mmHg; p=0.038), and peripheral diastolic blood pressure (+4.09 mmHg; p=0.044) signi cantly increased during only hyperglycemic-hyperinsulinemia. There were no signi cant changes in aortic or peripheral pulse pressure or systolic blood pressure within any protocol (all p>0.05).

Discussion
To our knowledge, this study is the rst to investigate the discrete effects of acute hyperglycemia and hyperinsulinemia on cfPWV, AIx, RM, Pf, and Pb in the same subjects, with several novel observations warranting discussion. First, the combination of hyperglycemia and hyperinsulinemia increased cfPWV, while isolated hyperglycemia or hyperinsulinemia did not. Second, hyperglycemic-hyperinsulinemia acutely increased heart rate, mean arterial pressure, and diastolic blood pressure. Finally, no signi cant changes were seen in AIx, RM, Pf, and Pb during hyperglycemic-hyperinsulinemia.
Prior work from Puzantian et al. found that acute hyperglycemia (using pancreatic clamping methodology in healthy subjects) did not alter cfPWV, but they noted that further studies were needed to determine the independent and combined roles of glucose and insulin on cfPWV 20 . To our knowledge, the current study is the rst to investigate this question. Infusion of OCT allowed us to isolate the effects of insulin and glucose and provide the rst evidence that moderate hyperglycemia unmasks an action of physiologic hyperinsulinemia to increase central aortic stiffness in healthy humans. Here we must note that heart rate, mean arterial pressure, and diastolic blood pressure also rose signi cantly during hyperglycemic-hyperinsulinemia. A positive association between heart rate and cfPWV has been demonstrated in recent studies 36,37 . Bikia et al. recently performed an in-silico evaluation of the impact of heart rate on cfPWV and con rmed that small heart rate changes only slightly alter cfPWV, but they suggested that a more clinically signi cant impact on cfPWV should be considered in cases wherein heart rate might vary to a greater extent 38 . The mean increase in heart rate and cfPWV during hyperglycemic-hyperinsulinemia in the current study were 3.62 bpm and 0.4 m/s, respectively. Prior reports show that the effect of heart rate on cfPWV is small and on the order of 0.02 m/s per 1 bpm change 36,37 , thus increased heart rate alone is unlikely to explain the increase in cfPWV observed during hyperglycemic-hyperinsulinemia. We also note that acute hyperglycemic-hyperinsulinemia signi cantly increased mean arterial pressure and diastolic blood pressure without altering systolic blood pressure.
These ndings mirror prior reports demonstrating that increases in heart rate [39][40][41] and arterial stiffness 42,43 precede the development of systolic hypertension. Similarly, acute stiffening of the thoracic aorta and its branches (i.e., the arterial distribution represented by cfPWV) corresponds to increases in mean arterial pressure 44 . Increased transmural pressure at the level of the central aorta causes the elastic lamellae of this large artery to stretch and become stiffer, resulting in higher cfPWV 44 . The increase of mean arterial pressure observed in the current study corresponded to an acute increase in cfPWV. Such increases of vascular tone, occurring repetitively, may contribute to vascular remodeling 45 . Further studies are needed to investigate and test this hypothesis.
We also found that AIx, RM, Pf, and Pb did not signi cantly change during hyperglycemichyperinsulinemia. This result aligns with previous studies reporting that increased cfPWV is not associated with elevated AIx in persons with DM 21-23 . Our prior work has also shown that insulin has opposing effects on cfPWV and AIx in various metabolic conditions. Speci cally, insulin (during euglycemia) acutely reduced AIx in both healthy and metabolic syndrome subjects, but increased cfPWV in metabolic syndrome subjects only 35 . In that study, metabolic syndrome subjects were insulin-resistant and had chronically higher fasting plasma glucose and insulin concentrations (i.e., the milieu of metabolic syndrome and type 2 DM). Lacy et al. hypothesized that dissipation of the energy of the incident pressure wave in persons with DM blunting wave re ections might explain this phenomenon 21 . In other words, it is reduced wave re ections, not the increased outgoing pressure wave, that result in the attenuation of AIx in persons with DM 21,22 . In contrast, Khoshdel et al. theorized that AIx underestimates arterial stiffness in persons with DM because of their wider pulse pressure compared to healthy controls 23 . In this hypothesis, the main determinant of AIx is pulse pressure and not the re ected wave 22 . The current study provides some clari cation for this debate. In our healthy subjects with intact vascular function, the acute combination of hyperglycemia with hyperinsulinemia increased heart rate, mean arterial pressure, diastolic blood pressure, and cfPWV without signi cantly altering pulse pressure. We note that cfPWV is a direct measure of arterial stiffness, while AIx and pulse pressure are surrogate indicators of arterial stiffness 46 . Pulse pressure did not signi cantly change within the hyperglycemichyperinsulinemia protocol (or within any protocol) of the current study. These results suggest blunted wave re ections may better explain why AIx likely underestimates arterial stiffness in type 2 DM 22 , and that the acute changes in cfPWV are independent of pulse pressure.
Taken together, our ndings suggest a potential mechanism whereby acute, repetitive metabolic alterations could impact hemodynamic function and aortic stiffness. These results may have implications for the accelerated vascular aging seen in persons with both types 1 and 2 DM who regularly experience the acute combination of hyperglycemia and hyperinsulinemia. We caution, however, that the chronic impact of glucose and insulin per se on arterial stiffness includes a more complex vascular remodeling not seen with the acute metabolic interventions studied here. For example, widened pulse pressure has been associated with DM in many studies [47][48][49] , but we did not see a change in pulse pressure within any protocol of the current study. Acute alterations in arterial stiffness result from changes in endothelial function, vascular tone, heart rate, and other causes, whereas chronic arterial stiffening is mainly due to vascular remodeling, brosis, and calci cation 50 . Whether repeated, acute measurable episodes of arterial stiffening contribute to long-term remodeling, brosis, and vascular injury remains a point of speculation. Recent work has shown that hyperglycemic-hyperinsulinemia signi cantly increases circulating norepinephrine (but not epinephrine) in healthy humans 51 . The increases in cfPWV, heart rate, mean arterial pressure, and diastolic blood pressure observed during hyperglycemic-hyperinsulinemia in the current study mimic the enhanced sympathetic nervous system activity seen in insulin-resistant states like obesity and type 2 DM 52, 53 . Our results, combined with these prior reports, point to sympathetic stimulation as a potential explanation for the hemodynamic changes seen during acute hyperglycemic-hyperinsulinemia. Unfortunately, we did not measure sympathetic nervous system activity in our study. Future work will focus on investigating this question and establishing a mechanistic basis for the hemodynamic effects observed during hyperglycemichyperinsulinemia.
While the current study provides novel insight into acute vascular changes during hyperglycemichyperinsulinemia, several study limitations should be mentioned. First, we studied a small number of healthy young adults and the study was powered to detect within-protocol responses to glucose and insulin. Thus, we did not examine between-protocol differences. Second, persons with DM or those who are older and/or less healthy might respond differently to our study protocols. Third, while AIx is widely used to evaluate arterial stiffness, con icting data exist in regards to its validity 54,55 . A recent scienti c statement from the American Heart Association recommended using wave separation analysis, as opposed to aortic AIx alone, when investigations are focused on the role of wave re ection 29 . These considerations led us to examine Pf, Pb, and RM in the current study. Fourth, mean cfPWV baseline values varied between protocols, though these differences were not statistically signi cant. This raises the issue of whether the signi cant within-protocol change seen during hyperglycemic-hyperinsulinemia was in uenced by some "regression to the mean." We considered this, but the concomitant changes seen in heart rate, diastolic blood pressure, and mean arterial pressure rather suggest the cfPWV change is legitimate and possibly due to the sympathetic mechanisms discussed above. Finally, we cannot rule out that OCT has in some unknown manner skewed the observed vascular responses, and this possibility should not be discounted. We do note, however, that no vasoactive effects have been identi ed in previous studies using a similar dose of OCT 51, 56-58 and that OCT infusion does not alter the hemodynamic effects of acute hyperglycemia 59 .

Conclusion
Hyperglycemic-hyperinsulinemia acutely increased cfPWV, heart rate, mean arterial pressure, and diastolic blood pressure in healthy humans, perhaps re ecting enhanced sympathetic tone. Whether repeated bouts of hyperglycemia with hyperinsulinemia (as occur regularly during the postprandial period in persons with DM and insulin-resistant syndromes) contribute to chronically-enhanced arterial stiffness remains unknown.

Declarations
Ethics Approval and Consent to Participate: All protocols were approved by the University of Virginia Institutional Review Board (#19948). Each subject gave written informed consent at their initial screening visit prior to study participation.
Availability of Data and Materials: The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Declaration of Con icting Interests: The authors declare that they have no con icts of interest to disclose.
Funding: This work was supported in part by National Institutes of Health research grants F32HL14230401, KL2TR003016, and UL1TR003015 (to WBH), and RO1DK073059 and RO1HL142250 (to EJB).