Lipid-Heparin Infusion Suppresses the IL-10 Response to Trauma in Subcutaneous Adipose Tissue in Humans

Authors


(dimas.ikeoka@healthsite.at)

Abstract

An imbalance between pro- and anti-inflammatory cytokine productions in adipose tissue is thought to contribute to chronic, systemic, low-grade inflammation and consequently to an increased risk of cardiovascular complications in obese and type 2 diabetic patients. Nonesterified fatty acids (NEFA), whose serum levels are elevated in such patients, have been shown to interfere with cytokine production in vitro. In order to evaluate the effects of elevated NEFA levels on cytokine production in adipose tissue in vivo we used an 18-gauge open-flow microperfusion (OFM) catheter to induce local inflammation in the subcutaneous adipose tissue (SAT) of healthy volunteers and to sample interstitial fluid (IF) specifically from the inflamed tissue. In two crossover studies, nine subjects received either an intravenous lipid-heparin infusion to elevate circulating NEFA levels or saline over a period of 28 h. The former increased the circulating levels of triglycerides (TGs), NEFA, glucose, and insulin over the study period. NEFA effects on locally induced inflammation were estimated by measuring the levels of a panel adipokines in the OFM probe effluent. Interleukin-6 (IL-6), IL-8, tumor necrosis factor-α (TNF-α) and monocyte chemoattractant protein-1 (MCP-1) levels increased during the study period but were not affected by lipid-heparin infusion. In contrast, the level of IL-10, an anti-inflammatory cytokine, was significantly reduced during the final hour of lipid-heparin infusion (saline: 449.2 ± 105.9 vs. lipid-heparin: 65.4 ± 15.4 pg/ml; P = 0.02). These data provide the first in vivo evidence that elevated NEFA can modulate cytokine production by adipose tissue.

Introduction

Adipose tissue was long considered to be solely a triglyceride (TG) storage depot that releases nonesterified fatty acids (NEFA) into the circulation in accordance with the bodies energy needs. It is now known to produce both pro- and anti-inflammatory cytokines as well as many other biological mediators, including resistin, leptin, and plasminogen activator inhibitor-1 (1,2,3,4) This moreover appears to be of pathophysiological importance because a combination of deregulated-adipose tissue cytokine production and elevated NEFA levels in obese individuals has been associated with the development of insulin resistance, type 2 diabetes, and an increased risk of cardiovascular events (5,6,7,8,9,10). Little is however known about the mechanisms underlying these associations.

Elevated NEFA levels could conceivably exert a direct influence on glucose uptake and/or contribute to the perturbation of cytokine production in the adipose tissue of obese persons. In support of this, it has recently been shown that NEFA can activate intracellular signaling molecules implicated in the reduction of insulin sensitivity and proinflammatory activation such as c-Jun N-terminal kinase and nuclear factor κB in cultured mouse adipocytes (11,12,13). Specific fatty acids have, moreover, been shown to differentially regulate the production of inflammatory and anti-inflammatory cytokines in these cells (14). On the other hand, in vivo studies that validate these in vitro data in subcutaneous adipose tissue (SAT) are lacking. Indeed this represents a significant technical challenge because the trauma induced by introduction of a sampling device per se will modulate the levels of some cytokines locally.

With this key issue of local sampling in mind we recently characterized the changes in the levels of a panel of cytokines in the interstitial fluid (IF) surrounding an open-flow microperfusion (OFM) catheter following its insertion into SAT of human volunteers (15). During the 8-h sampling period, time-dependent increases in the levels of interleukin-1β (IL-1β), IL-6, IL-8, and tumor necrosis factor-α (TNF-α) were observed, consistent with the induction of a local inflammatory response by the trauma caused by catheter insertion. In the present study, we use this as a model system to assess whether elevation of circulating NEFA level in healthy human subjects can modulate a local inflammatory process in adipose tissue.

Methods and Procedures

Subjects

Following approval of the study by the local ethics committee of the Medical University of Graz and informed consent, 10 male nondiabetic white subjects were selected to participate in the study. All subjects were healthy and not taking any regular medication. The study protocol was performed in accordance with the Declaration of Helsinki.

Study design

A randomized, open-label, placebo-controlled, crossover study (summarized in Figure 1) was conducted at the Center for Medical Research of the Medical University of Graz. Subjects were randomly assigned to undergo two experimental procedures during two study visits, separated from each other by at least a week, namely continuous venous infusions over 28 h of (i) an emulsified lipid solution containing heparin (intralipid 20%, 40 ml/h; Fresenius Kabi, Graz, Austria; heparin 250 U/h; IMMUNO Baxter AG, Vienna, Austria) or (ii) 0.9% saline, in both cases using the same infusion rate. Volunteers were instructed to ingest a low-fat and low-calorie diet and to avoid exercise during the day preceding each study visit. Study procedures were started in the morning at 0800 h following a 12-h overnight fasting period. One polyethylene venous catheter was inserted into an antecubital vein for venous infusions only and a second venous cannula was inserted into a dorsal vein of the contralateral arm and maintained in a thermoregulated box at 50 °C throughout the study to ensure arterialized venous blood sampling. Three hours later (1100 h) an 18-gauge open-flow OFM catheter was inserted into the periumbilical SAT of the abdominal wall after local anesthesia with procaine hydrochloride not containing a vasoconstrictor (Novanaest purum 2%; Gebro Pharma, Fieberbrunn, Austria). The OFM technique, described in detail elsewhere (16,17), uses a double lumen catheter with 0.5 mm macroperforations that allow the free exchange of solutes between the surrounding tissue and the internal lumen. The catheter and the afferent loop of the respective tubing system were prefilled with a perfusate consisting of 20% autologous serum in sterile isotonic multi-ionic solution (ELO-MEL isoton; Fresenius Kabi). An in vitro test confirmed that this perfusate effectively prevented nonspecific binding of cytokines to the OFM catheters and tubing systems used in the study (data not shown). Serum was collected from each volunteer in the morning of each study visit and added to the perfusate solution using strict sterile techniques. A peristaltic pump (Minipuls 3; Gilson, Middleton, WI), was used to maintain a continuous flow of the perfusate solution in a push-pull mode at 1 µl/min during the 24 h following OFM catheter insertion. Effluent probe, continuously pumped throughout the system, was collected in glass vials that were exchanged at predetermined intervals according to the sampling protocol (Figure 1). Blood sampling at 4 h intervals was initiated at 1200 h and continued until 1200 h the following day. After centrifugation, serum and plasma and SAT effluent samples were immediately transferred to a −80 °C freezer. Subjects were maintained under fasting conditions throughout the whole study period. For safety reasons, activated time of thromboplastin (aTTP) was measured during lipid-heparin infusion.

Figure 1.

Schematic overview of the study protocol. ISF, interstitial fluid; NEFA, nonesterified fatty acids; OFM, open-flow microperfusion.

Inulin infusion and recovery calculation for OFM catheters

As previously described, an isotonic perfusate was pumped through the OFM catheter. The collected effluent consisted of a blend of perfusate and IF. To estimate the proportion of a given solute that is collected from IF in this mixture, we used a continuous venous infusion of inulin at a known concentration, and measured this substance in parallel blood and OFM effluent samples. Inulin was first administered as a bolus (50 mg/kg of body weight) before starting the continuous intravenous infusion (Inutest 25%; Fresenius Pharma, Graz, Austria) at a rate corresponding to 0.25 × creatinine clearance mg/min which was then maintained for the remainder of the study (Cockcroft-Gault approximation formula) to ensure a steady-state serum concentration of 250 mg/ml, as previously described (18). Inulin and lipid-heparin or saline infusions were started simultaneously. Recovery efficiency (RE) was retrospectively calculated by dividing the inulin concentration in effluent samples (Ief) by the average serum concentration (Ise) (RR = Ief/Ise). Since inulin concentrations in serum and in IF are theoretically equal in the steady-state phase, RE reflects the ratio of inulin present in the SAT effluent samples after dilution with perfusate to the actual concentration in IF. As observed in the majority of our previous experiments, OFM catheter recovery estimated by this ratio was in the range of 10–20%, meaning that the concentration of inulin in the collected samples was 10–20% of the current concentration in the tissue. The actual cytokine concentrations were thus derived by dividing the levels measured in effluent samples by the RE. Since RE as calculated above does not differ substantially from the estimated recovery for other higher molecular weight molecules such as albumin (16) it was assumed that inulin recovery can reliably provide better estimates of actual cytokine concentrations in IF.

Analytical methods

Cytokine concentrations in SAT effluent samples were measured using multiplexed flow-cytometric bead-based enzyme-linked immunosorbent assay according to the manufacturers' instructions (R&D Systems, Minneapolis, MN). Briefly, microbeads coated with highly specific antibodies against IL-6, IL-8, IL-10, TNF-α, and monocyte chemoattractant protein-1 (MCP-1) were added to SAT effluent samples in duplicate in a 96-well microfilter plate. Following incubation for 3 h with continuous agitation, the bead/sample mixtures were incubated with biotinylated antibodies against each analyte and finally with streptavidin to produce luminescence signals proportional in magnitude to the analyte concentrations in the samples. Resuspended beads were read in a bead-based multiplex system (BioPlex200; Bio-Rad Laboratories, Hercules, CA) and data was analyzed using Bio-Plex Manager Software, version 4.4. For IL-6, IL-8, IL-10, TNF-α, and MCP-1 the average intra-assay coefficients of variability were respectively 4.5, 6.0, 5.6, 4.2, and 5.3%, whereas the interassay coefficients of variabilities were respectively 6.8, 13.9, 8.3, 6.8, and 7.8%, as stated by the manufacturer. Minimum detectable dose provided by the manufacturer for the same cytokines were 0.36, 0.39, 0.13, 0.47, and 0.95 pg/ml, respectively.

Sampling was carried out continuously. Consequently, cytokine concentrations in SAT effluent represented the average concentrations during the time period between vial exchanges.

The time period between catheter insertion and the obtaining of the first effluent drops, corresponding to the time necessary for the effluent to completely fill the efferent loop of the tubing system and reach the sampling vials, was, on average, 1 hour. Baseline perfusate cytokine concentrations were also tested before perfusion. Glucose concentrations in heparinized whole blood were measured locally using a standard point-of-care blood gas analyzer (Omni S6; Roche Diagnostics, Mannheim, Germany). Insulin concentrations were measured using an ultrasensitive solid-phase two-site enzyme-linked immunosorbent assay (Mercodia, Uppsala, Sweden; intra-assay precision 2.9%; interassay precision 2.2%). NEFA were measured in plasma using a fully automated, enzymatic, and photometric assay (Roche COBAS MIRA Chemistry Analyzer; Roche Diagnostics, Montclair, NJ; intra-assay precision 3.5%; interassay precision 8.8%). TGs and cholesterol were measured in plasma using an automated, enzymatic, colorimetric method (Cobas Integra System; Roche Diagnostics, Mannheim, Germany).

Statistical analysis

Continuous variables meeting the normality criteria for Shapiro-Wilk testing are indicated as mean ± s.e.m. or mean ± s.d. as specified in the text and were analyzed with parametric methods (“ANOVA” for repeated measurements and post hoc t-tests with Bonferroni correction). Analysis of variance for repeated measurements was set to detect significant effects of time (time effect), different treatments (lipid-heparin vs. saline infusion: treatment effect) and/or significant interactions between both of these variables (time × treatment: time-treatment effect). Skewed distributed data that were not normally distributed following log-transformation were analyzed by nonparametric methods. Descriptive levels below 0.05 were considered significant for all statistical tests. Statistical analysis was performed using SPSS 15.0 software (SPSS, Chicago, IL).

Results

Ten healthy volunteers were selected and completed the research protocol. The data obtained for one volunteer were however excluded from analysis due to the development of a toenail infection during the study. No adverse events occurred. No signs of local inflammation or infection were detected by clinical inspection of the site of insertion of the OFM catheter during or after each experiment. Baseline data collected from the investigated subjects at the screening visit are summarized in Table 1.

Table 1.  Baseline anthropometric and laboratorial characteristics of the healthy volunteers enrolled in the study
inline image

TGs, NEFA, glucose, and insulin

Plasma TG, NEFA, glucose and insulin concentrations measured during the study are presented in graphical form in Figure 2. Continuous infusion of lipid-heparin resulted in a steady and significant elevation of plasma TG concentrations (mean value ± s.d. 279.6 ± 40.5 vs. 68.1 ± 8.25 mg/dl for saline controls; treatment effect: P < 0.001 using ANOVA for repeated measurements; time effect: P = 0.008; time-treatment effect: P < 0.001). NEFA levels were however only higher in the treatment group than in the saline group during the first 8 h of continuous infusion (0400 h: saline 0.43 ± 0.15 mmol/l, lipid-heparin 0.86 ± 0.30 mmol/l; P = 0.005; 0800 h: saline 0.84 ± 0.21 mmol/l, lipid-heparin 0.81 ± 0.16 mmol/l; P = 0.78). Plasma glucose concentrations gradually decreased in both groups (time effect: P < 0.001), though more rapidly in the saline infusion group (time-treatment effect P = 0.02). Insulin concentrations initially decreased in both groups but returned to the first measured value in the group receiving lipid-heparin infusion (treatment effect: P = 0.02).

Figure 2.

Plasma levels of triglycerides, NEFA, glucose, and insulin of healthy volunteers submitted to 28 h saline (closed circles) or lipid heparin (open triangles) infusion are indicated as mean ± s.e.m. (error bars). Data were analyzed using ANOVA for repeated measurements. NEFA, nonesterified fatty acids.

Analyte recovery efficiency

The analyte recovery efficiency for each catheter was calculated using the inulin concentrations in serum and SAT effluent as described above and was shown in each case to be stable over time (saline 0.20 ± 11; lipid-heparin 0.21 ± 12; mean ± s.d.). Recovery efficiencies were estimated for each time point and used to correct cytokine concentrations from SAT effluent samples before statistical analysis.

Cytokines in perfusate and SAT effluent samples

Perfusate solutions before infusion. Baseline cytokine levels in the serum-containing perfusate solutions were determined before infusion and subtracted from the concentrations measured in the SAT effluent samples. Low concentrations of IL-6 (1.6 ± 0.2 vs. 1.5 ± 0.4 pg/ml; P = 0.36; mean ± s.d.), IL-8 (2.4 ± 0.9 vs. 2.6 ± 1.1 pg/ml; P = 0.31), and TNF-α (1.7 ± 0.9 vs. 1.4 ± 1.0 pg/ml; P = 0.39) were found in perfusates for saline and lipid-heparin experiments, respectively. MCP-1 concentrations were higher but also comparable for the two interventions (36.7 ± 15.5 vs. 33.8 ± 15.4 pg/ml; P = 0.66). IL-10 concentrations were below the detection limit.

Cytokine profiles in SAT effluent. Analysis of cytokine concentrations was performed on log-transformed data. Time profiles for the retransformed data are depicted in Figure 3. The profiles obtained for MCP-1 and IL-6 are both consistent with biphasic secretion, starting with low concentrations and an early peak during the first 8 h followed by a second and broader peak between 1000 h and 1400 h. Neither time nor treatment effects were not observed for IL-6 but significant time variations were shown in the case of MCP-1 (time effect: P < 0.001). Thus, in the control intervention, MCP-1 increased from 2.7 ± 0.4 ng/ml at 0530 h to 68.9 ± 29.6 ng/ml at 0830 h (P = 0.001; mean ± s.e.m.), decreased to 28.6 ± 8.6 ng/ml by 1130 h (P = 0.03) and increased again to 44.6 ± 10.9 at 1500 h (P = 0.02). IL-8 concentrations increased significantly after catheter insertion (time effect: P < 0.001) (i) from 0.6 ± 0.1 ng/ml at 0530 h to 128.2 ± 21.0 ng/ml at 15:00 h during saline infusion (P < 0.01) and (ii) from 0.5 ± 0.1 to 122.6 ± 29.5 ng/ml during the same period in the lipid-heparin group (P = 0.04). Thereafter, they remained more or less constant until the end of the study. No treatment effect was observed. In the control saline infusion group TNF-α levels increased (time effect: P < 0.001) between 0730 h (34.2 ± 14.5 pg/ml) and 1700 h (1181.6 ± 390.5 pg/ml; P = 0.04) and afterwards declined. Similar results were obtained for the subjects that received lipid-heparin (0730: 19.8 ± 6.4 pg/ml; 1700: 537.2 ± 199.4) without a treatment effect (P = 0.39). IL-10 concentrations, which were initially largely below the detection limit, significantly increased after 6–8 h (time effect: P < 0.001) and, in the comparison between both interventions, started to display a distinct time-variation between 1400 h and 1600 h, reaching significance in the last 6 h of the study (2400 h 462.3 ± 218.4 pg/ml vs. 77.4 ± 18.4 pg/ml P = 0.01 and 2600 h 436.1 ± 214.9 pg/ml vs. 53.5 ± 15.8 pg/ml P < 0.01; respectively for saline and lipid-heparin; means for the last 4 h saline: 449.2 ± 105.9 vs. lipid-heparin: 65.4 ± 15.4 pg/ml; P = 0.02).

Figure 3.

TNF-α, IL-6, IL-8, IL-10, and MCP-1 profiles in effluent samples from SAT of healthy volunteers submitted to 28 h saline (closed circles) and lipid heparin (open triangles) continuous infusion. Data analysis was performed on log-transformed data, but actual concentrations corrected for the recovery are indicated as mean ± s.e.m. (error bars); *P < 0.05. IL, interleukin; MCP-1, monocyte chemoattractant protein-1; TNF-α, tumor necrosis factor-α.

Discussion

In vitro studies have implicated NEFA as modulators of adipocyte cytokine production but are yet to be confirmed in SAT in vivo. In the present study, which builds on our previously published work (15), OFM was used (i) to produce a local trauma and serve as a model in vivo of inflammatory response in SAT of healthy human volunteers (through catheter insertion) and (ii) to sample pro- and anti-inflammatory cytokines specifically from the inflamed site from healthy subjects who received a lipid-heparin infusion to elevate plasma NEFA levels or a control saline infusion. As previously reported, for a shorter study, the cytokine expression profiles observed in IF surrounding the OFM probe in saline-infused subjects are consistent with the triggering of a local inflammatory reaction by the insertion of an OFM catheter into SAT. Thus, distinct and reproducible cytokine-specific time-dependent profiles were recorded. The elevated cytokine levels were moreover maintained beyond the 24th h of the present study. This effectively excludes the possibility that the results obtained reflect ongoing equilibration between the OFM system and the tissue and/or circadian modulation of basal cytokine secretion (19,20). With regard to the main study question, lipid-heparin infusion did not significantly alter the proinflammatory tone of the inflammatory response to catheter insertion. On the other hand, a clear inhibitory effect on the secretion of the anti-inflammatory cytokine IL-10 was observed. This result, which is consistent with recent data, showing that IL-10 secretion and expression are downregulated in mouse adipocyte cultures by palmitic acid, an 18-carbon saturated fatty acid (14), represents the first in vivo evidence in humans that elevated NEFA levels can disturb the balance between pro- and anti-inflammatory cytokines in SAT.

IL-10 inactivates inflammatory cells and inhibits the secretion of proinflammatory cytokines, such as TNF-α and IL-6 (21,22). It has moreover been positively associated with insulin sensitivity and has a protective effect against IL-6-induced insulin resistance in skeletal muscle (23). IL-10 inhibition in adipose tissue might thus represent an additional diabetogenic stimulus for individuals with high levels of circulating NEFA, such as obese and type 2 diabetic patients.

Two aspects of the apparent interplay between the metabolic changes and cytokine production in adipose tissue following lipid-heparin infusion particularly merit discussion. Firstly, the observed time-dependent changes in the plasma concentrations of glucose and insulin indicated that insulin sensitivity was lower in lipid-heparin infused subjects than in those that received a saline infusion, especially toward the end of the study. The coincidence of this reduced insulin sensitivity in lipid-heparin infused subjects with IL-10 suppression, makes it tempting to speculate that a cause and effect relationship exists between these variables. The second aspect relates to the timing of elevated NEFA levels with respect to saline infusion and IL-10 suppression.

Plasma NEFA concentrations in lipid-heparin and saline-infused subjects only differed significantly during the first 8 h of the experiment, which can be attributed to the combined influence of prolonged fasting on lipolysis (24,25), and the relatively low rate of lipid-heparin infusion that was used. The latter was chosen to ensure plasma NEFA concentrations within the range observed in obese and diabetic individuals in real life (600–800 µmol/l) (26) and because the effects of NEFA on carbohydrate metabolism can also be expected to occur at even lower rates (27). Since NEFA levels were similar in the late phases of both interventions, the suppression of SAT IL-10 secretion in lipid-heparin infused subjects would appear to be a delayed effect of exposure to elevated NEFA concentrations. Although such a delay cannot be easily explained, it is supported by a report of IL-10 inhibition by NEFA in cultured adipocytes (14). Our current data however do not rule out the alternative explanations that NEFA acted indirectly or indeed did not play an important role in the observed IL-10 inhibition. In principle, TG, glucose or insulin or a combination of these could have contributed to IL-10 suppression because their levels were also higher in lipid-heparin infused subjects. Elevated TG levels coincide with insulin resistance in metabolic syndrome (28), but have always been described as a consequence rather than a cause of insulin resistance and diabetes (29). It therefore seems unlikely that TG played a role in the observed modulation of IL-10 release. Insulin has consistently been shown to exhibit anti-inflammatory properties (30) and for this reason, is an unlikely promoter of inflammation in the present model, notwithstanding our single previous publication suggesting such an activity (15). A lack of support for the suppression of IL-10 in adipose tissue by insulin or by TG, thus reinforces the notion that NEFA are the most likely causative factor in this case, in agreement with in vitro studies. We nonetheless see a need for additional studies employing a different methodology to distinguish between these possible mechanisms.

It is further noteworthy that the measured cytokine time profiles appear to differ substantially from the classic description of acute inflammation, namely an early increase in TNF-α, followed by a secondary response comprising the expression of other cytokines such as IL-6 and IL-8 (31). A delayed TNF-α response, starting 6 hours after the initial stimulus and also biphasic profiles for IL-6 and MCP-1, with a second increase in the measured concentrations almost in parallel with TNF-α suggest the occurrence of a second inflammatory wave during this period. These biphasic profiles have not been described to date and might represent a delayed inflammatory reaction, analogous to a delayed type hypersensitivity response.

For completeness, some limitations of the present study must be addressed. The study population was designed to detect significant differences in IL-6 concentrations when using ANOVA for repeated measurements. Because of the small sample size, it is thus possible that the measurements of other cytokines did not share the same power and were thus a priori unable to achieve significance.

Elevated NEFA levels were eventually observed in both the test and control groups (at the 12 h time point). This can be attributed to prolonged fasting and could be avoided in future studies by offering the subjects regular meals. OFM technology allows the sampling of a very wide range of analytes in IF. It cannot however precisely pinpoint the cellular origin of these substances. Cytokine secretion in vivo, as measured by our method, can also not be completely paralleled with ex vivo cytokine production in the way that it was been done in the cited previous studies. The parallel use of other techniques, such as gene expression analysis of adipose tissue biopsies and measurement of arteriovenous differences in cytokine concentrations would provide a more accurate estimation of local secretion and also differentiate between locally released (paracrine) and secreted (endocrine) cytokines (32).

The present study focused specifically on SAT for reasons of technical feasibility, i.e., an analysis of visceral adipose tissue fat by OFM is too invasive a procedure to be used with healthy subjects. The relative importance of SAT and visceral adipose tissue in the induction of systemic inflammation, insulin resistance, and chronic atherosclerotic disease however remains controversial (20,33,34,35). The full significance of our results will thus only become clear following future studies comparing the responses of these tissue types. Finally, analysis may be limited by the difficulty of achieving homogeneity in such a large set of repeated measurements, as provided by the cytokine profiles, and to clearly define significant differences with reasonable physiological implications.

To conclude, the present study shows for the first time that lipid-heparin infusion modulates the local inflammatory response in SAT of healthy subjects elicited by the insertion of an OFM catheter.

No apparent change in the secretion of inflammatory cytokines was observed, rather the production of the anti-inflammatory cytokine, IL-10 was suppressed, providing support for a significant role of IL-10 in the response of adipose tissue to elevated circulating NEFA levels in vivo. The study does not however exclude the possibility that other substrates, such as glucose, TG, and insulin, individually or in combination, contribute to the observed effect on IL-10 levels. Future investigations that address these open questions are justified by their potential relevance to common diseases.

ACKNOWLEDGEMENT

The authors acknowledge Heimo Strohmaier, Sandra Klaschka, Karin Buerger, Maria Suppan, and Beate Tiran for their assistance with the analysis of samples, Andrea Wutte for supporting the study execution and Andrea Groselj-Strehle for statistical advice. The present study was partially financed by a Marie Curie Incoming International Fellowship program under the European Union Sixth Framework Program (“Survive ICU”; MIF1-CT-2005-007838).

DISCLOSURE

The authors declared no conflict of interest.

Ancillary