Chronic kidney disease accelerates endothelial barrier dysfunction in a mouse model of an arteriovenous fistula

Abstract

Hemodialysis patients depend on arteriovenous fistulas (AVF) for vascular access. Unfortunately, their 2-yr primary patency rate is only 60% because of AVF clog due to intimal hyperplasia at the venous anastomosis. Chronic kidney disease (CKD) can increase neointima formation by unknown mechanisms. A new AVF mouse model was created, and the mechanisms of CKD on neointima formation in AVFs were investigated. We created AVFs in mice by anastomosing the common carotid artery to the internal jugular vein. CKD was induced [BUN (blood urea nitrogen) in control and CKD mice, 33.3 ± 3.9 vs. 114.2 ± 12.1 mg/dl, P < 0.05]. After 1 day, there was endothelial cell loss and CD41-positive platelet aggregation, especially in the venous anastomosis. An invasion of macrophages and neutrophils peaked at 1 wk after surgery. Neointima formation (smooth muscle cell accumulation and extracellular matrix deposition) increased progressively over 4 wk. Mice with CKD had ∼45% (P < 0.05) more neointima formation than control mice. CKD decreased vascular endothelial-cadherin expression in endothelial cells and delayed regeneration of the endothelium. CKD also increased inflammatory cells (Mac-2-positive or CD45-positive) in AVFs at 2 wk. Finally, AVFs were “leakier” (increased accumulation of Evans blue) in CKD mice at 7 and 14 days than control mice. We find that CKD increases neointima formation and endothelial barrier dysfunction. We have created a mouse model of AVF with characteristics similar to failed AVFs in patients. The model will allow testing of strategies directed at improving AVF function in CKD patients.

approximately 470,000 u.s. patients with chronic kidney disease (CKD) depend on hemodialysis treatments. The success of hemodialysis requires a functioning arteriovenous fistula (AVF), initially created by Brescia et al. (4). Unfortunately, nearly 50% of the AVFs fail in the 2 years following their creation, generally because of proliferating vascular smooth muscle cells (SMCs) forming a neointima (2, 9, 25). Attempts to eliminate the neointima or create another AVF are costly, amounting to >$1 billion dollars per year in the United States. These strategies have variable success, and new methods are needed to understand neointima formation to improve the function of AVFs.

The endothelial cell (EC) monolayer is at the interface between the extravascular space and blood, playing a crucial role in vascular homeostasis (8). When there is EC dysfunction following vascular injury or EC loss, vascular permeability increases to promote extravasation of neutrophils and monocytes (28, 29). In a vein graft model in mice, we found that mechanical and shear stresses initially cause endothelial cell damage proceeding to denudation of ECs from the vein graft (7). Events following surgical placement of the AVF have not been defined.

Events in dialysis affecting the AVF are complicated because the patients are uremic, a condition that reportedly contributes to AVF failure (18, 20). For example, CKD is associated with a decrease in circulating endothelial progenitor cells, which could limit repair of the endothelium (19). It has also been reported that uremic serum causes dysfunction of endothelium and impairs EC barrier function (14). We propose that CKD augments processes that cause AVF failure. To evaluate this proposition, we created AVFs in CKD mice and uncovered a sequence of pathological changes that lead to AVF failure. The model was used to study underlying events causing CKD-induced endothelial dysfunction and neointima formation.

MATERIALS AND METHODS

Murine CKD model.

All studies were approved by the Institutional Animal Care and Use Committee of Baylor College of Medicine and were performed in accordance with National Institutes of Health guidelines. CKD was induced using the method as described (33). Mice aged 8–10 wk were used and divided into control and CKD groups. Briefly, mice housed in a 12:12-h light-dark cycle were fed 20% protein chow. After matching for body weight, CKD was produced by a two-step surgery in male C57BL6J mice. Mice were anesthetized (ketamine, 125 mg/kg body wt and xylazine, 6.4 mg/kg body wt). The left kidney was decapsulated to avoid ureter and adrenal damage, and two-thirds of the left kidney was removed. After surgery, the mice were kept on a warm blanket and were monitored; two doses of buprenorphine (0.1–2.5 mg/kg body wt sc) were administered after operation 12 h apart for pain relief. The diet was changed to 6% protein rodent diet chow (Harlan Teklad, Madison, WI) ad libitum to reduce mortality and limit hypertrophy (22).The right kidney was removed after 1 wk. Subsequently, the mice were pair-fed 40% protein chow with sham-operated control mice. In this model, the blood urea nitrogen (BUN) generally exceeds 80 mg/dl, which is four-fold greater than values in control mice, while serum creatinine is 3.8-fold higher (30). The blood pressure was measured as described previously (16). The BUN and relevant physiological factors, including the levels of hematocrit, were measured before AVF surgery by Comparative Pathology Laboratory Center in Baylor College of Medicine. After 2–3 wk of subtotal nephrectomy, AVFs were created in control and CKD mice. Measurements were performed in these AVFs after collection at different time points. The animals used in this study were summarized in Table 1.

Table 1.

Summary of the use of animals in the study

Surgery	Mice (n)	Measurement
Non-AVF control	6	Hct, BUN, body weight, blood pressure
1 mo AVF (including 3 mice used for Evans blue analysis)	7	H&E; trichrome staining, immunostaining for inflammatory markers, α-SMA, and CD31
1 wk AVF (including 3 mice used for Evan's blue analysis)	5	H&E; immunostaining for inflammatory markers, α-SMA, and CD31, CD41, and vWF
2 wk AVF	5	H&E; immunostaining for inflammatory markers, α-SMA, and CD31
3 wk AVF	5	H&E; immunostaining for inflammatory markers, α-SMA, and CD31
1 wk AVF	3	Enface analysis, VE-cadherin staining

AVF, arteriovenous fistula; Hct, hematocrit; BUN, blood urea nitrogen; H&E, hematoxylin and eosin; SMA, smooth muscle actin; vWF, von Willebrand factor; VE, vascular endothelial.

Murine AVF model.

An AVF was constructed by an end-to-end anastomosis between the right carotid artery and internal jugular vein (17). Briefly, male C57BL6J mice (Jackson Laboratory, Bar Harbor, ME) ranging from 12 wk of age were anesthetized with ketamine (125 mg/kg body wt) and xylazine (6.4 mg/kg body wt), and the right internal jugular vein was isolated using a dissecting microscope (Leica MZ6, Wetzlar, Germany). Its distal end was clamped and ligated. The common carotid artery was ligated below its bifurcation, and the proximal end was clamped. An end-to-end anastomosis was created using 11–0 nylon suture with an interrupted stitch. After unclamping, patency was confirmed visually. The mice were kept warm after surgery, and the analgesic (buprenorphine) was given two times 12 h apart. At 1, 2, 3, and 4 wk after surgery, the mice were anesthetized with ketamine (125 mg/kg body wt) and xylazine (6.4 mg/kg body wt) through intraperitoneal injection and euthanized by perfusion of the left ventricle with PBS and 10% formalin (to maintain the endothelium and morphology of the AVF) for 10 min. AVF was then collected. The neointima and media were defined as the regions between the lumen and the adventitia.

The vessel wall thickness was determined by measuring the difference between the area of the lumen and the neointima using NIS-Elements BR 3.0 program (Nikon). Five slides of cross sections were obtained by selecting the first of every 10 sections from each AVF; they were used to evaluate neointima formation in AVFs.

Materials and antibodies.

Antibodies against SMA-α were obtained from Abcam (Abcam, Cambridge, MA), antibodies against CD45, vascular-endothelial (VE)-cadherin, and CD31 were obtained from BD Biosciences (San Jose, CA). Antibodies against CD41 and von Willebrand factor (vWF) were obtained from Santa Cruz Biotechnology (Santa Cruz, Santa Cruz, CA) and anti-Mac-2 antibodies were from Cedarlane Laboratories (Burlington, Ontario, Canada). Anti-neutrophil antibodies were from AbD Serotec (Raleigh, NC). ABC kits and peroxidase substrate kit were purchased from Vector Laboratories (Burlingame, CA), and Evans blue was obtained from Sigma-Aldrich (St. Louis, MO).

Immunohistochemistry.

Morphology of AVFs was examined in tissues fixed in 10% neutral-buffered formalin and paraffin-embedded. Tissue sections were stained with hematoxylin and eosin (H&E) or Masson's-modified trichrome. Histology was assessed in AVFs that had been perfused, as described previously (6). Briefly, after wax was removed, sections were incubated (30 min in 3% H₂O₂ in methanol at room temperature). Subsequently, they were washed with PBS and incubated in 10 mM citrate buffer (pH 6.0) for 20 min after heating in a microwave. Sections were blocked with 10% goat serum (Vector Laboratories) for 30 min before being incubated with the primary antibodies. Staining was performed according to the ABC kit instructions (Vector Laboratories). Signals were visualized using a peroxidase substrate DAB kit (Vector Laboratories), and photographs were recorded with staining intensity analyzed using the NIS-Elements BR 3.0 program (Nikon).

En face analysis of AVF.

We analyzed the endothelium in AVFs using an en face technique with immunostaining, as described previously (31). AVF segments were cut longitudinally, mounted on glass slides with endothelium facing upward, and air dried for 1 to 2 h. AVFs were incubated with antibodies against VE-cadherin followed by immunofluorescently labeled secondary antibodies. The ratio of endothelial marker-positive cells to total cells was calculated; DAPI was used to stain nuclei.

Evans blue examination of endothelial barrier function of AVFs.

At the end of the experiment, 50 μl of 5% Evans blue diluted in saline was injected into the tail vein, and 10 min later, the mice were euthanized. AVFs were removed, washed with PBS, and fixed for 5 min in 10% neutral buffered formalin (27). AVFs were photographed with the contralateral jugular vein used as control. Evans blue accumulated extracellular matrix following disruption of the endothelium.

Statistics.

Cells with positive staining in the neointima were counted in defined areas in five slides taken from each AVF. The individuals who counted cells were blinded to treatments. Statistical analysis was performed using ANOVA. All data were expressed as mean ± SE, and P < 0.05 was deemed as statistically significant.

RESULTS

Neointima formation in experimental AVFs.

At 1 mo after surgery, there was a significant enlargement of internal jugular vein in the AVF (Fig. 1A). After 1 mo, H&E staining of the segment at 0 to 4 mm from the anastomosis revealed neointima formation with cells staining positively for SMA-α (Fig. 1, B and C). The maximal neointima hyperplasia was located at 1 mm from the anastomosis. At the far end of the AVF (4 mm distal to the anastomosis), neointima hyperplasia and SMA-α positive cells in the stenosis were reduced compared with findings at the anastomosis (Fig. 1C).

Fig. 1.

Characterization of a mouse arteriovenous fistula (AVF) model. A: photograph of the AVF after 1 mo. The anastomosis of the common carotid artery (CCA) with the internal jugular vein is indicated by black arrows. Hematoxylin and eosin (H&E; B) and α-smooth muscle actin (α-SMA; C) staining of 1 mo AVFs. The slides at different locations from the anastomosis were selected, and the morphology and neointima formation were detected by H&E and α-SMA staining, respectively. D: longitudinal analysis after 1 mo AVF. H&E staining (left) and α-SMA staining (brown) in the neoinitma at the venous end (right). E: α-SMA staining of cross sections of both the vein (red framed) and artery (green framed) at the anastomosis. Bottom: higher-magnification (×400) images are shown.

In longitudinal sections of AVF, relatively normal morphology was observed in the arterial segment (Fig. 1D). In the venous anastomosis, however, there was eccentric vascular remodeling and increased accumulation of SMA-α-positive cells (Fig. 1D). At the anastomotic site, sections that cross the artery (in green frame) and vein (in red frame) had more SMA-α-positive cells accumulated in the venous half (Fig. 1E). In contrast, there were two or three layers of neointima cells in the arterial segment near the anastomosis (elastin-positive, Fig. 1E).

Time course of neointima formation in AVFs.

Remodeling of the AVF increased in a time-dependent manner from 1 to 4 wk after placement (Fig. 2A). Neointima containing SMA-α-positive cells was detected at 2 wk; after 3 or 4 wk, there was substantial neointima formation at the anastomosis area (cellular layers increased vs. findings in normal veins, Fig. 2B). Notably, the neointima area increased from 0.5 ± 0.14 at 1 wk to 1.8 ± 0.5 (×10⁵ μm²) at 4 wk (Fig. 2C). The ratio of the lumen area to the lesion decreased from 7.12 ± 1.26 at 1 wk to 0.31 ± 0.05 at 4 wk (Fig. 2D).

Fig. 2.

Neointima formation increases in a time-dependent manner. H&E (A) and α-SMA staining (B) were used to quantify the sizes of neointimas (α-SMA-positive) (C). Ratios of lumen to neointima areas are shown (D). Results are mean ± SD (n = 5). The sections of AVFs were obtained within 0.5 mm to 1 mm of the anastomosis. *P < 0.05.

EC dysfunction and inflammatory cell infiltration.

To identify intact endothelium, cross sections of AVFs were stained with the EC marker, CD31. There was a dramatic decrease of CD31 signal soon after placing the AVF. These findings were fully recovered at 1 mo after AVF (Fig. 3, A and B).

Fig. 3.

Loss of endothelial junction molecules and inflammatory responses in AVFs. A: endothelial cell marker, CD31, was detected in cross sections of AVFs at different time points [1 wk (1W), 2W, and 4W]. B: positive cells were counted and shown as means ± SD; n = 5. C: platelet markers CD41 and vWF were positively stained in the subendothelial area of AVFs. D and E: macrophages and neutrophils were detected by anti-Mac-2 and MCA antibodies. F and G: macrophage- and neutrophil-positive cells were detected in 5 slides from each AVF. *P < 0.05.

At day 7, platelet markers, CD41 and vWF, were present in the lumen of the AVF (Fig. 3C). Even earlier (at 6 h after surgery), platelets were present in the AVF. As early as day 1, there was an inflammatory response, and after 1 wk, there was increasing infiltration of macrophages and neutrophils with the excess of macrophages detected at week 2 (Fig. 3, D and E). Neutrophil infiltration was at its peak at 1 wk after surgery and returned to basal levels at 4 wk (Fig. 3, F and G).

CKD increases neointima formation.

CKD was induced, and the biological indexes were analyzed in CKD and sham control mice (Table 2). The BUN level in the CKD group is significantly higher compared with the control group. The body weight and hematocrit level were decreased ∼8.9% and 20% in CKD mice, respectively. There was no significant difference in blood pressure between control and CKD mice. These results are consistent with a report showing that systemic arterial pressures were not affected by five-sixths nephrectomy-induced CKD, it is elevated after renal ablation produced by infarction (15).

Table 2.

Effects of CKD on BUN, body weight, Hct, and blood pressure

	BUN, mg/dl	Body Weight, g	Hct, %	Blood Pressure, SBP, mmHg
Control	33.3 ± 3.9	27.2 ± 0.2	52.5 ± 2.7	95.3 ± 12.3
CKD	114.2 ± 12.1*	24.8 ± 1.5*	42.1 ± 2.8*	99.6 ± 10.2

Data are expressed as means ± SD. CKD, chronic kidney disease; SBP, systolic blood pressure.

^*P < 0.05; n = 6.

To investigate how uremia affects the pathology of the AVF, AVFs were created in CKD and sham control mice. The time sequence for creating CKD and AVF were shown (Fig. 4A). Trichrome staining demonstrated that CKD significantly increased the deposition of matrix, the amount of fibrosis, and the number of cells in AVFs (Fig. 4B). In addition, the area of the neointima (SMA-α-positive staining) was increased by CKD vs. results in control mice (Fig. 4, C and D). The lumen area in AVFs of CKD mice was smaller than that in AVFs of control mice (Fig. 4D); the ratio of lumen/neointima was dramatically decreased in AVFs in CKD mice (Fig. 4E).

Fig. 4.

Chronic kidney disease (CKD) increases matrix deposition and neointima formation in AVFs. A: flow diagram showed the time points for creating CKD and AVF. AVFs were placed in CKD mice and control mice. One month later, sections were obtained within 0.5 to 1 mm of anastomosis, and the morphology and fibrosis of the AVFs were detected by trichrome staining (B); immunostaining of α-SMA is shown (C). The lumen and neointima areas (D) and the ratio of lumen to neointima were measured (D and E). The data are presented as means ± SD; n = 7. *P < 0.05.

CKD delays recovery of the endothelial barrier function in AVFs.

CKD increased fibrosis in both the vein and artery of the AVF, including increased collagen deposition in the adventitia after 1 mo. Compared with responses in the control mice, the inferior vena cavas of mice with CKD displayed thickening of the vessel adventitial area, which was trichrome positive. Compared with common carotid arteries of control mice, arteries of mice with CKD displayed marked thickening of the vessel wall due to fibrosis in the adventitia (Fig. 5A). In contrast, no neointima was found in vena cavas or arteries of control mice or mice with CKD (Fig. 5A). Increased fibrosis (trichrome-positive staining) was found in vessels of the heart and mesenteric arteries of CKD mice but not in vessels of control mice (data not shown).

Fig. 5.

CKD delays the recovery of the endothelial junction molecules and impairs endothelial barrier function. A: CKD induces fibrosis (Trichrome staining) in the CCA and control vein compared with result in control mice. B: an enface analysis of vascular endothelial (VE)-cadherin expression in AVFs. Immunofluorescent images showed VE-cadherin staining of normal vein and staining in the AVFs in CKD and control mice. The intensity of VE-cadherin expression was analyzed (C). D and E: infiltration of inflammatory cells in AVFs (1 wk) in control and CKD mice were detected by immunostaining with antibodies against macrophages (Mac-2), neutrophils (MCA), T cells (CD3), or monocytes (CD45) (D). Positive cells were counted and summarized (E). F and G: CKD induces endothelial barrier dysfunction in AVFs. Before collecting AVFs, Evans blue was administrated intravenously to mice, followed by washing with PBS. The AVFs were dissected, and photographs were obtained (F). The intensity of blue staining was analyzed (G). The data are represented as means ± SD; n = 3. *P < 0.05.

At day 1 after surgery, VE-cadherin levels significantly decreased in AVFs that were created in control or CKD mice. In this case, there were no significant differences between the two groups (Fig. 5C). At day 7, VE-cadherin expression was significantly higher in AVFs from control mice than in CKD mice (Fig. 5, B and C). Thus, CKD delays the recovery of VE-cadherin expression and hence, recovery of the endothelium. CKD also stimulated more robust inflammatory responses in AVFs: inflammatory cell infiltration was detected using antibodies against macrophages (Mac-2), neutrophils (MCA), T cells (CD3), or monocytes (CD45). In each case, there was significant cellular infiltration into AVFs in CKD mice compared with results in AVFs from control mice (Fig. 5, D and E).

To evaluate whether CKD affects EC barrier dysfunction, we examined Evans blue staining following its intravenous injection: the intensity of Evans blue in AVFs in CKD mice increased ∼1.5-fold when compared with that present in control mice after 1 wk (Fig. 5, F and G). After 4 wk, there was no detectable staining for Evans blue in the AVFs of control mice. However, the Evans blue signal was still positive in anastomosis of AVFs in CKD mice (Fig. 5, F and G). These results indicate that CKD induces endothelial barrier dysfunction in AVF.

DISCUSSION

There are a few reports of efforts to create small animal models (rats or mice) of AVFs (5, 23). We created a mouse AVF model by anastomosing the right common carotid artery to the right internal jugular vein. This would differ somewhat from that of Castier et al. (5), who created an end-to-side AVF by suturing the common carotid artery into the side of the jugular vein. In the venous arm of the AVF, we found neointima and adventitial developed SMA-α-positive cells and Mac-1-positive-macrophages (Figs. 1 and 4). These results are consistent in both models. There are several differences between the models of Castier and the one we studied: 1) We evaluated the effects of CKD on the histological changes in the AVF. Castier et al. (5) only evaluated changes in the AVF created in normal mice. 2) Our model is an end-to-end from the common carotid artery into the internal jugular vein. 3) We found there was more severe neointima hyperplasia in the venous arm near the anastomosis (Fig. 1). 4) Castier et al. (5) reported the development of neointima in both the artery and vein of their AVF model. In normal mice, we found a few layers of thickening of the arterial arm of the AVFs placed in normal mice. The more dramatic result was found in the neointima present in the venous arm of the AVF in normal mice. These responses were magnified in AVFs placed in mice with CKD. Notably, our AVF model in CD mice reflects events occurring in failed AVFs of dialysis patients: neointima in these subjects is principally in the venous side of the anastomosis (1, 24).

Several key pathological responses developed in the AVF model: 1) EC denudation occurred within hours of creating the AVF. At 1 day after creating the AVF, H&E staining revealed no nuclear staining in the perilumen wall of the internal jugular vein near the anastomosis. This indicates a loss of endothelial cells (Fig. 3A), platelets aggregated in the denuded AVF (Fig. 3C), severe neointima formation was located in the venous end of the anastomosis, and the presence of CKD was found to increase the formation of the neointima and was associated with CKD-induced dysfunction of the EC barrier (i.e., the leak of the Evans blue dye into the exposed subendothelial matrix). Consequently, the model that we created could be used to study mechanisms that lead to AVF failure in dialysis patients.

Multiple factors, including surgical trauma, mechanical stretch, and blood flow turbulence, have been proposed as factors promoting SMC proliferation. In our model, denudation of the endothelium, platelet aggregation, and impaired barrier function were prominent pathological changes that preceded neointima formation. Importantly, the presence of CKD augments the severity of these changes (Fig. 4).

The endothelial monolayer lies at the interface between blood and the extravascular space playing a crucial role in vascular homeostasis (8). In pathological conditions, injury to vascular endothelial cells results in increased permeability of the vascular barrier with increased extravasation of neutrophils and monocytes across the vessel wall (21). The factors in CKD that cause endothelial changes are unclear, but it is reported that treatment of ECs with uremic serum increases the expression of VCAM and tissue factor (26). In CKD mice with AVFs, there was increased migration of inflammatory cells into the AVF (Fig. 5D). They could be chemokines to stimulate SMC proliferation. This mechanism suggested that because elimination of MCP-1 (12), SDF-1 (32), or other chemokines can decrease neointima hyperplasia. Others suggest there is reduced bioavailability of nitric oxide in CKD, possibly because of increased oxidative stress to the vascular wall (28, 29). At 3 days after placement of the AVF, there was decreased expression of the endothelial junction molecule VE-cadherin (Fig. 5B) and CKD increased this response.

Another cause of EC dysfunction in AVFs in CKD mice could be related to a defect in endothelial progenitor cells, which impairs the ability of ECs to regenerate the denuded endothelium in the AVF. For example, there are reports that CKD limits mobilization of EPCs and impairs their differentiation into mature ECs (13, 14). Although we found that CKD suppresses EC regeneration in AVF, it did not affect the loss of ECs occurring in the newly formed AVF (day 1), there were regenerated ECs at day 7 in CKD mice but at significantly lower levels than that in control mice (Fig. 5C). These signs of an incomplete EC barrier could explain the increase in Evans blue-positive staining of extracellular areas of AVFs from CKD mice at 1 week. We conclude that CKD-induced endothelial barrier dysfunction is an initial factor that stimulates neointima formation.

Aggregation of platelets is an immediate response to EC denudation and could be linked to the greater inflammatory responses in AVFs that are placed in CKD mice (Fig. 5, D and E). In a vein graft model, we have shown that platelets are involved in attracting inflammatory cells via secretion of the chemokine, SDF-1α (7). Platelets store growth factors (PDGF and transforming growth factor-β1) that can affect SMC proliferation, and treatment of patients with AVFs using aspirin or clopidogrel can protect against early failure of AVFs from thrombosis and may improve long-term patency of hemodialysis grafts (10, 11).

A limitation of our model is that it is based on an end-to-end anastomosis rather than end-to-side anastomosis. Second, we selected the common carotid artery and jugular vein to place the AVF, instead of limb vessels. Possibly, the higher blood flow via the carotid artery could produce cardiac overload (3). However, the pathologic characteristics found in our model are similar to those in failed AV grafts of hemodialysis patients.

In summary, we have created a mouse model of AVF that loses function through a predictable sequence of events. Pathologic features of these events are similar to those in human AVFs. Early, there are cellular responses (endothelial denudation and platelet aggregation) that are followed by infiltration of inflammatory cells and proliferation of SMCs. Later, there is neointima formation associated with EC dysfunction at the venous anastomosis. We demonstrate that CKD accelerates these features by decreasing the expression of EC junction molecules, while delaying EC regeneration in AVFs. The platelet aggregation leads to SMC proliferation and neointima formation. Thus, our results point to different possible treatment strategies or areas to investigate why AVFs fail.

GRANTS

This study was supported by NIDDK Grant RO1-DK095867 and American Heart Association Grant 10SDG2780009 (to J. Cheng), support from National Institutes of Health Grants R37 and DK-37175, and a generous grant from Dr. and Mrs. Harold Selzman.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

Author contributions: A.L. and J.C. conception and design of research; A.L., Y.W., G.H., and J.C. performed experiments; A.L., Y.W., G.H., L.T., and J.C. analyzed data; A.L. and J.C. interpreted results of experiments; A.L., L.T., and J.C. prepared figures; A.L., L.T., and J.C. drafted manuscript; A.L. and J.C. edited and revised manuscript; A.L., Y.W., G.H., L.T., and J.C. approved final version of manuscript.