- Email: [email protected]
Endovascular Biopsy and Endothelial Cell Gene Expression Analysis of Dialysis Arteriovenous Fistulas: A Feasibility Study
CLINICAL STUDY
Endovascular Biopsy and Endothelial Cell Gene Expression Analysis of Dialysis Arteriovenous Fistulas: A Feasibility Study Hugh McGregor, MD, Zhengda Sun, PhD, David McCoy, MS, Vishal Kumar, MD, Miles Conrad, MD, Mark Wilson, MD, and Daniel Cooke, MD ABSTRACT Purpose: To demonstrate feasibility of endothelial cell (EC) biopsy from dialysis arteriovenous fistulas (AVFs) with the use of guidewires and to characterize gene expression differences between ECs from stenotic and nonstenotic outflow vein segments. Materials and Methods: Nine consecutive patients undergoing fistulography for AVF dysfunction from June to August 2016 were enrolled. ECs were biopsied with the use of guidewires from venous outflow stenoses and control outflow veins central to the stenoses. ECs were sorted with the use of flow cytometry, and the Fluidigm Biomark HD system was used for single-cell quantitative polymerase chain reaction (qPCR) analysis of gene expression. Forty-eight genes were assessed and were selected based on different cellular functions and previous literature. Linear mixed models (LMMs) were used to identify differential gene expression between the groups, and self-organizing maps (SOMs) were used to identify cell clusters based on gene coexpression profiles. Results: A total of 219 and 213 ECs were sampled from venous outflow stenoses and control vein segments, respectively. There were no immediate biopsy-related complications. Forty-eight cells per patient were sorted for qPCR analysis. LMM identified 7 genes with different levels of expression at stenotic segments (P < .05), including AGTR2, HMOX2, MTHFR, SERPINC1, SERPINE1, SMAD4, and VWF. SOM analysis identified 4 cell clusters with unique gene expression profiles, each containing stenotic and control ECs. Conclusions: EC biopsy from dialysis AVFs with the use of guidewires is feasible. Gene expression data suggest that genes involved in multiple cellular functions are dysregulated in stenotic areas. SOMs identified 4 unique clusters of cells, indicating EC phenotypic heterogeneity in outflow veins.
ABBREVIATIONS AVF ¼ arteriovenous fistula, EC ¼ endothelial cell, LMM ¼ linear mixed models, qPCR ¼ quantitative polymerase chain reaction, SOM ¼ self-organizing map
From the Department of Radiology and Biomedical Imaging, University of California, 505 Parnassus Avenue, M391, San Francisco, California 941430628. Received January 2, 2018; final revision received April 10, 2018; accepted April 22, 2018. Address correspondence to H.M.; E-mail: [email protected] H.M. and Z.S. contributed equally to this work. Table E1 can be found by accessing the online version of this article on www. jvir.org and clicking on the Supplemental Material tab. None of the authors have identified of conflict of interest. © SIR, 2018 J Vasc Interv Radiol 2018; ▪:1–7 https://doi.org/10.1016/j.jvir.2018.04.034
Venous outflow stenosis because of neointimal hyperplasia and medial fibrosis accounts for the majority of dialysis arteriovenous fistulas (AVFs) requiring intervention to maintain patency (1). The histologic hallmark of neointimal hyperplasia is smooth muscle cell and myofibroblast migration and proliferation within the neointima (2). The mechanisms mitigating this process are not well understood, with current knowledge largely based on immunohistochemical analysis of resected surgical specimens (3) and genetic analysis of in vitro (4) and nonhuman animal models (5–8). The lack of a safe and reliable means for cellular analysis of AVFs in humans has limited the genetic characterization of AVF failure. Consequently, the ability to predict AVF
2 ▪ Endovascular Biopsy and EC Gene Expression Analysis of Dialysis AVFs
EDITORS’ RESEARCH HIGHLIGHTS The authors demonstrated that a 1.5-mm-diameter J-type guidewire successfully harvested endothelial cells (ECs) from within stenotic dialysis AV fistulas for gene characterization. A number of dysregulated genes, ones not found in normal fistula segments, were identified. This suggests that heterogeneity of EC phenotypes at specific vessel segments may play a role in stenosis formation. The findings also suggested that diabetes was a possible modifier of endothelial gene function.
failure and identify pharmacologic therapies that may prevent or slow venous outflow stenosis are limited. To address this limitation, the present study expands on previous endovascular techniques (9–14) for the targeted biopsy of endothelial cells (ECs) in vivo in dialysis AVFs with the use of endovascular devices. The objectives of this study were: (i) to demonstrate feasibility of EC biopsy in dialysis AVFs with the use of guidewires; (ii) to characterize gene expressivity differences between ECs from stenotic and nonstenotic outflow vein segments; and (iii) to develop an analytic technique using linear mixed models (LMMs) to conduct unsupervised machine-learning statistical analysis of single-cell gene expression.
MATERIALS AND METHODS The study protocol was approved by the Institutional Review Board.
Patients Ten consecutive patients on hemodialysis referred to the interventional radiology service for fistulography from June to August 2016 were enrolled and consented to participate in this study. One patient was excluded from the final analysis owing to the presence of an AV graft. Inclusion criteria included end-stage renal disease, a dialysis AVF with a peripheral venous outflow stenosis, and consent for tissue analysis. Patient demographics and clinical data were obtained from the most recent primary care clinic notes in the electronic medical record (Table). Three patients were male. The mean patient age was 61 years (± 11 years). Seven patients had hypertension, 5 were diabetic, 1 was a current smoker, and 4 were past smokers. The mean BMI was 28.8 kg/m2 (± 10.9). Severity of stenosis was quantified from images saved in our Picture Archiving and Communication System. Percentage of stenosis was calculated by dividing the diameter of the stenosis by the mean of measurements taken from nonstenotic vessel segments immediately proximal and distal to the stenosis.
Endothelial Biopsy A 6-F sheath (Cordis, Milpitas, California) was placed in the peripheral venous outflow and fistulography was performed.
McGregor et al ▪ JVIR
After identification of a peripheral venous outflow stenosis, a 0.035-inch Glidewire (Terumo, Tokyo, Japan) was advanced central to the stenosis. The stenosis was then crossed with a 5-F angled catheter (Kumpe; Cook Medical, Bloomington, Indiana) and the Glidewire was removed. A 0.035-inch “J”-tipped guidewire (Rosen; Cook Medical) was then advanced to the distal aspect of the catheter (Fig 1). The catheter was withdrawn to expose the distal 1 cm of the guidewire, which was then advanced and retracted within the stenosis at 1-cm intervals ~20 times. The “J” tip of the wire opened relative to the degree of the stenosis. The guidewire was removed and the distal 5 cm was cut and immediately placed in a 50-mL tube (Falcon; Thermo Fisher Scientific, Waltham, Massachusetts) containing 10 mL EC dissociation buffer (Gibco; Thermo Fisher Scientific, Waltham, Massachusetts). Control samples from angiographically normal venous outflow vein central to the stenosis were obtained in a similar fashion. Care was taken during both experimental and control biopsies to ensure contact between the wire and the target vessel only. The sampling wires were otherwise sheathed within the delivery catheter.
Endothelial Sorting Endothelial cells were sorted according to the protocol described in detail by Sun et al (10). Forty-eight cells were sorted from each patient. Briefly, after dislodging cells by vortex and centrifuge, the sampled cells were labeled with 4 EC-specific antibodies (CD31, CD34, CD105, CD146), 1 myeloid-specific antibody (CD11b), and 1 platelet-specific antibody (CD42) to identify and sort single ECs. ECs were sorted based on CD31þ, CD34þ, CD105þ, CD146þ, CD11b, and CD42. Aria II fluorescence-activated cell sorting (FACS; BD Biosciences, Franklin Lakes, New Jersey) with a 100 nm nozzle in single-cell sort mode was used.
Reverse Transcription and cDNA Preamplification Each single EC was sorted directly into 1 well of a 96-well plate with 9 μL reverse transcription–specific target amplification (RT-STA) buffer. The RT-STA buffer contained 5 μL Cellsdirect 2 Reaction Mix (Thermo Fisher Scientific), 0.2 μL SuperScript III RT Platinum Taq Mix (Life Technologies, Carlsbad, California), 2.8 μL nuclease-free water, and 1 μL 10 primer mixture (Fluidigm, San Francisco, California). The primer pairs were selected based on previous nonhuman animal and in vitro studies identifying gene pathways that potentially affect AVF failure, including transforming growth factor (TGF) b–dependent pathways and pathways involved in vascular injury, repair, and thrombosis. Components of these genetic pathways with available drug inhibitors or promoters were also selected (Table E1 [on the article’s Supplemental Material page at www.jvir.org]) (5,8,10). Primers were custom designed by Fluidigm to cross introns and avoid amplifying genomic
23.7 ± 2.3 24.3 ± 2.3 66.0 ± 8.0 66.7% Right
Quantitative PCR and Gene Expression Analysis
Mean or %
60.7 ± 10.5
33.3% M
77.8% Y
55.6% Y
11.1% current
28.8 ± 10.9
DNA. The samples were incubated at 50 C for 15 minutes for reverse transcription and at 95 C for 2 minutes for reverse transcriptase inactivation and Taq polymerase activation. Specific target amplification was achieved during 18 polymerase chain reaction (PCR) cycles (95 C for 15 seconds and 60 C for 4 minutes per cycle).
BMI ¼ body mass index; HTN ¼ hypertension.
24
24
24
24
65
78
Cephalic arch Left
Right
21.2
Cephalic arch
Figure 1. Graphic depicting the endothelial biopsy technique. A “J”-shaped guidewire is advanced tip to tip with a catheter after crossing a stenosis. The catheter is withdrawn ~1 cm and the guidewire is retracted and advanced 20 times, collecting endothelial cells. The guidewire is then resheathed in the catheter, removed, and immediately placed in cell dissociation buffer.
23.7 No
No No
Yes Male
No Female
58 9
8
70
Yes
18
24
30
24
70
68
Cephalic arch
Cephalic arch
Left
Right
20
30.5
No
Past Yes
Yes
Female
Yes
Female
54 7
6
67
Yes
24
24
24
24
50
60 Right
Cephalic-forearm
Right 32 Past
Current No
No
Male
No
Female
44 5
4
66
Yes
Right
Right Left
34.2
32.9 39
Past
Past No Yes No
Yes
Male Female
Yes Yes
Female
78 51 2 3
1
59
Yes
25.5
Basilic
24 27 24 21 73 65
24 24 63 Cephalic arch
Brachiocephalic Cephalic arch
Control Stenosis Degree of Stenosis, % Site of Stenosis Side of Fistula BMI, kg/m2 Smoker Diabetes Sex
HTN
3
Age, y ID
Patient Characteristic
Table. Patient Demographics, Fistula Characteristics, and Number of Cells Sampled per Patient
Fistula Characteristic
Number of Harvested Cells
Volume ▪ ▪ Number ▪ ▪ Month ▪ 2018
The Fluidigm Biomark HD system was used for the singlecell quantitative PCR (qPCR) reaction and data collection. Each preamplified single-cell cDNA sample from above was diluted fivefold in TE Buffer; 2.25 μL was then mixed with 2.5 μL 2 Ssofast Evagreen Supermix with Low ROX (BioRad, Hercules, California) and 0.25 μL 20 DNA Binding Dye Sample Loading Reagent (Fluidigm). Five μL premixed sample was pipetted into the 48 sample inlets on the 48 48 Dynamic Array (Fluidigm). Five μL Assay Mix, composed of 2.5 μL 2 Assay Loading Reagent, 2.25 μL 1 DNA suspension buffer, and 0.25 μL of each primer pair (100 μmol/L), was then pipetted into the 48 assay inlets on the chip. The chip was loaded into the Biomark HD system (Fluidigm) and thermocycled through 35 cycles of 5 seconds at 96 C and 20 seconds at 60 C after a hot start phase of 60 seconds at 95 C. Fluorescence in the Evagreen channel was detected and collected by means of a charge-coupled device camera placed above the chip, and ROX (6-carboxy-X-rhodamine) intensity was used for normalization.
Data Processing and Statistical Analysis A series of LMMs, each with a single gene expression level (48 total) as the outcome, were performed. For each model, a number of fixed-effect predictors were used, including: cell location, smoking status (nonsmoker, past smoker, current smoker), age, sex, hypertension history,
4 ▪ Endovascular Biopsy and EC Gene Expression Analysis of Dialysis AVFs
McGregor et al ▪ JVIR
Figure 2. P value distributions for each fixed effect. The x-axis indicates P values and the y-axis the number of genes for each P value. Leftward skewing, as seen with diabetes and severe stenosis, indicates a trend toward significance.
diabetes history, degree of stenosis, and interaction between degree of stenosis and cell location. Results are reported as the genetic differences by location for mild and severe stenotic regions. Degree of stenosis was binarized at the mean stenosis percentage value (66%); stenoses < 66% were considered to be mild and those > 66% were considered to be severe. Gene expression levels were analyzed as Log2Ex units, defined as the limit of detection (set to 28) minus the gene expression cycle threshold (Ct) value (negative Log2Ex values were set to 0). Thus, each unit increase was interpreted as a twofold increase in gene expression; results were presented as fold change values.
Unsupervised Clustering To explore clustering of cells based on gene coexpression, self-organizing maps (SOMs) of the LMM residuals were used. The number of clusters in the residual data were
statistically validated with the use of the “clvalid” package in R (15–18).
Self-Organizing Maps SOMs were used for dimensional reduction and visualization of the distribution of gene weights throughout the network. In SOMs, nodes are used to build a feature map of the data. A predefined number of nodes are introduced into the dataspace with a predefined set of weights for each of the 48 genes. With the use of a competitive learning strategy, these nodes move closer to regions in the dataspace they are most similar to based on these weights. Once the distance between the nodes and their respective clusters of cells (which they are most similar to) no longer changes substantially with each additional epoch of training, the network locks in. The multidimensional data can then be represented as a network of nodes, each of which represents clusters of cells with similar genetic coexpression profiles.
Volume ▪ ▪ Number ▪ ▪ Month ▪ 2018
5
Each node was plotted as a circle with fan-plots in the circle reflecting the specific gene weights linking the similar cells within each node.
RESULTS Patients and Cell Collection Stenosis characteristics are presented in the Table. Six of 9 fistulas were right sided. Seven AVFs were brachiocephalic and 6 of 9 stenoses were located at the cephalic arch. Each patient had a single stenosis in the peripheral venous outflow. Severity of stenosis ranged from 50% to 78%, with mean stenosis of 66%. All stenoses were treated with angioplasty. One stenosis required stenting because of vessel rupture after angioplasty. Technical success was 100%. There were no immediate biopsy-related complications, including vessel dissection or vasospasm. Two hundred nineteen ECs from stenotic vessel segments and 213 control ECs were successfully collected and sorted for qPCR analysis. The fewest number of cells sampled for a site was 18 (control vein, patient 6). Cells for each patient by sampling site are listed in the Table.
Linear Mixed Models Table E1 (on the article’s Supplemental Material page at www.jvir.org) lists the fold-change coefficients and respective P values for each predictor used in the LMM of the 48 genes. Seven genes showed significant fold-change differences between stenotic and control vessel segments, including HMOX2, MTHFR, SERPINE1, SMAD4, and VWF (reduced expression in stenotic segments) and SERPINC1 and AGTR2 (increased expression in the stenotic segments). No other significant fold changes were found for the other risk factors. To examine the overall distribution of P values, bar plots (Fig 2) were created and median P values extracted for each fixed effect. The median P value for the 48 genes was 0.64 in the mild stenosis group and 0.25 in the severe stenosis group. Diabetes, the risk factor that demonstrated a left skew toward lower P values, had a median P value of 0.34.
Self-Organizing Maps A series of SOMs identified 10 outlier cells characterized by altered expression of cell type–specific validation genes (Table E1 [on the article’s Supplemental Material page at www.jvir.org]) and were considered to be nonendothelial in origin. A final model with the 10 cells removed was created. Figure 3 shows the within sum of squares for each number of clusters in the residual SOM network, indicating stability of the network at 4 cell clusters. Figure 4 shows the distribution of gene weights across the SOM network for each gene family group. Background colors indicate the clusters. Thirty-two cells, 15 control and 17 stenosis cells, linked to group 1 (pink), 35 cells, 20 control and 15 stenosis cells, linked to group 2 (green), 15 cells, 7 control and 8 stenosis cells, linked to group 3 (blue), and 340 cells, 164 control
Figure 3. Within sum of squares (wss) for each number of clusters in the self-organizing map network. Note the “elbow” at 4 clusters, which suggests stability of the network.
and 176 stenosis cells, linked to group 4 (purple). By gene family group, the SOM plot shows different gene expression patterns throughout the network, especially for those genes in the TGF-b1 family and for the EC injury family. No SOM group, clustered by gene coexpression, was composed of cells based solely on sampling location. The majority of cells clustered into group 4 (340 cells) and showed similar genetic expressivity for all 7 gene families.
DISCUSSION This study demonstrates that targeted EC biopsy from stenotic and control vessel segments in dialysis AVFs is feasible. In 9 consecutive patients, no immediate complications attributable to the EC biopsy occurred. The biopsy technique is within the skill set of interventional radiologists and was performed independently by multiple users with varying degrees of interventional experience (1–20 years) after a single instructional presentation and question-andanswer session. The device used, a Rosen guidewire, is inexpensive and ubiquitous in interventional radiology suites. The technical success rate was 100%. Adequate and consistent numbers of ECs were collected from each vessel segment and patient with only small variations in the number of cells in each group. EC identity determined by surface antigens (FACS) was confirmed by genetic expressivity (qPCR) in 98% of cells, with a total of 10 outlier cells initially sorted by FACS removed from the analysis. Differences in expression of 7 of 48 genes of interest were identified (P < .05). However, after adjusting for multiple comparisons with the Bonferroni correction (0.05/48; P ¼ .001) gene-specific over- or underexpression based on cell sampling location did not reach the corrected significance threshold, limiting the biologic impact of these findings. These results are predictable, considering the small number of cells collected and genes analyzed. Nonetheless, the 7 genes described in this study before Bonferroni correction
6 ▪ Endovascular Biopsy and EC Gene Expression Analysis of Dialysis AVFs
McGregor et al ▪ JVIR
Figure 4. Self-organizing map networks arranged by gene family. Background colors represent the 4 unique cell clusters. Distribution of gene weights are depicted by fan width within each node.
have been shown in in vitro and nonhuman animal studies to be involved in vascular disease and neointimal hyperplasia (19–24). In addition, the leftward skewing noted on the P value histogram with diabetes as the fixed effect suggests that diabetes may be a significant modifier of endothelial gene function. This is consistent with previous studies describing diabetic vasculopathy and is supported by a study demonstrating protein expression differences in diabetic and nondiabetic mice (25,26). This may have implications for future studies aimed at defining endothelial specific gene function between distinct AVF populations. These individual gene expression differences must also be viewed in a broader cellular context. With the emergence of single-cell analytics, researchers across a variety of disciplines have discovered that the traditional concept of cell identity based on morphology and immunohistochemistry oversimplifies categorization and contribution to disease. For example, Tirosh et al recently described 2 different transcriptional states within a melanoma population, each of which reflected different degrees of drug resistance (27).
The present study identified 4 different cell clusters, each containing cells from the stenotic and control locations. These results are consistent with current literature on singlecell heterogeneity and suggest that heterogeneity of EC phenotypes at specific vessel segments may contribute to stenosis formation. This provides valuable insight into the complex nature of this vascular disease on a genetic level. There are several limitations to this study. First, the small number of patients and genes studied limited the biologic significance of our findings and the number of clinical and angiographic variables included in the analysis. The multidimensionality of the data with multiple comparisons required a Bonferroni correction resulting in a significant P value of <.001. The 7 genes that met traditional significance criteria (< .05) are promising targets, but the biologic significance of the data is limited after Bonferroni correction. Future studies will require more patients and more cells collected, which will enable inclusion of additional clinical variable and computational fluid dynamic data in the analysis. Second, the technical limitation of studying only 48
Volume ▪ ▪ Number ▪ ▪ Month ▪ 2018
genes limits the biologic significance of our data. Entire functional classes of genes may be overlooked by this approach. In addition, gene expression is cyclic in nature, and the positive or negative up-regulation of certain genes may be related to the temporal phase during which they were sampled. A possible solution to this limitation is RNA sequencing, which allows for quantification of the entire transcriptome. This technique would ensure that potentially critical genes were not excluded from study by technical limitations and allow for better temporal analysis of gene expression. In summary, the present study accomplished 3 goals: (i) to prove the feasibility of a guidewire-based technique for the targeted biopsy of viable ECs from dialysis AVFs; (ii) to characterize gene expression differences between ECs collected from stenotic and control vessel segments; and (iii) to develop an analytic technique using residuals from linear mixed modeling to conduct unsupervised machine-learning statistical analysis of single-cell gene expression. Future studies should use more comprehensive techniques, such as RNAseq, to further characterize the genetic changes leading to venous outflow stenoses in AVFs.
ACKNOWLEDGMENTS This study was funded by RSNA Fellow Grant #RF1617 (McGregor, PI) and was presented at SIR 2017.
REFERENCES 1. Martinez L, Duque JC, Tabbara M, et al. Fibrotic venous remodeling and nonmaturation of arteriovenous fistulas. J Am Soc Nephrol 2018; 29: 1030–1040. 2. Lee T. Novel paradigms for dialysis vascular access: downstream vascular biology—is there a final common pathway? Clin J Am Soc Nephrol 2013; 8:2194–2201. 3. Roy-Chaudhury P, Arend L, Zhang J, et al. Neointimal hyperplasia in early arteriovenous fistula failure. Am J Kidney Dis 2007; 50:782–790. 4. Franzoni M, Cattaneo I, Longaretti L, et al. Endothelial cell activation by hemodynamic shear stress derived from arteriovenous fistula for hemodialysis access. Am J Physiol Heart Circ Physiol 2016; 310:H49–H59. 5. Kang L, Grande JP, Hillestad ML, et al. A new model of an arteriovenous fistula in chronic kidney disease in the mouse: beneficial effects of upregulated heme oxygenase1. Am J Physiol Renal Physiol 2016; 310: F466–F476. 6. Janardhanan R, Yang B, Kilari S, et al. The role of repeat administration of adventitial delivery of lentivirus–shRNA–VEGF-A in arteriovenous fistula to prevent venous stenosis formation. J Vasc Interv Radiol 2016; 27: 576–583.
7
7. Janardhanan R, Yang B, Vohra P, et al. Simvastatin reduces venous stenosis formation in a murine hemodialysis vascular access model. Kidney Int 2013; 84:338–352. 8. Misra S, Shergill U, Yang B, et al. Increased expression of HIF1alpha, VEGF-A and its receptors, MMP2, TIMP1, and ADAMTS1 at the venous stenosis of arteriovenous fistula in a mouse model with renal insufficiency. J Vasc Interv Radiol 2010; 21:1255–1261. 9. Cooke DL, Su H, Sun Z, et al. Endovascular biopsy: evaluating the feasibility of harvesting endothelial cells using detachable coils. Interv Neuroradiol 2013; 19:399–408. 10. Sun Z, Su H, Long B, et al. Endothelial cell high-enrichment from endovascular biopsy sample by laser capture microdissection and fluorescence activated cell sorting. J Biotechnol 2014; 192(Pt A):34–39. 11. Cooke DL, Bauer D, Sun Z, et al. Endovascular biopsy: Technical feasibility of novel endothelial cell harvesting devices assessed in a rabbit aneurysm model. Interv Neuroradiol 2015; 21:120–128. 12. Cooke DL, McCoy DB, Halbach VV, et al. Endovascular biopsy: in vivo cerebral aneurysm endothelial cell sampling and gene expression analysis. Transl Stroke Res 2018; 9:20–33. 13. Waldo SW, Brenner DA, McCabe JM, et al. A novel minimally-invasive method to sample human endothelial cells for molecular profiling. PLoS One 2015; 10:e0118081. 14. Feng L, Stern DM, Pile-Spellman J. Human endothelium: endovascular biopsy and molecular analysis. Radiology 1999; 212:655–664. 15. Datta S. Comparisons and validation of statistical clustering techniques for microarray gene expression data. Bioinformatics 2003; 19:459–466. 16. Datta S. Methods for evaluating clustering algorithms for gene expression data using a reference set of functional classes. BMC Bioinformatics 2006; 7:397. 17. Brock G, Pihur V, Datta S. clValid: An R package for cluster validation. J Stat Softw 2008; 25:1–22. 18. Handl J, Knowles J, Kell DB. Computational cluster validation in postgenomic data analysis. Bioinformatics 2005; 21:3201–3212. 19. Ayer A, Zarjou A, Agarwal A, et al. Heme oxygenases in cardiovascular health and disease. Physiol Rev 2016; 96:1449–1508. 20. Alhenc-Gelas M, Plu-Bureau G, Hugon-Rodin J, et al. Thrombotic risk according to SERPINC1 genotype in a large cohort of subjects with antithrombin inherited deficiency. Thromb Haemost 2017; 117: 1040–1051. 21. Mallamaci F, Bonanno G, Seminara G, et al. Hyperhomocysteinemia and arteriovenous fistula thrombosis in hemodialysis patients. Am J Kidney Dis 2005; 45:702–707. 22. Samarakoon R, Overstreet JM, Higgins PJ. TGF-b signaling in tissue fibrosis: redox controls, target genes and therapeutic opportunities. Cell Signal 2013; 25:264–268. 23. Takata H, Yamada H, Kawahito H, et al. Vascular angiotensin II type 2 receptor attenuates atherosclerosis via a kinin/NO–dependent mechanism. J Renin Angiotensin Aldosterone Syst 2015; 16:311–320. 24. Aparecida-Silva R, Borges LF, Kessler K, et al. Transforming growth factor-b1 SMAD effectors and medial cell number in ascending aorta diseases. Cardiovasc Pathol 2016; 25:240–246. 25. Beckman JA, Creager MA. Vascular complications of diabetes. Circ Res 2016; 118:1771–1785. 26. Jüllig M, Chen X, Middleditch MJ, et al. Illuminating the molecular basis of diabetic arteriopathy: a proteomic comparison of aortic tissue from diabetic and healthy rats. Proteomics 2010; 10:3367–3378. 27. Tirosh I, Izar B, Prakadan SM, et al. Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. Science 2016; 352(6282):189–196.
Symbol
Gene
Rationale
Location Location (Stenotic vs (Stenotic vs Control Site) Control Site) in Severein MildStenosis Stenosis Group Group FC
P Value
FC
P Value
Degree of Stenosis
FC
P Value
Age
FC
Sex
P Value
FC
P Value
Past Smoking Status
FC
P Value
Current Smoking Status
FC
P Value
Hypertension
FC
P Value
Diabetes
FC
P Value
Interaction Degree of Stenosis and Location
FC
P Value
ACVR1
activin A receptor type 1
TGF-b
0.136 .086
0.107 .057
0.063 .939
0.000
.990
0.371 .657
0.236 .751
0.421 .846 0.533 .654
ACVRL1
activin A receptor like type 1
TGF-b
1.316 .073
0.793 .127
1.781 .366
0.077
.371
0.061 .961
0.804 .547
3.304 .445
3.352 .257 1.600 .323 2.109 .019*
0.136 .857 0.012
.698
0.296 .650 0.296 .627
0.155 .926
0.694 .496 0.409 .513
0.111 .858
0.503 .527
0.029 .767
ADAMTS1 ADAM metallopeptidase with thrombospondin type 1 motif 1
Animal
0.287 .571
0.176 .625
AGTR1
angiotensin II receptor type 1
Drug target 0.058 .868
0.308 .217
0.773 .622
0.025
.723
1.715 .354
0.543 .679
4.450 .371
2.427 .355 2.028 .276
0.366 .396
AGTR2
angiotensin II receptor type 2
Drug target
0.919 .021*
0.975 .463
0.087
.260
4.216 .113
1.549 .269
13.511 .097
9.009 .075 4.713 .089
0.895 .193
ANGPT1
angiopoietin 1
EC injury
0.221 .707
0.122 .770
0.499 .661
0.000 1.000
0.455 .644
0.128 .880
2.186 .472
3.148 .202 3.286 .123
0.099 .891
ANGPT2
angiopoietin 2
EC injury
0.160 .657
0.375 .143
.740 0.282 .562 0.577 .319
0.366 .764
CCL2
C-C motif chemokine ligand 2
EC injury
0.316 .061
0.063 .597
0.716 .628
0.011
CD34
CD34 molecule
Validation
0.145 .778
0.354 .330
0.508 .618
0.016
ENG
endoglin
Validation
1.123 .121
0.875 .089
EPHB2
ephrin type-B receptor 2
Validation
0.157 .644
0.208 .389
0.091 .858
0.016
EPHB4
ephrin type-B receptor 4
Validation
0.594 .067
0.368 .109
0.456 .740
0.039
0.607 .610
0.225 .570
ETS1
ETS proto-oncogene 1
Animal
0.286 .401
0.104 .668
1.022 .580 0.102
.347 3.841 .198 0.597 .692
6.886 .291 0.483 .828 0.697 .637
0.182 .663
F2
thrombin
Thrombosis
0.116 .756
0.449 .091
0.509 .612 0.016
.715 0.590 .534 0.150 .846
0.536 .816
GDF2
growth differentiation factor 2
TGF-b
0.657 .280
0.541 .210
1.150 .353
0.038
.418
0.308 .667
4.982 .197
2.138 .233 1.186 .259
1.198 .109
HIF1A
hypoxia inducible factor 1 alpha
Animal
0.208 .481
0.034 .870
0.399 .455
0.024
.341 0.047 .895 0.321 .432
1.211 .362
1.486 .167 0.754 .202
0.174 .631
HMOX1
heme oxygenase 1
Animal
0.277 .630
0.130 .750
0.415 .741 0.019
.736 0.083 .938
0.909 .453
0.751 .802
0.639 .690 0.909 .439 0.147 .835
HMOX2
heme oxygenase 2
Animal
0.926 .027*
0.531 .074
1.453 .724 0.072
.714 1.275 .741
0.034 .992
0.721 .944
1.174 .826 2.219 .551
1.457 .005*
IL1B
interleukin 1 beta
EC injury
0.027 .941
0.092 .723
2.936 .341
0.152
.316
4.501 .222
3.135 .286
16.106 .174
7.650 .188 5.029 .178
0.119 .790
IL6
interleukin 6
EC injury
0.444 .341
0.515 .120
0.134 .851
0.044
.304
0.087 .881
0.382 .532
0.964 .584
1.457 .266 1.289 .193 0.959 .094
KLF2
Kruppel like factor 2
EC injury
0.096 .773
0.001 .996
2.740 .090
0.055
.178
2.109 .095
1.709 .107
7.602 .072
2.224 .126 1.662 .106
LRP1
LDL receptor related protein 1
EC injury
0.126 .837
0.735 .091
2.691 .673 0.160
.608 1.401 .808
0.029 .996
MAPK7
mitogen-activated protein kinase 7
TGF-b
0.135 .648
0.045 .830
0.145 .822
.724
0.128 .808
MAPK8
mitogen-activated protein kinase 8
TGF-b
0.261 .570
0.041 .899
0.241 .800 0.011
.789 0.325 .707 0.049 .949
1.293 .603 0.176 .880 0.083 .909 0.303 .592
MCAM
melanoma cell adhesion molecule
Validation
0.033 .956
0.581 .179
1.490 .321 0.030
.548 1.715 .246 0.159 .846
1.499 .578
MMP3
matrix metallopeptidase 3
EC injury
0.060 .703
0.079 .484
0.315 .720 0.023
.605 0.589 .521 0.364 .644
1.285 .593 0.113 .920
0.025 .965
0.097 .856 0.007
0.267 .820 0.005
0.031 .978
.713 0.075 .926 0.142 .849
0.821 .722
0.546 .654
0.581 .490 0.209 .739
.912 0.402 .693
0.861 .438
1.823 .546
2.666 .246 1.656 .247
.504 0.027 .947 0.408 .402
0.413 .724
.571
0.149 .904
2.508 .144
0.123 .829
0.008 .994
0.882 .306
0.600 .861 0.118 .946
0.253 .221
0.248 .780
0.019 .957 0.051 .903
0.798 .549 1.081 .313 0.565 .218
0.095 .816
5.471 .735 1.862 .819 1.888 .717 0.860 .252 0.534 .739
0.679 .478 0.473 .442 0.181 .620
2.432 .246 2.839 .136 0.547 .463 0.290 .694
0.139 .474 continued
McGregor et al ▪ JVIR
0.010
0.066 .863 0.535 .227
0.442 .900 0.524 .779
.866
0.195 .880 0.079 .947
0.406 .556
7.e1 ▪ Endovascular Biopsy and EC Gene Expression Analysis of Dialysis AVFs
Table E1. Gene Targets, Rationale, and Fold-Change Coefficient for Each Fixed Effect
Symbol
Gene
Rationale
Location Location (Stenotic vs (Stenotic vs Control Site) Control Site) in Severein MildStenosis Stenosis Group Group FC
MTHFR
methylenetetrahydrofolate reductase NOS3 nitric oxide synthase 3 PDGFB platelet derived growth factor subunit B PECAM1 platelet and endothelial cell adhesion molecule 1 PIK3CA phosphatidylinositol4,5bisphosphate 3-kinase catalytic subunit alpha PTEN phosphatase and tensin homolog PTGS1 prostaglandin-endoperoxide synthase 1 PTGS2 prostaglandin-endoperoxide synthase 2 PTPRC protein tyrosine phosphatase, receptor type C RASA1 RAS p21 protein activator 1 SERPINC1 serpin family C member 1 SERPINE1 serpin family E member 1 SMAD1 SMAD family member 1 SMAD2 SMAD family member 2 SMAD4 SMAD family member 4 TEK TEK receptor tyrosine kinase TGFB1 transforming growth factor beta 1 THBD thrombomodulin TNF tumor necrosis factor VEGFA vascular endothelial growth factor A VWF von Willebrand factor
P Value
FC
P Value
Degree of Stenosis
FC
P Value
Age
FC
Sex
P Value
FC
Past Smoking Status
P Value
FC
P Value
Current Smoking Status
FC
P Value
Hypertension
FC
P Value
Diabetes
FC
P Value
Interaction Degree of Stenosis and Location
FC
P Value
Thrombosis
0.113 .845
0.833 .042*
0.628 .782 0.059
.601 1.384 .554
0.551 .777
1.375 .812
0.745 .803 1.018 .609 0.946 .182
EC Injury Drug target
0.843 .165 0.130 .760
0.153 .722 0.442 .145
1.250 .549 0.165 .797
0.057 0.067
.550 0.409 .811 .188 0.193 .718
0.430 .786 0.042 .928
2.161 .662 1.637 .381
2.162 .458 2.024 .336 0.690 .354 1.156 .294 0.476 .411 0.573 .275
0.184 .825
0.686 .246
0.149 .904 0.056
.392 1.979 .242
0.169 .855
TGF-b
0.615 .377
0.443 .370
2.204 .300 0.120
.250 6.701 .093 1.315 .386 11.175 .152
0.582 .749
Cell cycle
1.162 .108
0.656 .202
0.823 .773 0.058
.674 1.803 .542 0.996 .692
4.053 .604
1.859 .638 2.089 .455 1.818 .041*
Drug target
0.005 .963
0.023 .771
0.918 .117 0.008
.470 1.005 .093
0.023 .895
0.792 .302
1.385 .096 1.415 .059 0.029 .837
EC injury
0.305 .647
0.182 .700
.468 1.149 .321
0.144 .846
2.976 .337
EC injury
0.073 .900
0.470 .256
Drug target Thrombosis EC Injury TGF-b TGF-b TGF-b Validation TGF-b
0.170 1.421 0.010 0.646 0.603 0.007 0.048 0.107
0.738 0.746 0.855 0.698 0.864 0.631 0.408 0.007
Validation EC injury Animal
0.188 .430 0.189 .705 0.643 .332
Validation
.763 .024* .984 .347 .353 .987 .946 .553
Thrombosis 0.797 .327
0.616 .574
0.035
1.429 .745 0.200
.067 0.399 .655 0.030 .094 5.180 .158 0.173 .017* 0.222 .772 0.041 .153 1.296 .354 0.076 .061 0.193 .842 0.096 .045* 0.244 .898 0.071 .423 1.225 .383 0.004 .955 0.522 .474 0.040
0.018 .916 0.408 .251 0.408 .387 1.479 .010*
1.923 .539 0.101 .943
1.294 2.444 0.300 0.990 2.726 2.275 0.411 1.119
.251 0.589 .446 .288 3.629 .187 .645 0.343 .580 .372 1.769 .209 .142 1.283 .266 .354 0.470 .780 .655 0.824 .415 .240 0.072 .889
1.705 .169 0.079 .941 0.902 .568
0.050 0.016 0.085
.255 .749 .337
2.435 .108 1.150 .206 0.034 .971 0.687 .505 0.388 .762 1.322 .384
0.018 .988
0.099
.239
1.269 .349
0.293 .752
1.618 14.785 1.471 0.429 4.318 4.705 2.308 1.547
0.236 .832 1.058 .215
4.646 .121 3.403 .103
.429 5.958 .313 1.473 .704 12.300 .394 1.213 .833 .466 .211 .341 .261 .198 .500 .925 .325
0.671 .512 0.502 .623
Volume ▪ ▪ Number ▪ ▪ Month ▪ 2018
Table E1. Gene Targets, Rationale, and Fold-Change Coefficient for Each Fixed Effect (continued)
0.123 .880
0.372 .916 0.396 .578
.472 0.264 .788 0.473 .138 5.621 .185 4.220 .466 0.766 .462 0.029 .851 3.635 .159 4.401 .239 1.449 .346 2.066 .440 0.487 .848 0.359 .434 1.319 .402 2.049 .425 0.649 .490 0.712
.501 .155 .957 .083 .164 .821 .182 .320
0.567 0.675 0.866 0.052 0.261 0.638 0.360 0.114
.414 .380 .163 .951 .744 .240 .682 .606
7.203 .101 0.031 .991 7.629 .218
2.206 .167 1.243 .183 0.170 .561 1.912 .325 1.523 .263 0.219 .721 6.964 .126 5.093 .108 1.050 .197
3.565 .342
3.064 .219 1.978 .211 0.682 .494
The interaction term compares the differences in expression levels of stenotic cells and control cells between high-stenosis areas and low-stenosis areas. EC, endothelial cell; FC, fold change; TGF ¼ transforming growth factor. *P < .05.
7.e2
Endovascular Biopsy and Endothelial Cell Gene Expression Analysis of Dialysis Arteriovenous Fistulas: A Feasibility Study Hugh McGregor, MD, Zhengda Sun, PhD, David McCoy, MS, Vishal Kumar, MD, Miles Conrad, MD, Mark Wilson, MD, and Daniel Cooke, MD ABSTRACT Purpose: To demonstrate feasibility of endothelial cell (EC) biopsy from dialysis arteriovenous fistulas (AVFs) with the use of guidewires and to characterize gene expression differences between ECs from stenotic and nonstenotic outflow vein segments. Materials and Methods: Nine consecutive patients undergoing fistulography for AVF dysfunction from June to August 2016 were enrolled. ECs were biopsied with the use of guidewires from venous outflow stenoses and control outflow veins central to the stenoses. ECs were sorted with the use of flow cytometry, and the Fluidigm Biomark HD system was used for single-cell quantitative polymerase chain reaction (qPCR) analysis of gene expression. Forty-eight genes were assessed and were selected based on different cellular functions and previous literature. Linear mixed models (LMMs) were used to identify differential gene expression between the groups, and self-organizing maps (SOMs) were used to identify cell clusters based on gene coexpression profiles. Results: A total of 219 and 213 ECs were sampled from venous outflow stenoses and control vein segments, respectively. There were no immediate biopsy-related complications. Forty-eight cells per patient were sorted for qPCR analysis. LMM identified 7 genes with different levels of expression at stenotic segments (P < .05), including AGTR2, HMOX2, MTHFR, SERPINC1, SERPINE1, SMAD4, and VWF. SOM analysis identified 4 cell clusters with unique gene expression profiles, each containing stenotic and control ECs. Conclusions: EC biopsy from dialysis AVFs with the use of guidewires is feasible. Gene expression data suggest that genes involved in multiple cellular functions are dysregulated in stenotic areas. SOMs identified 4 unique clusters of cells, indicating EC phenotypic heterogeneity in outflow veins.
ABBREVIATIONS AVF ¼ arteriovenous fistula, EC ¼ endothelial cell, LMM ¼ linear mixed models, qPCR ¼ quantitative polymerase chain reaction, SOM ¼ self-organizing map
From the Department of Radiology and Biomedical Imaging, University of California, 505 Parnassus Avenue, M391, San Francisco, California 941430628. Received January 2, 2018; final revision received April 10, 2018; accepted April 22, 2018. Address correspondence to H.M.; E-mail: [email protected] H.M. and Z.S. contributed equally to this work. Table E1 can be found by accessing the online version of this article on www. jvir.org and clicking on the Supplemental Material tab. None of the authors have identified of conflict of interest. © SIR, 2018 J Vasc Interv Radiol 2018; ▪:1–7 https://doi.org/10.1016/j.jvir.2018.04.034
Venous outflow stenosis because of neointimal hyperplasia and medial fibrosis accounts for the majority of dialysis arteriovenous fistulas (AVFs) requiring intervention to maintain patency (1). The histologic hallmark of neointimal hyperplasia is smooth muscle cell and myofibroblast migration and proliferation within the neointima (2). The mechanisms mitigating this process are not well understood, with current knowledge largely based on immunohistochemical analysis of resected surgical specimens (3) and genetic analysis of in vitro (4) and nonhuman animal models (5–8). The lack of a safe and reliable means for cellular analysis of AVFs in humans has limited the genetic characterization of AVF failure. Consequently, the ability to predict AVF
2 ▪ Endovascular Biopsy and EC Gene Expression Analysis of Dialysis AVFs
EDITORS’ RESEARCH HIGHLIGHTS The authors demonstrated that a 1.5-mm-diameter J-type guidewire successfully harvested endothelial cells (ECs) from within stenotic dialysis AV fistulas for gene characterization. A number of dysregulated genes, ones not found in normal fistula segments, were identified. This suggests that heterogeneity of EC phenotypes at specific vessel segments may play a role in stenosis formation. The findings also suggested that diabetes was a possible modifier of endothelial gene function.
failure and identify pharmacologic therapies that may prevent or slow venous outflow stenosis are limited. To address this limitation, the present study expands on previous endovascular techniques (9–14) for the targeted biopsy of endothelial cells (ECs) in vivo in dialysis AVFs with the use of endovascular devices. The objectives of this study were: (i) to demonstrate feasibility of EC biopsy in dialysis AVFs with the use of guidewires; (ii) to characterize gene expressivity differences between ECs from stenotic and nonstenotic outflow vein segments; and (iii) to develop an analytic technique using linear mixed models (LMMs) to conduct unsupervised machine-learning statistical analysis of single-cell gene expression.
MATERIALS AND METHODS The study protocol was approved by the Institutional Review Board.
Patients Ten consecutive patients on hemodialysis referred to the interventional radiology service for fistulography from June to August 2016 were enrolled and consented to participate in this study. One patient was excluded from the final analysis owing to the presence of an AV graft. Inclusion criteria included end-stage renal disease, a dialysis AVF with a peripheral venous outflow stenosis, and consent for tissue analysis. Patient demographics and clinical data were obtained from the most recent primary care clinic notes in the electronic medical record (Table). Three patients were male. The mean patient age was 61 years (± 11 years). Seven patients had hypertension, 5 were diabetic, 1 was a current smoker, and 4 were past smokers. The mean BMI was 28.8 kg/m2 (± 10.9). Severity of stenosis was quantified from images saved in our Picture Archiving and Communication System. Percentage of stenosis was calculated by dividing the diameter of the stenosis by the mean of measurements taken from nonstenotic vessel segments immediately proximal and distal to the stenosis.
Endothelial Biopsy A 6-F sheath (Cordis, Milpitas, California) was placed in the peripheral venous outflow and fistulography was performed.
McGregor et al ▪ JVIR
After identification of a peripheral venous outflow stenosis, a 0.035-inch Glidewire (Terumo, Tokyo, Japan) was advanced central to the stenosis. The stenosis was then crossed with a 5-F angled catheter (Kumpe; Cook Medical, Bloomington, Indiana) and the Glidewire was removed. A 0.035-inch “J”-tipped guidewire (Rosen; Cook Medical) was then advanced to the distal aspect of the catheter (Fig 1). The catheter was withdrawn to expose the distal 1 cm of the guidewire, which was then advanced and retracted within the stenosis at 1-cm intervals ~20 times. The “J” tip of the wire opened relative to the degree of the stenosis. The guidewire was removed and the distal 5 cm was cut and immediately placed in a 50-mL tube (Falcon; Thermo Fisher Scientific, Waltham, Massachusetts) containing 10 mL EC dissociation buffer (Gibco; Thermo Fisher Scientific, Waltham, Massachusetts). Control samples from angiographically normal venous outflow vein central to the stenosis were obtained in a similar fashion. Care was taken during both experimental and control biopsies to ensure contact between the wire and the target vessel only. The sampling wires were otherwise sheathed within the delivery catheter.
Endothelial Sorting Endothelial cells were sorted according to the protocol described in detail by Sun et al (10). Forty-eight cells were sorted from each patient. Briefly, after dislodging cells by vortex and centrifuge, the sampled cells were labeled with 4 EC-specific antibodies (CD31, CD34, CD105, CD146), 1 myeloid-specific antibody (CD11b), and 1 platelet-specific antibody (CD42) to identify and sort single ECs. ECs were sorted based on CD31þ, CD34þ, CD105þ, CD146þ, CD11b, and CD42. Aria II fluorescence-activated cell sorting (FACS; BD Biosciences, Franklin Lakes, New Jersey) with a 100 nm nozzle in single-cell sort mode was used.
Reverse Transcription and cDNA Preamplification Each single EC was sorted directly into 1 well of a 96-well plate with 9 μL reverse transcription–specific target amplification (RT-STA) buffer. The RT-STA buffer contained 5 μL Cellsdirect 2 Reaction Mix (Thermo Fisher Scientific), 0.2 μL SuperScript III RT Platinum Taq Mix (Life Technologies, Carlsbad, California), 2.8 μL nuclease-free water, and 1 μL 10 primer mixture (Fluidigm, San Francisco, California). The primer pairs were selected based on previous nonhuman animal and in vitro studies identifying gene pathways that potentially affect AVF failure, including transforming growth factor (TGF) b–dependent pathways and pathways involved in vascular injury, repair, and thrombosis. Components of these genetic pathways with available drug inhibitors or promoters were also selected (Table E1 [on the article’s Supplemental Material page at www.jvir.org]) (5,8,10). Primers were custom designed by Fluidigm to cross introns and avoid amplifying genomic
23.7 ± 2.3 24.3 ± 2.3 66.0 ± 8.0 66.7% Right
Quantitative PCR and Gene Expression Analysis
Mean or %
60.7 ± 10.5
33.3% M
77.8% Y
55.6% Y
11.1% current
28.8 ± 10.9
DNA. The samples were incubated at 50 C for 15 minutes for reverse transcription and at 95 C for 2 minutes for reverse transcriptase inactivation and Taq polymerase activation. Specific target amplification was achieved during 18 polymerase chain reaction (PCR) cycles (95 C for 15 seconds and 60 C for 4 minutes per cycle).
BMI ¼ body mass index; HTN ¼ hypertension.
24
24
24
24
65
78
Cephalic arch Left
Right
21.2
Cephalic arch
Figure 1. Graphic depicting the endothelial biopsy technique. A “J”-shaped guidewire is advanced tip to tip with a catheter after crossing a stenosis. The catheter is withdrawn ~1 cm and the guidewire is retracted and advanced 20 times, collecting endothelial cells. The guidewire is then resheathed in the catheter, removed, and immediately placed in cell dissociation buffer.
23.7 No
No No
Yes Male
No Female
58 9
8
70
Yes
18
24
30
24
70
68
Cephalic arch
Cephalic arch
Left
Right
20
30.5
No
Past Yes
Yes
Female
Yes
Female
54 7
6
67
Yes
24
24
24
24
50
60 Right
Cephalic-forearm
Right 32 Past
Current No
No
Male
No
Female
44 5
4
66
Yes
Right
Right Left
34.2
32.9 39
Past
Past No Yes No
Yes
Male Female
Yes Yes
Female
78 51 2 3
1
59
Yes
25.5
Basilic
24 27 24 21 73 65
24 24 63 Cephalic arch
Brachiocephalic Cephalic arch
Control Stenosis Degree of Stenosis, % Site of Stenosis Side of Fistula BMI, kg/m2 Smoker Diabetes Sex
HTN
3
Age, y ID
Patient Characteristic
Table. Patient Demographics, Fistula Characteristics, and Number of Cells Sampled per Patient
Fistula Characteristic
Number of Harvested Cells
Volume ▪ ▪ Number ▪ ▪ Month ▪ 2018
The Fluidigm Biomark HD system was used for the singlecell quantitative PCR (qPCR) reaction and data collection. Each preamplified single-cell cDNA sample from above was diluted fivefold in TE Buffer; 2.25 μL was then mixed with 2.5 μL 2 Ssofast Evagreen Supermix with Low ROX (BioRad, Hercules, California) and 0.25 μL 20 DNA Binding Dye Sample Loading Reagent (Fluidigm). Five μL premixed sample was pipetted into the 48 sample inlets on the 48 48 Dynamic Array (Fluidigm). Five μL Assay Mix, composed of 2.5 μL 2 Assay Loading Reagent, 2.25 μL 1 DNA suspension buffer, and 0.25 μL of each primer pair (100 μmol/L), was then pipetted into the 48 assay inlets on the chip. The chip was loaded into the Biomark HD system (Fluidigm) and thermocycled through 35 cycles of 5 seconds at 96 C and 20 seconds at 60 C after a hot start phase of 60 seconds at 95 C. Fluorescence in the Evagreen channel was detected and collected by means of a charge-coupled device camera placed above the chip, and ROX (6-carboxy-X-rhodamine) intensity was used for normalization.
Data Processing and Statistical Analysis A series of LMMs, each with a single gene expression level (48 total) as the outcome, were performed. For each model, a number of fixed-effect predictors were used, including: cell location, smoking status (nonsmoker, past smoker, current smoker), age, sex, hypertension history,
4 ▪ Endovascular Biopsy and EC Gene Expression Analysis of Dialysis AVFs
McGregor et al ▪ JVIR
Figure 2. P value distributions for each fixed effect. The x-axis indicates P values and the y-axis the number of genes for each P value. Leftward skewing, as seen with diabetes and severe stenosis, indicates a trend toward significance.
diabetes history, degree of stenosis, and interaction between degree of stenosis and cell location. Results are reported as the genetic differences by location for mild and severe stenotic regions. Degree of stenosis was binarized at the mean stenosis percentage value (66%); stenoses < 66% were considered to be mild and those > 66% were considered to be severe. Gene expression levels were analyzed as Log2Ex units, defined as the limit of detection (set to 28) minus the gene expression cycle threshold (Ct) value (negative Log2Ex values were set to 0). Thus, each unit increase was interpreted as a twofold increase in gene expression; results were presented as fold change values.
Unsupervised Clustering To explore clustering of cells based on gene coexpression, self-organizing maps (SOMs) of the LMM residuals were used. The number of clusters in the residual data were
statistically validated with the use of the “clvalid” package in R (15–18).
Self-Organizing Maps SOMs were used for dimensional reduction and visualization of the distribution of gene weights throughout the network. In SOMs, nodes are used to build a feature map of the data. A predefined number of nodes are introduced into the dataspace with a predefined set of weights for each of the 48 genes. With the use of a competitive learning strategy, these nodes move closer to regions in the dataspace they are most similar to based on these weights. Once the distance between the nodes and their respective clusters of cells (which they are most similar to) no longer changes substantially with each additional epoch of training, the network locks in. The multidimensional data can then be represented as a network of nodes, each of which represents clusters of cells with similar genetic coexpression profiles.
Volume ▪ ▪ Number ▪ ▪ Month ▪ 2018
5
Each node was plotted as a circle with fan-plots in the circle reflecting the specific gene weights linking the similar cells within each node.
RESULTS Patients and Cell Collection Stenosis characteristics are presented in the Table. Six of 9 fistulas were right sided. Seven AVFs were brachiocephalic and 6 of 9 stenoses were located at the cephalic arch. Each patient had a single stenosis in the peripheral venous outflow. Severity of stenosis ranged from 50% to 78%, with mean stenosis of 66%. All stenoses were treated with angioplasty. One stenosis required stenting because of vessel rupture after angioplasty. Technical success was 100%. There were no immediate biopsy-related complications, including vessel dissection or vasospasm. Two hundred nineteen ECs from stenotic vessel segments and 213 control ECs were successfully collected and sorted for qPCR analysis. The fewest number of cells sampled for a site was 18 (control vein, patient 6). Cells for each patient by sampling site are listed in the Table.
Linear Mixed Models Table E1 (on the article’s Supplemental Material page at www.jvir.org) lists the fold-change coefficients and respective P values for each predictor used in the LMM of the 48 genes. Seven genes showed significant fold-change differences between stenotic and control vessel segments, including HMOX2, MTHFR, SERPINE1, SMAD4, and VWF (reduced expression in stenotic segments) and SERPINC1 and AGTR2 (increased expression in the stenotic segments). No other significant fold changes were found for the other risk factors. To examine the overall distribution of P values, bar plots (Fig 2) were created and median P values extracted for each fixed effect. The median P value for the 48 genes was 0.64 in the mild stenosis group and 0.25 in the severe stenosis group. Diabetes, the risk factor that demonstrated a left skew toward lower P values, had a median P value of 0.34.
Self-Organizing Maps A series of SOMs identified 10 outlier cells characterized by altered expression of cell type–specific validation genes (Table E1 [on the article’s Supplemental Material page at www.jvir.org]) and were considered to be nonendothelial in origin. A final model with the 10 cells removed was created. Figure 3 shows the within sum of squares for each number of clusters in the residual SOM network, indicating stability of the network at 4 cell clusters. Figure 4 shows the distribution of gene weights across the SOM network for each gene family group. Background colors indicate the clusters. Thirty-two cells, 15 control and 17 stenosis cells, linked to group 1 (pink), 35 cells, 20 control and 15 stenosis cells, linked to group 2 (green), 15 cells, 7 control and 8 stenosis cells, linked to group 3 (blue), and 340 cells, 164 control
Figure 3. Within sum of squares (wss) for each number of clusters in the self-organizing map network. Note the “elbow” at 4 clusters, which suggests stability of the network.
and 176 stenosis cells, linked to group 4 (purple). By gene family group, the SOM plot shows different gene expression patterns throughout the network, especially for those genes in the TGF-b1 family and for the EC injury family. No SOM group, clustered by gene coexpression, was composed of cells based solely on sampling location. The majority of cells clustered into group 4 (340 cells) and showed similar genetic expressivity for all 7 gene families.
DISCUSSION This study demonstrates that targeted EC biopsy from stenotic and control vessel segments in dialysis AVFs is feasible. In 9 consecutive patients, no immediate complications attributable to the EC biopsy occurred. The biopsy technique is within the skill set of interventional radiologists and was performed independently by multiple users with varying degrees of interventional experience (1–20 years) after a single instructional presentation and question-andanswer session. The device used, a Rosen guidewire, is inexpensive and ubiquitous in interventional radiology suites. The technical success rate was 100%. Adequate and consistent numbers of ECs were collected from each vessel segment and patient with only small variations in the number of cells in each group. EC identity determined by surface antigens (FACS) was confirmed by genetic expressivity (qPCR) in 98% of cells, with a total of 10 outlier cells initially sorted by FACS removed from the analysis. Differences in expression of 7 of 48 genes of interest were identified (P < .05). However, after adjusting for multiple comparisons with the Bonferroni correction (0.05/48; P ¼ .001) gene-specific over- or underexpression based on cell sampling location did not reach the corrected significance threshold, limiting the biologic impact of these findings. These results are predictable, considering the small number of cells collected and genes analyzed. Nonetheless, the 7 genes described in this study before Bonferroni correction
6 ▪ Endovascular Biopsy and EC Gene Expression Analysis of Dialysis AVFs
McGregor et al ▪ JVIR
Figure 4. Self-organizing map networks arranged by gene family. Background colors represent the 4 unique cell clusters. Distribution of gene weights are depicted by fan width within each node.
have been shown in in vitro and nonhuman animal studies to be involved in vascular disease and neointimal hyperplasia (19–24). In addition, the leftward skewing noted on the P value histogram with diabetes as the fixed effect suggests that diabetes may be a significant modifier of endothelial gene function. This is consistent with previous studies describing diabetic vasculopathy and is supported by a study demonstrating protein expression differences in diabetic and nondiabetic mice (25,26). This may have implications for future studies aimed at defining endothelial specific gene function between distinct AVF populations. These individual gene expression differences must also be viewed in a broader cellular context. With the emergence of single-cell analytics, researchers across a variety of disciplines have discovered that the traditional concept of cell identity based on morphology and immunohistochemistry oversimplifies categorization and contribution to disease. For example, Tirosh et al recently described 2 different transcriptional states within a melanoma population, each of which reflected different degrees of drug resistance (27).
The present study identified 4 different cell clusters, each containing cells from the stenotic and control locations. These results are consistent with current literature on singlecell heterogeneity and suggest that heterogeneity of EC phenotypes at specific vessel segments may contribute to stenosis formation. This provides valuable insight into the complex nature of this vascular disease on a genetic level. There are several limitations to this study. First, the small number of patients and genes studied limited the biologic significance of our findings and the number of clinical and angiographic variables included in the analysis. The multidimensionality of the data with multiple comparisons required a Bonferroni correction resulting in a significant P value of <.001. The 7 genes that met traditional significance criteria (< .05) are promising targets, but the biologic significance of the data is limited after Bonferroni correction. Future studies will require more patients and more cells collected, which will enable inclusion of additional clinical variable and computational fluid dynamic data in the analysis. Second, the technical limitation of studying only 48
Volume ▪ ▪ Number ▪ ▪ Month ▪ 2018
genes limits the biologic significance of our data. Entire functional classes of genes may be overlooked by this approach. In addition, gene expression is cyclic in nature, and the positive or negative up-regulation of certain genes may be related to the temporal phase during which they were sampled. A possible solution to this limitation is RNA sequencing, which allows for quantification of the entire transcriptome. This technique would ensure that potentially critical genes were not excluded from study by technical limitations and allow for better temporal analysis of gene expression. In summary, the present study accomplished 3 goals: (i) to prove the feasibility of a guidewire-based technique for the targeted biopsy of viable ECs from dialysis AVFs; (ii) to characterize gene expression differences between ECs collected from stenotic and control vessel segments; and (iii) to develop an analytic technique using residuals from linear mixed modeling to conduct unsupervised machine-learning statistical analysis of single-cell gene expression. Future studies should use more comprehensive techniques, such as RNAseq, to further characterize the genetic changes leading to venous outflow stenoses in AVFs.
ACKNOWLEDGMENTS This study was funded by RSNA Fellow Grant #RF1617 (McGregor, PI) and was presented at SIR 2017.
REFERENCES 1. Martinez L, Duque JC, Tabbara M, et al. Fibrotic venous remodeling and nonmaturation of arteriovenous fistulas. J Am Soc Nephrol 2018; 29: 1030–1040. 2. Lee T. Novel paradigms for dialysis vascular access: downstream vascular biology—is there a final common pathway? Clin J Am Soc Nephrol 2013; 8:2194–2201. 3. Roy-Chaudhury P, Arend L, Zhang J, et al. Neointimal hyperplasia in early arteriovenous fistula failure. Am J Kidney Dis 2007; 50:782–790. 4. Franzoni M, Cattaneo I, Longaretti L, et al. Endothelial cell activation by hemodynamic shear stress derived from arteriovenous fistula for hemodialysis access. Am J Physiol Heart Circ Physiol 2016; 310:H49–H59. 5. Kang L, Grande JP, Hillestad ML, et al. A new model of an arteriovenous fistula in chronic kidney disease in the mouse: beneficial effects of upregulated heme oxygenase1. Am J Physiol Renal Physiol 2016; 310: F466–F476. 6. Janardhanan R, Yang B, Kilari S, et al. The role of repeat administration of adventitial delivery of lentivirus–shRNA–VEGF-A in arteriovenous fistula to prevent venous stenosis formation. J Vasc Interv Radiol 2016; 27: 576–583.
7
7. Janardhanan R, Yang B, Vohra P, et al. Simvastatin reduces venous stenosis formation in a murine hemodialysis vascular access model. Kidney Int 2013; 84:338–352. 8. Misra S, Shergill U, Yang B, et al. Increased expression of HIF1alpha, VEGF-A and its receptors, MMP2, TIMP1, and ADAMTS1 at the venous stenosis of arteriovenous fistula in a mouse model with renal insufficiency. J Vasc Interv Radiol 2010; 21:1255–1261. 9. Cooke DL, Su H, Sun Z, et al. Endovascular biopsy: evaluating the feasibility of harvesting endothelial cells using detachable coils. Interv Neuroradiol 2013; 19:399–408. 10. Sun Z, Su H, Long B, et al. Endothelial cell high-enrichment from endovascular biopsy sample by laser capture microdissection and fluorescence activated cell sorting. J Biotechnol 2014; 192(Pt A):34–39. 11. Cooke DL, Bauer D, Sun Z, et al. Endovascular biopsy: Technical feasibility of novel endothelial cell harvesting devices assessed in a rabbit aneurysm model. Interv Neuroradiol 2015; 21:120–128. 12. Cooke DL, McCoy DB, Halbach VV, et al. Endovascular biopsy: in vivo cerebral aneurysm endothelial cell sampling and gene expression analysis. Transl Stroke Res 2018; 9:20–33. 13. Waldo SW, Brenner DA, McCabe JM, et al. A novel minimally-invasive method to sample human endothelial cells for molecular profiling. PLoS One 2015; 10:e0118081. 14. Feng L, Stern DM, Pile-Spellman J. Human endothelium: endovascular biopsy and molecular analysis. Radiology 1999; 212:655–664. 15. Datta S. Comparisons and validation of statistical clustering techniques for microarray gene expression data. Bioinformatics 2003; 19:459–466. 16. Datta S. Methods for evaluating clustering algorithms for gene expression data using a reference set of functional classes. BMC Bioinformatics 2006; 7:397. 17. Brock G, Pihur V, Datta S. clValid: An R package for cluster validation. J Stat Softw 2008; 25:1–22. 18. Handl J, Knowles J, Kell DB. Computational cluster validation in postgenomic data analysis. Bioinformatics 2005; 21:3201–3212. 19. Ayer A, Zarjou A, Agarwal A, et al. Heme oxygenases in cardiovascular health and disease. Physiol Rev 2016; 96:1449–1508. 20. Alhenc-Gelas M, Plu-Bureau G, Hugon-Rodin J, et al. Thrombotic risk according to SERPINC1 genotype in a large cohort of subjects with antithrombin inherited deficiency. Thromb Haemost 2017; 117: 1040–1051. 21. Mallamaci F, Bonanno G, Seminara G, et al. Hyperhomocysteinemia and arteriovenous fistula thrombosis in hemodialysis patients. Am J Kidney Dis 2005; 45:702–707. 22. Samarakoon R, Overstreet JM, Higgins PJ. TGF-b signaling in tissue fibrosis: redox controls, target genes and therapeutic opportunities. Cell Signal 2013; 25:264–268. 23. Takata H, Yamada H, Kawahito H, et al. Vascular angiotensin II type 2 receptor attenuates atherosclerosis via a kinin/NO–dependent mechanism. J Renin Angiotensin Aldosterone Syst 2015; 16:311–320. 24. Aparecida-Silva R, Borges LF, Kessler K, et al. Transforming growth factor-b1 SMAD effectors and medial cell number in ascending aorta diseases. Cardiovasc Pathol 2016; 25:240–246. 25. Beckman JA, Creager MA. Vascular complications of diabetes. Circ Res 2016; 118:1771–1785. 26. Jüllig M, Chen X, Middleditch MJ, et al. Illuminating the molecular basis of diabetic arteriopathy: a proteomic comparison of aortic tissue from diabetic and healthy rats. Proteomics 2010; 10:3367–3378. 27. Tirosh I, Izar B, Prakadan SM, et al. Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. Science 2016; 352(6282):189–196.
Symbol
Gene
Rationale
Location Location (Stenotic vs (Stenotic vs Control Site) Control Site) in Severein MildStenosis Stenosis Group Group FC
P Value
FC
P Value
Degree of Stenosis
FC
P Value
Age
FC
Sex
P Value
FC
P Value
Past Smoking Status
FC
P Value
Current Smoking Status
FC
P Value
Hypertension
FC
P Value
Diabetes
FC
P Value
Interaction Degree of Stenosis and Location
FC
P Value
ACVR1
activin A receptor type 1
TGF-b
0.136 .086
0.107 .057
0.063 .939
0.000
.990
0.371 .657
0.236 .751
0.421 .846 0.533 .654
ACVRL1
activin A receptor like type 1
TGF-b
1.316 .073
0.793 .127
1.781 .366
0.077
.371
0.061 .961
0.804 .547
3.304 .445
3.352 .257 1.600 .323 2.109 .019*
0.136 .857 0.012
.698
0.296 .650 0.296 .627
0.155 .926
0.694 .496 0.409 .513
0.111 .858
0.503 .527
0.029 .767
ADAMTS1 ADAM metallopeptidase with thrombospondin type 1 motif 1
Animal
0.287 .571
0.176 .625
AGTR1
angiotensin II receptor type 1
Drug target 0.058 .868
0.308 .217
0.773 .622
0.025
.723
1.715 .354
0.543 .679
4.450 .371
2.427 .355 2.028 .276
0.366 .396
AGTR2
angiotensin II receptor type 2
Drug target
0.919 .021*
0.975 .463
0.087
.260
4.216 .113
1.549 .269
13.511 .097
9.009 .075 4.713 .089
0.895 .193
ANGPT1
angiopoietin 1
EC injury
0.221 .707
0.122 .770
0.499 .661
0.000 1.000
0.455 .644
0.128 .880
2.186 .472
3.148 .202 3.286 .123
0.099 .891
ANGPT2
angiopoietin 2
EC injury
0.160 .657
0.375 .143
.740 0.282 .562 0.577 .319
0.366 .764
CCL2
C-C motif chemokine ligand 2
EC injury
0.316 .061
0.063 .597
0.716 .628
0.011
CD34
CD34 molecule
Validation
0.145 .778
0.354 .330
0.508 .618
0.016
ENG
endoglin
Validation
1.123 .121
0.875 .089
EPHB2
ephrin type-B receptor 2
Validation
0.157 .644
0.208 .389
0.091 .858
0.016
EPHB4
ephrin type-B receptor 4
Validation
0.594 .067
0.368 .109
0.456 .740
0.039
0.607 .610
0.225 .570
ETS1
ETS proto-oncogene 1
Animal
0.286 .401
0.104 .668
1.022 .580 0.102
.347 3.841 .198 0.597 .692
6.886 .291 0.483 .828 0.697 .637
0.182 .663
F2
thrombin
Thrombosis
0.116 .756
0.449 .091
0.509 .612 0.016
.715 0.590 .534 0.150 .846
0.536 .816
GDF2
growth differentiation factor 2
TGF-b
0.657 .280
0.541 .210
1.150 .353
0.038
.418
0.308 .667
4.982 .197
2.138 .233 1.186 .259
1.198 .109
HIF1A
hypoxia inducible factor 1 alpha
Animal
0.208 .481
0.034 .870
0.399 .455
0.024
.341 0.047 .895 0.321 .432
1.211 .362
1.486 .167 0.754 .202
0.174 .631
HMOX1
heme oxygenase 1
Animal
0.277 .630
0.130 .750
0.415 .741 0.019
.736 0.083 .938
0.909 .453
0.751 .802
0.639 .690 0.909 .439 0.147 .835
HMOX2
heme oxygenase 2
Animal
0.926 .027*
0.531 .074
1.453 .724 0.072
.714 1.275 .741
0.034 .992
0.721 .944
1.174 .826 2.219 .551
1.457 .005*
IL1B
interleukin 1 beta
EC injury
0.027 .941
0.092 .723
2.936 .341
0.152
.316
4.501 .222
3.135 .286
16.106 .174
7.650 .188 5.029 .178
0.119 .790
IL6
interleukin 6
EC injury
0.444 .341
0.515 .120
0.134 .851
0.044
.304
0.087 .881
0.382 .532
0.964 .584
1.457 .266 1.289 .193 0.959 .094
KLF2
Kruppel like factor 2
EC injury
0.096 .773
0.001 .996
2.740 .090
0.055
.178
2.109 .095
1.709 .107
7.602 .072
2.224 .126 1.662 .106
LRP1
LDL receptor related protein 1
EC injury
0.126 .837
0.735 .091
2.691 .673 0.160
.608 1.401 .808
0.029 .996
MAPK7
mitogen-activated protein kinase 7
TGF-b
0.135 .648
0.045 .830
0.145 .822
.724
0.128 .808
MAPK8
mitogen-activated protein kinase 8
TGF-b
0.261 .570
0.041 .899
0.241 .800 0.011
.789 0.325 .707 0.049 .949
1.293 .603 0.176 .880 0.083 .909 0.303 .592
MCAM
melanoma cell adhesion molecule
Validation
0.033 .956
0.581 .179
1.490 .321 0.030
.548 1.715 .246 0.159 .846
1.499 .578
MMP3
matrix metallopeptidase 3
EC injury
0.060 .703
0.079 .484
0.315 .720 0.023
.605 0.589 .521 0.364 .644
1.285 .593 0.113 .920
0.025 .965
0.097 .856 0.007
0.267 .820 0.005
0.031 .978
.713 0.075 .926 0.142 .849
0.821 .722
0.546 .654
0.581 .490 0.209 .739
.912 0.402 .693
0.861 .438
1.823 .546
2.666 .246 1.656 .247
.504 0.027 .947 0.408 .402
0.413 .724
.571
0.149 .904
2.508 .144
0.123 .829
0.008 .994
0.882 .306
0.600 .861 0.118 .946
0.253 .221
0.248 .780
0.019 .957 0.051 .903
0.798 .549 1.081 .313 0.565 .218
0.095 .816
5.471 .735 1.862 .819 1.888 .717 0.860 .252 0.534 .739
0.679 .478 0.473 .442 0.181 .620
2.432 .246 2.839 .136 0.547 .463 0.290 .694
0.139 .474 continued
McGregor et al ▪ JVIR
0.010
0.066 .863 0.535 .227
0.442 .900 0.524 .779
.866
0.195 .880 0.079 .947
0.406 .556
7.e1 ▪ Endovascular Biopsy and EC Gene Expression Analysis of Dialysis AVFs
Table E1. Gene Targets, Rationale, and Fold-Change Coefficient for Each Fixed Effect
Symbol
Gene
Rationale
Location Location (Stenotic vs (Stenotic vs Control Site) Control Site) in Severein MildStenosis Stenosis Group Group FC
MTHFR
methylenetetrahydrofolate reductase NOS3 nitric oxide synthase 3 PDGFB platelet derived growth factor subunit B PECAM1 platelet and endothelial cell adhesion molecule 1 PIK3CA phosphatidylinositol4,5bisphosphate 3-kinase catalytic subunit alpha PTEN phosphatase and tensin homolog PTGS1 prostaglandin-endoperoxide synthase 1 PTGS2 prostaglandin-endoperoxide synthase 2 PTPRC protein tyrosine phosphatase, receptor type C RASA1 RAS p21 protein activator 1 SERPINC1 serpin family C member 1 SERPINE1 serpin family E member 1 SMAD1 SMAD family member 1 SMAD2 SMAD family member 2 SMAD4 SMAD family member 4 TEK TEK receptor tyrosine kinase TGFB1 transforming growth factor beta 1 THBD thrombomodulin TNF tumor necrosis factor VEGFA vascular endothelial growth factor A VWF von Willebrand factor
P Value
FC
P Value
Degree of Stenosis
FC
P Value
Age
FC
Sex
P Value
FC
Past Smoking Status
P Value
FC
P Value
Current Smoking Status
FC
P Value
Hypertension
FC
P Value
Diabetes
FC
P Value
Interaction Degree of Stenosis and Location
FC
P Value
Thrombosis
0.113 .845
0.833 .042*
0.628 .782 0.059
.601 1.384 .554
0.551 .777
1.375 .812
0.745 .803 1.018 .609 0.946 .182
EC Injury Drug target
0.843 .165 0.130 .760
0.153 .722 0.442 .145
1.250 .549 0.165 .797
0.057 0.067
.550 0.409 .811 .188 0.193 .718
0.430 .786 0.042 .928
2.161 .662 1.637 .381
2.162 .458 2.024 .336 0.690 .354 1.156 .294 0.476 .411 0.573 .275
0.184 .825
0.686 .246
0.149 .904 0.056
.392 1.979 .242
0.169 .855
TGF-b
0.615 .377
0.443 .370
2.204 .300 0.120
.250 6.701 .093 1.315 .386 11.175 .152
0.582 .749
Cell cycle
1.162 .108
0.656 .202
0.823 .773 0.058
.674 1.803 .542 0.996 .692
4.053 .604
1.859 .638 2.089 .455 1.818 .041*
Drug target
0.005 .963
0.023 .771
0.918 .117 0.008
.470 1.005 .093
0.023 .895
0.792 .302
1.385 .096 1.415 .059 0.029 .837
EC injury
0.305 .647
0.182 .700
.468 1.149 .321
0.144 .846
2.976 .337
EC injury
0.073 .900
0.470 .256
Drug target Thrombosis EC Injury TGF-b TGF-b TGF-b Validation TGF-b
0.170 1.421 0.010 0.646 0.603 0.007 0.048 0.107
0.738 0.746 0.855 0.698 0.864 0.631 0.408 0.007
Validation EC injury Animal
0.188 .430 0.189 .705 0.643 .332
Validation
.763 .024* .984 .347 .353 .987 .946 .553
Thrombosis 0.797 .327
0.616 .574
0.035
1.429 .745 0.200
.067 0.399 .655 0.030 .094 5.180 .158 0.173 .017* 0.222 .772 0.041 .153 1.296 .354 0.076 .061 0.193 .842 0.096 .045* 0.244 .898 0.071 .423 1.225 .383 0.004 .955 0.522 .474 0.040
0.018 .916 0.408 .251 0.408 .387 1.479 .010*
1.923 .539 0.101 .943
1.294 2.444 0.300 0.990 2.726 2.275 0.411 1.119
.251 0.589 .446 .288 3.629 .187 .645 0.343 .580 .372 1.769 .209 .142 1.283 .266 .354 0.470 .780 .655 0.824 .415 .240 0.072 .889
1.705 .169 0.079 .941 0.902 .568
0.050 0.016 0.085
.255 .749 .337
2.435 .108 1.150 .206 0.034 .971 0.687 .505 0.388 .762 1.322 .384
0.018 .988
0.099
.239
1.269 .349
0.293 .752
1.618 14.785 1.471 0.429 4.318 4.705 2.308 1.547
0.236 .832 1.058 .215
4.646 .121 3.403 .103
.429 5.958 .313 1.473 .704 12.300 .394 1.213 .833 .466 .211 .341 .261 .198 .500 .925 .325
0.671 .512 0.502 .623
Volume ▪ ▪ Number ▪ ▪ Month ▪ 2018
Table E1. Gene Targets, Rationale, and Fold-Change Coefficient for Each Fixed Effect (continued)
0.123 .880
0.372 .916 0.396 .578
.472 0.264 .788 0.473 .138 5.621 .185 4.220 .466 0.766 .462 0.029 .851 3.635 .159 4.401 .239 1.449 .346 2.066 .440 0.487 .848 0.359 .434 1.319 .402 2.049 .425 0.649 .490 0.712
.501 .155 .957 .083 .164 .821 .182 .320
0.567 0.675 0.866 0.052 0.261 0.638 0.360 0.114
.414 .380 .163 .951 .744 .240 .682 .606
7.203 .101 0.031 .991 7.629 .218
2.206 .167 1.243 .183 0.170 .561 1.912 .325 1.523 .263 0.219 .721 6.964 .126 5.093 .108 1.050 .197
3.565 .342
3.064 .219 1.978 .211 0.682 .494
The interaction term compares the differences in expression levels of stenotic cells and control cells between high-stenosis areas and low-stenosis areas. EC, endothelial cell; FC, fold change; TGF ¼ transforming growth factor. *P < .05.
7.e2