Abstract Background Native (pre-existing) collaterals are arteriole-to-arteriole anastomoses that interconnect adjacent arterial trees and serve as endogenous bypass vessels that limit tissue injury in ischemic stroke, myocardial infarction, coronary and peripheral artery disease. Their extent (number and diameter) varies widely among mouse strains and healthy humans. We previously identified a major quantitative trait locus on chromosome 7 (Canq1, LOD = 29) responsible for 37% of the heritable variation in collateral extent between C57BL/6 and BALB/c mice. We sought to identify candidate genes in Canq1 responsible for collateral variation in the cerebral pial circulation, a tissue whose strain-dependent variation is shared by similar variation in other tissues. Methods and Findings Collateral extent was intermediate in a recombinant inbred line that splits Canq1 between the C57BL/6 and BALB/c strains. Phenotyping and SNP-mapping of an expanded panel of twenty-one informative inbred strains narrowed the Canq1 locus, and genome-wide linkage analysis of a SWRxSJL-F2 cross confirmed its haplotype structure. Collateral extent, infarct volume after cerebral artery occlusion, bleeding time, and re-bleeding time did not differ in knockout mice for two vascular-related genes located in Canq1, IL4ra and Itgal. Transcript abundance of 6 out of 116 genes within the 95% confidence interval of Canq1 were differentially expressed >2-fold (p-value<0.05÷150) in the cortical pia mater from C57BL/6 and BALB/c embryos at E14.5, E16.5 and E18.5 time-points that span the period of collateral formation. Conclusions These findings refine the Canq1 locus and identify several genes as high-priority candidates important in specifying native collateral formation and its wide variation. Introduction Ischemic stroke, myocardial infarction and atherosclerotic disease of arteries supplying the brain, heart and lower extremities are leading causes of morbidity and mortality. Recently, variation in the density and diameter (extent) of the native pre-existing collaterals present in a tissue have become recognized as important determinants of the wide variation in severity of tissue injury caused by these diseases [31][1]–[32][4]. Collaterals are arteriole-to-arteriole anastomoses that are present in most tissues and cross-connect a small fraction of distal-most arterioles of adjacent arterial trees. After acute obstruction of an arterial trunk, collateral extent dictates the amount of retrograde perfusion from adjacent trees thus severity of ischemic injury. With chronic obstruction, native collaterals undergo anatomic lumen enlargement, a process termed collateral remodeling or arteriogenesis that requires days-to-weeks to reach completion, resulting in an increase in conductance of the collateral network [33][5]. Surprisingly, native collateral extent differs widely among “healthy” individuals, ie, those free of stenosis of arteries supplying the heart or brain. Thus coronary collateral flow (ie, CFI[p]) varies more than 10-fold in individuals without coronary artery disease [34][4], [35][6]. Likewise, large differences in collateral-dependent cerebral blood flow are evident in individuals after sudden embolic stroke [36][2], [37][3], [38][7]. Moreover, direct measurement of native collateral extent shows that it varies widely among healthy inbred mouse strains [39][8]–[40][10]. Variation in healthy humans and mice can be attributed to genetic differences affecting mechanisms that control formation of the collateral circulation, which in mouse occurs late in gestation [41][11], [42][12], and to differences in environmental influences and cardiovascular risk factors that recent studies are finding affect persistence of these vessels in adults [43][13], [44][14]. We previously identified a prominent QTL on chromosome 7 (Canq1, LOD = 29) responsible for 37% of the heritable variation in collateral extent (as well as collateral remodeling) in the cerebral circulation of C57BL/6 and BALB/c mice [45][15]—strains which exhibit the widest difference among 15 strains examined [46][9]. Moreover, these same 15 strains had an equally wide variation in infarct volume after middle cerebral artery (MCA) occlusion that was tightly and inversely correlated with collateral density and diameter, thus establishing a strong causal relationship between severity of stroke and collateral extent [47][9], [48][16]. Three other loci were responsible for an additional 35% of the variation in collateral extent [49][15]. Importantly, genetic-dependent differences in collateral extent in the cerebral circulation are shared by similar variation in other tissues, at least in mice [50][8], [51][10], [52][17], [53][18]. Consistent with this, Canq1 maps to the same location on chromosome 7 as recently reported QTL linked to hindlimb ischemia (LSq1) [54][19] and cerebral infarct volume (Civq1) [55][16]. Thus, our findings [56][15] identify the physiological substrate, variation in collateral extent that underlies these QTL. Efficient Mixed-model Association Mapping with high-density SNPs for the above 15 strains allowed us to narrow Canq1 to a region (“EMMA region”) [57][15]. In the present study we sought to further refine this locus and the candidate genes potentially responsible for collateral variation. Identification of the causal genetic element(s) will reveal key signaling pathways controlling collateral formation—about which little is known [58][1], [59][12], [60][17]. Moreover, if subsequently confirmed in humans, this information will aid patient stratification, clinical decision making, and the development of collaterogenic therapies. To this end, we strengthened our association mapping over Canq1 using additional inbred strains, performed linkage analysis of a second genetic mapping population, measured collateral extent and infarct volume in mice genetically deficient for several candidate genes, and examined expression of 116 genes spanning the EMMA region and Canq1 in the pial circulation of mouse embryos during the period when the collateral circulation forms. Materials and Methods Animals Mouse strains (male, ∼10 weeks-old): LP/J, NON/ShiLtJ (NON), LEWES/EiJ, 129X1/SvJ (129X1), CAST/EiJ, C57BL/6J (B6), BALB/cByJ (Bc), Itgal^−/− (#005257, B6 background), Il4^−/− (#002518, B6 background), Il4ra^−/− (#003514, Bc background) and CXB11.HiAJ (CxB11) were obtained from Jackson Laboratories. PWD were a gift of Dr. Fernando Pardo-Manuel, UNC. F1 progeny obtained from reciprocal mating of SWR/J (SWR) and SJL/J (SJL) were mated to produce a 10-week old F2 reciprocal population. This study was approved by the University of North Carolina's Institutional Animal Use and Care Committee and was performed in accordance with the National Institutes of Health guidelines. Phenotyping Mice were phenotyped for collateral number, diameter and cerebral artery tree territories as described previously [61][15] ([62]Materials and Methods S1). Association Mapping As detailed previously [63][15], the EMMA algorithm [64][20], [65][21] was applied in the R statistical package to collateral number, obtained from 21 inbred strains comprised of 15 previously reported strains [66][9] plus 6 newly phenotyped strains. Known and imputed dense SNP data were downloaded from [67]http://phenome.jax.org/db/q?rtn=snps/download and [68]http://compgen.unc.edu/wp/ and pooled. The kinship matrix (pair-wise relatedness) among the 21 inbred strains was calculated using the dense SNP-set within the Canq1 region, and modeled as random effects [69][22]. Each SNP within Canq1 was modeled as a fixed effect. After the mixed model was fitted, an F-test was conducted and a p-value obtained at each SNP location. DNA Isolation, Genotyping and Linkage Analysis Tail genomic DNA from SWRxSJL-F2 mice (n = 123) was genotyped using the 377-SNP GoldenGate genotyping array (Illumina, San Diego, CA). SNP positions were obtained from Build 37.1 of the NCBI SNP database. There are 324 informative markers on 19 autosomes in this array, including 29 on Chr 7 with 6 between 121.5 and 144.5 thus 1 every ∼4 Mb. Collateral number was subjected to linkage analysis using the single QTL model in the R statistical package [70][15]. Threshold for significant QTL was defined as p = 0.05. Bleeding and Re-bleeding Assays Bleeding (thrombosis) and re-bleeding (thrombolysis) assays were similar to previous methods [71][23]. Under 1.2% isoflurane anesthesia supplemented with oxygen and with rectal temperature maintained at 37°C, the tail was pre-warmed for 5 min in 50 mL of 37°C saline. A 5 mm length from the tip of the tail was amputated with a fresh sterile #11 scalpel blade, and the tail was immediately returned to the saline. The cut surface was positioned 10 mm below the ventral surface of the body. Bleeding time was the time between the beginning and cessation of bleeding. Re-bleeding time was the time between cessation and resumption of bleeding. Middle Cerebral Artery Occlusion and Measurement of Infarct Volume These were done as detailed previously [72][15] ([73]Materials and Methods S1). Gene Expression B6 and Bc breeders were paired at 10–12 weeks-age. The presence of a vaginal plug the morning after pairing was designated as 0.5 days post-coitus (E0.5). Embryos were collected at E14.5, E16.5 and E18.5 under deep anesthesia (ketamine+xylazine, 100+15 mg/kg) and staged according to crown-to-rump length. Brains were quickly removed into RNAlater® (Sigma-Aldrich Corp, St. Louis, MO) and stored at −20°C. Approximately 24 hours later the pia mater containing the pial circulation was peeled from the dorsal cerebral cortex of both hemispheres under a stereomicroscope and stored in RNAlater® at −20°C. Pia from 8–10 embryos of each strain were pooled for each RNA sample (≥2 litters per pool). Three pooled RNA samples were prepared for each strain and time-point (18 samples; ∼100 embryos/strain). Samples were thawed in RNAlater and homogenized (TH, Omni International, Marietta, GA) in Trizol Reagent (Invitrogen, Carlsbad, CA). Total RNA was purified using the RNeasy Micro Kit according to the manufacturer (Qiagen, Valencia, CA). RNA concentration and quality were determined by NanoDrop 1000 (Thermo Scientific, Wilmington, DE) and Bioanalyzer 2100 (Agilent, Foster City, CA), respectively. Measurement of transcript number was conducted for 139 selected genes and 11 splice variants by the genomics facility at UNC using NanoString custom-synthesized probes (NanoString, Seattle, WA) [74][24], [75][25]. Transcript number for each gene was normalized to the mean of 6 housekeeping genes: Gapdh, βactin, Tubb5, Hprt1, Ppia and Tbp. For quantitative RT-PCR, reverse transcription was performed with SuperScript™ First-Strand Synthesis System (Invitrogen, # 11904-018) following the manufacturer's instructions. Amplification was achieved with SYBR® Green JumpStart™ Taq ReadyMix™ (Sigma, #S4438) on a Rotor-Gene 3000 (Corbett Life Science). Samples were analyzed in triplicate and values averaged. Primer sequences are listed in [76]Table S1. Data were analyzed using the Relative Expression Software Tool 2009 (version 2.0.13) and calculations as described [77][26]. Statistical Analysis Values are mean±SEM. Significance was p<0.05 unless stated otherwise. ANOVA and Student's t-tests were used as indicated in the figures and tables. For analysis of expression: (1) transcript numbers for strain and embryonic time-point were subjected to 2-way ANOVA, with Bonferroni adjustment of p-values for multiple comparisons (p-values÷150, for 150 transcripts). (2) ANOVA p-values were also derived after correction with the Bonferroni inequality for pre-planned post-hoc tests for 2 strains and 3 time points. (3) Student's t-tests were used to test effect of strain independent of time, with and without Bonferroni adjustment (p-values÷150). Statistical treatment of association mapping data was as described [78][15]. Results Collateral Number in CXB11.HiAJ RIL Verifies Canq1 on Chromosome 7 The CxB recombinant inbred lines (RILs) originated from Bc and B6 parentals [79][27]. Among them, CxB11.HiAJ (CxB11) is particularly relevant. Given the unique mosaic genetic structure for each RIL, we hypothesized that if Canq1 in general, and the “EMMA region” [80][15] in particular, are determinants of collateral extent, an RIL that inherits this locus from either B6 (abundant large-diameter collaterals) or Bc (sparse small-diameter collaterals) [81][8]–[82][10], [83][15] will exhibit traits dominated by the parental phenotype. CxB11 splits Canq1 such that the centromeric half of Canq1 and the EMMA region are from B6 ([84] Figure 1A ). These results confirm our previous findings [85][15] showing the importance of Canq1 and the EMMA region for specifying collateral extent. Figure 1. Collateral extent in CXB11.HiAJ (CxB11) confirms importance of Canq1. [86]Figure 1 [87]Open in a new tab A, Inheritance patterns of Cang1 on chromosome 7 in CxB11 recombinant inbred line derived from BALB/c (Bc)×C57BL/6 (B6). Y axis, SNP positions (Build 37). CxB11 inherits from B6 the segment (B6, grey; Bc, black) that includes the EMMA region and extends through Itgal. B, Collateral number is intermediate between B6 and Bc in CxB11 and significantly different from B6xBc-F1. C, Territory of cerebral cortex supplied by the ACA, MCA, and PCA trees. B6 and Bc are identical (confirms Zhang et al. [88][9]), while CxB11 has smaller MCA and PCA and larger ACA territory than B6. These differences do not correlate with variation in collateral number or diameter in these or 15 other strains [89][9]. Numbers of mice given in parentheses in this and other figures. Because area (“size”) of the MCA, ACA, and PCA trees varies with genetic background and could contribute to variation in collateral extent [90][9], we measured the percentage of cortical territory supplied by each tree. Values for CxB11 ([91]Figure 1C) fall within the range of differences that have no correlation with variation in collateral extent; also, the parental strains are identical as reported previously [92][9]. Thus, variation in tree size cannot explain the CxB11 results. Additional Inbred Strains Strengthen and Refine the EMMA Region EMMA mapping is more efficient computationally and accounts better for population structure embedded in inbred strains, when compared to traditional association mapping algorithms [93][21]. We previously applied this algorithm to 15 inbred strains and obtained a most-significant 172 kb region (EMMA-A, p = 2.2×10^−5) and a second-most significant region (EMMA-B, 290 kb, p = 4.2×10^−4) [94][15]. To strengthen this analysis, we first generated a heatmap of the known and imputed SNPs within EMMA-A for 74 inbred strains ([95]Figure S1). The 15 strains lowest on the y-axis are those used previously [96][15]. The lowest 5 have significantly fewer collaterals than B6. Among the 15 strains, 9 out of 10 with high collateral number exhibit a DNA haplotype-like block similar to B6 ([97] Figure 2 , blue), whereas all 5 strains with low collateral number exhibit a haplotype-like block similar to Bc ([98]Figure 2, mostly green). The exception is SJL/J (discussed below). To test the robustness of the EMMA-A region for variation in collateral number, we chose the 6 most informative additional inbred strains from the 74 lines (denoted red in [99]Figure 2): 3 strains with haplotype structures predicting high (129X1, LP, NON) and low (CAST/Ei, PWD, LEWES/Ei) collateral number. Collateral numbers in 4 of these strains fit the prediction from their haplotype structures ([100]Figure S2). The remaining two are wild-derived strains that have more complex ancestry and haplotype structures. Figure 2. Collateral number for 21 inbred strains and heatmap of their SNPs within the EMMA region. [101]Figure 2 [102]Open in a new tab Left, Heatmap of known and imputed SNPs for 21 strains in EMMA region (blue, SNP same as B6; green, different from B6; blank, genotype unknown). Right, Collateral number per hemisphere for 21 strains. Names in red denote newly phenotyped strains, black names are strains from Zhang et al. [103][9]. N = 8–10/strain. Dashed line, reference to B6. An important consideration is whether the EMMA-A peak shifts position or if the significant SNPs change their p-values from our previous findings [104][15] after the additional 6 strains are remapped with the original 15. [105]Figure 3 shows that EMMA-A did not change position, that the previous EMMA-A peak was narrowed further, and that the new peak acquired increased significance for the same SNPs (p = 9×10^−6). Figure 3. Six more strains strengthen and narrow the EMMA interval. [106]Figure 3 [107]Open in a new tab The EMMA algorithm was applied to collateral number for the 21 strain data set, ie, 6 new strains (red strains in [108]Figure 2) plus 15 from Zhang et al. [109][9]. ∼49,000 high quality imputed SNPs under the Canq1 peak were tested. The previous highly significant region was strengthened and narrowed (p<9×10^−6, 4 SNPs). Black bars indicate the most (Group A) and second most (Group B) significant regions in the previous study [110][15]. Mapping allowed fewer than three SNPs with missing genotypes. The kinship matrix used to account for population structure will vary with use of different SNPs as new strains are considered [111][22]. Furthermore, we previously did not allow missing genotypes to avoid a decrease in statistical power [112][15]. We thus examined the effect of several parameters on our new mapping results by: allowing fewer than 3 missing genotypes in the strains; using only informative SNPs; excluding wild-derived strains. Irrespective of these mapping parameters, 4 SNPs always had the most significant p-values (rs32978185, rs32978627, rs32973294, and rs32973297; 132.502–132.504 Mb; [113]Figures S3, [114]S4, [115]S5, [116]S6, [117]S7). Genome-wide QTL Mapping of SWRxSJL-F2 Shows Absence of Canq1 Strains SJL and SWR have identical Bc-like haplotype structures in the EMMA regions, but SJL averages 9.6Á±0.4 collaterals per hemisphere while SWR averages 1.3±0.4 ([118] Figure 4 ). To test the validity of the EMMA regions and possibly identify additional QTL, we created a SWRxSJL-F2 cross and performed genome-wide LOD score profiling of collateral number-by-genotype. As predicted, we found no QTL on Chr 7 ([119]Figure 4). We also found no QTL elsewhere in the genome. We did not measure collateral diameter because 1) of the time required to obtain average collateral diameter for each of the 123 F2 mice, 2) we have shown that variation in collateral number and diameter map to the same Canq1 locus [120][15], 3) collateral diameter has a much smaller range of variation than number, thus lower LOD score [121][15], and 4) no significant QTL was found for collateral number ([122]Figure 4). Figure 4. Genome-wide mapping of collateral number in 123 SWRxSJL-F2 mice. [123]Figure 4 [124]Open in a new tab Top panel, Collateral number per hemisphere among 15 strains of mice (from [125]Figure 2). Middle panel, Green = known and high-quality imputed (p>0.90) SNPs common to both strains and BALB/c. White, known and low-quality SNPs different between the two strains, including two common regions with “no useful data”, per JAX website (red). Lower panel, LOD profiling using single QTL model. Locations of genotyping SNPs are shown as ticks on the abscissa. 95% confidence level (dashed line) was estimated using 1000 permutations. Insert shows the range of collateral number between the MCA and ACA trees per hemisphere on the abscissa (bin range = 3–14), and number of mice having a given collateral number bin on the ordinate (range 2–23). Collateral Extent, Hemostasis and Infarct Volume in Il4ra^−/−, Il4^−/− and Itgal^−/− Do Not Support these Genes as Candidates for Canq1 Collateral extent in the adult is determined by mechanisms that specify their formation in the embryo. Recent studies indicate that Vegfa, Clic4, Flk1, Adam10 and Adam17 are important in this process [126][11], [127][12], [128][17], [129][18]. However, none of these genes nor others known to impact their expression or signaling pathways are located in EMMA-B, with the exception of IL4ra (interleukin 4 receptor alpha chain). Expression of Il4ra is lower in Bc compared to B6 (see below). Moreover, IL4rα is present on endothelial cells (and other cell types) where it dimerizes with IL13rα or γc to mediate responses to IL4 and IL13. These include increased expression of VEGF-A; furthermore, hypoxia-induced VEGF-A expression and angiogenesis are reduced in Il4^−/− mice [130][28]–[131][30]. We thus examined Il4^−/− (B6 background) and Il4ra^−/− (only available on Bc background) mice. Collateral extent in Il4^−/− mice (9.5 collaterals, 21 um diameter) was not different from B6 (9.3 and 23 um) ([132] Figure 5 ). Likewise, collateral extent in Il4ra^−/− mice (0.2 collaterals, 17 um) was similar to Bc (0.2 and 13 um). These results do not support involvement of IL4ra in variation of collateral extent. Knockout mice are not available for the other genes (see below) in EMMA-B. Figure 5. Collateral extent is not altered in Itgal, IL4 or IL4ra deficient mice relative to host strains. [133]Figure 5 [134]Open in a new tab A, Itgal^−/− and Il4^−/− mice are B6 while Il4ra^−/ ^− (receptor alpha) are Bc background. Number and diameter are not significantly different among Itgal^−/−, Il4 ^−/− and B6, nor between Il4ra^−/− and Bc (t-tests). B, Territory (size) of ACA, MCA, and PCA trees are same in B6 and Bc but differ in Itgal^−/− and Il4ra^−/ ^−, confirming dissociation of collateral and tree territory phenotypes among inbred mouse strains [135][9]. C,D, Bleeding and infarct volume of knockouts are not different from B6; re-bleeding in Il4^−/− is not different from B6 but shorter in Itgal^−/− (t-tests). Given the limitations of any mapping algorithm, it is possible that a genetic element(s) outside of EMMA-B could underlie Canq1. Several findings identify Itgal (integrin α-L chain, CD11a) as a candidate. Itgal is located at 134 Mb near the peak of Canq1. Itgal dimerizes with integrin β2-chain (CD18) to form LFA-1 which is present on platelets and most leukocyte types and binds ICAM1 on endothelial cells, leading to firm adhesion and transmigration [136][31], [137][32]. Keum and Marchuk identified Itgal as a candidate gene within the Civq1 QTL for infarct volume measured 24 hours after permanent MCA occlusion in a B6xBc-F2 population [138][16], and Itgal^−/− mice have smaller infarct volumes 24 hours after 1-hour transient MCA occlusion model [139][32]. However, Itgal^−/− mice had the same collateral extent as B6 controls ([140]Figure 5). Consistent with this, infarct volume 24 hours after permanent MCA occlusion was not different in Itgal^−/− mice ([141]Figure 5). Infarct volume was also not affected in Il4^−/− mice (B6 background; [142]Figure 5), in agreement with no effect on collateral extent. These findings do not support Itgal or Il4ra as candidate genes for Canq1 [143][15] or Civq1 [144][16]. Variation in MCA tree size among strains is also a determinant of variation in infarct volume after permanent [145][9] or transient [146][33] MCA occlusion. MCA tree size was smaller in Itgal^−/− mice ([147]Figure 5), which may contribute to their smaller infarct volume after transient MCA occlusion [148][32]. The involvement of other factors besides collateral extent in determining severity of brain ischemia after acute occlusion could underlie previous [149][31] findings. Leukocyte/platelet adhesion and hemostasis have complex influences on infarct volume, especially after transient occlusion, through effects on thrombosis and thrombolysis (ie, hemostasis) and inflammation [150][33], [151][34]. Moreover, thrombosis and thrombolysis vary with genetic background [152][23]. Because these mechanisms could be affected in Itgal^−/− and IL4^−/− mice, we measured bleeding and re-bleeding time to assay thrombosis and thrombolysis, respectively. No differences were observed among knockouts and B6 controls ([153]Figure 5), with the exception of shorter time to re-bleeding in Itgal^−/− mice (discussed below). Depending on models and severity of stroke, increased thrombolysis can either reduce no-reflow and lessen infarct volume, or increase petechial hemorrhage and no-reflow and increase infarct volume [154][33], [155][34]. IL4ra^−/− mice were not analyzed for bleeding and re-bleeding because of their cost, because IL4^−/− showed no differences in these assays, and because the data in Panels A and B of [156]Figure 5, where both IL4^−/− and IL4ra^−/− mice were studied, provided no support for these genes in the difference in collateral extent in mice. We also studied hindlimb ischemia [157][10] in Il4^−/−, Itgal^−/− and their B6 background strain ([158]Figure S8) and obtained results in agreement with the above cerebral studies for lack of effect on native collateral extent in a second tissue—skeletal muscle. Although not the subject of this study, recovery of blood flow was less in Il4^−/− mice, suggesting reduced collateral remodeling (see [159]Figure S8 for relevant references). Blood differential cell count ([160]Table S2)