Four additional mouse crosses improve the lipid QTL landscape and identify Lipg as a QTL gene.

To identify genes controlling plasma HDL and triglyceride levels, quantitative trait locus (QTL) analysis was performed in one backcross, (NZO/H1Lt × NON/LtJ) × NON/LtJ, and three intercrosses, C57BL/6J × DBA/2J, C57BL/6J × C3H/HeJ, and NZB/B1NJ × NZW/LacJ. HDL concentrations were affected by 25 QTL distributed on most chromosomes (Chrs); those on Chrs 1, 8, 12, and 16 were newly identified, and the remainder were replications of previously identified QTL. Triglyceride concentrations were controlled by nine loci; those on Chrs 1, 2, 3, 7, 16, and 18 were newly identified QTL, and the remainder were replications. Combining mouse crosses with haplotype analysis for the HDL QTL on Chr 18 reduced the list of candidates to six genes. Further expression analysis, sequencing, and quantitative complementation testing of these six genes identified Lipg as the HDL QTL gene on distal Chr 18. The data from these crosses further increase the ability to perform haplotype analyses that can lead to the identification of causal lipid genes.

genic diet from 8-16 weeks of age. TGs for this cross have been published previously ( 32 ).
Quantitative complementation testing was performed for Lipg . Male B6. Lipg Ϫ / Ϫ and B6 Lipg +/+ mice were crossed with female B6, C3H, D2, NZB, or SM mice. Both sexes of the chow-fed F1 progeny from each mating were measured for HDL at 8 weeks of age.

Plasma lipid analysis
Crosses B6 × C3H and NZB × NZW were fasted for 4 h in the morning, while crosses for B6 × D2 and (NZO × NON) × NON were not fasted. Fasting does not usually affect HDL but may signifi cantly alter TG levels, which could reduce the ability to identify TG QTL (false negatives) but should not create false positives. Thus, TG QTL are also reported for the unfasted crosses because they may be informative for future positional mapping of these QTL genes.
Blood from the retro-orbital sinus was collected in tubes containing EDTA and centrifuged at 9,000 rpm for 5 min. Plasma was frozen at -20°C until assay. Plasma HDL, total cholesterol (TC), and TG concentrations were measured using enzymatic reagent kits (Beckman Coulter, Fullerton, CA) according to the manufacturer's recommendations on the Synchron CX Delta System (Beckman Coulter).

Genotyping
DNA was extracted from tail tips using phenol-chloroform. For the crosses B6 × D2 and NZB × NZW, SNPs were genotyped by the Allele-Typing Service at The Jackson Laboratory in conjunction with KBiosciences (Hoddesdon, UK). For the crosses B6 × C3H and (NZO × NON) × NON, polymorphic MIT microsatellite markers were genotyped using agarose gel electrophoresis (NuSieve 3:1; FMC BioProducts, Rockland, ME). Markers were chosen at evenly spaced intervals where possible, and further details of the markers used are available in supplementary Tables I-IV . All data from these crosses are available in the Mouse Phenome Database in QTL archive under Su1 (jax.org/phenome; qtlarchive).

QTL analysis
QTL mapping was performed using R/QTL (version 1.07-12, available at http://www.rqtl.org) as described previously ( 32,33 ). HDL, TC, and TG values were transformed using log base 10 to obtain a normal distribution (supplementary Fig. I ). Maineffect QTL were computed at 2 centimorgan (cM) increments over the entire genome. QTL were deemed signifi cant if they either met or exceeded the 95% genome-wide adjusted threshold, approaching saturation with few new loci being reported ( 27 ), the additional crosses provide important information that can be used for combining data sets and haplotype analysis to further narrow loci and identify the genes ( 29 ). In comparison, few TG QTL have been reported, and the seven novel TG QTL reported here will likely aid the identifi cation of TG QTL genes in the future as more crosses are performed.
This study reports plasma HDL and TG QTL from four new mouse crosses involving seven different strains. Furthermore, six additional HDL QTL crosses have been published since the last review ( 27 ) ( Table 1 ). These separate crosses will aid haplotype analysis, and where possible, raw data from multiple crosses can be combined to narrow the QTL region and identify the underlying genes. In this study, some of the new QTL were used for combined-cross analysis with haplotype analysis, mRNA expression analysis, and quantitative complementation testing to identify and prove that endothelial lipase ( Lipg ) is an HDL quantitative trait gene on distal mouse chromosome (Chr) 18.
Each cross, carried out for different reasons, is summarized here. B6 × D2 is a cross originally studied for albuminuria ( 30 ) phenotyped 335 chow-fed male F2 mice at 8 weeks of age. B6 × C3H, a cross with 277 female F2 mice generated from reciprocal mating, fed mice an atherogenic diet (Ath) as described previously from 8 to 14 weeks of age when they were phenotyped. (NZO × NON) × NON, a cross previously described for diabesity phenotypes ( 31 ), measured HDL and TG levels in 146 chow-fed males at 24 weeks of age. NZB × NZW, a cross with 264 F2 male and female mice, measured HDL and TG in mice fed the athero-

Chr 18/ Lipg locus expression analysis
Liver mRNA expression levels were interrogated for the genes within the Chr 18 QTL region, narrowed after haplotype analysis to determine if parental strains for the three crosses (B6 × D2, B6 × C3H, and NZB × SM) differed in expression. The livers were obtained from fi ve male and fi ve female 12 week old mice of each strain fed chow or atherogenic diet. Total RNA was extracted from fi ve mice of each strain-sex-diet group; three were chosen using random numbers from each group of fi ve for microarrays, and all fi ve were used for real-time PCR. Expression analysis was performed using Mouse Genome 430 2.0 Gene-Chip arrays (Affymetrix). The array data has been deposited at the Gene Expression Omnibus ( www.ncbi.nlm.nih.gov/geo/ ; series number: GSE10493). RNA extraction, quantifi cation, cDNA synthesis, and data analysis were carried out as described previously ( 40 ).
For real-time PCR, cDNA samples were mixed with SYBR Green Master Mix (Applied Biosystems, Foster City, CA) and gene-specifi c primers in a total volume of 25 µl. The primer pairs are as follows: Lipg forward 5 ′ -TGGCTGCAGGAGAAGGAAGA-3 ′ and reverse 5 ′ -CAGCGTGTAGGTATGCAGGA-3 ′ and ␤ -Actin forward 5 ′ -CTTCTTGGGTATGGAATCC-3 ′ and reverse 5 ′ -GCT CA-GGAGGAGCGGTGAT-3 ′ . PCR was performed in 96-well optical reaction plates with an ABI PRISM 7500 sequence detection system (Applied Biosystems). Cycling parameters were 2 min at 50°C, 10 min at 95°C, and 40 cycles of 15 s at 95°C, and 1 min at 60°C. After PCR, a dissociation curve was constructed by increasing temperature from 65°C to 95°C for detection of PCR product specifi city. PCR reactions were set up in triplicate for each strainsex-diet group, and the expression of Lipg was normalized to the expression of ␤ -actin.

Statistical analysis for complementation test
A least square means analysis was used to examine the interaction in F1 mice between Lipg (knockout or wild-type) by "allele" (high or low HDL in the crosses) for the quantitative complementation analysis. The null hypothesis for this test is that the difference in HDL between F1 mice carrying the wild-type alleles of Lipg and crossed to either the low HDL allele or high HDL allele strain at this QTL locus ( ⌬ 2) should not differ from the difference in HDL between F1 mice carrying Lipg Ϫ /+ alleles ( ⌬ 1). If there was a signifi cant difference ( P < 0.01), then the candidate gene is considered to successfully complement. Data were analyzed using JMP version 7.0 (SAS Institute, Cary, NC).

HDL and TC QTL mapping
The four crosses (B6 × C3H, B6 × DBA/2, NZO × NON, and NZW × NZB) revealed 14 signifi cant ( P < 0.05) and 11 suggestive ( P < 0.63) HDL QTL on Chrs 1-6, 8,11,12,15,16,18, and 19 ( Fig. 1 ; Table 2 ). All HDL QTL have been identifi ed in other mouse crosses ( 27 ), except those on Chrs 1, 8, 12, and 16 observed in the NZB × NZW cross. Distal Chr 1 was the most frequently identifi ed QTL, observed in three crosses, including B6 × D2, B6 × C3H, and (NZO × NON) × NON, and its causal gene is most likely Apoa2 ( 41 ). A second locus with a peak at 49 cM on Chr 1 was identifi ed in the cross NZB × NZW. Recently, Farp2 and/or Stk25 were reported as the candidate genes for this QTL using the NZB × NZW cross ( 42 ); both genes have some evidence, but further evidence is needed to choose between the two. Three signifi cant QTL were identifi ed which was assessed by 1,000 permutations; they were deemed suggestive if they either met or exceeded the 37% genome-wide adjusted threshold but were not signifi cant ( 34 ). Approximate cM coordinates for markers were obtained by dividing base pair positions (mouse genome build 36) by a factor of 2 except for Chr 19, where Mb/1.04 was used. The validity of this approximation was confi rmed by comparison to estimated map positions in R/QTL and also from previous cM-to-Mb comparisons in mouse ( 35 ). Simultaneous pair-wise genome scans were performed to detect gene interactions; however, no signifi cant interactions were identifi ed.
Combining crosses was performed as previously described ( 36 ). Briefl y, genotypes from Chr 18 in the B6 × C3H and NZB × SM crosses were recoded, so that the B6 and SM genotypes became L for low HDL alleles and the C3H and NZB genotypes became H for high HDL alleles. A logarithm (base 10 of odds (LOD) score was computed at 2 cM intervals across the QTL interval for each cross separately and then for both crosses combined. The combined data were analyzed with "sex" and "cross" as additive covariates.
The raw data sets for each of these crosses, as well as multiple previously published HDL QTL studies, are available online in the Mouse Phenome Database (http://jax.org/phenome/ qtlarchive).

SNP and haplotype analysis
Recently, >8 million SNPs were released for 16 mouse inbred strains ( 37 ). These data and several other sources were used to infer genotypes for a 50 strain set using a Hidden Markov Model; this imputed SNP resource is available at http://cgd.jax.org/ ImputedSNPData/v1.1/ ( 38 ) and also in the Mouse Phenome Database on the center for genome dynamics (CGD) set of SNPs. SNPs for Chr 18 (n = 291,266) were downloaded, and locations that met the haplotype conditions were identifi ed using a script implemented in the R statistical package. To carry out haplotype analysis, genomic regions within the QTL were excluded if the pair of strains that gave rise to a QTL had an identical haplotype pattern (B6 = C3H, B6 = DBA/2, and NZB = SM). Such regions are deemed identical by descent and are very unlikely to contain the causal genetic polymorphism underlying a common QTL ( 39 ). The remaining QTL regions were selected if the strains carrying the allele that increased HDL (C3H = D2 = NZB) were identical, if the strains carrying the allele that decreased HDL (B6 = SM) were identical, and if the haplotype pattern differed between the high and low allele strains (C3H, D2, NZB ≠ B6, SM). Gene lists for these regions were extracted from Ensembl ( www.ensembl.org ). These genes were examined in the Mouse Phenome Database ( http://www.jax.org/phenome ) for coding differences.

Sequencing
Additional sequencing was performed to confi rm sequence variants between the different strains for Mro and Lipg . The genomic sequences of Mro (ENSMUST00000120033) and Lipg (ENSMUST00000066532) were obtained from the Ensembl (http://www.ensembl.org) mouse genome assembly, and primers were designed to amplify each exon, including at least 50 nucleotides of the adjacent introns. Sequences around noncoding SNPs were obtained from http://www.ncbi.nlm.nih.gov/ SNP/ . Standard PCR was performed using primers listed in supplementary Table V . Purifi ed PCR products were subjected to thermocycle sequencing on capillary-based machines by the Jackson Laboratory DNA Sequence Laboratory. The sequence was analyzed using Sequencher software ( TG levels were positively correlated with HDL levels in three crosses; however, for all three crosses, the proportion of TG explained by the relationship of TG to HDL was small; B6 × C3H ( r 2 = 0.14, P < 0.0001), B6 × D2 ( r 2 = 0.05, P < 0.0001), and NZB × NZW ( r 2 = 0.03, P = 0.0068). TG levels were not correlated with HDL in the (NZO × NON) × NON cross. A stronger correlation between TG and HDL was not predictive of more overlapping HDL and TG QTL. Nonetheless, fi ve signifi cant TG QTL (Chrs 1, 2, 8, 12, and 18) had overlapping 95% confi dence intervals with HDL QTL ( Table 2 ).

Combining crosses and haplotype analysis reduced the Chr 18 HDL QTL to six genes
Several crosses detected HDL QTLs on Chr 18 as listed in Table 3 . The LOD score plots for the crosses, where available, are reproduced in Fig. 3A . These LOD score plots indicate as many as four Chr 18 QTL at 23, 55, 77, and 85 Mb. Table 3 lists which crosses contain each QTL (if the LOD score plots are not available, the reported peak and confi dence intervals were used to estimate the presence or absence of the QTL with surrounding regions denoted by "-?"). Further analysis presented here focuses on Chr 5 at 26 cM (LOD = 9.0) and 57 cM (LOD = 12.7) in the NZB × NZW cross and at 41 cM (LOD = 3.4) in the (NZO × NON) × NON cross. Interestingly, HDL levels are raised by heterozygous alleles for the Chr 5 QTL in the (NZO × NON) × NON cross. It has been previously shown that heterotic effects in an NZO × NON outcross exacerbated both diabesity and perturbations in lipid metabolism ( 43 ). Two signifi cant QTL were identifi ed on Chr 3 at 26 cM (LOD = 4.5, NZB × NZW cross) and at 64 cM (LOD = 3.0, B6 × D2 cross). All these QTL have been observed multiple times in other crosses ( 27 ).
For TC, in the chow-fed crosses B6 × D2 and (NZO × NON) × NON, three suggestive QTL distributed on Chrs 2, 7, and 8 were identifi ed specifi cally for TC; the other seven QTL were identifi ed for both TC and HDL. This indicates that the majority of the cholesterol QTL observed in mice maintained on a chow diet are represented by HDL loci. In the atherogenic diet-fed crosses, B6 × C3H and NZB × NZW, 11 QTL were identifi ed for TC, six of them (54.5%) were independent of HDL QTL ( Table 2 ;  see supplementary Fig. II ). This indicates that nonHDL cholesterol levels differ between strains after consuming atherogenic diet and yield QTL different from HDL loci.

TG QTL mapping
Five signifi cant TG QTL were identifi ed ( Fig. 2 ; Table  2 ). Two were mapped at approximately the same location on Chr 2 at 73 cM (LOD = 2.4) in (NZO × NON) × NON and 80 cM (LOD = 4.7) in B6 × C3H, and these shared approximately the same confi dence interval. Three addi-   the same gene, then combining crosses will increase the LOD score and narrow the confi dence interval; if the hypothesis that the QTL are caused by the same gene is incorrect, the LOD score will not increase and the confi dence interval remains the same or increases. Combining crosses requires access to the raw data, and for the Chr 18 QTL at 77 Mb, data are available only for the crosses B6 × C3H and NZB × SM. To combine the data, the genotype information was recoded from a strain-specifi c code to a phenotype-specifi c code; B6 and SM genotypes became L for the low HDL allele, while C3H and NZB genotypes became H for the high HDL allele. This analysis reduced the QTL to a 9 Mb region spanning from 72-81 Mb ( Figs. 3B and 4A ). The QTL was further narrowed by reducing the region to that overlapping the human QTL at 18q12. 1-22.2 ( 46 ), thus reducing the region to 6.9 Mb and 42 genes ( Fig. 4A , step c). The QTL was further narrowed by haplotype analysis using the same two crosses as well as adding the B6 × D2 cross ( 44 ). The B6 × CAST cross could also be included; on the Chr 18 QTL at 77 Mb, which is found in at least four crosses: B6 × C3H, B6 × D2, NZB × SM, and B6 × CAST. The allele causing high HDL in these crosses was C3H, D2, NZB, or CAST. From the LOD score curves, it can be seen that the B6 × 129 and the NZB × NZW crosses detect a QTL at 55 Mb but not at 77 Mb ( Fig. 3A ).
The B6 × D2 cross reported in this article did not detect the Chr 18 QTL, but another B6 × D2 cross in the literature carried out in female mice did detect a Chr 18 QTL ( 44 ). It is not clear why the B6 × D2 cross reported here failed to detect a Chr 18 QTL; however, the complementation test discussed below shows some consistency with this observation. The mode of inheritance differs for some of these crosses, but the mode of inheritance may be infl uenced or obscured by a closely linked QTL, especially if it is stronger, or by the genetic background. Thus, the different modes of inheritance should not hinder combined cross analysis ( 36,45 ).
Combining crosses can be a powerful method for narrowing QTL. If the QTL in separate crosses are caused by a No QTL detected indicated by "-," while an X indicates the presence of a QTL with an interval spanning a common peak location found in other crosses. For some published QTL, data were lacking to determine the exact confi dence interval of the QTL; therefore, the surrounding regions are by denoted by "-?". b QTL are reported for the fi rst time in this article. c The QTL was found in chow-fed animals only.
other regulatory regions. Analysis of the coding SNPs in these six genes (Mouse Phenome Database; www.jax. org/phenome/SNP) revealed two genes with SNPs that caused nonsynonymous amino acid changes: Asp218Glu in maestro ( Mro ) and Tyr262Cys in endothelial lipase ( Lipg ). These two SNPs were resequenced in B6, C3H, and D2 to confi rm the public data and also in NZB and SM, strains that had not previously been genotyped for these SNPs. However, sequencing showed that strains NZB and SM did not differ at the SNP in Mro , making Mro an unlikely candidate gene. Following resequencing of Lipg , the reported SNP in D2 causing the Tyr262Cys change, originally identifi ed by Mural et al. ( 47 ), was shown to be an error; all fi ve strains have Tyr-262. Sequencing Mro and Lipg exons for all fi ve strains showed that except for the Asp218Glu Mro SNP (B6 = Asp; C3H, D2, NZB, and SM = Glu), there was no other coding variants among these strains for Mro and Lipg coding regions. Consequently, a coding region sequence variant causing a functional change is not likely to be the cause of this common QTL. Since these six genes do not appear to have a coding region polymorphism that alters protein function, the QTL may be mediated by a sequence variant affecting differences in the mRNA. To further characterize the Chr 18 QTL, microarrays were used to examine gene expression of the six genes ( Table 4 ). Parental expression of these genes from mouse livers from each parental strain-sex-diet group (n = 3) giving rise to the original QTL were examined. The three original Chr 18 QTL yield six strain-sexdiet comparisons: NZB × SM (males and females on chow and high-fat atherogenic diet), D2 × B6 (females on chow), and C3H × B6 (females on high-fat atherogenic diet). Of however, strains recently derived from the wild such as CAST may not share the identical by descent regions of the classic inbred strains so haplotyping is often uninformative, as in this case (data not shown). Haplotyping is based on the assumptions that the gene causing the QTL is the same in all three crosses and that the mutation is an ancestral mutation, which is reasonable since it was found in multiple crosses. Of the approximately 300,000 SNPs available for Chr 18, approximately 30,000 mapped within the QTL confi dence interval. These SNPs were interrogated to identify regions that fi t the following criteria: the strains that carry the allele that increased HDL had identical haplotypes (C3H, D2, and NZB), the strains that carry the allele that decreased HDL had identical haplotypes (B6 and SM), and these two haplotypes differed from each other. This haplotype analysis reduced the QTL region to six genes as potential candidates: Smad4 , Mro , Ccdc11 , Lipg , Dym , and Gm672 ( Fig. 4 ).

Further analysis of the six genes by sequence and expression differences
It is expected that the QTL gene should have a difference in function caused by an amino acid change in the coding region or a noncoding sequence difference that alters expression levels, mRNA stability, splice sites, or  Table 3 are not available for reproduction. B: The LOD score plot obtained by combining the B6 × C3H and the NZB × SM crosses is shown as a solid line. The horizontal dashed line represents the signifi cance threshold level ( P = 0.05) as determined by 1,000 permutation tests. are near transcription factor binding sites. These SNPs were also sequenced for NZB and SM if the public databases were missing data for these strains. These SNPs are arranged in genomic order for all the strains that were parents of QTL crosses that detected a QTL on Chr 18, including the two crosses B6 × 129 and NZB × NZW that do not detect the QTL at 77Mb, but rather detecting an independent QTL at the 55 Mb locus ( Table 5 ). Strains B6, SM, and 129 share the same haplotype block, while strains NZB, NZW, C3H, and D2 fall into a second haplotype block (gray region of Table 5 ). This shows that B6 × 129 and NZW × NZB crosses could not have detected the QTL caused by Lipg because the parental strains are identical to each other. CAST has its own unique haplotype typical of a wild-derived strain (data not shown).

Defi ciency complementation of Lipg
Lipg has a known role in HDL metabolism, and its knockout and transgene alter HDL levels. However, as a further test of its role as the causal gene for this QTL, we carried out a defi ciency complementation test ( 48,49 ). In such a test, the B6.129-Lipg Ϫ / Ϫ strain is mated to both parents of a QTL cross. As a control, the B6 strain is mated to the same parents. The four different F1 populations are tested for HDL levels. The null hypothesis for a defi ciency complementation test is that the difference ( ⌬ 2 in Fig. 6 ) between F1 mice that carry the Lipg wild-type gene is the same as the difference ( ⌬ 1 in Fig. 6 ) between F1 mice that are heterozygous for the Lipg knockout. If ⌬ 1 is equal to ⌬ 2, then reduced expression of Lipg in the Lipg Ϫ /+ F1 has no effect on HDL and there is no complementation between Lipg expression levels and the high or low HDL allele. However, if ⌬ 1 differs signifi cantly from ⌬ 2, then the Lipg allele from the knockout complements the Lipg allele in one parental strain, indicating that the knockout gene and the QTL gene are the same. We crossed B6.129-Lipg Ϫ / Ϫ and strain B6 (B6. Lipg +/+ ) to each of the parents of the three crosses that detected the QTL, to strains B6 and C3H ( Fig. 6A ), to strains B6 and D2 ( Fig. 6B ), and to strains NZB and SM ( Fig. 6C ). As shown in Fig. 6 , in each case, ⌬ 2 was signifi cantly larger than ⌬ 1 ( P < 0.001), indicating that Lipg did complement and is therefore the QTL gene. For the B6 × D2 complementation test ( Fig. 6B ), the change between ⌬ 2 and ⌬ 1 in females is considerably greater than it is between males, which may explain why this QTL was the six genes identifi ed through haplotyping, only Lipg is consistently signifi cantly differentially expressed (fi ve of six possible comparisons, false discovery rate < 0.05). Two other probe sets have two of six possible comparisons that are signifi cant ( Table 4 ). Lipg expression was confi rmed by real-time PCR in mouse livers (n = 5/strain-sex-diet group) among all the parental strain combinations that detected the Chr 18 QTL. In this case, Lipg expression was observed to be signifi cantly different in all six comparisons with lower expression in the NZB, D2, and C3H strains carrying the allele that increased HDL compared with the B6 and SM strains carrying the allele that decreased HDL levels ( Fig. 5 ). Thus, Lipg is the most likely candidate gene, and there must be a sequence polymorphism that affects its expression.
To fi nd the polymorphism that regulates Lipg expression, transcription factor binding sites in the upstream region of Lipg were examined using public SNPs with TRANSFAC software ( http://www.biobase.de/pages/index. php?id = 271 ). Two SNPs between the low-allele strain B6 and high-allele strains C3H and D2 are inside or right next to putative transcription factor binding sites for c-ETS-1 and v-MYB ( Table 5 ), and an additional 10 SNPs  GSE10493). Each sex-diet-strain is composed of n = 3 mice, giving rise to a total of 18 mice surveyed with the low-allele compared with high-allele strains. Comparisons with a false discovery rate (FDR) < 0.05 are indicated in bold. The allele for high HDL is listed fi rst in each comparison. FC, fold change; n.s. not signifi cant; F, female; M, male.

Fig. 5.
Lipg mRNA analysis in parental strains involved in the HDL QTL crosses. Real-time gene expression to confi rm microarray observations; total RNA was extracted from mouse livers (n = 5 per sex-strain-diet group leading to a total of n = 30 for low-allele mice and n = 30 high-allele mice for a total of 60 independent mice). Expression levels of Lipg are normalized to ␤ -actin and expressed as mRNA copies per 1,000 copies of ␤ -actin. * P < 0.01 compared with the expression in high allele strain.

DISCUSSION
To identify QTL genes, traditional techniques have relied upon genetic methods such as further mapping through the creation of congenic strains ( 53,54 ). However, with the availability of dense SNPs for most of the common inbred strains ( 37,38 ), fi ne mapping QTL should be easier since many of the polymorphic regions in any cross will be known. Furthermore, the ability to combine crosses based on the assumption that the causal alleles are ancestral demonstrates a need for further QTL studies. Generating an F2 or backcross population may be more effi cient than constructing a set of congenic strains and can help map multiple common QTL ( 55 ).
In this study, four additional mouse crosses are reported for HDL, bringing the total number of HDL QTL studies to 35. HDL is one of the most characterized quantitative phenotypes in mouse models, second only to obesity and body weight phenotypes ( 56 ). However, as most crosses were originally performed with varying purposes, focusing on specifi c diets or sexes or optimizing the study for atherosclerosis phenotypes ( 57 ), identifying causal genes through haplotyping and combined crosses remains a challenge. The four crosses described here identify HDL detected in a female cross and not in the male cross reported here.
The complementation test works best when both the QTL and the knockout are fully recessive ( 50 ). However, heterozygous Lipg Ϫ / + mice are additive for the HDL phenotype ( 51 ). Although in the B6 × C3H QTL, the C3H allele is recessive, the QTL shows additive inheritance in the B6 × D2 and the NZB × SM cross. It is interesting that the complementation test worked just as well for these crosses, showing additive inheritance as for the cross showing the recessive inheritance. This indicates that while it is not optimal to have both the knockout gene and the QTL gene show additive inheritance, it does not rule out the possibility of observing complementation as we did for all three crosses.
It is also important to note that the Lipg Ϫ / Ϫ mice are not on a pure B6 background. This raises the possibility that passenger genes from the 129 strain could have caused the results we observe in the complementation test. However, this is unlikely because no QTL for HDL levels is present in the Lipg region, as demonstrated recently in a large (n = 528) F2 cross between B6 and 129/S ( 52 ) and by the haplotype pattern in Table 5 showing that B6 and 129 are identical in the Lipg region.   ( Table 4 ). Furthermore, for Lipg , there is a consistent pattern whereby the strains with the high HDL allele expressed lower Lipg mRNA compared with the strains with the low HDL allele ( Fig. 5 ). Finally, a deficiency complementation test implicates Lipg as the gene involved in three of the QTL on distal Chr 18. In this test, the interaction of Lipg genotype with either high-or low-allele strains was signifi cant, demonstrating complementation of Lipg with the Chr 18 HDL QTL. Likely it is this consistent difference in expression levels, caused by a noncoding polymorphism that causes the distal Chr 18 HDL QTL. The relevant haplotype region contains at least two transcription factor binding sites with such a polymorphism.
The results presented here for Lipg are consistent with the gene function and the phenotypes observed both in Lipg knockout mice and transgenic mice. Lipg encodes endothelial lipase, and endothelial lipase hydrolyzes HDL phospholipids in vivo, thus generating smaller phospholipid-depleted HDL particles that are more rapidly catabolized by both the kidney and the liver ( 58 ). Lipg knockout mice have increased HDL levels along with increased plasma cholesterol, phospholipids, and associated apolipoproteins ( 51,59 ). Overexpression of human Lipg in mice results in a signifi cant decrease in HDL cholesterol and apolipoprotein A-I levels ( 59 ).
In humans, Lipg maps to 18q21.32 and is included in a human QTL for HDL ( 46 ) and apolipoprotein A-I levels ( 60 ). It has been shown that apolipoprotein A-I is involved in the pathway modulating endothelial lipase function and HDL regulation ( 61 ). In addition, associations between LIPG SNPs or haplotypes and HDL have been found in several human studies ( 51,62,63 ) with indications that SNPs in regulatory regions are important ( 64 ). Exactly which and how distant the regulatory regions operate to regulate Lipg remains an important question that further mouse studies could help address in the future.
The number of QTL crosses available for meta-analysis is a limiting factor; however, as more data sets become available, the ability to fi ne map the causal genes should also increase. This strategy relies heavily on the quality of databases, in particular the accuracy of SNPs. Currently, it is estimated that there is a ‫ف‬ 57% false-negative rate of discovery in the mouse inbred strain data set ( 37 ). Even a single SNP between two strains is suffi cient for a QTL, and this can be overlooked by such an approach, requiring a broader application of tools such as complementation testing to prove candidate genes. However, as SNP databases increase in both size and accuracy, the false-negative rate should decline, and the success rate of identifying QTL genes should improve.
The authors thank Dr. Keith DiPetrillo who performed the B6 × D2 cross, Peter Reifsnyder and Pam Stanley who performed the (NZO × NON) × NON cross, and Harry Whitmore and Susan Sheehan for excellent technical assistance. QTL, most of which have been replicated in previous studies, indicating that the QTL are controlled by ancestral mutations common among inbred strains.
The additional crosses allow for one more HDL QTL gene, Lipg , to be identifi ed. For the Chr 18 QTL, combining crosses and haplotype analysis based primarily on the dense resequencing of B6, D2, and C3H (http://mouse. perlegen.com/mouse/) identifi ed six likely candidates. Sequence analysis revealed that these genes did not contain coding sequence variants that could explain the QTL. However, an examination of expression analysis reveals that of the six genes narrowed by haplotype, only Lipg is consistently differentially expressed between relevant strains. Indeed, when examining all 165 probe sets in the 95% combined-cross confi dence interval (data not shown), Lipg remains the most consistently differentially regulated . Data for HDL presented as means (mg/dl) ± SD. The number of F1 mice used for each group is n = 10-15. Open circles are the high allele strains (C3H, D2, and NZB); closed circles are the low allele strains (B6 and SM). A: B6 × C3H cross. B: B6 × D2 cross. C: NZB × SM cross. The defi ciency complementation consists of comparing ⌬ 1 with ⌬ 2; if there is no complementation, ⌬ 1 = ⌬ 2 and lines are parallel. If there is complementation, the two deltas are significantly different from each other and the lines are not parallel. P values were calculated using two-way ANOVA test in JMP software. For all comparisons, P < 0.001.