Reduced cholesterol and triglycerides in mice with a mutation in Mia2, a liver protein that localizes to ER exit sites.

Through forward genetic screening in the mouse, a recessive mutation (couch potato, cpto) has been discovered that dramatically reduces plasma cholesterol levels across all lipoprotein classes. The cpto mutation altered a highly conserved residue in the Src homology domain 3 (SH3) domain of the Mia2 protein. Full-length hepatic Mia2 structurally and functionally resembled the related Mia3 protein. Mia2 localized to endoplasmic reticulum (ER) exit sites, suggesting a role in guiding proteins from the ER to the Golgi. Similarly to the Mia3 protein, Mia2’s cytosolic C terminus interacted directly with COPII proteins Sec23 and Sec24, whereas its lumenal SH3 domain may facilitate interactions with secretory cargo. Fractionation of plasma revealed that Mia2cpto/cpto mice had lower circulating VLDL, LDL, HDL, and triglycerides. Mia2 is thus a novel, hepatic, ER-to-Golgi trafficking protein that regulates cholesterol metabolism.

In an effort to reveal novel pathways of cholesterol homeostasis in vivo, N -ethyl-N -nitrosourea-(ENU) mutagenized, C57BL/6 (B6) mice were screened, and affected families showing altered levels of plasma cholesterol were selected. One family derived from this screen, couch potato ( cpto ) mutant mice, had dramatically low plasma levels of both total cholesterol and HDL-cholesterol (HDL-C). The cpto phenotype was associated with a Phe-to-Ser substitution at a highly conserved position within the Src homology domain 3 (SH3)-encoding domain of the Mia2 gene.
MIA2 was originally described as a human gene that encodes a protein with an N-terminal SH3 domain homologous to those of two other genes, melanoma inhibitory activity difl uoride membranes. The proteins were detected by Western blot, using the above antibodies, or mouse monoclonal anti-FLAG M2 (Sigma), followed by HRP-coupled secondary antibodies and ECL Plus Western blot detection reagent (Amersham).

Immunocytochemistry
Cells (HuH-7, NIH-3T3) were grown overnight in 12-well culture dishes on glass coverslips, rinsed in 1× PBS, then fi xed in 10% formalin for 15 min at room temperature. Cells were then permeabilized in either dilution buffer (1× TBS, 0.05% Tween-20, 0.5 M NaCl, 3% BSA, 0.3% Triton X-100) or detergent-free dilution buffer plus digitonin (1× TBS, 0.5 M NaCl, 3% BSA plus 50 g/ml digitonin) for 30 min at room temperature. Primary antibodies were diluted 1:200 in the same buffers, and incubated on the cells for 1 h at room temperature. Following two washes in 1× PBS, Alexa 488-or Alexa 546-conjugated secondary antibodies were added to the cells for 30 min at room temperature. After two washes in 1× PBS, coverslips were inverted on a drop of ProLong Gold with DAPI (Invitrogen) on a glass microscope slide for mounting, and staining was examined via fl uorescent microscopy.

Yeast two-hybrid assay
The yeast reporter strain L40, which contains an integrated LacZ reporter under the control of the LexA operon ( 14 ), was transformed with plasmids expressing an LexA-Mia2 1106-1397 fusion protein (bearing the TRP1 selectable marker) and one of several VP16 fusion proteins (bearing the LEU2 selectable marker). Yeast, grown on solid media lacking tryptophan and leucine to maintain the plasmids, was transferred to nitrocellulose and lysed by freeze/thaw in liquid nitrogen. The yeast-containing nitrocellulose was then incubated in X-gal substrate, and the appearance of blue precipitate was recorded as described ( 15 ).

Body composition and plasma metabolite analysis
Glucose, triglyceride, total and HDL-C, and nonesterifi ed FA levels in plasma from individual, 4 hour-fasted mice were measured on a clinical blood chemistry analyzer (AU400e; Olympus). Insulin levels in duplicate samples were determined by enzymelinked immunosorbent assay (Crystalchem). Body composition analysis was performed using the EchoMRI-100 whole-body composition analyzer (EchoMRI).

Plasma fractionation and cholesterol and triglyceride assay
Fresh mouse plasma was pooled by Mia2 genotype (four to fi ve animals per pool), and frozen until use. Thawed plasma pools were fractionated by fast-protein liquid chromatography (FPLC), over Superdex 200 10/300 GL and Superdex 200 HR10/30 columns in series. Forty 0.5 ml fractions were collected per genotype pool. Total cholesterol in FPLC fractions was quantitated using a ThermoDMA 2350-400H kit, following the manufacturer's instructions. Triglycerides were measured with a Raichem 84098 kit, following the manufacturer's instructions. Apolipoproteins in the fractions were estimated after visualization in SDS-PAGE gels and protein scanning.

Liver lipid extraction and quantitation
Hepatic lipids were extracted following an adaptation of published methods ( 16 ). Briefl y, pieces of mouse liver were weighed, then homogenized in 10 vols of 1× PBS. Homogenate (120 l) was then extracted with 0.8 ml of a 2:1 (v:v) mixture of chloroform-methanol, with vortexing to mix thoroughly. After adding 100 l more of 1× PBS and remixing, samples were spun at 4,200 g at 4°C for 10 min. Two hundred microliters of the organic (lower) phase was air-dried in a fresh tube then resuspended in 100 l of

Animals
All mice were housed, fed, bred, and maintained under standard conditions. All procedures were conducted in conformity with the Public Health Service Policy on the Humane Care and Use of Laboratory Animals and were approved by the Scripps Research Institute's Institutional Animal Care and Use Committee. The mouse N -ethyl-N -nitrosourea screen was performed as described ( 11,12 ), using C57BL/6J animals for the initial mutagenesis. For mapping studies, progeny of ENU-mutagenized founders were bred to the 129S1/SvIMJ background.

Histology
Animals were euthanized for tissues as adults. Livers were fi xed in formalin at 4°C, then processed for paraffi n embedding by the Scripps Histology Core Facility, and sectioned, mounted, and stained with hematoxylin and eosin.

Mapping and identifying the gene responsible for the cpto phenotype
The affected ( cpto ) family was identifi ed in the G3 generation, and backcrossed to B6 to demonstrate heritability. Affected ( cpto ) mice in the B6 background were bred to 129S1/SvIMJ mice, and the resulting (heterozygous) progeny were intercrossed to generate the F2 generation. Genomes of affected and unaffected F2 mice were screened for correlation of the low-cholesterol phenotype with inheritance of B6 strain-specifi c , single-nucleotide polymorphisms (SNPs), as described ( 13 ). Further recombination refi ned the interval to chromosome 12, at 58.2-60.3 Mbases. Exons from all genes within the interval were amplifi ed and sequenced until the mutation in Mia2 was identifi ed.

cDNA construction
Mouse Mia2 was amplifi ed in three pieces (a 5 ′ fragment from bp 1-984, a 2,334 bp segment spanning the internal Eco RI and SbfI sites, and a 1,275 bp fragment from the SbfI site to the 3 ′ end of the open reading frame) that were ligated together using overlapping Eco RI and SbfI sites. For wild-type (WT) and cpto mutant versions of the fi rst two fragments, RNA was prepared from WT and Mia2 cpto livers using Trizol (Sigma), and cDNA was prepared using Super-Script III reverse transcriptase (Invitrogen) before amplifi cation by PCR. For the third fragment (SbfI site to 3 ′ end) of mouse Mia2 , a mouse Ctage5 cDNA clone (ID BC064355) was obtained from the Origene full-length cDNA collection, and used as template for PCR. Mouse Mia2 x1-4 and Mia2 x1-6 cDNAs were generated using the full-length Mia2 cDNA as template in PCR, using primers containing the alternative 3 ′ coding sequences. For human MIA2 , RNA was prepared from HepG2 cells. cDNA was reverse transcribed using SuperScript III (Invitrogen), and used as template for PCR. All PCR-generated clones were verifi ed by double-stranded sequencing.

Western blots
Polyclonal rabbit anti-human MIA2 antibodies (Antigenix America; Huntington Station, NY), polyclonal rabbit anti-human CTAGE5 antibodies (Atlas Antibodies/Sigma), polyclonal rabbit anti-human apoA-I antibodies (Calbiochem; San Diego, CA), polyclonal goat anti-human apoB antibodies (Calbiochem), and monoclonal mouse anti-GM130 and anti-Sec31A antibodies (both from BD Biosciences Pharmingen) were purchased and used according to the manufacturers' instructions. Proteins were routinely separated by electrophoresis on 3-8% NuPAGE Tris-Acetate or 4-12% NuPAGE Bis-Tris precast gels (Invitrogen) following the manufacturer's instructions, and transferred to polyvinylidene cpto phenotypes in a total of 32 mice, both female and male ( Fig. 1F, G ). Genotypically, in over 11 litters from HET by HET crosses, mice were obtained at roughly the expected 1:2:1 Mendelian ratio (25 MUT: 32 HET: 20 WT). We therefore concluded that the Mia2 F91S mutation probably played a causative role in the cpto phenotype, and will henceforth refer to the Mia2 F91S mutant allele as Mia2 cpto .
Plasma triglycerides, which like cholesterol are transported within lipoproteins, also were substantially lower in

Mia2 shows extended homology to Mia3 and Drosophila Tango1
The original report of Mia2 cloned from mice described a gene with four coding exons ( 1 ). In examining the sequence of Mia2 , we observed that the "gene" immediately distal to Mia2 , Ctage5 / Mgea6 , encoded both coiled-coil and proline-rich motifs highly homologous to the carboxyl termini of both Mia3 and Drosophila Tango1 ( Microarray-based analysis of Mia2 expression in mice, using a probe set directed against the Mia2 SH3-encoding exons (viewable at http://biogps.org), detected Mia2 expression robustly in liver, with additional signals in kidney and small intestine ( 21 ). This agreed with published reports of liver-specifi c Mia2 expression in mouse and human tissues ( 1,7,8 ). In contrast, analysis using a probe set targeting the last two exons of Ctage5 showed expression in mice to be widespread. Similarly, our own qRT-PCR analysis using primer/probe combinations specifi c for Mia2 or Ctage5 showed that Ctage5 -containing transcripts were widely expressed, whereas Mia2 -containing transcripts were restricted to liver, intestine, kidney, and heart (see supplementary Fig. IIID).
1% Triton X-100 in ethanol. This was then air dried and resuspended in 500 l 1× PBS, for the fi nal lipid extract. Fifty microliter aliquots of this extract were used to measure either cholesterol, using the Amplex Red cholesterol assay kit (#A12216, Life Technologies; Carlsbad, CA) or triglycerides (Triglyceride Assay Kit #10010303, Cayman Chemicals; Ann Arbor, MI).

RNA quantitation
RNA was isolated from various tissues with Trizol (Invitrogen) according to the manufacturer's instructions. Primer/probe combinations for quantitative reverse-transcriptase PCR (qRT-PCR) were purchased from Applied Biosystems (ABI), catalog number Mm00616368_m1 for mouse Mia2 and Mm00524692_m1 for mouse Ctage5 . qRT-PCR was done on an ABI 7900HT sequence detection system. RNA was quantifi ed by the comparative C T method using m 36b4 as an internal standard. For each primer/ probe combination ( Mia2 or Ctage5 ), values for different tissues were normalized to values for liver.

The couch potato mutation alters the SH3 domain of Mia2
The couch potato ( cpto ) mutation was initially identifi ed in a screen of ENU-mutagenized B6 mice, designed to reveal recessive mutations that confer signifi cant changes in plasma cholesterol. One family derived from a single ENU-mutant founder displayed a distinctly bimodal distribution in cholesterol levels, with roughly one-fourth of the animals having dramatically lower cholesterol levels ("affected," Fig. 1A ), consistent with inheritance of a single recessive allele. Similar proportions of animals also had low plasma HDL-C ( Fig. 1B ). This phenotype was dubbed couch potato , and the family was selected for further analysis: affected mice were crossed to the 129S1/ SvImJ (129) background for mapping, or backcrossed to B6 for physiological studies. The cpto phenotype was initially mapped to a 13.7 Mb interval on mouse chromosome 12 in 99 B6.129 hybrid F2 progeny, using a genome-wide array of SNPs as molecular markers. Continued backcrosses to 129 and mapping with a denser set of SNP markers revealed a tight correlation between the cpto low-cholesterol/ low HDL-C phenotype and inheritance of an interval from 52.0 to 54.6 Mb (mouse genome build NCBIM33) from the B6 (mutant) background ( Fig. 1C ).
Sequencing of candidate genes within this interval (see supplementary Fig. I) revealed a T-to-C transition in the third exon of the Mia2 gene, which replaces the WT phenylalanine-91 with a serine in the Mia2 protein (F91S, Fig.  1D ). This substitution occurs within Mia2's N-terminal SH3 domain. Comparison of Mia2's protein sequence with its three other mouse paralogs (Mia, Otor, and Mia3) showed that this position is otherwise invariant ( Fig. 1E ). Furthermore, in both the solution structure ( 17 ) and crystal structure ( 18 ) of MIA's atypical SH3 domain, the analogous residue is located at or adjacent to the surface, in a region shown in other SH3 structures to comprise the ligand-binding interface, indicating that an F91S substitution could be functionally signifi cant. After backcrossing for fi ve generations into the B6 background, this Mia2 F91S allele segregated precisely with both the low total cholesterol and low HDL-C Fig. 1. Mapping, identifi cation and phenotype of the cpto mutation. Distribution of total plasma cholesterol (A), and HDL-C levels in couch potato ( cpto ) mice (B). Two peaks were visible in each distribution, with "affected" animals defi ned as those covered by the lower cholesterol or lower HDL peaks. C: In animals bred with 129S1/SvImJ, phenotypically affected ( cpto ) mice were homozygous for C57BL6/J-specifi c SNP markers between 52 and 54.6 Mb of mouse chromosome 12 (genome build NCBIM33). "H"/orange represents heterozygous (one copy B6, one copy 129) and "B"/green represents homozygous B6 at each SNP. Animals were sorted left-to-right by ascending HDL levels. D: Electropherograms and corresponding nucleotide and amino acid sequence of exon 3 of Mia2 from wild-type (WT) mice or mice heterozygous (HET) or homozygous mutant (MUT) for the Mia2 F91S mutation. E: Alignment of the N-termini of all four mouse Mia2homologous proteins. Highlighting shows that F91 of Mia2 is conserved in the other three paralogs (Mia, Otor, Mia3). F, G: The cpto lowcholesterol/low-HDL phenotype followed inheritance of the Mia2 F91S mutation in female (F) and male (G) cpto mice, after fi ve generations of backcrossing into the B6 background. Data are expressed as mean ±SEM. Student's t -test, * P < 0.05, *** P < 0.001.
anti-MIA2 and anti-CTAGE5 was ‫ف‬ 240 kDa, and was present in both mouse liver and human hepatoma HuH-7 and HepG2 cell lines ( Fig. 2C ). This protein also comigrated with a recombinant protein (Mia2 x1-30 , detected with anti-FLAG) based on the longest transcript class, containing all six Mia2 exons and all but the fi rst Ctage5 exons ( Fig. 2C , Mia2 p240). We refer to this recombinant cDNA as fulllength Mia2 ; the sequence of the protein it encodes is shown in supplementary Fig. IVA. Other mouse and human anti-MIA2-reactive protein species were evident as well, although Because the cpto mutation affects Mia2 's SH3 domain, this limited " Mia2 " expression profi le should refl ect the tissues in which Mia2 function affects cholesterol levels. Notably, the major site of Mia2 expression is the liver, the organ from which nascent HDL and VLDL are secreted.

Mia2 produces 240 and 120 kDa protein isoforms
Detection of multiple alternative Mia2 transcript classes made it important to investigate Mia2 protein isoforms in vivo. The largest endogenous protein detected with both Numbers in corresponding domains of mMia3 and dTango1 refl ect percent amino acid identity to Mia2 within each region. Although primary sequence conservation is low, amino acid composition in acidic and proline-rich regions is highly similar. B: RT-PCR-based detection of Mia2 splicing events terminating in Mia2 exon 4 (ex4UTR products) and exon 6 (ex6UTR products) and continuing through to Ctage5 -specifi c exons (ex9R products) in total RNA from livers of Mia2 +/+ (WT), Mia2 cpto/+ (HET) and Mia2 cpto/cpto (MUT) mice. Ethidium bromide-stained gel shows RT-PCR products; cartoon shows relative positions of PCR primers. C: Anti-MIA2 and anti-CTAGE5 both detected a ‫ف‬ 240 kDa protein that comigrated with the protein produced by full-length Mia2 x1-30 expression vector. D: HuH-7 cells transfected with control (CON) or pooled MIA2 -specifi c siRNAs were harvested 2 or 3 days posttransfection. Western blotting with anti-CTAGE5 showed reduced Mia2 p240 isoform, whereas the Mia2 p120 isoform (which should not be regulated by the MIA2 -specifi c siRNAs) was unaffected. E: In whole-cell lysates from mouse livers, full-length (mutant SH3-containing) Mia2 p240 was present at lower levels in MUT animals relative to HET animals, whereas Mia2 p120 levels did not vary.

MIA2 is an ER exit site protein
The structural homology between Mia2 p240 and Drosophila Tango1 and vertebrate Mia3/Tango1 suggested that Mia2 may also have a similar function. Mia3/Tango1 was identifi ed as an ER exit site protein that interacts with the COPII coat proteins Sec23 and Sec24 through its C-terminal proline-rich domain, and with select Golgi-bound cargo, such as collagen VII, through its N-terminus ( 4 ). Similarly, dTango1 is ER exit site-localized ( 23 ), and is required for secretion of some proteins in a Drosophila tissue culture system ( 9 ). If Mia2 functionally resembles these proteins, Mia2 should fi rst have a similar subcellular location.
Human HuH-7 cells were used, because they express primarily a full-length MIA2 protein similar to the major species in mouse liver ( Fig. 2C-E ). Immunostaining of these cells with MIA2-specifi c polyclonal antibodies ( Fig. 3 ) identifi ed two discrete subcellular locations: bright, asymmetric staining immediately adjacent to the nucleus, and a dispersed, punctate cytosolic signal. The juxtanuclear MIA2-positive signal was adjacent to but distinct from the early Golgi marker, GM130 ( Fig. 3A ). Anti-CTAGE5 staining of HuH-7 cells also produced perinuclear and punctate cytosolic signals, but in contrast to anti-MIA2, the anti-CTAGE5 perinuclear signal largely colocalized with GM130 (see supplementary Fig. VIA), although it was unclear whether this staining refl ected specifi c binding of MIA2 p240, MIA2 p120, or of other human CTAGE-family pseudogene proteins (see supplementary Fig. V).
The majority of the anti-MIA2 signal, however, was in a punctate staining pattern in the cytosol that resembles that of ER exit sites. Indeed, MIA2 staining was coincident with the signal for the COPII subunit, Sec31A ( Fig. 3B ). Substantially less overlapping signal was seen with anti-CTAGE5 (see supplementary Fig. VIB). MIA2's subcellular localization thus closely resembled that of MIA3/ TANGO1, which remains in the ER, and does not accompany COPII vesicles to the Golgi ( 4 ). This argues that lower plasma cholesterol levels resulting from mutations in Mia2 involve some aspect of cholesterol homeostasis that is dependent on ER-to-Golgi traffi cking.

Mia2 protein topology and COPII binding resemble that of Mia3/Tango1
Because Mia2's SH3 domain carries the cpto mutation, its position relative to the ER is an important component of its function. Differential permeabilization ( 24 ) of HuH-7 cells was used to identify regions of MIA2 that were exposed to the cytosol ( Fig. 4A ). Digitonin preferentially permeabilizes the cellular membrane, but not the internal (Golgi, ER) membranes ( 25 ). To allow complete antibody access to all internal compartments, a parallel set of cells was incubated in Triton X-100. As shown in Fig. 4A , Golgiand/or ER-localized apoA-I proteins were detectable solely with Triton X-100 permeabilization, and not with digitonin, indicating that these proteins were entirely enclosed within organelle membranes. In contrast, Golgi protein GM130 and COPII protein Sec31A were detected under both conditions, indicating that these epitopes resided in their identity is unclear, inasmuch as the majority of the additional species appear to be much larger than the proteins containing Mia2 exons 1-4 or exons 1-6 ( Fig. 2C ).
Additional anti-CTAGE5-reactive proteins were detectable, particularly at or below 120 kDa, in most tissues and cell lines examined ( Fig. 2C and supplementary Fig.  IVB). It should be noted that in human cells, but not in mice, anti-CTAGE5 may cross-react with a family of expressed pseudogenes, including the previously described CTAGE1, CTAGE2, and CTAGE4 ( 20,22 ), all of which should encode proteins of similar size to MIA2 p120 (see supplementary Fig. VA). These proteins are highly homologous to human MIA2/CTAGE5 in the region used as an immunogen for anti-CTAGE5 (see supplementary Fig. VB), but should not react with anti-MIA2. Full-length Mia2 p240 in mice was detectable in liver, and may also be present in lung, but was not detectable in muscle, brain, adipose tissue, or testis (see supplementary Because the anti-MIA2-reactive proteins, all of which are larger than 120 kDa, should contain the SH3 domain bearing the cpto mutation, the abundant p240 MIA2 isoform was examined in further detail. A pool of siRNAs specifi c for MIA2 -coding exons was used to knock down MIA2 expression in HuH-7 cells. An anti-CTAGE5 Western blot showed that the MIA2 siRNAs substantially diminished the abundance of the p240, full-length MIA2 isoform, while producing no detectable change in the abundance of the 120 kDa species ( Fig. 2D ). This confi rmed that the p240 isoform was encoded by both MIA2 and CTAGE5 exons, inasmuch as it is recognized by a CTAGE5-specifi c antibody, but it was also sensitive to the quantities of Mia2 transcripts. It should also be noted that in Mia2 cpto mice, compared with heterozygotes, homozygous mutant livers often contained lower amounts of the p240 Mia2 isoform, whereas the abundance of the p120 Mia2 isoform was not altered by the cpto mutation ( Fig. 2E ). Thus, the cpto mutation may alter the stability of SH3 domain-containing isoforms, in addition to effects of the primary amino acid change on protein function.
Based on this evidence, we propose that Mia2 and Ctage5 should be considered a single gene, with multiple alternative 5 ′ exons, and that this gene be referred to as Mia2, to acknowledge its homology to Mia and Mia3 . Support for this arrangement is seen in the predicted structures of Mia2 / Ctage5 genes in other species, such as chimp (XP_001147938), orangutan (NP_001128150), marmoset (XP_002753890), platypus (XP_001513694), and fi nch (XP_002199906), all of which include isoforms encoded by Mia2 -specifi c exons spliced onto downstream Ctage5 exons. Like Mia2 , the Mia3 gene appears to give rise to transcripts analogous to the Ctage5 -only transcripts shown in supplementary Fig. IIIC (e.g., EST sequence CR990232). Because both Mia2 and Mia3 are capable of generating both SH3-containing and SH3-lacking protein isoforms, there may be a functional signifi cance to this organization.
with Sec24C, components of the COPII inner vesicular coat ( Fig. 4B ). This suggests that, like MIA3/TANGO1, Mia2 functions in assembly of COPII vesicles, and that its lumenally localized SH3 domain, altered by the cpto mutation, might interact with outgoing cargo.

couch potato mutant mice have low plasma cholesterol and triglycerides, and increased hepatic triglycerides
Whole plasma had reduced total cholesterol and HDL-C in Mia2 cpto/cpto mice ( Fig. 1F ). To more directly assess the different lipoproteins and their lipid content, size-fractionated plasma was examined. FPLC plasma fractionation revealed that Mia2 mutant mice had lower cholesterol levels in all lipoprotein fractions, not just in HDL ( Fig. 5A ). As expected, cholesterol in the HDL fractions (samples 18-22, Fig. 5A ) was severely reduced among Mia2 cpto/cpto animals (MUT), to approximately 25-30% of the levels observed in WT, whereas cholesterol levels in Mia2 cpto/+ animals (HET) were indistinguishable from those in WT. Furthermore, cholesterol levels in fractions from the VLDL/LDL peak (samples 5-9, Fig. 5A ) were also decreased to 25-30% of WT levels in MUT animals. Triglyceride, which is largely carried in VLDL particles, was also affected by the cpto mutation, with a signifi cant reduction among homozygous mutants in the VLDL/LDL fractions (samples 5-9, Fig. 5B ). The cpto mutation thus affected the lipid content of multiple plasma lipoproteins. It should be noted that the degree of reduction in triglycerides in the fractionated plasma is similar to that of cholesterol in VLDL/LDL and HDL, although the reduction is larger than that observed in whole plasma (see supplementary Fig. IIA). This may be because free glycerol is present in whole plasma, but not the fractionated samples, although this has not been directly tested. the cytosol ( Fig. 4A ). Anti-MIA2 recognized epitopes including Mia2's N-terminal SH3 domain and the adjacent acidic region (see supplementary Fig. VIIB) and produced a signal with Triton X-100 permeabilization, but not with digitonin ( Fig. 4A ). In contrast, anti-CTAGE5, which recognized epitopes in the C-terminal coiled-coil domain (see supplementary Fig. IVB), was positive under both permeabilization conditions ( Fig. 4A ). Thus the N-terminal half of MIA2, including the SH3 and acidic region, was lumenally localized, whereas the C-terminal coiled-coil and proline-rich portions of MIA2 probably resided in the cytosol. This orientation is identical to the topology recently demonstrated for MIA3/TANGO1 ( 4 ).
Because anti-CTAGE5 may also have recognized expressed pseudogenes in human cells (see supplementary Fig. V), HuH-7 cells were transfected with expression vectors producing C-terminally FLAG epitope-tagged Mia2 proteins of various lengths (see supplementary Fig. VIIA). In agreement with the anti-MIA2 and anti-CTAGE5 staining, the C terminus of full-length Mia2 x1-30 (both WT and cpto mutant versions) was detectable after permeabilization with both digitonin and Triton X-100. A shorter version, Mia2 x1-6 , which encoded a protein with a truncated/ablated transmembrane domain, appeared to have a lumenally located carboxy terminus (see supplementary Fig. VIIA). It should be noted, however, that all overexpressed MIA2 proteins, including the shorter versions shown in supplementary Fig. VII, appear to mislocalize throughout the ER, and not to ER exit sites.
Because immunostaining indicated Mia2 localized to ER exit sites, Mia2's ability to interact directly with components of the COPII vesicular coat, as observed for MIA3/ TANGO1 ( 4 ), was also examined. In a yeast two-hybrid assay, the proline-rich C-terminal region of Mia2 (LexA-Mia2 1106-1397 ) interacted specifi cally with both Sec23A and formed HDL, and apoB-100 is the primary protein in VLDL and LDL ( 26 ). Despite apparent increased hepatic triglyceride levels and decreased plasma levels of cholesterol and triglycerides, however, no consistent, Mia2 cptodependent alteration in circulating apolipoproteins was observed. Analyzing whole, unfractionated plasma from individual mice revealed wide variation in the levels of apoA-I from mouse to mouse, but low levels of apoA-I in whole plasma did not consistently correlate with the Mia2 cpto genotype (see supplementary Fig. VIIIA). Levels of plasma apoE, apoB-100, and apoB-48 were similarly variable (see supplementary Fig. VIIIA).
To control for inter-mouse variability, and to determine whether differences in apolipoprotein levels were associated with specifi c lipoproteins, we identifi ed the proteins in individual fractions of the fractionated plasma pools ( Fig. 5A, B ) by Western blotting ( Fig. 5D-F ). Whereas apoA-I was modestly reduced in fractions of the MUT pool corresponding to HDL, relative to the HET and WT pools (60% of WT levels), this reduction was less dramatic than the overall reduction seen in mutant HDL-C ( Fig. 5A, D ). Similarly, cholesterol was dramatically reduced in VLDL/ LDL fractions of the MUT pool relative to both HET and Because Mia2 resembles secretory pathway-regulating proteins such as Mia3 and Tango1, hepatic cholesterol and triglyceride levels were directly measured to determine whether lowered plasma lipids coincides with increased retention of lipids within the liver ( Fig. 5C ). Interestingly, no signifi cant difference was evident between HET and MUT hepatic cholesterol levels ( Fig. 5C , left panel). In contrast, however, there was a distinct trend toward increased triglyceride levels in the liver ( Fig. 5C , right panel), although this fell short of statistical signifi cance ( P = 0.07). Some inherent regulation of hepatic triglyceride levels in Mia2 cpto mice must still be operative, however, because mutant mice did not progress to an overtly steatotic state histologically (see supplementary Fig. IIF).

cpto mutation does not appear to directly alter apolipoprotein secretion
Both Mia3/Tango1 and fl y Tango1 appear to regulate secretory processes, and depletion of either protein results in demonstrable defects in the secretion of specifi c proteins ( 4,9 ). It is possible, then, that defective Mia2 could lower plasma cholesterol levels through reduced secretion of apolipoproteins. ApoA-I is the major protein in newly Fig. 4. MIA2 topology as revealed by differential permeabilization and staining of MIA2 and CTAGE5 epitopes. A: HuH-7 cells were grown on coverslips, fi xed in 4% formaldehyde/1× PBS for 10 min at room temperature, then permeabilized for 15 min at room temperature in either TBST (containing 0.1% Tween-20) plus 0.3% Triton X-100 (bottom row), or TBS (without Tween-20) plus 50 g/ml digitonin (top row). Both anti-GM130 and anti-Sec31 functioned comparably under either permeabilization condition, indicating a cytosolically exposed epitope. In contrast, apoA-I was detectable in the ER and Golgi only with Triton X-100 permeabilization, indicating that it was lumenally enclosed. Anti-MIA2, which recognizes the N-terminal half of MIA2, matched the lumenal pattern, whereas anti-CTAGE5, which recognizes the C-terminal portion of MIA2, stained a cytosolic epitope. B: Yeast two-hybrid results ( ␤ -galactoside assay) showing interaction between mouse Mia2's C-terminal proline-rich domain (Mia2 1106-1397 ) and the human COPII proteins SEC23A and SEC24C. Colonies containing both an LexA-Mia2 1106-1397 bait vector and either empty VP16 prey vector (as a negative control) or vectors encoding VP16-SEC23A or -SEC24C fusion proteins were streaked and grown together, lifted on a single fi lter and processed. Blue color indicates protein-protein interaction.
Levels of other apolipoproteins, such as apoE (found on HDL and VLDL/LDL) and apoB-48 (chylomicron and LDL in mouse) also did not correlate with the Mia2 cpto genotype ( Fig. 5F and data not shown). Furthermore, no differences were observed in the levels of apoA-I, apoB-100, or apoE secreted into the media of HuH-7 cells either WT pools ( Fig. 5A ), yet no obvious reduction in apoB-100 levels was evident in MUT pool fractions relative to HET pool fractions ( Fig. 5E ). Both MUT and HET pools contained less apoB-100 than did WT pools, but because HET animals were phenotypically unaffected by the Mia2 cpto mutation, this difference was not related to the Mia2 cpto mutation. defects are the mechanism responsible for the lowering of cholesterol levels in Mia2 cpto/cpto mice, and that we have not found the cargo that is altered. Attempts to detect proteinprotein interactions with the Mia2 N-terminal portion have been hindered by mislocalization of the overexpressed protein (throughout the ER; see supplementary Fig. VIIA), and inability of the human-specifi c antibodies (anti-MIA2 and anti-CTAGE5) to immunoprecipitate native mouse or human protein. A pilot immunoprecipitation-mass spectrometry approach using an overexpressed, epitope-tagged Mia2 N-terminal fragment failed to detect interaction with any proteins known to be associated with cholesterol metabolism (data not shown).
The Mia2 cpto mutation apparently did not directly alter general or cholesterol-specifi c hepatic protein secretion, although these mice did display a trend toward increased triglyceride levels, but not cholesterol levels, in their livers. Two other possibilities for how Mia2 affected plasma cholesterol levels are an effect at the level of cholesterol and/ or lipoprotein synthesis/assembly, or an effect on lipoproteins after they are secreted, via enzymatic activity, distal uptake, or catabolism. Although we cannot rule out an effect on lipoprotein assembly or postsecreted lipoproteins, we have evidence that fails to support an effect on lipid synthesis. Cholesterol synthesis is regulated by the SREBP family of transcription factors, whose own activity is regulated by ER-to-Golgi traffi cking ( 27 ). If Mia2 is also involved in SREBP regulation, the Mia2 cpto mutation could decrease SREBP processing, and thus downregulate SREBP target genes in the liver. The end result of this cascade in the mouse would be reduced circulating cholesterol and triglyceride levels.  ( Fig. 5 ). Regardless of the mechanism, Mia2 represents a novel gene regulating hepatic cholesterol metabolism.

Mia2, Mia3/Tango1, and human disease
Whereas the Mia2 cpto mutation produces a dramatic cholesterol phenotype in mice, it is not necessarily the case that MIA2 performs a similar role in humans. Some aspects of cholesterol metabolism differ signifi cantly between rodents and humans, including, for example, the relative levels of LDL and HDL cholesterol, which are nearly inverse between mice and humans. Nonetheless, several aspects of Mia2 biology suggest that human MIA2 could function analogously to its murine counterpart. First, as in mice, MIA2 expression in humans is liver specifi c ( 1 ), and we detected an MIA2 protein in human hepatocyte-like HuH-7 and HepG2 cells that was similar to the major Mia2 overexpressing WT versus cpto mutant full-length Mia2 cD-NAs, or of HuH-7 cells transfected with MIA2 siRNA pools. Furthermore, these proteins did not detectably accumulate within the cells (see supplementary Fig. IXA, B, and data not shown).
Arguing against a general defect in secretion, albumin, a liver-secreted blood protein, was not signifi cantly different between mutant and WT as measured by Coomassie staining of fractionated plasma shown in Fig. 5 (data not  shown). Another liver-secreted plasma protein, Rbp4, was not reduced in mutant plasma, as measured by ELISA (see supplementary Fig. VIIIB). We determined whether the Mia2 protein or its isolated SH3 domain interacted with known apolipoproteins, and if these interactions were affected by the cpto mutation. In coimmunoprecipitation experiments using mouse liver membrane extracts from Mia2 cpto/+ or Mia2 cpto/cpto animals, no direct binding of endogenous Mia2 to apoA-I, apoB-100, apoB-48, or apoE was observed (see supplementary Fig. VIIIC). Similar results were observed in HuH-7 cells (data not shown). Together, although there is a slight reduction in some apolipoproteins in Mia2 cpto/cpto plasma, the low cholesterol levels in Mia2 cpto/cpto mice did not result from either a general or an apolipoprotein-specifi c defect in Mia2 -dependent secretion from the liver.

DISCUSSION
Couch potato , a novel ENU-induced mutation that reduced circulating VLDL-C, LDL-C, HDL-C, and triglyceride levels, has been identifi ed in mice. The mutation resides in a poorly characterized gene not previously related to cholesterol metabolism, Mia2 . The Mia2 cpto mutation alters a highly conserved residue in Mia2's N-terminal SH3 domain (F91S). Mia2 is a largely liver-specifi c protein, encoded by exons previously annotated separately as the Mia2 and Ctage5/Mgea6 genes. The transcript that encodes full-length Mia2 (p240) spans both sets of exons, and like the related MIA3/TANGO1 protein ( 4 ), localized to ER exit sites, with a lumenally localized SH3 connected to a cytosolic carboxy-terminal proline-rich domain that interacted with the COP II proteins Sec23 and Sec24.
Mia2 is the fi rst non-SH3-only member of this family of genes to have a mutant phenotype in whole animals. The Mia2 cpto point mutation modestly reduced the overall steadystate level of full-length Mia2 protein in the liver, although the exact mechanism by which this resulted in lower levels of VLDL-C/LDL-C, HDL-C, and triglycerides in the plasma remains an open question. In vitro reduction of the levels of either fl y Tango1 or human MIA3 results in demonstrable, cargo-specifi c, defects in protein secretion ( 4,9 ). In Mia2 cpto/cpto mice, there were variably reduced levels of some liver-secreted apolipoproteins, such as apoA-I and apoE ( Fig. 5 ), yet the degree of reduction in apoA-I protein was inconsistent and smaller than the overall reduction in HDL-C. In addition, we observed a reduction in VLDL-C, but not of the major protein component of VLDL, apoB-100. It remains possible that cargo-specifi c secretion synthesis ( HMG-CoA reductase , HMG-CoA synthase , and SCAP again), also produced reductions in HRP secretion similar to or greater than those observed for d Tango1 , along with similar ER retention of Mannosidase II-GFP ( 9 ). These genes were excluded from further analysis as "transport and Golgi organization" genes, because they were known regulators of lipid metabolism. Whereas MIA3 binds the COL7A1 protein in an SH3-dependent fashion, and either MIA3 or CTAGE5 depletion results in impaired COL7A1 secretion ( 4,29 ), the conclusion that MIA3 acts as a specifi c adaptor for large cargo at ER exit sites is at odds with the requirement for both dTango1 and mammalian MIA3 for secretion of the 305 amino acid HRP protein ( 4,9 ). An alternative hypothesis could be that, as with the Drosophila metabolic genes above, the observed secretion defects when CTAGE5 , MIA3 , or d Tango1 are mutated or depleted may be indirect, and instead result from alteration of lipid and/or cholesterol metabolism ( 30 ).

Note added in proof
During review of this manuscript, a relevant paper was published by Wilson et al. describing the phenotype of Mia3 knockout mice, which die neonatally due to a defect in collagen secretion. Importantly, this work supports our claim that alternative transcripts arise from the Mia3 gene similar to those that arise from the Mia2 gene. species in mouse liver. Still, no studies to date have linked MIA2 / CTAGE5 variants to alterations in human cholesterol levels. Multiple genome-wide association studies have, however, identifi ed an association between variants within the human MIA3 locus and cardiovascular disease, although this association does not appear to be mediated by plasma lipids ( 5,6,28 ). On the basis of our results in mice, we would expect a mutation in human MIA2 to similarly be associated with cholesterol levels and/or cardiovascular disease; although no such human variants have yet been identifi ed or reported.
It was recently shown that a carboxy-terminal portion of MIA2 (CTAGE5, or MIA2 p120) forms heterooligomers with MIA3/TANGO1 in human tissue culture cells, via the two proteins' cytosolic coiled-coil regions ( 29 ). Although we have no direct evidence, we would expect full-length, SH3-containing MIA2 p240 to also dimerize with MIA3/ TANGO1, inasmuch as the cytosolic portions of MIA2 p240 and MIA2 p120 are identical. Depletion of CTAGE5 is suffi cient to cause cellular retention of the MIA3 cargo, COL7A1 ( 29 ), which binds directly to MIA3 ′ s SH3 domain within the ER lumen ( 4 ). This argues that the integrity of the heterodimer is important for MIA3 activity, and thus we would expect that changes in the levels or function of either MIA2 or MIA3 should alter heterodimer function.
It is also possible that given that MIA2 and MIA3 are both highly similar in their domain structure, bind to each other, and share some protein partners at their carboxyterminal ends, MIA3 genetic variants might directly contribute to cardiovascular disease via a mechanism similar to the (as yet unknown) one altered by the Mia2 cpto mutation in mice. Although modulating protein-protein interactions that occur within the ER lumen may present a formidable hurdle for drug development, MIA2 , MIA3 , or both genes could potentially represent novel targets in the ongoing struggle against human cardiovascular disease.

Mia2/Mia3/dTango1: regulators of secretion, metabolism, or both?
Drosophila Tango1 was originally identifi ed as a regulator of secretion and/or Golgi organization because dTango1 knockdown in Drosophila S2 cells resulted in an 80% reduction in secretion of HRP, along with retention in the ER of a green fl uorescent protein-tagged version of the Golgi protein Mannosidase II ( 9 ). Subsequent work showed that dTango1 regulates secretion of a secretable luciferase, and localized the protein to S2 cell ER exit sites ( 23 ). Similarly, knockdown of mammalian MIA3/TANGO1 results in defective secretion of collagen VII (encoded by COL7A1 ) and HRP, although notably not of collagen I, alkaline phosphatase, or VSV-G; nor is there a detectable reduction in gross protein secretion ( 9 ). Interestingly, similar results have also recently been reported for knockdown of a CTAGE5 -only transcript in HeLa cells ( 29 ). All of these results are consistent with the interpretation that dTango1, Mia3, and related proteins such as Mia2 regulate ER-to-Golgi traffi cking of a discrete set of proteins.
Yet knockdown of genes critical either for lipid biogenesis ( ACC , Drosophila FASN , SCAP ) or for cholesterol/steroid