Isolation of Novel Animal Cell Lines Defective in Glycerolipid Biosynthesis Reveals Mutations in Glucose-6-phosphate Isomerase*

Glycerolipids are structural components for membranes and serve in energy storage. We describe here the use of a photodynamic selection technique to generate a population of Chinese hamster ovary cells that display a global deficiency in glycerolipid biosynthesis. One isolate from this population, GroD1, displayed a profound reduction in the synthesis of phosphatidylcholine, phosphatidylethanolamine, and triglycerides but presented high levels of phosphatidic acid and normal levels of phosphatidylinositol synthesis. This was accompanied by a reduction in phosphatidate phosphatase 1 (PAP1) activity. Expression cloning and sequencing of the cDNA obtained from GroD1 revealed a point mutation, Gly-189 → Glu, in glucose-6-phosphate isomerase (GPI), a glycolytic enzyme involved in an inherited disorder that results in anemia and neuromuscular symptoms in humans. GPI activity was reduced by 87% in GroD1. No significant differences were found in DNA synthesis, protein synthesis, and ATP levels, whereas glycerol 3-phosphate levels were increased in the mutant. Expression of wild-type hamster GPI restored GPI activity, glycerolipid biosynthesis, and PAP1 activity in GroD1. Two additional, independently isolated GPI-deficient mutants displayed similar phenotypes with respect to PAP1 activity and glycerolipid biosynthesis. These findings uncover a novel relationship between GPI, involved in carbohydrate metabolism, and PAP1, a lipogenic enzyme. These results may also help to explain neuromuscular symptoms associated with inherited GPI deficiency.

The production of glycerolipids, including glycerophospholipids and triacylglycerols, is important for the supply of structural components for membrane production and the storage of excess calories in animal cells. This process must be regulated to adjust to the demands of the cell and the environmental conditions. For example, dividing cells require a constant supply of glycerophospholipids for membrane production, whereas diets rich in carbohydrates result in increased production of fatty acids, which are stored as triglycerides. Dysregulation of glycerolipid biosynthesis results in a number of pathologies, including diabetes and obesity (1,2). Metabolic syndromes have become a prominent health issue in the United States as well as in other developed countries (3,4).
The mechanism by which glycerolipid biosynthesis is regulated is still not well defined. This is due, in part, to the fact that, until recently, few of the biosynthetic enzymes had been purified to homogeneity and many of the genes that code for these proteins had yet to be identified. Besides the enzymes directly involved in assembling glycerolipids, there are undoubtedly many genes and gene products involved in controlling glycerolipid biosynthesis.
Forward genetics (mutant isolation) is used to identify genes required for a certain phenotype. This way, one can identify factors important for that phenotype, which may not have been predicted given the existing knowledge about the biochemistry in that area. A number of animal cell mutants have been isolated using screening or selection techniques that target cell lines deficient in the synthesis of specific phospholipid species such as phosphatidylcholine (PC), 2 sphingomyelin (SM), or phosphatidylserine (5)(6)(7)(8). These mutants have helped in our understanding of the synthesis and importance of these phospholipids. In an effort to identify genes important for regulation of glycerolipid biosynthesis in general, we wanted to generate and select for mutants from an established animal cell line that would display global defects in glycerolipid biosynthesis.
Using a photodynamic selection technique we have generated a population of CHO cells that display a decreased ability to incorporate fatty acids into complex lipids. Isolates from this population displayed a reduction in glycerolipid biosynthesis accompanied by a reduction in a phosphatidate phosphatase activity (PAP1, sn-3-phosphatidate phosphohydrolase, EC 3.1.3.4). However, expression cloning and sequencing revealed the primary defect to be a point mutation in glucose-6-phosphate isomerase (GPI, D-glucose-6-phosphate aldoseketose isomerase, EC 5.3.1.9), a glycolytic enzyme with moonlighting cytokine functions (9). Mutations in GPI are the second most frequent cause of inherited glycolytic erythroenzymopathy in humans (10). This autosomal recessive disorder is characterized by a non-spherocytic hemolytic anemia of variable severity, which can present with neurological dysfunctions (10).
Our findings presented here reveal an unexpected relationship between GPI and PAP1. They may also yield insight into the nature of communication between carbohydrate metabolism and glycerolipid biosynthesis and may help explain the neurological dysfunctions associated with the human GPI disorders.
Mutagenesis-Cells were plated into 100-mm diameter tissue culture dishes at 5 ϫ 10 5 cells/dish and allowed to attach overnight. The cells were exposed to 200 g/ml ethylmethane sulfonate for 20 h at 37°C, the ethylmethane sulfonate-containing medium was removed, and the cells were allowed to grow to confluence at 33°C (the permissive temperature) to allow establishment of phenotypes prior to selection.
P12/UV Selection-Mutagenized cells (10 7 cells total) were plated into 100-mm diameter tissue culture dishes (10 6 cells/ dish) and allowed to attach overnight at 33°C. The following day, the cells were shifted to 40°C. After 2 h at 40°C, medium containing P12 was added to a final concentration of 5 M. After 3 h at 40°C, medium was removed, regular growth medium was added, and the cells were incubated for another hour at 40°C to remove any P12 fatty acid not assimilated into complex lipids. The cells were then suspended on a 1.5-mmthick glass plate over a long wavelength (Ͼ300 nm) UV source (Black-Ray UV lamp, Model XX-15L, UVP, Inc., San Gabriel, CA) and irradiated from underneath. The distance from the UV source and the duration of the UV irradiation was predetermined to result in Ͼ95% cell death. The dishes were then returned to 33°C to allow growth of the survivors. The surviving cells were pooled and subjected to the same selection proc-ess twice more. In the third round of selection, the majority of the cells survived. In all cases, the cells were grown at 33°C, and the selections were performed at 40°C. Clonal isolates were generated from the surviving population using limiting dilution at 33°C. Each isolate was tested for phospholipid biosynthesis as described below.
Phospholipid Biosynthesis and Composition-For phospholipid biosynthesis, short term labeling with 32 P i was used. Cells were plated into sterile glass scintillation vials (2.5 ϫ 10 5 cells/ vial) and allowed to attach overnight at 33°C. The next day, some vials were placed at 40°C for 2 h while others were left at 33°C. An aliquot of medium containing 32 P i (25 Ci/ml or 50 Ci/ml) was added, and cells were incubated for 3 h at the appropriate temperature. Medium was removed, and lipids were extracted by the method of Bligh and Dyer (12) in the presence of 300 g of carrier lipid (total bovine heart extract). An aliquot was taken to determine total chloroform-soluble radioactivity. Phospholipids were separated by two-dimensional TLC using Silica Gel 60 plates, developing in the first dimension with chloroform:methanol:29% ammonium hydroxide (60:30:5, v/v) followed by development in the second dimension using chloroform:methanol:acetic acid:H 2 O (50:30:6:1, v/v). The labeled species were localized by autoradiography, the labeled bands were scraped into scintillation vials, and radioactivity was quantitated by liquid scintillation spectrometry. Identification of the labeled species was determined by co-migration with authentic standards. Parallel, unlabeled vials were used for protein determination.
For phospholipid composition, steady-state labeling with 32 P i was used (13). Cells were plated into sterile glass scintillation vials (2.5 ϫ 10 4 cells/vial) at 33°C and allowed to attach overnight. An aliquot of medium containing 32 P i was then added to achieve a final concentration of 4 Ci/ml. Cells were incubated for 96 h at 33°C prior to analysis as described above.
Phospholipid Turnover-Cells were labeled with 32 P i (5 Ci/ ml) for 4 days (several generations) to uniformly label the phospholipids. Cells were then harvested and plated into unlabeled medium in 6-well dishes at 2 ϫ 10 5 cells/well. At various times, medium was removed, the cell monolayer was washed once with 3 ml of phosphate-buffered saline, and phospholipids were extracted into 2 ml of methanol containing 300 g of carrier lipid. After recovery using the method of Bligh & Dyer (12), samples were blown to dryness, resuspended in 1 ml of chloroform, and an aliquot was taken for determination of total phospholipid label using liquid scintillation spectrometry. The individual phospholipids were isolated, and radioactivity was quantitated as described above.

Labeling with [ 3 H]Oleic Acid and [ 3 H]Palmitic
Acid-Cells were plated into sterile glass scintillation vials (2.5 ϫ 10 5 cells/ vial) and allowed to attach overnight at 33°C. The next day, vials were shifted to 40°C for 2 h. An aliquot of medium containing [9,10-3 H]oleic acid or [9,10-3 H]palmitic acid was added to achieve a final concentration of 2 M at 2 Ci/ml, and the cells were incubated for 3 h at 40°C. Labeling medium was removed, and the cell monolayer was incubated for 30 min in F12c medium prior to extraction of cellular lipids as described above. Labeled lipids were separated using single dimension TLC on Silica Gel 60 plates using hexane:ethyl ether:acetic acid (70:30:1; v/v) as the development system. The plates were sprayed with EN 3 HANCE (PerkinElmer Life Sciences) prior to exposure to x-ray film at Ϫ80°C to localize labeled species. Identification of labeled species was determined by co-migration with authentic standards. Labeled bands were scraped into scintillation vials, and radioactivity was quantified as described above. Parallel, unlabeled vials were used for protein determinations.
Measurement of Neutral Lipid Levels-Cells were plated into 100-mm diameter tissue culture dishes at 10 6 cells/dish and allowed to attach overnight at 33°C. The next morning, the medium was changed to F12c containing 100 M oleic acid, and cells were grown at 33°C for 48 h after which they were harvested with trypsin and resuspended in 1 ml of phosphate-buffered saline. An aliquot of 50 l was used for protein determination, and 0.75 ml of this cell suspension was transferred to a glass tube containing 3 ml of methanol: chloroform (2:1, v/v). Lipids were extracted and separated on a Silica Gel G plates using the same single dimension system describe for fatty acid labeling. Plates were charred on a hot plate after spraying the plate with 50% sulfuric acid. Charred plates were scanned, and the densities of the bands of interest were determined using the National Institutes of Health ImageJ program. 3 Quantitation was performed by comparison to a standard curve for each neutral lipid class, run in adjacent lanes on the same TLC plate.
Cell Growth Measurements-Cells were plated in 60-mm diameter tissue culture dishes at 10 5 cells/dish and allowed to attach overnight at 33°C. The next day, dishes were transferred to 33°C or 40°C and allowed to grow. Cells were harvested each day with trypsin and counted using a hemocytometer. Cell growth was also visualized by staining cell colonies grown on tissue culture plates with 0.5% Coomassie Blue in methanol: water:acetic acid (45:45:10, v/v) for 1 h, followed by three washes with methanol:water:acetic acid (45:45:10, v/v).
Determinations of DNA and Protein Synthesis-DNA and protein synthesis was determined by measuring the incorporation of [methyl-3 H]thymidine and [ 35 S]methionine, respectively, into trichloroacetic acid-insoluble material. Cells were plated into 24-well dishes (2 ϫ 10 4 cells/well) and FIGURE 1. Phospholipid biosynthesis in ZR-82 and GroD1. Cells were incubated at either 33°C or 40°C with 32 P i for 3 h. A, after extracting the lipids, an aliquot was measured in a scintillation counter to quantify total labeled lipids. B, labeled phospholipids from 40°C were separated using a twodimensional TLC system and exposed to x-ray film. C, bands of interest were scraped from TLC plates, and the radioactivity was quantified using scintillation spectrometry. Bar graphs represent the average Ϯ S.D. of three independent samples. O, origin; PC, phosphatidylcholine; PE, phosphatidylethanolamine; PI, phosphatidylinositol; PA, phosphatidic acid; CL, cardiolipin; SM, sphingomyelin; PG, phosphatidylglycerol. *, p Յ 0.05, and the value for GroD1 is the same as the value obtained for ZR-82. allowed to attach overnight at 33°C. One dish was shifted the next morning to 40°C 3 h prior to the labeling. [methyl-3 H]Thymidine or [ 35 S]methionine was added to yield final radioactive concentrations of 2.5 Ci/ml and 1 Ci/ml, respectively. After 2 h at either 33°C or 40°C medium was removed, and 0.5 ml of ice-cold 10% trichloroacetic acid was added to each well. The cell monolayers were washed 5 times with 1 ml of 10% trichloroacetic acid and twice using ice-cold ether:ethanol (1:3, v/v). After drying, cellular material was solubilized in 0.3 ml of 0.5 N NaOH and incubated for 2 h at 37°C. Aliquots were counted by liquid scintillation spectrometry following neutralization with HCl.
Preparation of Subcellular Fractions-Cells were grown at 33°C prior to harvest without exposure to 40°C. Cell monolayers were washed twice using Tris-buffered saline and then scraped into Tris-buffered saline using a rubber policeman. Cells were pelleted and resuspended in ice-cold homogenization buffer (50 mM Tris-HCl, pH 7.2, 1 mM EDTA, 1 mM dithiothreitol) and disrupted by sonication, using a Branson cell disrupter (Branson Sonifier Co.), with two 10-s bursts at a power setting of 30% using a microtip. The sonicated cell suspension was centrifuged for 30 min at 13,000 ϫ g. The resulting supernatant was centrifuged for 90 min at 100,000 ϫ g to pellet membranes and membrane-associated proteins. The supernatant ("soluble fraction") was collected, and the pellet was resuspended in 0.5 M NaCl in homogenization buffer and centrifuged again for 90 min at 100,000 ϫ g to extract any remnants of soluble, membrane-associated proteins from the membranes. The final pellet was resuspended in homogenization buffer ("membrane fraction"). All fractions were aliquoted and stored at Ϫ80°C prior to assay.
Preparation of Labeled PA-[ 32 P] PA was synthesized enzymatically, using diacylglycerol kinase, diolein, and [␥-32 P]ATP as described previously (16) with the exception that TLC purification was omitted for some PAP assays after observing no significant difference in PAP activity compared with the non-TLC-purified substrate.
Assay of PAP-PAP activity was measured by following the release of water-soluble 32 P i from chloroform-soluble [ 32 P]PA for 20 min at 37°C. The reaction mixture contained 50 mM Tris-maleate buffer (pH 7.0), 2 mM MgCl 2 , 0.1 mM [ 32 P]PA (10,000 cpm/nmol), 0.5 mM Triton X-100, and cellular protein (0.1 mg) in a total volume of 0.1 ml. Mg 2ϩ -dependent PAP activity (PAP1) represents the difference in activity obtained with or without the addition of 5 mM EDTA to the assay mixture. All enzyme assays were conducted in triplicate and were linear with time and protein concentration.
Expression Cloning-A retroviral cDNA library from human substantia nigra (ViraPort, Stratagene, La Jolla, CA) was used to infect a GroD1 population. 10 separate pools of 10 6 cells were infected at a 1:5 (virus:cell) ratio to optimize chances of single gene insertion. After 2 days at 33°C cells were shifted to 40°C to obtain temperature-resistant colonies. Genomic DNA was isolated from temperature-resistant clones using the Genomic tip 100/G kit (Qiagen, Valencia, CA), and human cDNA inserts were amplified by PCR using primers against the flanking regions of the cDNA insert in the vector. PCR amplification of genomic DNA was performed by cycling 1 min at 95°C, 1 min at 64°C, and 4 min at 72°C for 40 cycles. Primers were as follows: 5Ј Retro, 5Ј-GGC TGC CGA CCC CGG GGG TGG-3Ј; 3ЈpBF, 5Ј-CGA ACC CCA GAG TCC CGC TCA-3Ј. PCR products were purified using the QIAquick Spin kit (Qiagen) and sequenced at Tufts University Core Facility (Boston, MA).
GPI Activity and Native Gel Zymogram-GPI activity was measured at room temperature in whole cell homogenates, as described previously (17) by monitoring the production of NADPH from fructose 6-phosphate at 340 nm. Thermal inactivation of the residual GPI activity was performed similarly as described (18); aliquots of whole cell lysate were incubated in a 60°C water bath for different periods of times and chilled immediately on ice, then assayed for GPI activity.
For zymograms, a polyacrylamide native Phastagel TM (Amersham Biosciences) with a gradient of 8 -25% was used to separate proteins obtained from soluble fractions of ZR-82 and GroD1, prepared as described above. The gel was stained by immersion in a GPI-specific staining solution (19) for 30 min at 37°C. The staining solution was removed by rinsing the gel with deionized water and was fixed with 7% (v/v) acetic acid.
GPI cDNA Sequencing-Total cellular RNA was isolated from cells with the RNeasy kit (Qiagen), and first strand synthe-sis of the RNA was performed with SuperScript III reverse transcriptase (Invitrogen, Carlsbad, CA). The cDNA coding region for GPI was amplified using the GoTag Green master mix (Promega) and primers 5Ј-TCG ACC GTC CGG CTC CGT GT-3Ј and 5Ј-TTT GGC TGG CTC GGC TGA CTT TAT TC-3Ј. The PCR product was purified and sequenced as described above.
GPI cDNA Retroviral Infection-The entire coding region of GPI cDNA was amplified from cDNA of CHO-K1 cells using Phusion DNA polymerase (New England BioLabs Inc., Ipswich, MA) and the primers 5Ј-GTT GTT GAA TTC TCT CCG CTC CCG CCA TGA-3Ј and 5Ј-GTT GTT GAA TTC TTA TTC TAT TCT GAT GTC CCG C-3Ј, and cloned as an EcoRI fragment into the pBABEpuro vector (20) to generate pBABE(GPI)puro. Insert directionality was confirmed by PCR and sequencing using pBABE 5Ј and pBABE 3Ј primers (20). HEK 293T cells were co-transfected, using FuGENE 6, with helper virus DNA pCL-10A1 (Imgenex, San Diego, CA) and either equal amounts of pBABE(GPI)puro or pBABEpuro. Supernatants containing retrovirus were harvested 48 h after transfection, filtered, and stored at Ϫ80°C. Cells were infected for 3 h with virus-containing supernatant containing 10 g/ml Polybrene. Medium was changed to F12c containing 6 g/ml puromycin 24 h after infection.
Protein Determination-Protein determinations were performed using the BCA kit and Coomassie protein assay reagent (Pierce).
Statistical Analysis-Two-tailed Student's t tests were performed, and values of p Յ 0.05 were considered significant differences.

Generation of a P12/UV-resistant Isolate Defective in Phospholipid Biosynthesis-
The ZR-82 cell line is a peroxisome/ plasmalogen-deficient variant of the CHO-K1 cell line. This cell line was chosen as the parent strain due to the fact that it accumulates the fluorescent fatty acid P12 almost 3-fold compared with CHO-K1 and is hypersensitive to P12/UV treatment (21). ZR-82 cells were mutagenized, and a P12/UV-resistant population of cells was generated as described under "Experimental Procedures." We assumed that major deficiencies in glycerolipid biosynthesis would be lethal, similar to previously reported cell lines with severe phospholipid biosynthesis deficiencies (5). To obtain conditionally lethal (temperature-  sensitive) mutants, once mutagenized, the cells were grown at 33°C (the permissive temperature). One isolate, GroD1, displayed reduced phospholipid biosynthesis; incorporation of 32 P i into phospholipids was reduced by 60% at 33°C and reduced by 68% at 40°C in comparison to the parent cell line (Fig. 1A). Separation of the labeled phospholipid classes by two-dimensional TLC revealed that, when compared with the parent strain, the most affected classes in GroD1 were PC and phosphatidylethanolamine (PE) (Fig. 1, B and C), reduced by 85 and 73%, respectively, when measured at 33°C. The differences were more pronounced at 40°C. Both of these species are synthesized in mammalian cells using diacylglycerol (DAG) as an intermediate (Fig. 2) (22). The labeling of SM, which depends on PC for donation of the phosphocholine head group (6,23), was also severely reduced in GroD1 (Fig. 1C). In contrast, phosphatidylinositol (PI), which does not require DAG formation for its synthesis (24), did not show a significant reduction over ZR-82 at 33°C and was only reduced by 22% at 40°C. Labeling of cardiolipin (CL) and phosphatidylglycerol (PG) was also severely reduced in GroD1 compared with ZR-82. Interestingly, although there appeared to be a severe decrease in the synthesis of phospholipids in GroD1, the labeling of phosphatidic acid (PA), an intermediate for the synthesis of all glycerolipids (25,26), was increased 80% over the parent strain.
Triglyceride Biosynthesis and Triglyceride Levels Are Reduced in GroD1-Labeling of cellular lipids with either [9,10-3 H]oleate or [9,10-3 H]palmitate was consistent with a decrease in glycerolipid biosynthesis (Fig. 3A). Oleate labeling of the phospholipids was reduced by 55% in GroD1, whereas palmitate labeling was reduced by only 25%. This decrease is less severe than that obtained using 32 P i . This is likely due to the fact that fatty acids can be incorporated into pre-existing phospholipids through deacylation/reacylation reactions (27) (supplemental Fig. S1). The labeling of triglycerides was dramatically reduced in GroD1 using either oleate or palmitate (ϳ11% of ZR-82 when using either fatty acid). In contrast to triglycerides and phospholipids, the labeling of cholesterol esters was increased in the mutant strain.
To confirm the fatty acid labeling data for TG and cholesterol ester, we also measured the cellular accumulation of neutral lipids after 48 h of incubation with 100 M oleate (Fig. 3, B and C). During this period GroD1 was able to accumulate only 20% of the TG found in ZR-82. However, the mutant cell line was capable of depositing fatty acids in the form of cholesterol esters, displaying increased levels (70% greater than ZR-82). Free cholesterol levels were not significantly different between the two strains under these conditions. We could not detect the accumulation of DAG in either cell line under these conditions. When we measured DAG levels using [ 32 P]ATP and diacylglycerol kinase (28), they were similar in GroD1 (7.58 Ϯ 0.85 nmol/mg of protein) and ZR-82 (8.08 Ϯ 0.96 nmol/mg of protein; n ϭ 3 for both values).
Phospholipid Composition-The phospholipid compositions of ZR-82 and GroD1 were determined at 33°C using steadystate labeling with 32 P i ( Table 1). The distribution of label was not radically altered in GroD1, although the trends did reflect the short term labeling data in the less abundant species. Both PC and PE showed similar contributions to the total labeling in GroD1 and ZR-82. The most notable change was in PA, which showed a 100% increase in relative 32 P content in GroD1. CL showed a 33% decrease in labeling relative to the other phospholipids.
Phospholipid Turnover-To evaluate if the discrepancy between the short term 32 P i labeling and the steady-state labeling was due to a difference in phospholipid turnover, we monitored the loss of radioactivity from phospholipids in cells, which had been labeled for several generations with 32 P i . After 2 days in unlabeled medium, turnover from the total phospholipid pool was 1.4 times slower in GroD1 in comparison to ZR-82 (Fig. 4A). When individual phospholipid classes were  examined (Fig. 4B), all, with the exception of PI, presented a slower turnover in GroD1 than in ZR-82. The most dramatic differences were observed in PG and CL. PI turnover remained the same for ZR-82 and GroD1.
GroD1 Displays a Temperature-sensitive Growth Phenotype-Although the lipogenesis deficiency displayed by GroD1 was independent of temperature, growth of GroD1 was greatly affected at 40°C. At 33°C, the growth rate of GroD1 was reduced by ϳ50% compared with ZR-82 (Fig. 5A). At 40°C, GroD1 divided twice and stopped growing. These cells eventually died after several days at this temperature as observed by the absence of cells when staining plates with Coomassie Blue after 10 days of growth at 40°C (Fig. 5B).
DNA Synthesis, Protein Synthesis, and ATP and G3P Levels-Protein and DNA synthesis, measured as the incorporation of [ 35 S]methionine and [methyl-3 H]thymidine, respectively, into trichloroacetic acid-precipitable radioactivity, was similar in GroD1 and ZR-82 at both 33°C and 40°C (Table 2). ATP levels were also normal in GroD1 when compared with the parent strain. Levels of G3P, a precursor for glycerolipid biosynthesis, were 25% greater in GroD1 than in ZR-82.
PAP1 Activity Is Decreased in GroD1-Phosphatidate phosphatases are responsible for the removal of the phosphate from phosphatidic acid to supply diacylglycerol required for the biosynthesis of PC, PE, and TG (29). Moreover, the biosynthesis of PI does not require this activity (Fig. 2). The decreased ability of GroD1 to synthesize the former three glycerolipid species, coupled with the relative lack of effect upon PI synthesis and the accumulation of PA, suggested a reduction in this activity in GroD1.
There are two groups of enzymes with PAP activity, PAP1 and PAP2. These differ in their cellular localization, Mg 2ϩ dependence, and sensitivity to N-ethylmaleimide (26,29,30). PAP2 includes a set of integral membrane proteins whose activities are not thought to be important for glycerolipid biosynthesis. We found no significant difference between GroD1 and ZR-82 when PAP2 was assayed in membrane fractions (Fig. 6A).
PAP1 has soluble activity, which can also be found loosely associated with cell membranes (31). It is Mg 2ϩ -dependent and, in animals, is sensitive to inhibition by N-ethylmaleimide. It is this activity that is thought to be important for glycerolipid biosynthesis (26,32). We found that in GroD1 PAP1 activity was reduced by ϳ70% (Fig. 6B).
We examined the initial steps leading to PA formation, glycerol-3-phosphate acyltransferase and lysophosphatidate acyl-  . Phospholipid biosynthesis and PAP1 activity are also reduced in GroD2 and DL7. A, phospholipid biosynthesis, measured by 32 P i short term labeling and B, PAP1 activity are reduced in GroD2, a mutant cell line isolated using the P12/UV selection from ZR-87, an independently isolated plasmalogen/peroxisome-deficient strain (different from ZR-82). Similar results were observed for DL7 (C and D), another mutant cell line isolated from wild-type CHO-K1. *, p Յ 0.05, and the value for the mutant cell line is the same as the value obtained for the parent strain.

TABLE 2 DNA synthesis, protein synthesis, and ATP and glycerol 3-posphate levels in ZR-82 and GroD1 cells
Cells were shifted from 33°C to the indicated temperature 2 h prior to analysis at that temperature. DNA and protein synthesis were measured by the incorporation of 2.5 Ci/ml [methyl-3 H]thymidine or 1 Ci/ml [ 35 S]methionine into trichloroacetic acid-insoluble radioactivity in 2 h. Cellular ATP levels were measured using a commercial ELITEN luciferase bioluminescent assay (Promega). Glycerol 3-phosphate was measured using an enzymatic spectrophotometric assay with glycerol-3-phosphate dehydrogenase and MTT. No significant differences were observed in DNA, protein synthesis, or ATP levels between GroD1 and ZR-82 at either 33°C or 40°C. Glycerol 3-phosphate levels were slightly increased in GroD1  transferase (11), by measuring the incorporation of labeled G3P into lysoPA and PA in whole cell homogenates (supplemental Fig. S2). No major difference was observed in the formation of PA or lysoPA when comparing GroD1 to ZR-82. Together these results indicated a defect in PAP1 activity in GroD1 cells that could contribute to the accumulation of PA and reduction in PC, PE, and TG biosynthesis. Isolation of Two Additional Glycerolipid-deficient CHO Mutants-To further establish the selection procedure we repeated the P12/UV mutant selection using two additional CHO cell lines, ZR-87 and wild-type CHO-K1, as the parent strains. Both cell lines were mutagenized and selected as described for ZR-82/GroD1. One glycerolipid-deficient mutant was isolated from each strain and examined for phospholipid synthesis and PAP1 activity.
ZR-87 is similar to ZR-82, in that it is also a peroxisome/ plasmalogen-deficient strain derived from the CHO-K1 cell line (11). As with ZR-82, it also accumulates more fluorescent P12 than CHO-K1 and is also hypersensitive to P12/UV treatment. The isolate obtained from ZR-87, GroD2, grew slowly at 33°C and did not survive at 40°C. It presented with a similar reduction in phospholipid biosynthesis as GroD1, with major reductions in PC, PE, and SM, but no significant difference in PI and increased PA labeling (Fig. 7A). PAP1 activity was also reduced in GroD2 to 25% of ZR-87 (Fig. 7B).
We repeated the selection using wild-type CHO-K1 to test the selection method on a non-peroxisome/plasmalogen deficiency cell line. The isolate derived from CHO-K1, DL7, displayed an identical phenotype with respect to PAP1 activity and phospholipid biosynthesis as the other two isolates (Fig. 7,  C and D).
Expression Cloning Reveals a Mutation in Glucose-6-phosphate Isomerase as the Primary Defect in GroD1-In an attempt to identify the gene responsible for the GroD1 phenotype we infected GroD1 cells with a human cDNA retroviral library. The infected cell populations were selected for growth at 40°C to obtain temperature-resistant colonies. Six colonies resulted from this infection, each from a separate pool of infected cells. Genomic DNA was isolated from each, and the viral insert was amplified using PCR. All amplifications resulted in a single band of ϳ1.8 kb. Sequencing of one of these PCR products (temperature-resistant isolate 1) revealed a sequence with 100% identity to human GPI.
GPI activity was measured in whole cell homogenates of GroD1 (Fig. 8A), and it was found to be decreased to 13% of ZR-82. Measurement of GPI in the temperature-resistant isolate showed activity that was 6-fold greater than the parent strain. To differentiate whether the decrease in GPI activity in GroD1 was due to a structural change or a modulation in GPI expression level we examined the thermal stability of GPI. GPI FIGURE 8. Glucose-6-phosphate isomerase activities. A, GPI activity was measured in whole cell homogenates at room temperature as described under "Experimental Procedures." B, thermal inactivation of the GPI activity; whole cell lysates were incubated at 60°C for the indicated periods, then assayed for GPI activity. The values in parentheses indicate the time required at 60°C to reach half of the initial activity. C, kinetic parameters obtained by fitting the Michaelis-Menten equation to experimental initial velocity values obtained at different fructose 6-phosphate concentrations. D, native PAGE of 0.5 g and 2.5 g of protein from the soluble fractions from ZR-82 or GroD1, stained specifically for GPI activity (in-gel colorimetric assay), note the difference in band intensity and migration. E and F, GPI activities measured in ZR-87 and GroD2, and CHO-K1 and DL7, respectively. All values represent the average Ϯ S.D. of three independent samples. *, p Յ 0.05, and the value for mutant isolate is the same as the value obtained for the parent strain.
activity from ZR-82 presented a thermal stability with a half-life of 31 s (Fig. 8B). In GroD1, GPI activity was highly sensitive to thermal inactivation losing half of its activity in only 7 s. The specific activity of GPI in GroD1 was linear with protein concentration (data not shown). Kinetic analysis revealed a V max 4.4 times lower and a K m 23 times higher in GroD1 in comparison to ZR-82 (Fig. 8C). When equal amounts of the soluble protein fraction were run on a native gel zymogram, GPI activity from GroD1 was notably reduced and displayed increased electrophoretic mobility (Fig. 8D). These results suggested a change in the polypeptide chain of GPI in GroD1 cells.
To identify lesion(s) in the GPI gene, we sequenced GPI cDNA from ZR-82 and GroD1. This revealed two mutations in GroD1, a silent mutation at valine 53 (GTG to GTA) that does not change the polypeptide sequence and a missense mutation at glycine 189 (GGA to GAA) changing this residue to a glutamate. Modeling this mutation, based on sequence homology with the solved atomic structure of human GPI (33), positioned this mutated residue, Gly-189, at the GPI homodimer interface close to the catalytic pocket of this enzyme (supplemental Fig.  S3). Taken together, these data suggest this insertion of a larger and negative charged residue at the homodimer interface affects the structural stability and kinetics of GPI in GroD1.
GroD2 displayed only 6% of its parent strain GPI activity (Fig.  8E). Sequencing of GPI cDNA from GroD2 identified a point mutation resulting in an amino acid change at residue 377, from proline to leucine. The residue Pro-377 is also located at the interface of the GPI homodimer and in close proximity to the enzymatic active site (supplemental Fig. S3).
CHO-K1 derived mutant DL7 displayed a dramatic reduction in GPI activity (2% of CHO-K1, Fig. 8F), and sequencing of GPI cDNA from DL7 revealed a single point mutation resulting in Pro-317 to Leu in GPI cDNA. This proline is located in the core of the GPI monomer (supplemental Fig. S3) and is the first residue of an ␣-helix in a helix turn helix region. Proline residues at such a position are usually very important for the overall secondary structure of proteins (34).
The generation of three independently isolated GPI-deficient mutants, GroD1, GroD2, and DL7, presenting different point mutations in GPI cDNA sequence and all displaying a glycerolipid biosynthesis deficiency with decreased PAP1 activity, strongly suggests that there is a correlation between GPI, PAP1 activity, and glycerolipid biosynthesis in CHO cells.
Expression of Wild-type GPI Corrects the Growth Phenotype, Glycerolipid Synthesis, and PAP1 Activity in GroD1-To further establish the role of GPI as the primary lesion in GroD1 we infected GroD1 and ZR-82 cells with a retroviral vector containing wild-type hamster GPI cDNA to generate GroD1 t(GPI) and ZR-82 t(GPI) . We also infected cells with the same retroviral vector not containing GPI cDNA to generate GroD1 t(Puro) and ZR-82 t(Puro) as controls. After 10 days of selection with puromycin, GPI activity in cells infected with GPI-bearing vectors was ϳ6-fold greater than sham-infected ZR-82 (Fig. 9A).
GroD1 t(GPI) cells were able to grow at 40°C, whereas the control GroD1 t(Puro) was not (Fig. 9B). Incorporation of 32 P i into phospholipids in GroD1 t(GPI) was also restored to levels similar to ZR-82 t(GPI) and ZR-82 t(Puro) (Fig. 9C). Expression of wild-type GPI in GroD1 restored short term 32 P i labeling of PC and PE and reduced the labeling of PA to values similar to ZR-82 t(GPI) . Expression of GPI in both ZR-82 and GroD1, resulted in equal PC and PE biosynthesis, although there were somewhat lower and greater, respectively, than ZR-82 t(Puro) . No dramatic changes were observed in the incorporation of 32 P i into PI among the different infected cell populations.
Neutral lipid analyses also showed that recovery of GPI activity in GroD1 cells resulted in restoration of normal levels of triglycerides and cholesterol esters (Fig. 9D). For GroD1 t(Puro) , however, levels of triglycerides remained low and cholesterol ester levels increased with no significant difference from uninfected GroD1. PAP1 activity in GroD1 t(GPI) , measured as Mg 2ϩdependent activity in soluble fractions (Fig. 9E), was restored to values similar to ZR-82. Taken together, these results indicate that GPI is the primary lesion in GroD1; the correction of this lesion restores PAP1 activity and glycerolipid biosynthesis.

DISCUSSION
We developed a selection technique to identify and isolate mutants with global deficiencies in glycerolipid biosynthesis. Characterization of three mutants, independently isolated through this technique, uncovered a novel connection between GPI and glycerolipid biosynthesis. Analysis of these three isolates, which displayed severely reduced glycerolipid biosynthesis, showed severely decreased GPI activity as well. In each mutant, the decreased GPI activity was the result of a separate, significant single amino acid substitution in the GPI polypeptide.
Glycolysis is responsible for the supply of both G3P and dihydroxyacetone phosphate, which are the building blocks for the backbone of glycerolipids. A decrease in the activity of a glycolytic enzyme such as GPI may limit the supply of these crucial building blocks, impacting glycerolipid biosynthesis. This may well be the case in the GPI-deficient cells. However, evidence suggests that the decrease in glycerolipid biosynthesis is not due solely to a decrease in glycolysis. First, the cellular levels of two glycolytic byproducts, G3P and ATP, are increased or unaffected in the mutant cells; ATP-dependent processes, such as DNA and protein synthesis, are unaffected. In support of these data, elevations of fructose 6-phosphate, ATP, and glyceralde-hyde 3-phosphate are observed in red blood cells in human patients with GPI deficiency (10,35), as well as in GPI-deficient mice (36), and are interpreted as being caused by an acceleration of the pentose phosphate shunt, activated by the blockage of glucose 6-phosphate isomerization. Additionally, we would expect that a reduced supply of G3P or ATP would result in an overall decrease in glycerolipid biosynthesis. However, PI and PA biosynthesis are unaffected or increased in all three mutant cell lines.
The LPIN genes (LPIN1, -2, and -3) constitute a family of genes in mammalian cells, which codes for PAP1 (37). Our initial characterization of these mutant cell lines led us to believe that the primary lesion was a defect in one of these LPIN genes or in an as yet uncharacterized PAP1 enzyme (31). The synthesis of glycerolipid classes (Fig. 2) that require PAP1 for their biosynthesis, PC, PE, and TG (22,38), was severely reduced while the biosynthesis of PI, not dependent on PAP1 activity (24), was relatively unaffected. In addition, cells accumulated phosphatidic acid. Finally, a measure of the cytosolic fractions revealed a very significant decrease in PAP1 activity in all three mutant cell lines. We were quite surprised to see that the genetic lesion associated with the loss of PAP1 activity and glycerolipid biosynthesis was, in fact, a defect in GPI. Together, our molecular biology and biochemical data point to a scenario in which a defect in GPI resulted in the loss of PAP1 activity.
Although we feel that the reduction in PAP1 activity likely contributes significantly to the reduction in glycerolipid synthesis in the GPI-deficient mutants, the reduced PAP1 activity is probably only one of a number of pleiotropic effects. Other, yet to be examined, lipogenic activities may be affected as well and contribute to the lipodystrophy.
One phenotype that was not directly explained by a PAP1 deficiency was the reduction in PG and CL biosynthesis. These phospholipids are produced in the mitochondria, and their synthesis is not thought to be dependent on PAP1 activity (39). The reduction in biosynthesis may be a secondary effect due to effects on mitochondrial membranes. It is interesting to note that a CHO-K1 mutant, defective in PC biosynthesis (5), also displays a 60% decrease in both PG and CL biosynthesis as judged by short term 32 P i labeling of these phospholipids. 4 Alternatively, the biosynthesis of both PG and CL requires an uncharacterized phosphatase, phosphatidylglycerolphosphate phosphatase (40). It is possible that a decrease in this activity is another pleiotropic effect of the GPI mutation. It is also possible that the enzyme that catalyzes PAP1 activity catalyzes the dephosphorylation of phosphatidylglycerolphosphate as well. Treatment of the parent strain with bromoenol lactone, an inhibitor of PAP1 activity (41), also resulted in a decrease in cardiolipin synthesis in CHO cells (supplemental Fig. S4). The isolated GPI mutant cell lines may be useful tools with which to examine this issue.
Although it is clear that there is a relationship between GPI and PAP1 activity, the nature of this relationship is not yet understood. The decrease in GPI activity probably alters the levels of glycolytic intermediates. Nutrient sensing signaling pathways, such as the mammalian target of rapamycin pathway (42), the hexosamine pathway (43), or the AMP kinase pathway (44), act by sensing changes in the level of cellular intermediates to function as cellular homeostatic regulators. Upon sensing changes in nutrient levels, these pathways produce covalent modification of protein and/or changes in protein expression that can modulate lipid metabolism. Perhaps CHO cells erroneously perceive the defect of GPI as a change in nutrient levels. This may well have an influence on PAP1 activity by stimulating covalent modification of the PAP1 enzyme or expression of one of the LPIN genes. The PAP1 protein can be modified by phosphorylation in a mammalian target of rapamycin-dependent manner (45), although it is unclear what effect this has on its activity. On the other hand, it has recently been shown that the LPIN1 gene is regulated by TORC2 (46). TORC2 is a target of phosphorylation by AMP kinase, controlling TORC2 gene regulator activity (47,48).
GPI also has cytokine activities (49); in addition to being a housekeeping enzyme of glycolysis, GPI is released by certain cell types and tumors (50 -52). GPI is also known as neuroleukin, a neurotrophic growth factor (53), autocrine motility factor, a stimulator of cell motility and growth (9), and maturation factor, an inducer of differentiation (50). An extracellular receptor for GPI/autocrine motility factor, gp78, has been identified (54). It has been proposed that GPI/autocrine motility factor binds to its receptor through its conserved active site or in a region close to the active site (55). Therefore, changes in the active site would most probably interfere with receptor binding. It is possible that structural modification of GPI in these cells may interfere with its role as an autocrine agent, and this may affect expression or function of PAP1 and cell growth.
Finally, the findings presented herein may shed light on some of the symptoms associated with an inheritable defect in GPI (10,56). Patients presenting with a GPI deficiency invariably display a non-spherocytic, hemolytic anemia. The erythrocytes are adversely affected in these patients probably due to their dependence on glycolysis for ATP production. In some severe GPI deficiencies, neuromuscular dysfunction is observed (57). The link between GPI deficiency and neuromuscular dysfunction has not been established (10). The reduced potency of GPI as a cytokine has been proposed to account for this (57). Alternatively, the marked hyperbilirubinemia and tissue hypoxia as a result of severe anemia have been implicated (58). Given the data presented here, we must also consider the effect of GPI deficiency on glycerolipid biosynthesis. A decrease in glycerolipid biosynthesis could have significant effects on membrane formation, membrane function, and axonal migration; PAP1deficient mice (fld mice) display a pronounced peripheral neuropathy (14,59).
In summary, we have presented the first mutant animal cell lines with a deficit in the biosynthesis of a number of glycerolipids. We have also uncovered a novel relationship between a glycolytic enzyme and PAP1 in CHO-derived cell lines. Further analysis of these mutants should shed light on the interrelationship between carbohydrate metabolism and lipogenesis. Also, the generation of additional mutants of