Proc. Natl. Acad. Sci. USAVol. 90, pp. 11658-11662, December 1993Biophysics

Evidence that facilitative glucose transporters may foldas 8-barrels

(membrane proteins/structure prediction/antibody/channels/porins)

JORGE FISCHBARG*t, MIN CHEUNG*, FERENC CZEGLEDY*, JUN LI*, PAVEL ISEROVICH*, KUNYAN KUANG*,JOHN HUBBARD*, MARGARET GARNERt, ORA M. ROSEN§¶, DAVID W. GOLDE§, AND JUAN CARLOS VERA§*Departments of Physiology, Ophthalmology, Medicine, and Biochemistry, College of Physicians and Surgeons, Columbia University, New York, NY 10032;tDepartment of Ophthalmology, Southwestern Medical Center, University of Texas, Dallas, TX 75235; and §Programs in Molecular Biology and MolecularPharmacology and Therapeutics, Memorial Sloan-Kettering Cancer Center, New York, NY 10021

Communicated by Alfred G. Redfield, September 9, 1993 (received for review February 15, 1993)

ABSTRACT A widely accepted model for the structure ofthe facilitative glucose transporters (GLUTs) predicts that theyform 12 transmembrane a-helices and that the highly con-served sequence Ile-386-Ala-405 in GLUT1 is intracellular.We raised a polyclonal antibody against a synthetic peptideencompassing this conserved sequence and found that antibodytreatment increased 2-deoxy-D-glucose (DOG) uptake in Xe-nopus oocytes expressing GLUT1, GLUT2, or GLUT4 onlywhen applied to the extracellular side. This effect was dosedependent and was specifically blocked by competition with thepeptide Ile-386-Ala-405; it was due to a decrease in the Km forthe transport of DOG. To ascertain GLUT orientation, weraised anti-peptide antibodies against the last 21 and 25 C-ter-minal amino acids of GLUT1 and GLUT4, respectively, whichwere previously shown to be intracellular. These antibodiesincreased DOG uptake when iniected into oocytes expressingGLUT1 and GLUT4, but not when added extracellularly.Prompted by the noted discrepancy, we found sequence simi-larity between GLUTs and porins, two of which are knownfrom crystallography to form 16-stranded transmembraneantiparallel -barrels. Analysis of the hydrophobicity, am-phiphllicity, and turn propensity ofGLUT1 leads us to proposethat GLUTs fold as porin-like transmembrane (barrels. Thismodel is consistent with the results of the present antibodystudies and also with previously published experimental evi-dence inconsistent with the 12-helix model.

Mammalian cells express a family of facilitative hexosetransporters (1) (GLUTs). These proteins have been wellcharacterized biochemically and their primary structure isknown (1, 2). As with many other membrane proteins,however, the secondary and tertiary structures ofthe GLUTsare unknown. When GLUT1 was molecularly cloned andsequenced, 12 of its hydrophobic regions were predicted toconstitute transmembrane (tm) helices (2). Several of them(helices 3, 5, 7, 8, and 9) were termed amphiphilic (2). Thismodel (hereafter 12H) has not yet led to a clear understandingof how glucose traverses GLUT1.We performed antibody (Ab) studies that yield evidence

inconsistent with the 12H model, and we report similaritybetween the secondary structure we predict for GLUTs andthat known to exist in two porins. Porins are trimeric proteinsin Gram-negative bacteria (3) that allow polar solutes to crosstheir outer membrane (3). Each porin monomer folds as a tm16-stranded antiparallel ,3-barrel (13B) (4, 5); we propose thatGLUTs fold like porins.

The publication costs of this article were defrayed in part by page chargepayment. This article must therefore be hereby marked "advertisement"in accordance with 18 U.S.C. §1734 solely to indicate this fact.

METHODS

Ab Studies. We raised three polyclonal Abs in rabbits andused the IgG fractions. They were Ab-1, against the last 21C-terminal amino acids of the GLUT1 protein; Ab-4, againstthe last 25 C-terminal amino acids of the GLUT4 protein(Ab-1 specifically reacted with GLUT1 but not with GLUT2or GLUT4, and Ab-4 reacted with GLUT4 but not withGLUT1 or GLUT2 as assessed by immunoprecipitation andimmunoblotting; data not shown); and Ab-c, raised against asynthetic peptide containing the sequence Ile-386-Ala-405 inGLUT1, a sequence that is highly conserved in all membersofthe GLUT family. Ab-c reacted with the GLUT1, GLUT2,and GLUT4 isoforms of mammalian facilitative transportersas assessed by immunoprecipitation and immunoblotting,and the reactivity was specifically blocked by competitionwith an excess of the peptide used to generate the Ab but notby an unrelated peptide (data not shown). For the experi-ments all Abs were suspended at a final concentration of 100jig of IgG per ml in modified (6) Barth's solution (MBS).Xenopus laevis oocytes were isolated as described (6) and

injected with 50 nl of water containing 10-20 ng of in vitrosynthesized capped RNA (6) encoding either GLUT1,GLUT2, or GLUT4 and incubated in MBS. Three days afterRNA injection, uptake of 2-deoxy-[1,2(n)-3H]D-glucose (3H-DOG) was measured using a 10-min uptake assay (6).Oocytes were placed into 1 ml of MBS containing 0.5 mMDOG and 1-5 ,&Ci of 3H-DOG per ml (10 Ci/mmol; 1 Ci = 37GBq; NEN/DuPont). Ten pooled oocytes yielded an uptakevalue; values were consistent within agiven batch ofoocytes.

Alignments. We used the BESTFIT and PILEUP routines ofthe GCG (Genetics Computer Group; Version 7.0) programpackage, with gap weight = 3.0 and length weight = 0.1 (7).We aligned the sequences of Rhodobacter capsulatus porin(4), Escherichia coli porin (Sw:Ompf_Ecoli), and GLUT1(Sw:Gtrl_Human).

Predictions. We developed an algorithm ("Union") topredict protein segments with relatively high hydrophobicityand propensity to form amphiphilic a or (3 structures. For aresidue span length i, Union (U) is:

U,j = Hi + i- (pt). [1]

Abbreviations: GLUT1 to -4, facilitative glucose transporters 1-4;tm, transmembrane; 12H, 12-helical model; POR, Rhodobactercapsulatus porin; OmpF, Escherichia coli porin; CFP, Chou-Fasman-Prevelige; PHD, profile neural network prediction Heidel-berg; j3B, ,-barrel; Ab, antibody; DOG, 2-deoxy-D-glucose; ATB-BMPA, 2-N-[4-(1-azi-2,2,2-trifluoroethyl)benzoyl]-1,3-bis-(D-man-nos-4-yloxy)-2-propylamine.tTo whom reprint requests should be addressed.'Deceased May 31, 1990.

11658

Dow

nloa

ded

by g

uest

on

Nov

embe

r 30

, 202

0

Proc. Natl. Acad. Sci. USA 90 (1993) 11659

Depending on the structure for which U is calculated, thesubindex stands for either a or ,3. Hi is the average hydro-phobicity for a span of i residues using the Kyte-Doolitte scale(8); p,i is the hydrophobic moment (9) (span i) for either a or

,B structures; the angles between a residue and the next for a

and structures were 1000 and 1600, respectively, usingstandard values for a-helices and the generic twist of 3-sheets.Hi and u,i values were assigned to the center residue of givenodd-valued spans. (pt) is the position-dependent turn propen-sity (10) (assigned to residue 2 in the 4-point turn). Wecalculated values of Hi, L,i, and (pt) for a given sequence andscaled their ranges to -4.5 to +4.5 in each case. Afteralgebraic addition (Eq. 1), the U,i values obtained were in turnrescaled to -4.5 to +4.5. We used union profiles to mark theapproximate locations of secondary structures. Segmentswere then refined by using (i) the Chou-Fasman-Prevelige (10)prediction method (CFP), which requires judgments by theoperator, and (ii) the results from a neural network predictionprogram [PHD; profile neural network prediction Heidelberg(11)], which runs unbiased, without human intervention. Wefound it convenient to display propensity profiles using theprogram PSAAM (12). Three-dimensional modeling was done inthe Insight and Discover graphical environments (BiosymTechnologies, San Diego).

RESULTS AND DISCUSSIONEffect of Abs on the Function of Mammalian Hexose Trans-

porters Expressed in X. laevis Oocytes. The highly conservedsequence (Ile-386-Ala-405 in GLUT1) is predicted to beintracellular in the 12H model (2), which locates it betweenits putative tm regions 10 and 11. Given the evidence for an

important functional role for the region between tm domains9 and 12 in GLUT1 (1), we reasoned that an Ab against thatconserved sequence might elicit inhibition or activation of thetransporter. After verifying its reactivity (see above), we

used X. laevis oocytes expressing different members of themammalian GLUT family to study the effect of this anti-peptide Ab on the uptake of DOG. Incubation with Ab for 1hr induced a measurable increase in the ability of oocytesexpressing any of the three mammalian transporters tested,GLUT1 (Fig. 1A, c), GLUT2 (Fig. 1B, c), and GLUT4 (Fig.1C, c) to take up DOG. The Ab, however, acted only whenpresent in the extracellular medium (Fig. 1 A-C, c). No effecton uptake was observed when the Ab was injected into theoocytes 1 hr before the uptake measurements (Fig. 1 A-C).The effect of Ab was dose dependent (Fig. 1D) and was

specifically blocked by competition with excess peptideduring the incubation period (Fig. 1E). The effect of the Abon DOG uptake was evident after a short incubation period;near-maximal levels of activation were reached in =30 min(Fig. 1F). Incubation for several hours induced an additionalincrease in uptake (Fig. 1F).To determine whether the GLUTs were expressed with the

correct orientation in the membrane ofthe oocytes, we testedthe effect oftwo other anti-peptide Abs we elicited against theC-terminal regions of GLUT1 and GLUT4. It was knownfrom previous studies that this region of the transporters islocated intracellularly (13). As expected, the Abs did notaffect the capacity ofthe oocytes to take up DOG when addedextracellularly (Fig. 1 A-C) but caused a specific and mea-surable increase in the ability of oocytes expressing GLUT1or GLUT4 (but not GLUT2), to take up DOG when injectedintracellularly (Fig. 1 A-C). These observations are consis-tent with previous indications that the C-terminal region iscentral to the function of the transporter (13).

Since both the Ab (Ab-c) and insulin (6) increase DOGuptake in oocytes, we investigated whether Ab could act bymimicking insulin rather than by specifically binding toGLUTs. The results in Fig. 1 G-I suggest instead that the Aband insulin have different mechanisms of action. Incubation

0.010.1 1 (IgG (mg/ml)

I'-

E 0.08 *H

0.0410

8o.oo o2 1/S (10- 1/S (1)

1.0 0.5 1.0I/DOG, mM-1)

00.04

E o.ooI°.-l 0.0 0.5 1.02 1/S (1/DOG, mM-1)

FIG. 1. (D, E, H, and I) Data represent averaged values from two10-oocyte groups; other data are averages from three such groups.Individual values differed with each other by <20%o. For 60 min beforethe uptake assay, one group of oocytes (intracellular Ab, solid bars)was injected with 20-30 nl of a solution containing either Ab-1, Ab-4,or Ab-c (1 ng of Ab/1 nl of water). A second group of oocytes(extracellular Ab, shaded bars) was incubated for 60 min in MBScontaining the same Abs before measuring 3H-DOG uptake. Controloocytes (open bars) were incubated in MBS. (D) Oocytes were

incubated for 60 min with Ab in the outside incubation medium; theAb concentration was varied as indicated. Solid circles, Ab-c; open

circles (controls), Ab-4. (E) Oocytes incubated with Ab-c plus theaddition ofvarious concentrations ofa peptide. The following peptideswere used: solid circles, the conserved peptide Ile-386-Ala405; opencircles, the last 20 amino acids at the C-terminal end of GLUT4. (F)Oocytes incubated with Abs in the outside medium. Solid circles,Ab-c; open circles, Ab-4. (G) Open circles, oocytes incubated initiallyin medium containing 1 ,uM insulin; arrow, the medium was replacedby another one containing insulin plus Ab-c (100 ,g/ml). Solid circles,Ab-c in the initial incubation medium, Ab-c plus insulin after thearrow. (H and I) Lineweaver-Burk plots of 3H-DOG uptake inoocytes expressing GLUT1 and GLUT4, respectively, and incubatedin the following media: open circles, MBS (controls); solid circles,MBS plus Ab-c (100 Ag/ml); triangles, MBS plus 1 ,uM insulin.

of the oocytes with insulin did not affect the Km of thetransporters for DOG, increasing instead the Vma (Fig. 1 Hand I; Table 1). This is consistent with insulin inducing thetranslocation of transporters to the cell membrane. On theother hand, the Ab induced a measurable decrease in the Kmfor DOG in oocytes expressing either GLUT1 or GLUT4,without changing the Vma, (Fig. 1 H and I; Table 1). Theshort-term effect of the Ab on uptake (Fig. 1F) can beaccounted for by an increased affinity of the transporters forDOG. The additional increase in uptake observed after long

Table 1. Effects of insulin and Ab on Vma,, and Km valuesVmax ± SE,

Complementary pmol per Km ± SE,RNA oocyte per min mM

GLUT1 Control 109 ± 23 8.6 ± 3.2(from Fig. 1H) Antibody 90 ± 2 2.8 ± 0.2

Insulin 163 ± 22 6.1 ± 1.7GLUT4 Control 76 ± 10 8.0 ± 1.9

(from Fig. 11) Antibody 60 ± 2 2.7 ± 0.2Insulin 119 ± 9 7.8 ± 1.1

Biophysics: Fischbarg et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 30

, 202

0

Proc. Natl. Acad. Sci. USA 90 (1993)

incubation periods with the Ab (Fig. 1F) may be due to theentrapment of the transporters at the level of the cell mem-brane. Additional evidence for the different modes of actionof the Ab and insulin came from experiments in whichoocytes were first treated with insulin and then with the Ab,and vice versa. Under the first condition, the Ab induced afurther 2-fold increase in uptake in oocytes pretreated withinsulin (for a total 4-fold increase; Fig. 1G). Quantitatively,this result is consistent with the effect ofthe Ab on the affinityofthe transporter for DOG. On the other hand, insulin did notaffect the uptake ofDOG in oocytes previously treated withthe Ab (Fig. 1G). One explanation for this finding is that thebinding of the Ab to the transporter may "anchor" it to theplasma membrane and disrupt the dynamic equilibrium thatallows insulin to modify the ratio of transporters locatedintracellularly versus those located at the plasma membrane.The Topology Induced from the Ab Findings Compromises

12H. A possible explanation for the effect of Ab recognitionof the sequence Phe-389-Ala-403 in terms of the 12H modelis to argue that perhaps tm helices 10 and 11 are in a highlymobile segment of the protein, leading to the exposure of theinternal loop between them to the extracellular medium.There is an a-helical membrane protein, colicin, which ap-pears to externalize some of its a-helices during large scaleconformational changes (14). Externalization, however,shuts offthe colicin channel, while in the present case uptakeby GLUTs is enhanced by the Ab-c, militating against acolicin-type mechanism. Moreover, the Ab-c had no effectwhen injected intracellularly, further evidence against theintracellular location of Phe-389-Ala-403. The simplest ex-planation for our findings is that the loop comprising thesegment Phe-389-Ala-403 is normally located on the extra-cellular side of the membrane, suggesting a topology incon-sistent with the 12H model. If GLUTs are multihelical, withtm helices -20 residues long, and if putative helices 11 and12 exist, then the conserved loop could only be intracellular,being separated from the intracellular C-terminal loop by thehairpin of these two helices (see Fig. 4).An Alternative Scheme: GLUT1 and the Porins. Given the

foregoing, we searched for an alternative secondary structurefor the transporter. We considered the structures ofthose fewmembrane proteins that have been solved by crystallographyso far, and we came upon porins. In contrast to a-helicalmembrane proteins crystallized earlier, porin monomersform 16-stranded antiparallel PBs (4, 5). When we aligned(Fig. 2) the sequences of R. capsulatus porin (POR), E. coliporin (OmpF), and GLUT1 (using BESTFIT), we found pair-wise scores for identity and similarity as follows: POR-OmpF, 20.0 and 45.7; POR-GLUT1, 19.9 and 46.6; OmpF-GLUT1, 18.2 and 42.9. Porins in general show little overallprimary sequence similarity (15). In particular, although thesecondary structures of POR and OmpF are the same, thescores for their alignment are modest. The alignments ofGLUT1 with the porins, however, elicit about the samescores as the alignment of the two porins. Hence, we set outto evaluate a possible porin-fold for GLUT1.

Prediction of Multiple tm (3-Strands in Porins. From ex-ploratory work, we chose a span of 7 residues to examinePOR, OmpF, and GLUT1 profiles. We found that the union,37 (Un) peaks identified the approximate location and lengthof the p-strands in both porins (Fig. 3). The thresholds in Fig.3 (1.83 for POR; 2.15 for OmpF) were selected so as not tomiss any strand; they result in only minimal overprediction.Segments were then refined by the CFP procedure. Incomparing the porin structures thus predicted with thoseknown from x-ray crystallography (4, 5), we found successrates [Q3 (16)] of 0.?7 and 0.75 for POR and OmpF, respec-tively. The correlation coefficients (17) for our predictionswere as follows-for POR: a, 0.56; (3, 0.70; turns, 0.28;random, 0.48; for OmpF: a, 0.25; 13, 0.64; turns, 0.30;

OmpfS16070 .. ........

Gtrl_Human MEPSSKKLTG RFMLAVGGAV LG%LQFGYNT GVIIAPOKVI EEFYNOTvVH 50

Ompf .... IGNKVD LYGKA.VGLH Y SKGNGENS YG.....IGNG DMTYARLGFSS16070 ...... L3AEVKSGDARMGVM ..NGDAVN FS..... SRS RVLFTGtil_Human RYGESILPTT ILTTLWSLSVA I SVGIMIGS FSVGLFV 4RF GRRN, 100

Ompf DLT GYGWEYF N ...... NS EGADAQTGNK TREAPAMT.YS16070 DSH ... EFGASPAHGEDVTAKGtrl_Human-IAVSAVLM GFSKLGKS E MLILGRFIIG V CGL*GFV PMYVGEV-PT 150

Ompf -IV~ R N .... YG\VY DALG DMLP EFGGDTAY.. SDDFFVGRVGS16070 ....DAk_ASE A1.... FGDLY E.VGYTDLDD RGGNDIPYLT GDERLTAEDNGtrl_Human AFR('LH QLGEVVGILI AQV|FGLDSIM GNKDLIWPLLL SIIFIp 200

Ompf IGVAT..YROS NFFGLV.. LNFAVOYLGK _RDTARRANGDGVGGSRRYS16070 PL. ..YTYS J..GAIFSVAAS.MdD GKVGETSEDD A[OEMAVAAAYGtrl_Human -C PFCPES PRFLLINRNE HENRAKSVLKS LRGTADVTHD LOEMKESRQ 250

Ompf -dGV. FGIVAYG AAdRIQPtQPLGNHG KEWAL- DNIYLAS16070 -G-N SIYTVGLGYEKF SPD .. TALMAD HMEQLELAAIA KFJGAT.Gtrl_Human 9REKKVTIL ELiORSPAYRQ IPILIAWLOLI SQQLSGNAZMSTSIFEK 300

Ompf |ANYGETRAT PITNKFTNTS GFAN QDVL LVAQYQFPFLRPSI AYTKS16070 -AYYAD DLT PVAAA D ... YGLS VDSTF;mGtrl_Human A*QQPVYAT .GSGIV G RAGVRTLHLIA GLD.AGCQ 350

Ompf SKADVEGIG DVD VNYFEV GATY.... Y NKMSTYVDY IQIDSDNKS16070 QGGYVOVLI D T DDVTYYGL GAS .... DL GGO AS.... IVGGI.ADDGtrl_Human IALAL QLPWMSY SI VAIFGFVAFFPEVGPGPIE LVA QGP 400

Ompf LGVGSD . ....... AVG IVYQF AEIYN K......... ..........

S16070 LP.NSDA . ....... DLG VKFK .. .......... ..........

Gtrl_Human RPAAIAV O SHPIVG H P QLC GPY| V LLVLFF 450

OmpfS16070Gtrl_Human -BPETKG IASG Q GGASQSDKI 3PLGAD SQV 493

FIG. 2. Multiple sequence alignment of two porins and GLUT1.S16070 stands for POR. Rectangles, existing (OmpF, POR) andpredicted (GLUT1) 13strands. Rounded rectangles, existing (OmpF,POR) and predicted (GLUT1) a-helices.

random, 0.44. The PHD method (available only for OmpF)predicted regions with secondary structure similar to ours(data not shown; Q3 = 0.68).The paucity of membrane proteins with known crystallo-

graphic structure does not allow meaningful statistical pre-dictive procedures (data not shown). Necessarily, we usedinstead prediction methods (CFP and PHD) derived fornonmembrane proteins. While it could be fortuitous that suchmethods yield sharply defined segments that correspond withPOR and OmpF known structure (and with the knownstructure of other membrane proteins such as bacteri-orhodopsin, the reaction center, and the pore-forming seg-ment of colicin; data not shown), the correspondence isnonetheless noteworthy (Fig. 3). Despite the inherent uncer-tainty, we applied these methods to GLUT1.

AL

=lio*

1.

cQ.cQ

0 50 100 150 200 250 300 350

50 100 150 200 250 300 350

Residue number

FIG. 3. Prediction of porin structures using union. Area graphs,Up prediction profiles. Structures known from crystallography(cryst.) (4, 5) or predicted (prd.) are shown above the profiles in eachcase.

11660 Biophysics: Fischbarg et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 30

, 202

0

Proc. Natl. Acad. Sci. USA 90 (1993) 11661

Prediction of Multiple tm ,-Strands in GLUT1. We identi-fied 16 predicted tm ,3strands in GLUT1 (Fig. 4). All were insegments that had been allocated as tm helices in the 12Hmodel (Fig. 2). Using only H21 profiles, several of the peaksseen (Fig. 2) appeared wide enough to be interpretable as tma-helices with spans of 21 residues (2). However, four ofthem (arrows in Fig. 4) were split by predicted turns. Theresulting segments were too short to bridge the membrane asa-helices but had the correct length for tm 13strands. Wetermed such patterns "1-hairpin signatures." Similarly, inthe remaining 8 segments previously predicted as 20-residuehelices (Fig. 2) we predict tm 1-strands some 10 residueslong, with the rest of the residues sometimes forming shorthelices. Our predictions for the location and length of seg-ments with secondary structure are in reasonable agreementwith those from the PHD program (Fig. 2).Given these predictions, we reexamined the alignment of

the sequences ofPOR, OmpF, and GLUT1. We verified thatsegments known to have secondary structure in one or bothporins aligned well with segments for which we predictedsecondary structure in GLUT1 (Fig. 2). Eleven of the 16predicted 13strands in GLUT1 overlapped partially with1-strands in porins. The paucity of gaps in these regions withconserved secondary structure is noteworthy. Some of theremaining 1-strands in the porins correspond to segmentspredicted as helices in GLUT1 and vice versa. The alignmentin Fig. 2 comprised about the last 400 residues in GLUT1;based on additional alignments (data not shown), the N-ter-minal region of GLUT1 might have originated in partialduplication of a porin gene. In addition, there is a high degreeof sequence conservation among members of the GLUTfamily, and hence a multi-,B-strand motifmay be applicable toall of them (data not shown).

Three-Dimensional Model of the fiB in GLUT1. The pre-dictions above suggested to us that GLUT1 might fold as theporins, forming a ,1B. To visualize whether such an idealizedconstruct was compatible with GLUT function, we built athree-dimensional model ofthe putative GLUT1,3B, with the

HUMAN GLUT1 FACILITATIVEGLUCOSETRANSPORTER0 50 100 150 200 250 300 350 400 450 500

more hydrophilic sides of the tm 1-strands facing the barrelpore. To ensure that there were no bad Van der Waalscontacts, limited energy minimization was performed (300iterations, conjugate descent algorithm, DISCOVER program).Fig. 5 shows an end-view photograph of the barrel (frominside the cell) including ,B-D-glucopyranose in its lumen. TheVan der Waals inside diameter of the barrel, while irregular,was at least 11 A, which is more than enough to allowhydrated hexoses to pass through the channel.

Prior Evidence Consistent with a fBB Fold. The 2-N-[4-(1-azi-2,2,2-trifluoroethyl)benzoylJ-1,3-bis-(D-mannos4-yloxy)-2-propylamine (ATB-BMPA) binding site. Peptide 217-272appears intracellular, since a specific Ab binds to it only whenthe cell membrane is permeabilized (18). This segment is veryhydrophilic, so that the more hydrophobic tm segment thatfollows it is likely to begin only at or near residue 273 (ineither the 12H, 13B, or PHD predictions; Fig. 2). The nextmarker along the chain is residue 282, which has beenrecently placed extracellularly, since mutation of it (Gln -)

Leu) decreases ATB-BMPA exofacial binding by 95% (19).Hence, segment 273-281 likely spans the membrane; thissegment (9 residues) is too short to be a tm a-helix (21residues) but has the correct length for a tm 1-strand (strand9, residues 271-280, Figs. 4 and 6). In the 12H model, residue282 was placed at the center of tm a-helix 7, where it wouldbe inaccessible to ATB-BMPA. In the 13B model, residue 282is instead in an extracellular connecting loop.

The proportions of a and 13 structures in GLUT based onCD and FTIR spectroscopy. This issue is unsettled. FromFTIR spectroscopic evidence, it was concluded that GLUT1displays distinct vibrations for a-helical structure while thosefor 1-structure are absent (21). This was partly challenged bya later FTIR study, which also found GLUT1 to be predom-inantly a-helical but in addition found evidence stronglysuggesting the presence of some 13-structure, with a portionof it forming antiparallel strands (22). Interpretations of CDevidence also appear divided. In one case, CD was said toindicate the presence in GLUT1 of some 82% a-helices, 10%o1-turns, and 8% random structure, with no 1-strands (23).However, more recently, use of an algorithm (24) to analyzeCD data led to predictions (25) of 13-structure in GLUT1,POR, and OmpF, among other membrane proteins. Our

42n

X- -21~~~WII-4

o 4-

X 2- ,P

>,-4- 2 3 4 5 67 9 10 11 12 HD

2 4 Glycos. A TB-BMPA Ab a

2 -7

-4-

0 50 100 150 200 250 300 350 400 450 500

Residue number

FIG. 4. Our prediction for GLUT1. From top down, predictionprofiles of hydrophobicity; turn; and union propensity for amphi-pathic a-helices and /-strands, respectively. Spans: 4 for <pt>;others are indicated in label subindices. For comparison, predictedstructures are shown at the top and bottom panels. For the 12Hmodel and for our prediction, symbols are shown angled so that theirlower and higher ends correspond to their intra- and extracellularsides, respectively.

FIG. 5. Putative ,BB of GLUT1 viewed from inside the cell. Amolecule of f-D-glucopyranose is shown in the center of the pore asa size marker (viewed from Cl).

Biophysics: Fischbarg et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 30

, 202

0

11662 Biophysics: Fischbarg et al. Proc. Natl. Acad. Sci. USA 90 (1993)

EXTRACELLULAR

[~22.

173V.

C) ''A

..L

::

G.:V

V,

6.164::

186 280.W L

P

L L

L V

L V

S: A

' I1L:

F

195 271

303:.V

P.:V

Y.AT-

311:1

343

L:

.R

.334 :

)217-

386-405

C½433

348 379 425

A FF 400°° F.FY:

L A 0

A F

MT F A TG M. V

A G L

L V: L

A VV FL G L

N

L F

358T FIL T

367 { 413 446

*272

FIG. 6. Model of secondary structure of GLUT1. Putative 16transmembrane (3-strands are represented by rectangles. The morehydrophilic sides of the /-strands (presumably lining the pore) arefacing right. In the extramembrane loops, triangles denote predictedturns, and rectangles mark predicted a-helices. Of the two possibleN-linked glycosylation sites N 45 andN 411, mutagenesis points to thefirst (20). Two epitopes, 217-272 (18) and 386-405 (this paper) and asugar binding site, Q 282 (19), mentioned in the text are boldface.

assignments for GLUT1 structure appear to be in line withthe more recent FTIR and CD studies (22, 25).

Solvent accessibility of the GLUT backbone is betterexplained by the /3B model. Others and ourselves havereported evidence for the existence of a water-filled poreacross GLUTs (22, 26, 27). Such an open pathway wouldhave to coexist with an apparent enzyme-type tight-fittingstructure, since GLUTs display steric selectivity for sub-strates (1). This apparent contradiction may be resolved bynoting that the water permeability of GLUTs (28) is onlysome 7% that of specific water channels (29), as if watertraverses an open pathway through GLUTs only during partof a cycle of conformational changes. Both the 12H and 13Bmodels imply a hydrophilic pore in GLUT. On the basis ofhydrogen-deuterium exchange, however, -90% of theGLUT1 amide protons are exchanged almost immediately(22, 30). These exchange data can be explained more readilyif GLUT1 is a B3B with a solvent-filled pore, as in that casemost backbone amide hydrogens lining the pore and formingconnecting loops would be accessible to solvent.GLUT1 as a Multifunctional 13B Transporter. From recent

evidence, compounds other than sugars such as water (27,31), nicotinamide (32), and dehydroascorbic acid (33) tra-verse GLUTs, suggesting that GLUTs are multifunctional(32). Since a barrel framework is essentially fixed, as arguedfor porins (4, 5), the GLUT1 connecting loops might operateas molecular gates and might also be involved in bindingsolutes and discriminating among them. The putative longintracellular GLUT1 loop (residues 204-270) may be anexample, since glucose binding to the loop induces a confor-mational change in it (34) and antibodies against the peptideAsn-217-Ile-272 inhibit the binding of cytochalasin B to theprotein (18). This loop may also have a binding site for ATP(Lys-225-Lys-229) (35) and protein kinase C phosphorylationsites (Ser-226, Ser-248) (36), all with potential functionalroles. Lastly, all three antibodies we tested bind to putativemobile loops and enhance DOG uptake. The topology wepropose is summarized in Fig. 6.A B3B Transporter Superfamily? Evolutionarily distant po-

rins are common in bacteria and share a well-conserved ,BBmotif (3). Predictive analysis of bacterial transporters of

several families has recently described in them (37) a com-mon translocation unit onto which energy coupling compo-nents may have been evolutionarily grafted. From our ownpredictions, recent conjecture that membrane proteins otherthan porins may belong to a ,BB superfamily (3) appearsplausible; 83Bs could form the translocation units of othertransporters and channels. Our analysis of GLUTs may helpfocus attention on this possibility.

This work was supported by National Institutes of Health GrantsEY06178, EY08918, EY07010, CA08748, and HL42107, and byResearch to Prevent Blindness, Inc. We used the following ColumbiaUniversity facilities: (i) molecular modeling (National Science Foun-dation Grant DIR8720229); (ii) computer (Cancer Center, NationalInstitutes of Health Grant P30 CA13696).1. Carruthers, A. (1990) Physiol. Rev. 70, 1135-1175.2. Mueckler, M., Caruso, C., Baldwin, S. A., Panico, M., Blench, I.,

Morris, H. R., Allard, W. J., Lienhard, G. E. & Lodish, H. F. (1985)Science 229, 941-945.

3. Pauptit, R. A., Schirmer, T., Jansonius, J. N., Rosenbusch, J. P.,Parker, M. W., Tucker, A. D., Tsernoglou, D., Weiss, M. S. & Schulz,G. E. (1991) J. Struct. Biol. 107, 136-145.

4. Weiss, M. S., Kreusch, A., Schiltz, E., Nestel, U., Welte, W., Weck-esser, J. & Schulz, G. E. (1991) FEBS Lett. 280, 379-382.

5. Cowan, S. W., Shirmer, T., Rummel, G., Steiert, M., Ghosh, R., Pauptit,R. A., Jansonius, J. N. & Rosenbusch, J. P. (1992) Nature (London) 358,727-733.

6. Vera, J. C. & Rosen, 0. M. (1990) Mol. Cell. Biol. 10, 743-751.7. Needleman, S. B. & Wunsch, C. D. (1970) J. Mol. Biol. 48, 443-453.8. Kyte, J. & Doolittle, R. F. (1982) J. Mol. Biol. 157, 105-132.9. Eisenberg, D., Weiss, R. M. & Terwilliger, T. C. (1984) Proc. Natl.

Acad. Sci. USA 81, 140-144.10. Prevelige, P. & Fasman, G. D. (1989) in Prediction ofProtein Structure

and the Principles ofProtein Conformation, ed. Fasman, G. D. (Plenum,New York), pp. 391-416.

11. Rost, B. & Sander, C. (1992) Nature (London) 360, 540.12. Crofts, A. R. (1992) Ph.D. Dissertation (Univ. of Illinois, Urbana).13. Oka, Y., Asano, T., Shibasaki, Y., Lin, J.-L., Tsukuda, K., Katagiri, H.,

Akanuma, Y. & Takaku, F. (1990) Nature (London) 345, 550-553.14. Parker, M. W., Postma, J. P. M., Pattus, F., Tucker, A. D. &Tsernoglu,

D. (1992) J. Mol. Biol. 224, 639-657.15. Welte, W., Weiss, M. S., Nestel, U., Weckesser, J., Schiltz, E. &

Schulz, G. E. (1991) Biochim. Biophys. Acta 1080, 271-274.16. Qian, N. & Sejnowski, T. J. (1988) J. Mol. Biol. 202, 865-884.17. Mathews, B. W. (1975) Biochim. Biophys. Acta 405, 442-451.18. Davis, A., Ciardelli, T. L., Lienhard, G. E., Boyle, J. M., Whetton,

A. D. & Baldwin, S. A. (1990) Biochem. J. 266, 799-808.19. Hashiramoto, M., Kadowaki, T., Clark, A. E., Muraoka, A., Momo-

mura, K., Sakura, H., Tobe, K., Akanuma, Y., Yazaki, Y., Holman,G. D. & Kasuga, M. (1992) J. Biol. Chem. 267, 17502-17507.

20. Asano, T., Katagiri, H., Takata, K., Lin, J. L., Ishihara, H., Inukai, K.,Tsukuda, K., Kikuchi, M., Hirano, H., Yazaki, Y. & Oka, Y. (1991) J.Biol. Chem. 266, 24632-24636.

21. Chin, J. J., Jung, E. K. Y. & Jung, C. Y. (1986) J. Biol. Chem. 261,7101-7104.

22. Alvarez, J., Lee, D. C., Baldwin, S. A. & Chapman, D. (1987) J. Biol.Chem. 262, 3502-3509.

23. Chin, J. J., Jung, E. K. Y., Chen, V. & Jung, C. Y. (1987) Proc. Natl.Acad. Sci. USA 84, 4113-4116.

24. Perczel, A., Hollosi, A. M., Tusnady, G. & Fasman, G. D. (1991) ProteinEng. 4, 669-679.

25. Park, K., Perczel, A. & Fasman, G. D. (1992) Protein Sci. 1, 1032-1049.26. Jung, E. K. Y., Chin, J. J. & Jung, C. Y. (1986) J. Biol. Chem. 261,

9155-9160.27. Fischbarg, J., Kuang, K., Vera, J. C., Arant, S., Silverstein, S. C.,

Loike, J. & Rosen, 0. M. (1990) Proc. Natl. Acad. Sci. USA 87,3244-3247.

28. Fischbarg, J., Kuang, K., Li, J., Arant-Hickman, S., Vera, J. C., Silver-stein, S. C. & Loike, J. D. (1993) Alfred Benzon Symp. 34, 432-446.

29. Preston, G. M., Carroll, W. B., Guggino, W. B. & Agre, P. (1992)Science 256, 385-387.

30. Haris, P. I. & Chapman, D. (1992) Trends Biochem. Sci. 17, 328-333.31. Zhang, R., Alper, S., Thorens, B. & Verkman, A. S. (1991) J. Clin.

Invest. 88, 1553-1558.32. Sofue, M., Yoshimura, Y., Nishida, M. & Kawada, J. (1992) Biochem.

J. 288, 669-674.33. Vera, J. C., Rivas, C. E., Fischbarg, J. & Golde, D. W. (1993) Nature

(London) 364, 79-82.34. Asano, T., Katagiri, H., Tsukuda, K., Lin, J. L., Ishihara, H., Inukai,

K., Yazaki, Y. & Oka, Y. (1992) FEBS Lett. 298, 129-132.35. Carruthers, A. & Helgerson, A. L. (1989) Biochemistry 28, 8337-8346.36. Deziel, M. R., Lippes, H. A., Rampal, A. L. & Jung, C. Y. (1989) Int.

J. Biochem. 21, 807-814.37. Nikaido, H. & Saier, M. H. (1992) Science 258, 936-942.

45

22 61 87 132 137T V 96 T

L F M V GTV L L G F.

LA~ GMP.

G V MS. F

:: M: P:G0- S L F M

L F L R-SA S: A G VV

L G F L GM: V E

L F M ::105 :: V:12 77 M 147

72E::120 E

I

Dow

nloa

ded

by g

uest

on

Nov

embe

r 30

, 202

0