Crystal Structure of the Interferon-induced Ubiquitin-like Protein ISG15* □ S

The biological effects of the ISG15 protein arise in part from its conjugation to cellular targets as a primary response to interferon- (cid:1) / (cid:2) induction and other markers of viral or parasitic infection. Recombinant full-length ISG15 has been produced for the first time in high yield by mutating Cys 78 to stabilize the protein and by cloning in a C-terminal arginine cap to protect the C terminus against proteolytic inactivation. The cap is subsequently removed with carboxypeptidase B to yield mature biologically active ISG15 capable of stoichiomet-ric ATP-dependent thiolester formation with its human UbE1L activating enzyme. The three-dimensional structure of recombinant ISG15C78S was determined at 2.4-Å resolution. The ISG15 structure comprises two (cid:2) -grasp folds having main chain root mean square deviation (r.m.s.d.) values from ubiquitin of 1.7 Å (N-terminal) and 1.0 Å (C-terminal). The (cid:2) -grasp domains pack across two conserved 3 10 helices to bury 627 Å 2 that accounts for 7% of the total solvent-accessible surface area. The distribution of ISG15 surface charge forms a ridge of negative charge extending nearly the full-length of the molecule. Additionally, the N-terminal The latter activity the carboxypeptidase B by dissolving the commercial enzyme in 50 m M Tris-HCl (pH 7.6) then isocratically passing the sample through a Mono S HR5/5 cation exchange column equilibrated with the same. Car- boxypeptidase B free of the contaminating carboxypeptidase A-like activity appeared in the unadsorbed fraction of the Mono S column. The resulting mature ISG15, ISG15C78S, or ISG15N13Y/C78S proteins were resolved from carboxypeptidase B by Superdex 75 HR 10/30 analytical gel filtration chromatography in 50 m M Tris-HCl (pH 7.5) containing 50 m M NaCl and 1 m M DTT at 1 ml/min flow rate. Mature recombinant ISG15 or its point mutants eluted as single apparently homogeneous peaks at their predicted monomer molecular masses of (cid:2) 17 kDa. Concentrations of recombinant ISG15 and ISG15C78S were spectrophotometrically quantitated using an empirically determined 280 nm extinction coefficient of 0.91 ml/mg (cid:2) cm (13). The concentration of recombinant ISG15N13Y/C78S was quantitated similarly using an empirical extinction coefficient of 0.98 ml/mg (cid:2) cm. Typical final yields for all three recombinant forms of ISG15 were 2–3 mg/liter of culture. Samples were flash frozen in small aliquots and stored at (cid:4) 80 °C. A 1-mg aliquot of ISG15N13Y/C78S was radioiodinated by the chloram- ine-T protocol developed previously for labeling of ubiquitin and Nedd8 (31, 32), yielding a specific radioactivity of 51,900 cpm/pmol. CD Spectroscopy Studies— Circular dichroism was monitored using a Jasco spectropolarimeter equipped with a thermostatted cell holder and interfaced A cylindrical with a was used for The CD four was from to 260 nm for containing the indicated concentration of protein dissolved in 10 m M Tris-HCl (pH and 0.2 m M DTT equilibrated to °C. Secondary structure calculations obtained from the CD spectra were based on the self-consistent method of Sreerama and Woody For measurement of protein unfolding by urea denaturation, equimo-lar concentrations of recombinant protein were prepared in 10 m M Tris-HCl (pH containing 0.2 m M DTT then mixed with an appro-priate concentration of 9 M urea (ultrapure DNA sequencing grade) dissolved in the same buffer to produce the indicated range of final urea concentrations. Solutions were allowed to equilibrate for at least 30 min before collecting spectra, which were constant with time thereafter. The free energy of denaturation was calculated assuming a two-state model The equilibrium constant K was determined by the change in ellipticity at 222 nm according to the [ ] D ), where [ (cid:5) ] N and [ (cid:5) ] D are the ellipticities for the native and denatured states, respectively, which were obtained by a least squares analysis of the pre- and post-transition regions, and [ (cid:5) ] is the residue ellipticity at each urea concentration apparent free energy of denaturation (cid:6) G app , linearly with denaturant concentration in the transition region about the inflection point the denatured, F D denaturant is by graphics software Seven per-cent of the data were withheld from the refinement process and used to calculate R free for cross-validation. After each round of CNS refinement, both 3 (cid:2) F o (cid:2) (cid:4) 2 (cid:2) F c (cid:2) and (cid:2) F o (cid:2) (cid:4) (cid:2) F c (cid:2) difference Fourier maps were calculated and examined for manual fitting of the model using TURBO-FRODO. The final model of the I222 structure revealed an osmium atom bound to the protein with a low occupancy (estimated occupancy of (cid:2) 0.3). To confirm that the osmium soaking did not result in any conformational changes, the structure of the P2 1 2 1 2 1 crystal form was solved by the molecular replacement methods using the I222 structure as the start-ing model. The resulting model was further refined using CNS and TURBO-FRODO. Ramachandran plot analyses indicate that all residues lie in either most-favored or additionally allowed regions for the structures of both crystal forms. The data collection and the final refinement statistics are given in Table I.

The broad pleiotropic and anti-viral effects of the interferons are mediated through signaling pathways that culminate in the enhanced expression of a temporally coordinated subset of cell type-specific proteins (1,2). Induction of ISG15 marks the earliest and most universal consequence of exposure to various interferon isoforms, suggesting that elevated levels of this 17-kDa polypeptide are essential for the subsequent response of cells to these cytokines (3,4) and their ability to mount an effective anti-viral defense (3)(4)(5)(6). The ISG15 polypeptide is induced by lipopolysaccharide and double-stranded RNA, both potent markers of parasitic and viral infection (7,8). Ectopic overexpression of ISG15 directly inhibits human immunodeficiency virus replication by abrogating nuclear processing of unspliced viral RNA precursors, leading to accumulation of nascent transcripts (9). Similarly, influenza B virus replication is blocked by endogenous ISG15 unless its expression is prevented by a specific transcriptional block mediated by the viral NS1 protein (10). Most recently, murine knock-out studies implicate an essential role for ISG15 in innate immunity, a critical defense against infectious agents that guides subsequent host adaptive immune responses (11).
Like most interferon-induced proteins, ISG15 is constitutively present in higher eukaryotes and serves additional function(s) unrelated to its role in anti-viral defenses (4,12). Recognition that ISG15 harbors tandem ubiquitin-like domains and terminates in a conserved 152 LRLRGG 157 ubiquitin C-terminal motif suggested some of the biological effects attributed to the polypeptide were directed in part by its conjugation to cellular targets (4,13). Immunochemical studies confirm that ISG15 conjugates are abundant in eukaryotic cells and increase significantly in response to interferon induction (13,14). The ISG15 polypeptide was the first example identified of the superfamily of Class 1 ubiquitin-like proteins, defined by their ability to undergo covalent ligation reactions to specific cellular targets (reviewed in Refs. 15 and 16). Nascent ISG15 is synthesized as an inactive precursor in which the mature glycine C terminus that participates in isopeptide bonds to cellular targets is blocked by an extension peptide that is post-translationally processed by the human ortholog of Ubp1 to yield the active ISG15 protein (17). Interestingly, an extracellular cytokine activity of ISG15 in stimulating interferon-␥ production and natural killer cell proliferation also requires an intact mature ISG15 C terminus (18).
Conjugation of ISG15 utilizes a parallel but distinct mechanism to that of ubiquitin (19). The ATP-dependent activation of ISG15 that is required for subsequent conjugation to cellular protein targets is catalyzed by UbE1L 1 (10), a 112-kDa paralog of the E1/Uba1 ubiquitin-activating enzyme (20). Marked ablation of UbE1L levels in various small cell lung carcinoma lines suggests loss of ISG15 conjugation is a contributing factor in malignant transformation (20,21). Similarly, induction of UbE1L is directly implicated in retinoic acid-mediated remission during acute promyelocytic leukemia (22), evidence for which includes the observation that either ectopic over expression of UbE1L or its induction by all-trans retinoic acid triggers 26 S proteasome-dependent degradation of promyelocytic leukemia/retinoic acid receptor-␣ protein and apoptosis in promyelocytic leukemia (23). More recently, human UbcH8 has been identified as the E2 conjugating enzyme for ISG15 and requires the action of UbE1L for ATP-dependent formation of the obligate UbcH8-ISG15 thiolester intermediate for conjugation (24). Cognate isopeptide ligases required for ISG15 conjugation have not been definitively identified to date but are suggested to overlap with a subset of putative ubiquitin-dependant ligases requiring UbcH8 (24). A limited number of protein targets for ISG15 conjugation have been reported, including the murine serpin 2a induced in response to parasite infection (25) and components of the Janus tyrosine kinase-STAT signaling pathway (26). However, observation that disruption of intracellular ISG15 conjugation dynamics by loss of function mutations in the ISG15-specific isopeptidase UBP43 results in severe developmental neurological defects hints at more global roles for ISG15 ligation (27).
Although ISG15 was the first ubiquitin-like protein identified (4,13), much less is known about the functional roles of this polypeptide than for other members of the superfamily. In part this derives from the remarkable absence of ISG15 orthologs in lower eukaryotes that precludes the powerful genetic approaches successfully exploited previously with its more ubiquitously distributed paralogs (13). In addition, the marked instability of ISG15 noted in early work continues as a major impediment to detailed in vitro analysis (28). In the present studies we demonstrate that the inherent instability of recombinant ISG15 derives from disulfide-linked dimerization through Cys 78 of the polypeptide, point mutation of which to serine markedly increases the stability of the polypeptide to that approaching ubiquitin. Enhanced stability has allowed the crystallization and structural determination of recombinant ISG15 for the first time. The 2.4-Å crystal structure for ISG15 confirms the tandem ubiquitin-like domains of ␤-grasp folds originally proposed on the basis of sequence analysis (13). The structure for ISG15 reveals a novel surface charge distribution between the ␤-grasp domains and an extensive solvent-exposed surface of low polarity encompassing a substantial fraction of the N-terminal ␤-grasp domain. Finally, docking simulations using the recent structure for Nedd8 bound to its cognate activating enzyme, the AppBp1-Uba3 heterodimer (29), suggests the basis for the specificity of UbE1L for ISG15 activation.

MATERIALS AND METHODS
Bovine ubiquitin was purchased from Sigma and purified to apparent homogeneity by fast protein liquid chromatography (30). Human recombinant proISG15 was that described previously (17). The complete coding sequence for human ISG15-activating enzyme UbE1L (generous gift of Robert M. Krug) was subcloned by PCR from a modified pVL1393 baculovirus transfer vector (10) into the BamH1/EcoR1 sites of pGEX to yield pGEX-UbE1L. Human GST-UbE1L was expressed in Escherichia coli BL21(DE3) cells harboring pGEX-UBE1L following isopropyl 1-thio-␤-D-galactopyranoside induction. The resulting recombinant GST-UbE1L was purified on glutathione-Sepharose following the manufacturer's instructions.
Expression and Purification of Recombinant ISG15-Ubiquitin and other type 1 ubiquitin-like proteins are sensitive to proteolytic inactivation through cleavage of their C-terminal glycine dipeptide by a widely expressed bacterial periplasmic carboxypeptidase (19). To cir-cumvent this problem in the present studies, a CGT codon for arginine was inserted between the mature human ISG15 C-terminal glycine GGC codon and the TAA STOP codon of pETUCRP (13) by PCR using an appropriately designed 3Ј-external primer containing a BamH1 restriction site. The resulting full-length PCR-amplified DNA was restricted with NcoI/BamH1 then ligated into a similarly restricted pET11d plasmid to yield pET11d-ISG15-R158. Because the bacterial carboxypeptidase that inactivates ubiquitin-like proteins is incapable of removing a C-terminal arginine residue, the presence of Arg 158 provided a cap residue to protect the integrity of the mature ISG15 C terminus during purification. In addition, the shift in pI from 6.7 for wild type mature polypeptide to 8.5 for recombinant ISG15-R158 enhanced resolution of the latter from contaminating proteins that were otherwise difficult to remove without additional chromatographic steps that reduced yield. Circular dichroism studies described under "Results" demonstrated that recombinant wild-type ISG15 exhibited significant structural instability resulting from its propensity spontaneously to form a disulfide-linked dimer during expression and purification. Disulfide-dependent instability was addressed by mutating the single Cys 78 of human mature ISG15-R158 to serine, yielding ISG15C78S-R158. The Cys 78 codon within pET11d-ISG15-R178 was mutated by overlap extension PCR using appropriately designed oligonucleotide primers, then the complete coding region was restricted with NcoI/BamH1 and ligated into similarly restricted pET11d to yield pET11d-ISG15C88S-R158. In subsequent functional studies, we found that recombinant mature ISG15 and ISG15C78S, from which the protective arginine caps had been proteolytically processed, were poorly radioiodinated by chloramine T using a protocol previously successful with ubiquitin and Nedd8 (31,32). Poor radioiodination presumably resulted from the relative solvent inaccessibility of the tyrosine residues within the ISG15 structure compared with the other two Type 1 ubiquitin-like proteins, ultimately confirmed by the refined crystal structure of the polypeptide. Overlap extension PCR was exploited to mutate the Asn 13 codon of human mature pET11d-ISG15C78S-R158 to tyrosine, yielding pET11d-ISG15N13Y/C78S-R158, to generate a solvent-accessible site for radioiodination, predicted from the subsequent structure for ISG15C78S.
Five liters of E. coli strain BL21(DE3) harboring either pET11d-ISG15-R158, pET11d-ISG15C88S-R158, or pET11d-ISG15N13Y/C78S-R158 were grown at 30°C to A 600 nm of ϳ0.6 in Luria broth containing 100 g/ml ampicillin then induced by addition of isopropyl 1-thio-␤-Dgalactopyranoside to a final concentration of 0.4 mM. After additional 1.5-h incubation at 30°C, cells were harvested by centrifugation. The resulting cell pellet was suspended in 100 ml of ice-cold 50 mM Tris-HCl (pH 7.5) containing 1 mM DTT, then lysed at 12,000 p.s.i. in a French pressure cell (all subsequent steps were conducted at 4°C). Crude extract was centrifuged at 10 5 ϫ g for 60 min, after which supernatant proteins precipitating between 30 and 50% saturated ammonium sulfate were collected by centrifugation, resuspended in 50 mM Tris-HCl (pH 7.5) containing 1 mM DTT, and dialyzed against 2 ϫ 4 liters of the same using dialysis tubing having a 3.5-kDa exclusion limit. The sample was then applied to a 5 ϫ 20 cm column of DEAE-52 anion exchange cellulose (Whatman) equilibrated with 50 mM Tris-HCl (pH 7.5) containing 1 mM DTT. Fractions containing the unadsorbed fraction were pooled then adjusted to 1 M ammonium sulfate and applied to a 3-ϫ 29-cm column of Phenyl-Superose FAST FLOW equilibrated with 20 mM sodium phosphate buffer (pH 7.2) containing 1.0 M ammonium sulfate and 1 mM DTT. Bound proteins were eluted by a 1.0 -0 M negative salt gradient (5 mM/ml) at 2 ml/min. Recombinant ISG15-R158, ISG15C78S-R158, and ISG15N13Y/C78S-R158 elute as single peaks at 0.4 M ammonium sulfate. Fractions containing ISG15-R158 or the two point mutants were pooled and concentrated by ultrafiltration to ϳ5 ml then resolved at 1 ml/min on a 1.5-ϫ 30-cm column of Sepharose 6 FAST FLOW equilibrated with 50 mM Tris-HCl (pH 7.5) containing 25 mM NaCl and 1 mM DTT. Fractions containing the recombinant proteins were pooled then dialyzed against 1 mM DTT using 3.5-kDa exclusion tubing.
The protective Arg 158 cap was removed from purified recombinant ISG15 polypeptides by exploiting the absolute specificity of porcine pancreatic carboxypeptidase B (EC 3.4.17.2) for cleaving C-terminal lysine and arginine residues. Recombinant ISG15-R158, ISG15C78S-R158, or ISG15N13Y/C78S-R158 at a concentration of 3-5 mg/ml in 50 mM Tris-HCl (pH 7.6) containing 150 mm NaCl and 1 mM DTT was incubated for 4 -6 h at 20°C in the presence of 2 IU of carboxypeptidase B (Sigma). Quantitative processing was assessed by analytical isoelectric focusing (19) and confirmed by MALDI-TOF mass spectrometry. In preliminary experiments we determined that the commercial carboxypeptidase B enzyme contained trace contaminating carboxypepti-dase A-like activity that slowly cleaved the exposed glycine dipeptide following processing. The latter activity could be resolved from the carboxypeptidase B by dissolving the commercial enzyme in 50 mM Tris-HCl (pH 7.6) then isocratically passing the sample through a Mono S HR5/5 cation exchange column equilibrated with the same. Carboxypeptidase B free of the contaminating carboxypeptidase A-like activity appeared in the unadsorbed fraction of the Mono S column.
The resulting mature ISG15, ISG15C78S, or ISG15N13Y/C78S proteins were resolved from carboxypeptidase B by Superdex 75 HR 10/30 analytical gel filtration chromatography in 50 mM Tris-HCl (pH 7.5) containing 50 mM NaCl and 1 mM DTT at 1 ml/min flow rate. Mature recombinant ISG15 or its point mutants eluted as single apparently homogeneous peaks at their predicted monomer molecular masses of ϳ17 kDa. Concentrations of recombinant ISG15 and ISG15C78S were spectrophotometrically quantitated using an empirically determined 280 nm extinction coefficient of 0.91 ml/mg⅐cm (13). The concentration of recombinant ISG15N13Y/C78S was quantitated similarly using an empirical extinction coefficient of 0.98 ml/mg⅐cm. Typical final yields for all three recombinant forms of ISG15 were 2-3 mg/liter of culture. Samples were flash frozen in small aliquots and stored at Ϫ80°C. A 1-mg aliquot of ISG15N13Y/C78S was radioiodinated by the chloramine-T protocol developed previously for labeling of ubiquitin and Nedd8 (31,32), yielding a specific radioactivity of 51,900 cpm/pmol.
CD Spectroscopy Studies-Circular dichroism was monitored using a Jasco J710 spectropolarimeter equipped with a thermostatted cell holder and interfaced with a Neslab RTE-111 water bath. A thermostatted cylindrical cuvette with a 1-mm path length was used for all measurements. The CD signal (average of four scans) was measured from 190 to 260 nm for solutions containing the indicated concentration of protein dissolved in 10 mM Tris-HCl (pH 7.5) and 0.2 mM DTT equilibrated to 20°C. Secondary structure calculations obtained from the CD spectra were based on the self-consistent method of Sreerama and Woody (33,34).
For measurement of protein unfolding by urea denaturation, equimolar concentrations of recombinant protein were prepared in 10 mM Tris-HCl (pH 7.5) containing 0.2 mM DTT then mixed with an appropriate concentration of 9 M urea (ultrapure DNA sequencing grade) dissolved in the same buffer to produce the indicated range of final urea concentrations. Solutions were allowed to equilibrate for at least 30 min before collecting spectra, which were constant with time thereafter. The free energy of denaturation was calculated assuming a two-state model (35). The equilibrium constant K was determined by the change in ellipticity at 222 nm according to the relationship, where [] N and [] D are the ellipticities for the native and denatured states, respectively, which were obtained by a least squares analysis of the pre-and post-transition regions, and [] is the residue ellipticity at each urea concentration (35). The apparent free energy of denaturation ⌬G app , varies linearly with denaturant concentration in the transition region about the inflection point when the fraction denatured, F D , is plotted versus denaturant concentration and is given by the relationship, ⌬G app ϭ ⌬G H 2 O Ϫ m[denaturant], in which ⌬G H 2 O is the apparent free energy for denaturation in the absence of denaturant. The value for ⌬G H 2 O is obtained by linear extrapolation of ⌬G app to [denaturant] ϭ 0. The slope, m, of the resulting plot is formally a function of the difference in the denaturant molecules bound to the denatured and native states of the protein (35).
Crystallization of ISG15-An initial crystallization screen was performed by vapor diffusion using the hanging drop method (36) following the sparse matrix protocol (37). One of the conditions yielded microcrystals within 1-2 days, and refinement of this condition resulted in thin plate-like crystals. The refined crystallization condition involved mixing equal volumes of the ISG15C78S protein (12 mg/ml) with the precipitating solution (0.1 M Tris-HCl (pH 8.0), 15% (w/v) polyethylene glycol 4000, and 0.2 M MgCl 2 ) and equilibrating against the precipitating solution. Larger single crystals were subsequently obtained by seeding the thin small crystals in the same solution but containing a lower concentration of the protein (10 mg/ml). Diffraction quality crystals grew to 0.3 ϫ 0.3 ϫ 0.05 mm within 2 weeks after seeding. In contrast, no crystals were obtained from wild-type ISG15 under similar conditions, presumably because of disulfide induced instability of the protein.
Data Collection-The crystal was mounted in a glass capillary, and diffraction data were collected at 4°C on an R-axis IIC image plate detector system with a Rigaku rotating anode generator. Data were collected by the oscillation method and processed using the HKL program package (38). Native ISG15C78S crystals belonged to the space group P2 1 2 1 2 1 with unit cell parameters a ϭ 55.2 Å, b ϭ 57.1 Å, and c ϭ 104.2 Å, containing two ISG15 molecules per asymmetric unit (V m ϭ 2.4 Å 3 /Da). However, upon addition of heavy atom compounds, P2 1 2 1 2 1 crystals were converted to another crystal form, making it difficult to obtain a heavy atom derivative. Soaking crystals belonging to the P2 1 2 1 2 1 form in a solution containing 5 mM K 2 OsCl 6 converted the crystals to the space group I222 (cell dimensions, a ϭ 55.8 Å, b ϭ 56.9 Å, and c ϭ 103.7 Å). After data collection the osmium-soaked I222 crystal was recovered from the capillary and soaked again in a solution of 2 mM K 2 PtCl 4 for 3 h, after which the derivative data were collected from a different portion of the same crystal that had not been exposed to the x-ray beam. The resulting platinum derivative was isomorphous with the osmium derivative, hence the osmium data were treated as a native data set for subsequent structure determination (Table I).
Structure Determination and Refinement-The structure of the I222 form was solved by a combination of molecular replacement and single isomorphous replacement methods. The molecular replacement procedure was carried out using the AMORE program (39) from the CCP4 suite (40) and the structure of ubiquitin (PDB code, 1ubq) as a search model (41). The initial rotation/translation results gave an R value of 53.2%, and the subsequent rigid body refinement lowered the R value to 50.9% for the data between 20 and 3 Å of resolution. The position of the platinum was obtained by the difference Patterson method. The phases obtained from the single isomorphous platinum derivative and the molecular replacement method were combined using the SIGMAA method (42). At this stage the side chains of the two ubiquitin molecules were replaced with those of the domains of ISG15, and the resulting model was further refined using the CNS program package (43) with alternating rounds of positional refinement and simulated annealing followed by manual fitting and rebuilding on a Silicon Graphics work station using the TURBO-FRODO graphics software (44). Seven percent of the data were withheld from the refinement process and used to calculate R free for cross-validation. After each round of CNS refinement, both 3͉F o ͉ Ϫ 2͉F c ͉ and ͉F o ͉ Ϫ ͉F c ͉ difference Fourier maps were calculated and examined for manual fitting of the model using TURBO-FRODO. The final model of the I222 structure revealed an osmium atom bound to the protein with a low occupancy (estimated occupancy of ϳ0.3). To confirm that the osmium soaking did not result in any conformational changes, the structure of the P2 1 2 1 2 1 crystal form was solved by the molecular replacement methods using the I222 structure as the starting model. The resulting model was further refined using CNS and TURBO-FRODO. Ramachandran plot analyses indicate that all residues lie in either most-favored or additionally allowed regions for the structures of both crystal forms. The data collection and the final refinement statistics are given in Table I.

RESULTS
Stabilization of Full-length Recombinant ISG15-We have previously reported that expression of recombinant Class 1 ubiquitin-like proteins is accompanied by rapid proteolytic inactivation through the action of a periplasmic carboxypeptidase that is largely refractory to known proteolytic inhibitors (19). In the present work we have exploited the inability of this bacterial protease to excise C-terminal arginyl residues by expressing recombinant ISG15 bearing a protective Arg 158 cap. The efficacy of this maneuver is demonstrated by the quantitative recovery in good yield of an intact ISG15 polypeptide of the predicted molecular mass by MALDI-TOF mass spectrometry (not shown). We also noted early in these studies that purified ISG15 was markedly sensitive to slow precipitation over time or upon freezing, particularly at concentrations above ϳ0.2 mg/ml. Non-reducing SDS-PAGE revealed that the precipitated form of the protein comprised a disulfide-linked dimer of ISG15 (not shown). Inclusion of DTT (10 mM) had little effect in preventing spontaneous precipitation of the ISG15 protein; however, mutation of the single cysteine residue at position 78 abrogated dimerization and precipitation. Subsequent urea denaturation studies have corroborated the enhanced stability of ISG15C78S compared with wild type protein. Fig. 1A compares the averaged CD spectra for ubiquitin, wild-type ISG15, and ISG15C78S at pH 7.5 and 20°C. Averaged spectra for the three polypeptides were qualitatively similar. The spectrum for human proISG15 was indistinguishable from that of the mature polypeptide (not shown), indicating that the C-terminal octapeptide extension has negligible effect on the structure of the protein. Secondary structure composi-tions for the proteins were calculated from the averaged spectra using the SELCON algorithm, which is based on the selfconsistent method of Sreerama and Woody (33,34), the results of which are summarized in Table II. The algorithm reasonably well predicts secondary structure content for ubiquitin com-pared with that determined from the published high-resolution crystal structure of the polypeptide (Table II). Importantly, the calculated secondary structure composition for wild-type ISG15 is comparable to that calculated for ubiquitin, suggesting that the two proteins possess similar folds. In addition, convergence of the spectral fit to unique solutions for the recombinant ISG15 forms requires that the protein molecules possess a uniform structure (33,34), precluding a significant fraction of denatured polypeptide in the preparations. The nearly identical fits for secondary structure compositions for ISG15, ISG15C78S, and proISG15 indicate that neither the point mutation at Cys 78 nor the C-terminal peptide extension of the precursor, respectively, have significant effects on overall structure. Inclusion of the eight additional unordered residues of the C-terminal propeptide extension accounts for the proportional apparent decrease in secondary structure composition for proISG15 compared with either wild-type ISG15 or ISG15C78S (17).
The ellipticity at 222 nm was measured as a function of urea concentration, and the fraction denatured (F D ) was calculated assuming a two-state model (35), as described under "Materials and Methods." All of the recombinant ISG15 proteins denatured in the range of 3-6 M urea (not shown). By comparison, the averaged CD spectrum for ubiquitin under the same conditions showed no change even at 9 M urea (not shown), as found earlier (45). Fig. 1B illustrates that values of F D for wild-type ISG15 (open circles and solid line) and ISG15C78S

Structure of ISG15 Ubiquitin-like Protein
(open squares and dashed line) follow the predicted sigmoidal dependence for a two-state model. Reversibility between the native and denatured states, a requirement of the two-state model (35), was tested by quantitatively denaturing the proteins in 6 M urea then allowing the proteins to refold by diluting the samples to an intermediate urea concentration near the respective inflection points of the unfolding curves. The values for F D observed after dilution matched the predicted values from the nonlinear regression analyses, consistent with the reversibility of the structural transitions (Fig. 1B, solid symbols). The apparent free energies of denaturation, ⌬G app , calculated from F D at each urea concentration followed a linear dependence on denaturant concentration from which the slope, m, and the extrapolated y intercept corresponding to the free energy of unfolding, ⌬G H 2 O , were calculated (35) (Fig. 1C). The urea denaturation studies indicated that blocking disulfide-linked dimerization of ISG15 by mutating the single Cys 78 residue significantly stabilized the protein. Because Cys 78 is absolutely conserved among ISG15 orthologs, we were concerned that the covalent dimer might represent the biologically active form of the polypeptide or that the point mutant might otherwise be inactive. Therefore we tested the ability of 125 I-ISG15N13Y/C78S to support the reaction of the ISG15-activating enzyme UbE1L (10). Mutation of Asn 13 to tyrosine was necessary to provide a surface-accessible tyrosine for iodination, because recombinant ISG15C78S was poorly radiolabeled by chloramine T. The autoradiogram of Fig. 1D demonstrates that 125 I-ISG15N13Y/C78S forms an ATP-dependent thiolester with UbE1L that is resolved from free radiolabeled polypeptide by non-reducing SDS-PAGE. That the adduct is a thiolester was demonstrated by loss of the band when resolved under reducing conditions (not shown). Therefore, monomeric ISG15 is biologically active in forming the obligate thiolester intermediate to its cognate activating enzyme.
Overall Structure of ISG15-Stabilization of ISG15 allowed the recombinant protein to be concentrated considerably without the denaturation and protein losses observed with wildtype polypeptide. This enabled us readily to obtain crystals of recombinant ISG15C78S suitable for structural determination. The final structural model contains the entire polypeptide chain from Trp 3 to Leu 154 with the first two N-terminal residues and the last four C-terminal residues being disordered ( Fig. 2A). The I222 structure and those of the two P2 1 2 1 2 1 structures (in the asymmetric unit) are virtually identical with r.m.s.d. values ranging from 0.4 Å between the two P2 1 2 1 2 1 structures to 0.7 Å between the I222 and P2 1 2 1 2 1 structures. Therefore, unless stated otherwise, references to native ISG15 refer to the I222 structure in the remainder of the text. As expected from the sequence similarity (4,13), the overall structure of ISG15 contains two easily discernable domains, each of which assumes a ␤-grasp fold that is nearly identical to that found in ubiquitin (41) (Fig. 2). Both domains of ISG15 contain a five-strand mixed ␤-sheet into which is intercalated a single three-turn ␣-helix (Fig. 2). Each domain of ISG15 also retains the two 3 10 helices characteristic of ubiquitin, one of which occurs between the ␣-helix and the ␤3 strand (Ala 40 -Gln 43 and Asp 119 -Leu 121 ) and the other of which occurs in a turn-rich region between the ␤4 and ␤5 strands (Pro 59 -Ser 62 and Gly 138 -Gly 141 ) (Fig. 2). The two ␤-grasp domains are connected by a six-residue extended linker peptide comprising Asp 76 -Pro 81 and encompassing (Cys/Ser) 78 . The two ␤-grasp domains of ISG15 are remarkably similar in structure, as predicted earlier by their intra-domain sequence conservation (4). The r.m.s.d. between the N-terminal domain (residues 4 -76, excluding 49 -50) and the C-terminal domain (residues 83-153) of ISG15 is 1.9 Å for the corresponding 70 C␣ atoms, and the r.m.s.d. values between ubiquitin and either the N-or C-terminal domains of ISG15 are 1.7 Å and 1.0 Å, respectively. Fig. 2B shows an overlay of the three structures, revealing their marked similarity in folding.
Relative Orientation of the Two Domains- Fig. 2A also shows the relative orientation between the two domains in the ISG15 structure. The N-terminal domain can be superimposed upon the C-terminal domain by rotating the former along the y-axis The last four C-terminal residues of ISG15 are disordered and not resolved, indicating the flexibility of the C-terminal tail. B, overlay of ribbon diagrams for ubiquitin (pink) with the amino-(blue) and Cterminal (green) domains of ISG15 to emphasize the marked similarities in their respective ␤-grasp folds. The graphics were generated using Molscript (62) and rendered by Raster3D (63). by 45°and by ϳ60°clockwise rotation along the z-axis (the axis perpendicular to the plane of the paper or monitor screen). The two domains are arranged so that the 3 10 element between the ␣-helix and the ␤3 strand of the N-terminal domain (Ala 40 -Gln 43 ) interacts with the 3 10 element in the turn-rich region between the ␤4 and ␤5 strand of the C-terminal domain (Gly 138 -Gly 141 ). The contact surface between the 3 10 helix segments of the two domains involves mainly van der Waals interactions between His 39 , Phe 41 , Pro 136 , and Gly 138 . There is also a weak hydrogen bond (3.4 Å) between the O␦ of Glu 139 and the main-chain amide of Phe 41 . The contact area between the two domains is 627 Å 2 , corresponding to 7% of the total solvent-accessible surface area. Because the structures of the I222 and P2 1 2 1 2 1 forms are virtually identical, it is reasonable to conclude that the observed crystal structure of ISG15, in particular, the relative orientation of the two domains is not due to crystal packing, but rather represents the most stable solution structure. However, it is entirely possible that, when ISG15 interacts with other proteins, the molecule might adopt a different relative orientation between the two domains by main-chain bond rotations within the linker peptide connecting them (residues Asp 76 -Lys 77 -Cys 78 /Ser 78 -Asp 79 -Glu 80 ). Notably, the linker peptide consists mainly of highly charged hydrophilic residues that might facilitate such a transition.
Surface Charge Distribution of ISG15-The distinct functional roles of the Class 1 ubiquitin-like proteins require their respective interacting proteins to recognize unique surface features of the polypeptides. Several distinguishing features of the ISG15 surface stand out in this structural comparison. The C-terminal ␤-grasp domain of ISG15 has a much less pronounced basic face (blue residues) than is present on ubiquitin. This difference is principally due to replacement of Arg 42 of ubiquitin with Trp 123 of ISG15, which disrupts an otherwise relatively continuous basic face, the remaining residues of which are conserved between the two proteins. In addition, ISG15 contains a slight bulge in the basic face, owing to a larger loop between ␤1 and ␤2 of the C-terminal domain compared with the orthologous region of ubiquitin, which harbors an additional basic residue at Lys 90 corresponding to Thr 9 of ubiquitin. The acidic residues of ISG15 are organized into a distinct ridge of negative charge along the molecule that comprise Asp 119 (Pro 38 ), Asp 120 (Asp 39 ), Asp 133 (Asp 52 ), Glu 132 (Glu 51 ), Glu 139 (Asp 58 ), and Glu 27 (Asn 25 ), with paralogous ubiquitin residues in parentheses. Most notably, the sequence in the N-terminal ␤-grasp domain encompassing the ␤3 through ␤5 segments comprise a large hydrophobic region, shown in white, that covers nearly half of the domain. The function of this large apolar region is unclear; however, the extended hydrophobic surface could account for the marked propensity of the purified protein to adsorbed non-  ubiquitin (B). The potential surfaces were generated using GRASP (64), with potentials ranging from Ϫ10kT/e (red) to ϩ10kT/e (blue). To the right of each electrostatic potential structure is shown the corresponding ribbon diagram for each molecule with color ramped from blue (N terminus) to red (C terminus). In both representations, the molecules are arranged so that the C-terminal domain of ISG15 is oriented identical to that of ubiquitin, with the putative respective E1-binding surfaces of each projecting toward the viewer. Relevant amino acids are labeled as indicated.
specifically to surfaces in relatively dilute solutions and to undergo dimerization prior to disulfide formation.
A Model For ISG15 Binding to UbE1L-Binding of the different ubiquitin-like proteins to their cognate activating enzymes represents the entry point for these polypeptides into their respective, mechanistically parallel pathways for ligation (46). The fidelity with which these recognition events occurs prevents deleterious cross-talk between pathways and competitive inhibition by the significantly higher steady-state concentrations of free ubiquitin found within cells (47,48), as discussed previously (13,19). To identify potential interacting residues that might be important for ISG15 recognition and binding, we modeled the polypeptide onto the recently published structure for human Nedd8 bound to heterodimeric human AppBp1-Uba3 (29). The C-terminal ubiquitin-like domain of ISG15 and Nedd8 share 26% identity over their C-terminal 76 residues and 51% side-chain conservation, as defined by the empirical substitution criteria of Bordo and Argos (49). In addition, the r.m.s.d. in C␣ positions between human Nedd8 and the C-terminal domain of human ISG15 is 0.84 Å over 67 residues (Fig. 4C) compared with 1.0 Å between human Nedd8 and human ubiquitin (not shown). Therefore, it was a relatively straightforward procedure to overlay the structure for the Cterminal domain of ISG15 onto that of Nedd8 bound to Ap-pBp1-Uba3 (29) to examine potential points of interaction for orthologous positions between Uba3 and Uba1 (Fig. 4, A and  B). The area of the interface between the docked molecules for the ISG15 complex is 2077 Å 2 and is very similar to the corresponding interface area found in the structure of the complex between Nedd8 and AppBp1-Uba3 (2214 Å 2 ). This docking model assumes that the global structural features in the overall interaction surfaces between ubiquitin-like proteins and their cognate activating enzymes are evolutionarily retained. This assumption appears to hold for both the structures for Nedd8 bound to AppBp1-Uba3 (29) and for Sumo bound to Sae1-Sae2 (50). The r.m.s.d. between the two known structures of the E1 paralogs, Uba3 (Nedd8-binding domain) and Sae2 (Sumo-binding domain), is 1.25 Å for 437 C␣ atoms, strongly suggesting that the structure of the corresponding domain of UbE1L would also be very similar to those of Uba3 and Sae2.
Within the docked structure, the C-terminal ubiquitin-like domain of ISG15 makes extensive contacts with the cleft defining the adenylate active site of AppBp1-Uba3, whereas the N-terminal ubiquitin-like domain of ISG15 projects well away from the surface of the Uba3 subunit of the activating enzyme (Fig. 4, A and B). The fit of the C-terminal domain of ISG15 into the active site cleft of AppBp1-Uba3 is remarkably good, with few regions of overlap between the solvent-accessible surfaces of the two structures, providing confidence in the general veracity of the docked model. The fit also benefits from the absence of major insertions or deletions in human UbE1L within the sequence segments defining the adenylate active site of Uba3 (49); in addition, human UbE1L is 16.9% identical and 52.4% similar to human Uba3 over residues 406 -1011 of human UbE1L (see Supplementary Data). Within these assumptions, the docked structure predicts seven unambiguous interactions between human ISG15 and human UbE1L, summarized in Table III. These putative interacting residues occur within regions of the Uba3 structural model that are reasonably well conserved among the four human paralogs (Fig. S1 of the Supplementary Data). Also listed in Table III are the paralogous interactions predicted between human ubiquitin and human Uba1a. Three of these interactions with UbE1L, involving Arg 153 , Glu 132 , and Arg 92 of ISG15, should be conserved between ubiquitin and Uba1a (Table III), based on sequence conservation between the interacting pairs. Four other interactions are predicted to be unique to ISG15 and, therefore, probably define the specificity of UbE1L for its cog- nate polypeptide. Two of the latter interactions stand out as potentially critical in defining the specificity of UbE1l and Uba1 for their cognate polypeptides. A predicted Coulombic interaction between Lys 90 of ISG15 and Glu 889 of UbE1L is not conserved in the paralogous residues of Thr 9 of ubiquitin and Leu 947 of Uba1a (Table III). We have previously shown that Arg 42 of ubiquitin represents an important residue for binding to its cognate activating enzyme (30); however, Trp 123 occupies the paralogous position within the C-terminal domain of ISG15. The docking model suggests that Trp 123 potentially engages in a hydrophobic ring stacking interaction with Phe 905 of UbE1L that is not conserved between the paralogous Arg 42 of ubiquitin and Tyr 947 of Uba1. DISCUSSION The present study provides the first structure for a tandem ␤-grasp domain protein, a specialized subclass of Class 1 ubiquitin-like polypeptides that includes the ISG15 and FAT10 families (4,13,51,52). In addition to their unique structure, predicted from sequence alignments (13,53), their more restricted phylogenetic distribution suggests roles in cell regulation distinct from their more widely distributed paralogs. Indeed, the appearance of ISG15 in evolution roughly approximates the emergence of the interferon response and immune surveillance.
The 2.4-Å structure of ISG15 reveals two classic ␤-grasp folds tightly packed in an end-on orientation ( Fig. 2A), the secondary structural elements of which are graphically superimposed on a sequence comparison of ISG15 and FAT 10 orthologs (Fig. 5). The C-terminal four residues of ISG15 are not resolved, reflecting their solvent exposure and resulting flexibility. Similar conformational freedom has been observed for the other Class 1 ubiquitin-like proteins for which structures are known (41, 54 -57). A detailed comparison of all bona fide ␤-grasp fold proteins by Laleli-Sahin (58) reveals a conserved spacing of six essential aliphatic residues that constitute the hydrophobic core of the protein and that defines the interface between the buried faces of the single intercalated ␣-helix and the ␤-sheet. These residues and their approximate spacings are conserved in each domain of ISG15 (Fig. 5, red dots). Not surprisingly, the resulting overall fold of each domain thus conforms closely to that observed for ubiquitin when the three structures are overlaid (Fig. 2B). The slightly better r.m.s.d. of 1.0 Å for the C-terminal domain of ISG15 relative to ubiquitin compared with the N-terminal domain (1.7 Å) is consistent with the greater sequence similarity noted earlier for the Cterminal domain. More stringent selective pressure for sequence and structural conservation within the C-terminal domain of ISG15 probably results from requirements imposed by binding of this domain to its cognate UbE1L activating enzyme (10) and ISG15-specific proteases (27). The two 3 10 helices of ubiquitin (41) are conserved in each domain of ISG15, two of which contribute to the packing interface between the N-terminal and C-terminal domains (Fig. 2). The two ␤-grasp domains of ISG15 are connected by a random coil that extends along the surface of the molecule, Fig. 2. Because the paralogous region of ubiquitin represents an important recognition motif for ubiquitin-specific proteases that process pro-and isopeptide-linked forms of ubiquitin and other ubiquitin-like proteins (17,59,60), sequence divergence within this surfaceexposed segment from the canonical C-terminal LRLRGG of ubiquitin must be necessary to preclude cleavage of the domains, as noted earlier (4).
The exposed linker region harbors the single cysteine residue located at position 78 (human ISG15 numbering). Our early experiments using non-reducing SDS-PAGE indicated that concentrated solutions of wild-type mature recombinant ISG15 readily formed a disulfide-linked dimer at neutral pH that subsequently denatured and precipitated from solution (not shown). Similar behavior had been noted previously with ISG15 isolated from natural sources (28,61). Reversible urea denaturation studies allowed us to graphically measure the free energy of unfolding, ⌬G H 2 O , quantitatively illustrating that the wild-type mature ISG15 polypeptide is markedly unstable at 20°C (Fig. 1). Mutation of Cys 78 to serine, to maintain the overall polarity at this position, obviated disulfide formation and prevented precipitation of the protein upon concentration. Urea denaturation studies demonstrated that the mutation substantially stabilized the polypeptide as well (⌬⌬G H 2 O ϭ Ϫ3.3 kcal/mol) (Fig. 1). The enhanced stability of this point mutant probably represents a lower limit to the actual effect of the Cys 78 mutation, because the inherent insolubility of the oxidized dimer precluded our generating ISG15 quantitatively in this form for analysis in the urea unfolding experiments illustrated in Fig. 1. Instead, we relied on the spontaneous oxidation of wild type polypeptide in these unfolding experiments. The reversibility of the denaturation at low concentrations of DTT and the stabilization of the protein that accompanies mutation of Cys 78 suggests a mechanism in which ISG15 first reversibly dimerizes prior to oxidation of Cys 78 to yield a metastable disulfide-linked intermediate that is subject to irreversible denaturation in the absence of urea. The instability undoubtedly accounts for precipitation and loss of activity that accompanies concentrating wild-type ISG15 solutions noted earlier (13,28,61). Potentially, dimerization could occur through association of the hydrophobic region within the Nterminal domain that was noted in the charge density map of ISG15 (Fig. 3).
Neither mutation of Cys 78 to serine to enhance the stability of the polypeptide nor the subsequent mutation of Asn 13 to tyrosine to provide a surface-accessible site for radioiodination qualitatively affects the function of the polypeptide, because 125 I-ISG15N13Y/C78S undergoes ATP-dependent activation and thiolester formation with human recombinant UbE1L, the ISG15-activating enzyme (Fig. 1D) (10). Most important, the ability to protect the C terminus of mature ISG15 from proteolytic inactivation during expression and purification by employing an arginine cap facilitates future detailed mechanistic and functional studies of this ligation pathway.
Earlier sequence analysis accurately predicted the tandem ␤-grasp folds of ISG15; however, such analyses could not anticipate the potential packing interactions, if any, between the domains. The structure of human ISG15 reveals that the two domains interact largely through hydrophobic interactions between the first and fourth 3 10 helices (3 10 -1 and 3 10 -4, respectively) (Fig. 5). The interaction surface buries 7% (627 Å 2 ) of the total solvent-accessible surface area (8957 Å 2 ) and represents a considerable stabilizing force for maintaining the overall struc- The predicted interactions are based on the "docked" structure of ISG15 with human UbE1L generated by overlaying the ISG15 structure onto that of Nedd8 in the crystal structure of the complex between Nedd8 and human AppBp1-Uba3 (29) ture depicted in Fig. 2. This suggests that the structure shown in Fig. 2 approximates the stable solution conformation of the polypeptide. This conclusion is consistent with sequence comparison among the known ISG15 orthologs and ubiquitin that show the first and fourth 3 10 helices are significantly more conserved among ISG15 orthologs than the second and third 3 10 helices that are solvent-exposed in the structure of Fig. 2 and do not interact (Fig. 5). In addition, the Glu 139 residue that participates in a side-chain hydrogen bond with the main-chain amide of Phe 41 is also conserved among ISG15 orthologs (Fig. 5). The FAT10 polypeptide is the only other tandem domain Class 1 ubiquitin-like protein identified (52,53). The sequences for the FAT10 family of ubiquitin-like proteins have areas of sequence similarity that are distinct from the ISG15 paralogs, particularly in the critical C-terminal sequences, suggesting that they serve different functions (Fig. 5). Induction of FAT10 by Type 2 interferon ␥, whereas ISG15 is induced principally by Type 1 interferon ␣/␤, supports a role for these proteins in different cellular processes (4,52). The pattern of sequence conservation within the two putative ␤-grasp domains of FAT10, particularly among the key aliphatic residues (red dots), is consistent with the protein assuming a ubiquitin-like fold (Fig. 5). Conservation among the FAT10 sequence segments corresponding to the first and fourth 3 10 helices of ISG15 compared with the second and third 3 10 helices are consistent with FAT10 orthologs assuming the same overall domain packing as ISG15 (Fig. 5). We also note that the Glu 139 residue that forms a hydrogen bond with the amide of Phe 41 at the domain interface of ISG15 is conserved as Asp 147 among the FAT10 orthologs (Fig. 5), consistent with this conclusion.
Finally, the overall conservation in fold between Nedd8 and the C-terminal domain of ISG15 allowed us, with some confidence, to model binding of ISG15 within the adenylate active site of the UbE1L-activating enzyme (Fig. 4). The model suggests that the C-terminal domain of ISG15 makes extensive contacts with UbE1L, whereas the amphipathic N-terminal domain (Fig. 3) remains solvent-exposed and not in contact with the enzyme (Fig. 4). This requires that conservation among ISG15 orthologs within the N-terminal domain serves functions distinct from those required for C-terminal activa-tion. The docking model allowed us to identify potential residues critical for binding to and discrimination of ISG15 by UbE1L to the exclusion of other Type 1 ubiquitin-like paralogs, summarized in Table III. Several of these predicted ISG15 binding residues are conserved in ubiquitin (in parentheses), including Arg 153 (Arg 72 ), Glu 132 (Glu 51 ), and Arg 92 (Lys 11 ) (Table III). Notably, Arg 72 of ubiquitin serves as a major specificity determinant that allows the ubiquitin (30,32), Nedd8 (29,55), and Sumo (50) activating enzymes to distinguish cognate from non-cognate polypeptides. Conservation at the Arg 153 (Arg 72 ) residue of ISG15 and ubiquitin, respectively, precludes a similar role in defining the specificity of UbE1L for ISG15. Instead, Lys 90 , Trp 123 , and Phe 149 of ISG15 appear to provide unique interaction "hot spots" that may allow UbE1L to recognize its cognate substrate to the exclusion of ubiquitin and other paralogs (Table III). We are currently using kinetic approaches with point mutants of ISG15 and UbE1L quantitatively to assess the role of these putative interacting residues in defining the substrate specificity of UbE1L.