Two Families of GTPases Dominate the Complex Cellular Response to IFN-γ¹

Interferon-γ is an immunomodulatory cytokine secreted mainly by activated thymus-derived (Th1 and TC₁) cells and NK cells. IFN-γ orchestrates a variety of cellular programs in different target cells (reviewed in Refs. 1 and 2). Important IFN-γ-stimulated cellular programs are the quantitative and qualitative enhancement of Ag presentation by MHC class I and class II pathways, nitric oxide and respiratory bursts involved in direct antimicrobial activation of macrophages, and translational inhibition directed against viral replication. The cellular response to IFN-γ is extraordinarily complex. Recent estimates suggest that >500 genes are inducible (3, 4), of which ∼300 have been identified (Ref. 5 and T. Klamp, unpublished observations). Thus, as many as 200 genes active in the cellular response to IFN-γ remain to be described.

Partially nonidentical programs are triggered by IFN-γ in different cell types. In macrophages, IFN-γ stimulates numerous pathways which reflect the specialized role of this cell type in early innate immunity (e.g., inducible nitric oxide synthase, IL-12, phox, complement components, FcγR, etc.), whereas in fibroblasts a more limited range of functions mainly focused on Ag presentation and direct antiviral action is induced. In the present study, two independent analyses of the complexity of IFN-γ-regulated gene induction in macrophages and fibroblasts have been conducted. To this end, the suppression-subtractive hybridization (SSH)³ technique (6) was used to identify cytokine-inducible genes in the mouse macrophage cell line ANA-1 and in mouse primary embryonic fibroblasts (MEF). The most abundant classes of differential cDNAs recovered from both cell types were members of two cytokine-inducible guanylate-binding protein (GBP) families, the 65-kDa and 47-kDa families. In the mouse, the 65-kDa GBP family consists of two previously known genes cloned and described full-length, mGBP1/mag-1 (7, 8) and mag-2 (8). Previously known members of the 47-kDa family are the proteins encoded by the IRG-47 (9), TGTP (10)/Mg21 (11), LRG-47 (12), and IGTP (13) cDNAs.

New members of each family were identified in both cell types and are cloned and characterized in this study. We show that all members of the two families are strongly induced by IFN-γ in the livers of mice infected with the facultative intracellular bacterium Listeria monocytogenes. Furthermore, we present additional evidence that a significant portion of the IFN-γ response in fibroblasts as well as in macrophages remains to be described, since at least 15 fragments of putative new IFN-γ-inducible genes were recovered from our screens.

C57BL/6J mice were obtained from the animal house at the Institute for Genetics, University of Cologne, Cologne, Germany. Mice deficient for IRF-1 (IRF-1^−/−), or the IFN-γ receptor (IFN-γR^0/0) have been described elsewhere (14, 15) and were kindly provided by M. Aguet and C. Weissmann. Mice lacking TNFRp55 (TNFRp55^−/−) were described by Pfeffer et al. (16). All mice were maintained in a conventional animal facility.

C57BL/6J, IRF-1^−/−, IFN-γR^0/0, and TNFRp55^−/− mice were infected i.p. with one-tenth of the LD₅₀ of Listeria monocytogenes (strain EGD). Twenty-four hours after infection, the mice were killed, and livers were removed and snap-frozen in liquid N₂.

Primary embryonic fibroblasts were isolated from C57BL/6J, IRF-1^−/−, and IFN-γR^0/0 mice at day 14 post coitum as described (17) and were grown in DMEM (Life Technologies, Gaithersburg, MD) supplemented with 10% heat-inactivated, low endotoxin FCS (HyClone, Logan, UT), 2 mM l-glutamine (Life Technologies), 1 mM sodium pyruvate (ICN, Aurora, OH), 100 U/ml penicillin (Life Technologies), 100 μg/ml streptomycin (Life Technologies), and 1× nonessential amino acids (Life Technologies).

Given that macrophage lines differ in their differentiation stages and responsiveness to IFN-γ, the inducibility of the well-characterized IFN-γ-induced inducible nitric oxide synthase mRNA was compared by RT-PCR in various macrophage cell lines. The C57BL/6J-derived macrophage cell line ANA-1 (18) was the strongest responder among those tested (data not shown). ANA-1 was grown in low endotoxin RPMI (Biochrom, Berlin, Germany), supplemented with 10% heat-inactivated, low endotoxin FCS (HyClone) and 50 μM 2-ME (Life Technologies).

For the preparation of SSH cDNA libraries (see below), MEFs were cultured as indicated for 3 or 24 h in medium containing 1000 U/ml recombinant mouse IFN-γ (Genzyme Diagnostics, Cambridge, MA). ANA-1 cells were stimulated with 2000 U/ml recombinant human TNF-α (Genzyme) and 55 U/ml recombinant mouse IFN-γ (Genzyme) for 16 h. For Northern blot analysis, both MEFs and ANA-1 were stimulated with 1000 U/ml IFN-γ for 24 h.

Total RNA was extracted from cell lines and tissues using the RNeasy mini kit (Qiagen, Hilden, Germany), the DNA/RNA isolation kit (Amersham, Little Chalfont, U.K.) or the acid guanidinium thiocyanate-phenol-chloroform extraction method (19). Poly(A)⁺ RNA was isolated from total RNA with the Oligotex mRNA kit (Qiagen) for the generation of the PCR-Select libraries and the poly(A)⁺ RNA isolation kit (Amersham) for the generation of the λ-cDNA library according to the manufacturers’ instructions. The amount of RNA was quantified spectrophotometrically.

Ten micrograms of total RNA were electrophoresed in 1% denaturing agarose-formaldehyde gels (20) and then transferred onto Hybond-N nylon membranes (Amersham Life Sciences) according to standard protocols. Equal loading, transfer, and quality of the RNA was examined by staining the membranes with methylene blue (21, 22). [α-³²P]dCTP (Amersham)-labeled probes were generated by the random priming method (23) using the Rediprime DNA labeling system (Amersham Life Sciences). Hybridizations were performed overnight at 42°C in a buffer containing 50% formamide, 5× Denhardt’s solution, 5× standard saline-phosphate-EDTA (SSPE), 1% SDS, and 10% dextran sulfate. Membranes were washed under stringent conditions. The hybridization signal was detected by autoradiography, using Kodak X-OMAT AR films. Labeled cDNA of mouse glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used as a control probe to reveal the amount of loaded total RNA, and the RNA Millenium Marker (Ambion) was used as the RNA size standard.

mRNA prepared from C57BL/6J embryonic fibroblasts stimulated for 24 h with 1000 U/ml recombinant IFN-γ was reverse-transcribed with Moloney murine leukemia virus reverse transcriptase (Boehringer Mannheim) in the presence of an oligo(dT)-anchor primer (Boehringer Mannheim). After degradation of the mRNA template with RNase H (Boehringer Mannheim), this library was used to amplify mGBP-2 with primers mGBP-2 5′ (5′-GAAAAGCTGCTTCTTCTTCTTCTCT-3′) and mGBP-2 3′ (5′-TCAAGACATGTTGTCACAGTGG-3′) (nucleotides 100–124 and 2405–2384, respectively) and mGBP-4 with primers MM44 5′ (5′-GGAAGCTCAGGGAAAGAACTAA-3′) and MM44 3′ (5′-CATACATTCTCTTATGAAAACCCAC-3′) (nucleotides 23–44 and 2443–2419, respectively).

SSH to generate subtracted cDNA libraries was performed essentially as described (6), using reagents and procedures provided by Clontech (Palo Alto, CA; PCR-Select cDNA Subtraction Kit). Briefly, double-stranded cDNA (tester cDNA) was synthesized from 2 μg of mRNA isolated from cytokine-stimulated MEF and ANA-1 cells (see above). Driver cDNA was prepared from unstimulated MEF and ANA-1 cells. The differential PCR products obtained after two rounds of subtraction were cloned without further purification into pGEM-T Easy (Promega, Madison, WI) or pCR 2.1 (Invitrogen, San Diego, CA) vectors. The SSH protocol generates clonal fragments of differential cDNAs bounded at both ends by RsaI restriction sites.

C57BL/6J embryonic fibroblasts were stimulated with 1000 U/ml IFN-γ for 24 h. After mRNA isolation (U.S. Biochemical, Cleveland, OH), cDNA was synthesized from 5 μg of mRNA (ZAP-cDNA synthesis protocol; Stratagene, Heidelberg, Germany), and the libraries were constructed using the Uni-Zap XR vector kit from Stratagene according to the manufacturer’s instructions with the following alteration. Before ligation of the cDNA into the Uni-Zap XR vector arms, the cDNA was electrophoresed in a 1% agarose gel and three size fractions (0.4–1.4 kb, 1.4–4 kb, and >4 kb) were eluted according to standard procedures (22) and ligated into the vector arms to generate three independent libraries. For the screening of the new members of the two GBP-families, the medium size library was used (size range, 1.4–4 kb).

The library was screened with [α-³²P]dCTP-labeled probes (Rediprime; Amersham). Filters were hybridized overnight at 65°C in HYB-9 DNA hybridization solution (Biozym, Gwent, U.K.). After rescreening, cDNA inserts were isolated by in vivo excision of the pBluescript phagemid from the Uni-Zap XR vector, according to the manufacturer’s instructions.

All solutions for molecular biology were made with ultrapure water (0.055 μS/cm) derived from a combined reverse osmosis/ultrapure water system (β 75/Δ UV/UF; USF Seral Reinstwassersysteme Gmbh, Ransbach-Baumbach, Germany) equipped with UV (185/254 nm) and ultrafiltration (5000-kDa cutoff).

Sequencing was performed in our sequencing facility on an ABI sequencer (Applied Biosystems). Sequences were aligned manually using SeqPup version 0.6 by Don Gilbert (available at ftp://iubio.bio.indiana.edu/molbio/seqpup/). Dot plot analyses and hydrophobicity plots were performed using the Wisconsin Package version 9.1, Genetics Computer Group, Madison, WI. Phylogenetic trees were constructed using PHYLIP version 3.572c (24). The EMBL Data Library accession numbers are mGBP-1/mag-1 (M55544), mGBP-2 (AJ007970), mag-2 (M81128), mGBP-4 (MM44731), hGBP-1 (M55542), hGBP-2 (M55543), rat GBP (M80367), chicken GBP (X92112), pig GBP (F14838), IGTP (U53219), LRG-47 (U19119), IRG-47 (M63630), Mg21/TGTP (L38444), GTPI (AJ007972), IIGP (AJ007971).

Differential libraries of cytokine-induced MEFs and ANA-1 macrophages were generated using the SSH protocol as described (Materials and Methods). MEFs were stimulated with 1000 U of IFN-γ per ml of medium for 3 and 24 h to generate SSH libraries representing an early and a late time point in the IFN-γ response. A total of 182 clones from the 2 MEF libraries were sequenced. Cells of the C57BL/6J macrophage cell line ANA-1 (18) were stimulated with a mixture of 55 U of IFN-γ and 2000 U of TNF-α per ml of medium for 16 h, and a third SSH library was generated from which 70 clones were sequenced.

The identities of cloned cDNA fragments recovered from MEF and ANA-1 macrophage-differential libraries are summarized in Table I. Fragments belonging to the 65-kDa and 47-kDa GBP families were strikingly overrepresented in both libraries (Table IB), with the new 65-kDa family mGBP-2 sequence (see below) with 36 fragments recovered being by far the most abundantly represented single gene. Of the other known IFN-γ-inducible genes recovered from the SSH libraries, the transcription factor IRF-1 was represented five times in total and the chemokine IP-10 nine times. Fragments of new members of the 47-kDa family, IIGP and GTPI (see below), were recovered from both MEF libraries. The two GBP families were abundant in both MEF and ANA-1 libraries. There was a possibly significant bias in the MEF libraries toward a higher abundance of GBP family members relative to other induced sequences.

Besides the new members of the 47-kDa and 65-kDa GBP families, cDNA fragments representing an unknown number of further new cytokine-inducible genes were found in the MEF and ANA-1 SSH libraries. A nonoverlapping set of 10 MEF and 5 ANA-1 fragments not thus far reported in the public databases were recovered (Table IA), all of which were strongly induced by IFN-γ in the respective cell type over low or undetectable resting levels (data not shown). Ongoing experiments will determine whether the yield of MEF and ANA-1-specific clones is a statistical accident or due to differential regulation.

The abundant cDNA named mGBP-2 in Table IB was contained in fragments spanning a full-length sequence similar to but distinct from the published mGBP-1/mag-1 (7, 8). Since C57BL/6J mice do not express mGBP-1 (25), it was possible that these fragments represented a homologous gene of which the promoter and 5′-end of the first exon has been described from BALB/c mice and named mGBP-2 (26), but of which no complete sequence has yet been described. We therefore performed PCR on mRNA from IFN-γ-stimulated MEFs using primers based on our fragment sequences representing the putative 5′- and 3′-ends of mGBP-2 (see Materials and Methods). The single 2.5-kb product yields an apparently full-length coding sequence confirmed in full-length clones from a MEF λ-cDNA library (data not shown). The 5′-end of the new sequence is identical with the published short segment of mGBP-2 (26); this new cDNA thus represents the first full-length mGBP-2 sequence. The high yield of mGBP-2 fragments from the subtractive libraries is consistent with the high frequency of mGBP-2 in the 1.4–4 kb size-selected primary MEF λ-cDNA library, in which mGBP-2 clones were represented at a frequency of 1.8%. The open reading frame (ORF) is 1767 nucleotides long, encoding a putative protein of 589 amino acids with a predicted molecular mass of 66.7 kDa and an isoelectric point of 5.52. Fig. 1 shows the new mGBP-2 sequence aligned with family members mGBP-1/mag-1 and mag-2 and a sequence (MM44731; deposited by B. H. Han, D. J. Park, R. W. Lim, and H. D. Kim) of another member of the 65-kDa family for which we propose the name mGBP-3. This sequence, extracted from a BLAST search of the GenBank database, arises from a cDNA isolated from erythroid progenitor cells infected with Friend virus. It contains an ORF of 1863 nucleotides and codes for a putative protein of 620 residues with a predicted molecular mass of 70.8 kDa and an isoelectric point of 6.6. No cDNA fragments representing mGBP-3 were recovered from MEF or ANA-1 SSH libraries (Table I), but a 2.4-kb product identical with the database sequence was identified from RT-PCR on mRNA isolated from MEFs stimulated with IFN-γ (see Materials and Methods). All members of the 65-kDa GBP family possess the canonical G(X₄)GKS and D(X₂)G motifs conserved in most GBPs (27), while all lack the (N/T)(K/Q)XD motif, as previously noted for mGBP-1 (7). A stretch of six amino acids (VVVAIV) immediately preceding the G(X₄)GKS motif is perfectly conserved in all members of the family including all known 65-kDa homologues from human, rat, pig, and chicken (7, 28, 29, 30). The sequences of the 65-kDa GBP family are further analyzed below (Table II, Figs. 6 and 7).

Fragments representing all known members of the 47-kDa GBP family (LRG-47, IRG-47, Mg21/TGTP, and IGTP) were recovered from the SSH libraries. Two adjacent segments of an apparently new 47-kDa family member (labeled IIGP in Table IB) were represented by 5 SSH fragments. Multiple apparently full-length clones containing both sequence segments were recovered from the λ-cDNA library following screening with one SSH fragment. The ORF is 1239 nucleotides in length, generating a protein of 413 residues, predicted molecular mass of 47.5 kDa, and isoelectric point of 6.23. Three adjacent segments of a second apparently new 47-kDa member (labeled GTPI in Table IB) were represented by 4 fragments. GTPI λ-cDNA library clones obtained by screening with one SSH fragment contain a 1185-nucleotide ORF, generating a putative protein of 395 residues with a predicted molecular mass of 45.2 kDa and an isoelectric point of 6.95. The IIGP and GTPI putative amino acid sequences are aligned to known members of the 47-kDa GBP family in Fig. 2. The sequence of IIGP is closer to Mg21/TGTP and IRG-47, whereas the sequence of GTPI is closer to LRG-47 and IGTP (Table III). All canonical GBP motifs are present in all members of the 47-kDa family, although LRG-47, IGTP, and GTPI have an unusual methionine in the first motif, turning the canonical GKS sequence into GMS.

Members of the 65-kDa and 47-kDa GBP families were shown by Northern blot analysis of MEF mRNA to be massively induced from undetectable constitutive levels by IFN-γ alone after 24 h of stimulation (Fig. 3, A and B, tracks 1 and 2). No induction could be detected in MEFs from IFN-γR^0/0-mice (Fig. 3, A and B, track 3). Of the 47-kDa members, all except GTPI were also strongly induced in IFN-γ-stimulated ANA-1 macrophages. Of the 65-kDa members, we were unable to detect induction of mag-2 in ANA-1 even though the original cDNA was cloned from IFN-γ-induced RAW 264.7, another macrophage cell line (8). Furthermore, the mag-2 Northern blot shows two bands at 2.8 and 4.1 kb, as originally described by Wynn et al. (8), while the mag-2 sequence deposited in GenBank is 3.2 kb long. These discrepancies remain to be resolved. The inducibility of mGBP-3 is still under investigation.

The 65-kDa and 47-kDa GBP families differed strikingly in their inducibility by IFN-γ in MEFs lacking a functional IRF-1 gene. No detectable induction of 65-kDa family member mRNA was found after 24 h of IFN-γ induction in MEFs from IRF-1^−/− mice (Fig. 3,A, track 4). All 47-kDa mRNAs were strongly induced in IRF-1^−/− MEFs, although IGTP and IRG-47 showed some dependence on IRF-1 (Fig. 3 B, track 4).

Twenty-four hours after i.p. infection of normal C57BL/6J mice with Listeria, mRNAs of mGBP-2, mag-2, and all 47-kDa GBPs were strongly induced in the liver (Fig. 4, A and B, tracks 1and 2). In IFN-γR^0/0 mice, no induction of any transcript was detectable (Fig. 4, A and B, tracks 3 and 4), showing the absolute dependence of this induction on IFN-γ. All GBP genes were also strongly induced in Listeria-infected TNFRp55^−/− mice, suggesting that induced TNF-α does not markedly contribute to the induction of these genes in vivo via the TNF-α p55 receptor, which mediates the TNF-α-dependent protective response to Listeria (31). Additional experiments are needed to reveal whether the small reduction in signal observed for mGBP-2, mag-2, and possibly LRG-47 and IIGP in the TNFRp55^−/− track is a consistent and significant effect.

Nucleotide and amino acid sequence comparisons (Tables II and III) show the 47-kDa family to be more divergent among its members and therefore either older or more rapidly evolving than the 65-kDa family. This observation is consistent with the lack of known members of this family recovered thus far from human or indeed any other material. A more extended analysis would be required to distinguish between these two possibilities. Both families can also be seen from Tables II and III to break down into two subfamilies, and this is confirmed by the unrooted trees of protein sequences shown in Fig. 5. Thus, mGBP-1 and -2 cluster closely together near human GBP-1 and 2, and distant from the mag-2/mGBP-3 cluster among the 65-kDa proteins, while IGTP, GTPI, and LRG-47 cluster together and distant from IIGP/IRG-47/Mg21/TGTP among the 47-kDa proteins. Dot plot comparisons between and within subfamily members (Fig. 6, A and B) demonstrate the complete colinearity of the 65-kDa family at both high and low stringency: the only characteristic distinction between the subfamilies is at the extreme C terminus. Structural divergence between the subfamilies in the C-terminal half of the 65-kDa proteins, however, is clearly apparent from the hydrophobicity plots (Fig. 7). Among the 47-kDa GTPases, sequence divergence at the C terminus is more marked than elsewhere both within and between subfamilies. The hydrophobicity plots show distinct structural features at individual, subfamily, and family levels. The dot plot shown in Fig. 6 C compares a 65-kDa (mGBP-2) and a 47-kDa (IIGP) family member, documenting the absence of any sequence relatedness between the two families.

The initial motivation for the experiments described in this paper was to use modern differential screening techniques to analyze the full complexity of the cellular responses to IFN-γ in primary MEF and to a synergistic mixture of IFN-γ and TNF-α in a mouse macrophage cell line, ANA-1. The SSH technique of Diatchenko et al. (6) was used in preference to differential display (32), firstly because of the built in focus on the recovery of cDNAs representing mRNAs with large ratios of induced to noninduced levels and secondly because of the relative ease of processing the differentially expressed material.

From the analysis of a total of 182 MEF clones and 70 ANA-1 clones, our results (Table I) show the striking and unexpectedly high yield of GTPases of the 65-kDa and 47-kDa families in both cell types. Altogether nine different GTPases of these two families were identified, including new members of each family, and members of these families were individually the most abundant cDNA fragments recovered. Subtractive hybridization differential cloning, in which high molar ratios of driver (i.e., noninduced in our case) to tester (i.e., induced in our case) cDNA are used to remove sequences found in both pools, will reduce, if not eliminate, sequences that may be differential but have nevertheless a significant constitutive level of expression. The results of our analysis are consistent with this expectation. Thus no component of the classical class I Ag presentation pathway (including the structural proteins H-2Kb, Db, β₂-microglobulin; the endoplasmic reticulum components TAP1, TAP2, gp96, and tapasin; and the proteasome components LMP2, LMP7, PA28α, and PA28β) was recovered from either cell type. Although all these components are inducible by IFN-γ, the pathway is constitutively active at a certain level in both cell types. The recovery of GTPase fragments in such abundance thus reflects the high abundance of these messages in the induced mRNA pool and their remarkably low level of expression in noninduced cells (Fig. 3). In confirmation of this view, human GBP-1 was identified in early experiments in human fibroblasts as one of the most abundantly induced proteins, with about 3 × 10⁵ protein molecules per cell 24 h after IFN-γ stimulation (33, 34), while, as noted above, the newly identified homologue mGBP-2 represented 1.8% of a primary cDNA library constructed from IFN-γ-stimulated MEFs.

In the two cell types studied, therefore, these GTPases are extreme specialists of the IFN-γ-induced state in the sense that they are virtually absent in resting cells and highly abundant after induction. That the same is also true for liver in vivo is suggested by the Northern blots shown in Fig. 4, where the low signal apparent in some tracks from noninfected mice is consistent with a low level of induction by endogenous cytokine production induced by background pathogens in these conventionally reared animals. The 47-kDa GTPase IGTP has been reported to show high constitutive levels of mRNA in tissues rich in lymphoid cells including the thymus and small intestine (13). Whether this expression can be generalized for other IFN-inducible GTPases and whether it reflects constitutive transcription rather than transcription induced by local production of IFN-γ should be determined by further analysis of expression of these gene families in tissues of IFN-γ- or IFN-γ receptor-deficient mice.

As noted in Results, cDNA fragments representing members of both the 47-kDa and 65-kDa GTPase families were recovered from both the 3-h and 24-h SSH libraries, showing that mRNAs from both families are significantly elevated both relatively early and relatively late in the cellular response to IFN-γ. Nevertheless, the induction of the two families was radically different in MEFs from IRF-1^−/− mice (Fig. 3). Both the 65-kDa family members, mGBP-2 and mag-2, were absolutely IRF-1 dependent, as documented elsewhere also for the promoters of mGBP-1 and mGBP-2 (26, 35), while the 47-kDa family members showed only more or less trivial reductions in signal in the IRF-1-deficient fibroblasts. Promoter sequences of both mGBP-1 and mGBP-2 show an apparently inactive STAT1-binding IFN-γ activation site (26), and two functional IFN-stimulated regulatory elements expected to act as IRF-1 binding sites. It is therefore likely that all 65-kDa GTPases are classical secondary response genes, relying for their induction on prior synthesis of the main secondary transcription factor IRF-1 induced by IFN-γ. Transcription of IGTP as a representative of the 47-kDa family, on the other hand, has been shown to be significant within 1 h of induction, and this early transcription is independent of de novo protein synthesis (13), consistent with our demonstration of significant independence from IRF-1. Thus, IGTP, and by implication the other 47-kDa family members, is at least partly a primary response gene with induction of mRNA presumably by activated STAT1 within minutes after IFN-γ stimulation. Transcription of the 47-kDa GTPases is presumably sustained by subsequent induction through IRF-1 or other secondary transcription factors, explaining the significant but incomplete dependence of 24-h mRNA levels on IRF-1 in our experiments, and the significant sensitivity of IGTP mRNA levels at all except the earliest times to cycloheximide (13). Although all the 47-kDa GTPase family members show a degree of independence of IRF-1, IRG-47 appears to be the most sensitive, perhaps consistent with an earlier study of the IRG-47 promoter in which an IFN-stimulated regulatory element was described and which concluded that IRF-1 was an essential transcription factor for this gene (36).

Both the MEFs and ANA-1 macrophages used to generate the SSH libraries in our study are derived from C57BL/6J mice. Neither library yielded any copies of the 65-kDa GTPase prototype, mGBP-1, while the most abundant fragments from both cell types proved to represent the close homologue, mGBP-2. This result is consistent with early studies from Staeheli et al. (25, 37), who showed that most inbred strains of mouse, including C57BL/6J, made no detectable mGBP-1 protein. Although a short 5′-cDNA fragment corresponding to a mGBP-1 homologue was subsequently identified, sequenced, and named mGBP-2 from BALB/c material (26), no full-length sequence of mGBP-2 has been available. The full-length sequence of mGBP-2 presented here shows it to be the closest known relative to mGBP-1 with 80.5% amino acid identity (Table II). Among the 65-kDa GTPases, mag-2 presents an anomaly in that despite strong induction in MEFs (Fig. 3) and in the livers of Listeria-infected mice (Fig. 4), no induction was detected in ANA-1 macrophages. Since mag-2 was first identified as an IFN-γ-inducible gene in a macrophage cell line, RAW264.7, this failure cannot be attributed to the difference in cell type alone. However, RAW264.7 is derived from the BALB/c strain, unlike the C57BL/6J-derived ANA-1. Polymorphism in expression of members of the 65-kDa GTPase family is reminiscent of that observed in mice in another IFN-inducible GTPase, namely the Mx protein, which confers resistance to influenza virus. Polymorphism among mouse strains in expression or function of members of the 47-kDa GTPase family has not been reported.

The existence of polymorphism between members of the 65-kDa GTPase family suggests the existence of some undisclosed but persistent selection pressure on these proteins. A similar conclusion may also be reached from the fine structure of the two GTPase families which are both divided into distinct subfamilies. That these subdivisions imply functional divergence is suggested in both families of GTPases. In the case of the 65-kDa family, the mGBP-1 and -2 proteins share a C-terminal motif CTIL which functions as a modification signal for isoprenylation (38) and presumably membrane attachment, while this motif is absent from mag-2 and mGBP-4. For the 47-kDa proteins, IGTP, GTPI, and LRG-47 share the unusual amino acid, methionine, in the first canonical GTP-binding motif, while this residue is replaced by the typical lysine in the members of the second subfamily. Although IGTP has been shown to be a GTPase (13), no GTPase activity of other 47-kDa GTPases has been documented. Divergence in the products of GTP hydrolysis has recently been demonstrated between the human GBP-1 (favoring GMP) and GBP-2 (favoring GDP) proteins of the 65-kDa family (39), suggesting that GTPase activity per se may be a target for evolutionary modification.

The ubiquitous and abundant presence of these two families of GTPases in the cellular response to IFN-γ, and their structural and biochemical diversification all point to their playing well-defined and specific roles in pathogen resistance. The present experiments provide a hint that if either family of proteins is involved in activation of cellular defense against intracellular pathogens such as L. monocytogenes, it is more likely to be the IRF-1-independent, largely primary response 47-kDa family, rather than the IRF-1-dependent, exclusively secondary response 65-kDa family. Although the molecular mechanisms by which IFN-γR and TNFRp55 trigger such defenses are not yet clear (15, 16, 40), gene products induced by STAT1 and interferon consensus sequence binding protein may be more significant than those induced by IRF-1 in defense against Listeria (41, 42, 43). A distinct perspective has been offered recently by Taylor et al. (44) who proposed, on the basis of finding the IGTP 47-kDa GTPase associated with the endoplasmic reticulum membrane, that this GTPase might function in protein processing for Ag presentation. Perhaps the strongest clue to the function of these enigmatic proteins is from the Mx protein, a large GTPase inducible by type I IFN, which is known to interfere directly with the replication of a number of RNA viruses, especially influenza (45). Very recent experiments have suggested specific antiviral activity in the Mg21/TGTP 47-kDa GTPase (46). Mg21/TGTP-transfected l-cells showed resistance to viral plaque formation by an RNA virus, vesicular stomatitis virus, but not to the DNA virus for herpes simplex.

While our manuscript was in review, Han et al. (47) published a characterization of mGBP-3, including the demonstration that it is induced by IFN-γ.

We thank A. Egert for help in the preparation of embryonic fibroblasts, R. Lange for sequencing, and S. Könen-Waisman for providing us with the IRF-1^−/− MEFs. We are especially grateful to Dr. Douglas Carlow for valuable discussions and for making data available on the putative antiviral activity of Mg21/TGTP before publication. We are grateful to C. Kocks and K. Rajewsky for bringing together the two groups collaborating in this study.