Identification of Amino Acid Residues Critical for Aggregation of Human CC Chemokines Macrophage Inflammatory Protein (MIP)-1α, MIP-1β, and RANTES

Human CC chemokines macrophage inflammatory protein (MIP)-1α, MIP-1β, and RANTES (regulated on activation normal T cell expressed) self-associate to form high-molecular mass aggregates. To explore the biological significance of chemokine aggregation, nonaggregating variants were sought. The phenotypes of 105 hMIP-1α variants generated by systematic mutagenesis and expression in yeast were determined. hMIP-1α residues Asp26and Glu66 were critical to the self-association process. Substitution at either residue resulted in the formation of essentially homogenous tetramers at 0.5 mg/ml. Substitution of identical or analogous residues in homologous positions in both hMIP-1β and RANTES demonstrated that they were also critical to aggregation. Our analysis suggests that a single charged residue at either position 26 or 66 is insufficient to support extensive aggregation and that two charged residues must be present. Solution of the three-dimensional NMR structure of hMIP-1α has enabled comparison of these residues in hMIP-1β and RANTES. Aggregated and disaggregated forms of hMIP-1α, hMIP-1β, and RANTES generally have equivalent G-protein-coupled receptor-mediated biological potencies. We have therefore generated novel reagents to evaluate the role of hMIP-1α, hMIP-1β, and RANTES aggregation in vitro and in vivo. The disaggregated chemokines retained their human immunodeficiency virus (HIV) inhibitory activities. Surprisingly, high concentrations of RANTES, but not disaggregated RANTES variants, enhanced infection of cells by both M- and T-tropic HIV isolates/strains. This observation has important implications for potential therapeutic uses of chemokines implying that disaggregated forms may be necessary for safe clinical investigation.

gation. Our analysis suggests that a single charged residue at either position 26 or 66 is insufficient to support extensive aggregation and that two charged residues must be present. Solution of the three-dimensional NMR structure of hMIP-1␣ has enabled comparison of these residues in hMIP-1␤ and RANTES. Aggregated and disaggregated forms of hMIP-1␣, hMIP-1␤, and RANTES generally have equivalent G-protein-coupled receptormediated biological potencies. We have therefore generated novel reagents to evaluate the role of hMIP-1␣, hMIP-1␤, and RANTES aggregation in vitro and in vivo.

The disaggregated chemokines retained their human immunodeficiency virus (HIV) inhibitory activities. Surprisingly, high concentrations of RANTES, but not disaggregated RANTES variants, enhanced infection of cells by both M-and T-tropic HIV isolates/strains. This observation has important implications for potential therapeutic uses of chemokines implying that disaggregated forms may be necessary for safe clinical investigation.
Chemokines (chemotactic cytokines) are a family of proteins primarily involved in the coordination of cellular trafficking and activation during immune responses (1). They have been characterized according to the sequence context around the first cysteine residue giving rise to the C, CC, CXC, and CX 3 C groups. Chemokines act via a family of G-protein-coupled receptors (GPCRs) 1 and are unusually large ligands (Ͼ60 amino acids) for this type of receptor. There is some controversy concerning the quaternary structure of biologically active chemokines when they bind to receptors, with evidence to suggest either monomers or dimers (2)(3)(4). The three-dimensional structures of the human CC chemokines MIP-1␤ (5), RANTES (6, 7), * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Human CC chemokines MIP-1␣, MIP-1␤, and RANTES (13,14) self-associate to form high-molecular mass aggregates (6,15). Not all chemokines self-associate, leading us to question whether self-association has any biological significance. Such self-association limits the range of purification and formulation conditions available for the clinical development of these proteins. Furthermore, the interpretation of bioassays is complicated, since different protein preparations may vary in their extent of aggregation (16 -18). The primary goal of this work was to identify biologically active variants of hMIP-1␣ that no longer self-associated, thereby allowing large scale production of protein for preclinical and clinical investigation.
hMIP-1␣ is a multifunctional CC chemokine that binds with high affinity to chemokine receptors CCR1 and CCR5 (19,20). The early description of MIP-1␣ as a proinflammatory cytokine (21) has not been confirmed and it has not restricted its use in a number of in vitro (22) and in vivo models (23,24). hMIP-1␣ has a number of potential clinical uses: as a myeloprotectant during aggressive cytotoxic therapy (25)(26)(27), for the treatment of psoriasis (28), for stem cell mobilization (24,29), and for ex vivo hematopoietic stem cell expansion (30).
The human immunodeficiency virus (HIV) inhibitory activities of MIP-1␣, MIP-1␤, and RANTES (31,32) and the identification of chemokine receptors as the essential co-factors for HIV infection of CD4 ϩ cells (33,34) suggested that chemokines, or small molecule mimics, could be considered as potential antiviral agents. HIV is generally classified according to its ability to replicate in macrophages (M-tropic) or established T-cell lines (T-tropic). Viruses isolated over the course of disease progression have been reported to change their in vitro properties from M-to T-tropic, suggesting that M-tropic (CCR5-utilizing) viruses are responsible for infection and the T-tropic (CXCR4-utilizing) viruses may be associated with the rapid CD4 ϩ cell decline and onset of symptoms (35). Thus a second goal of our study was to compare the antiviral properties of aggregating and nonaggregating hMIP-1␣, hMIP-1␤, and RANTES variants.
In this paper we describe the identification of two hMIP-1␣ amino acid residues critical for aggregation, substitution of which yielded biologically active disaggregated proteins. Furthermore, we show that substitution of the homologous residues in hMIP-1␤ and RANTES inhibits their aggregation without loss of bioactivity. Solution of the three-dimensional structure of hMIP-1␣ by NMR has enabled comparison of these residues in hMIP-1␣, hMIP-1␤, and RANTES. The G-proteincoupled receptor-mediated biological activities of the disaggregated chemokines, including inhibition of HIV-1 infection, are unaffected by these substitutions. However, RANTES was found to stimulate M-and T-tropic HIV infection in a concentration-dependent manner. In contrast nonaggregating RANTES did not enhance HIV-1 infection.

MATERIALS AND METHODS
Construction of Synthetic hMIP-1␣, hMIP-1␤, and RANTES Genes, Yeast Expression Vectors, and Mutants-Synthetic genes for hMIP-1␣, hMIP-1␤, and RANTES were designed by back-translation from their protein sequences with codon usage optimized for Saccharomyces cerevisiae (41). The genes were divided into 10 oligonucleotides (synthesized by RϩD Systems Ltd.), and the oligonucleotides were annealed and cloned into M13 vectors. M13 clones were used as templates for mutagenesis to create the hMIP-1␤ D27A, hMIP-1␤ E67S, RANTES E26A, RANTES E66S, and the library of hMIP-1␣ mutants. Mutagenesis was performed according to the method of Kunkel et al. (42). Oligonucleotides used for mutagenesis were 17-36 nucleotides in length (41). Dideoxy sequencing was used to identify the correct clone for each variant in the M13 vectors. The wild-type and variant genes were transferred as HindIII/BamHI DNA fragments into the yeast expression vector pSW6 to drive protein expression under the control of the inducible galactose promoter (43). Recombinant pSW6 clones containing the mutated chemokine genes were sequenced using the dideoxy sequencing method for double-stranded DNA.
Production of Recombinant Chemokine Proteins from S. cerevisiae-S. cerevisiae strain MC2 was transformed with DNA containing the wild-type and mutant chemokine genes by the LiAc/polyethylene glycol protocol (44) and representative clones identified. All yeast media were as described previously (45). Transformant cultures were induced after 24 h of growth by harvesting synthetic complete-Glu broth cultures and resuspension in synthetic complete-Glu/Gal-broth. All cultures were tested for their ability to express chemokine protein by reverse-phase HPLC analysis of cell-free supernatants 2 days after induction. Clones were considered expression-competent if a clear recombinant protein elution peak was observed (approximately Ն20% of wild-type chemokine yield). Three days after induction the proteins were purified from the yeast culture supernatants according to previously published methods (15). Both ion-exchange and reverse-phase HPLC chromatography were used to obtain the protein at 95% purity.
Native Polyacrylamide Gel Electrophoresis and Sedimentation Equilibrium Analytical Ultracentrifugation (AUC)-Nondenaturing 4 -20% (w/v) gradient polyacrylamide gels (Novex) were run under nondenaturing conditions in Tris-glycine buffer. See Blue markers (Novex) and wild-type hMIP-1␣ (2 g) were analyzed on each gel. All variants were scored according to their mobility relative to the wild-type protein.
AUC analysis was performed using a Beckman XLA analytical ultracentrifuge with absorbance optics as described previously (15,385). Protein samples at 0.1 or 0.5 mg/ml were loaded in a six-sector cell to enable simultaneous measurements and comparisons of variants. PBS "A" was loaded in the reference sector. Samples were analyzed at 15,000 rpm at 25°C for 18 h to reach equilibrium. Final solute distributions were recorded at 235, 280, and 295 nm as a function of radius. Scans were repeated after a further 3 h of centrifugation to confirm that equilibrium had been achieved. The final solute distribution ASCII data was fitted using Microcal Origin software to a model assuming a single ideal species (Ideal1) to provide an estimate of the weight average molecular weight. Data sets shown are an average of 20 scans in 0.001-cm increments across the radius of the rotor.
Competitive Receptor Binding Assays-Competitive receptor binding assays using the FDCP-mix A4 cell line (Spooncer et al. (37)) have been described previously (38). Briefly, 1 ϫ 10 6 murine FDCP-mix A4 cells, various amounts of unlabeled hMIP-1␣ or variants, and 3.85 ng/ml 125 I-hMIP-1␣ (250 Ci/g; Amersham Pharmacia Biotech) were incubated in a total volume of 250 l of binding medium (RPMI 1640, 20 mM HEPES, 1 mg/ml bovine serum albumin) at 25°C for 60 min. Phosphate-buffered saline (1 ml) was added, mixed, and the cells harvested by centrifugation. The supernatants were removed and the cells washed twice with PBS. Radioactivity in the cell pellet was determined and IC 50 values for the variants estimated.
Chemotaxis Assay-Cell migration was evaluated using 24-well tissue culture plate inserts with polyethylene terephthalate membranes (8-m pore size, Falcon). Chemokines were diluted (10 Ϫ7 to 10 Ϫ12 M) in fetal calf serum-free RPMI 1640 medium and 0.5 ml of each dilution placed into the companion plate. Freshly purified human mononuclear cells were suspended at 2.5 ϫ 10 6 cells/ml in fetal calf serum-free RPMI 1640 and 0.2 ml placed onto the insert filters prior to setting the inserts into the wells. The filters were removed after 2-h incubation, the medium discarded, and the filters dried. The cells were fixed in 4% (v/v) formaldehyde for 10 min, washed in PBS, stained with 2% (w/v) cresyl violet acetate and washed in PBS. Migration was assessed by averaging the count from two high-powered fields (ϫ 400 magnification) per assay point in each experiment. The data presented are the mean number of cells/high-powered fields from four independent experiments.
Determination of hMIP-1␣ Three-dimensional Structure by Nuclear Magnetic Resonance Spectroscopy-15 N/ 13 C-labeled hMIP-1␣ and hMIP-1␣ D26A was prepared by established methodology using the Pichia pastoris expression system (46). Minor modifications reduced the concentration of [ 13 C]methanol to 4% (w/v) used in the induction to reduce costs. hMIP-1␣ D26A was purified from the culture supernatant by sequential cation exchange and reverse-phase high-pressure liquid chromatography. Samples for NMR were at ϳ3 mM concentration, pH 3.5, 10% (v/v) D 2 O. All spectra were collected at 318 K.
Homonuclear assignment was performed as described by Wü thrich (47), and heteronuclear assignments were performed as reviewed by Whitehead et al. (48). 15 N relaxation data were acquired as described by Kördel et al. (49). Constraints for structure calculations were obtained from a 150-ms homonuclear NOE experiment (50), a 150-ms simultaneous 15 N/ 13 C edited NOESY experiment (51), and a 150-ms doublefiltered NOESY experiment. All data were collected and analyzed using FELIX (Micron Separations, San Diego, CA). The constraints were divided into three ranges with upper bounds of 2.5, 3.5, and 5 Å, based on a calibration from elements of regular secondary structure. constraints were obtained from an heteronuclear single quantum coherence-J experiment (52), constraining to either Ϫ60 (Ϯ40) o or Ϫ120 (Ϯ50) o . The structure calculation protocol consisted of calculation of approximate monomer structures, followed by filtering out of probable dimer NOEs using a filter cutoff of 10 Å. This process was cycled several times. Dimer NOEs were also checked against the double filtered NOESY experiment, although this was insufficiently sensitive to detect all NOEs, and acted mainly as a rigorous check that the correct dimer interface was present. The monomer structures were oriented as rigid entities via the dimer NOEs while being maintained at a separation of 100 Å. The monomers were then annealed together and the structures refined by further annealing cycles. All calculations were performed using modified variants of the protocol of Nilges et al. (53), using XPLOR (54). Fig. 4, A and B, were created using RASMOL (55). An ensemble of 10 structures and a minimized mean structure of hMIP-1␣ D26A have been submitted to the Brookhaven Protein Data Bank, entries 1B50 and 1B53, respectively.
HIV-1 Infectivity Assays-Neutralization was assessed using PHAstimulated PBMC as target cells with determination of soluble p24 antigen production as the end point (56). Chemokines were tested for their ability to inhibit HIV infection by incubation of a known concentration with 100,000 PBMC (mixed from two blood donors) in a final volume of 75 l of RPMI, 10% fetal calf serum, IL-2 (5 units/ml) for 2 h at 37°C. Cells were infected with a known infectious titer of virus, 100 TCID 50 , and the virus/ligand incubation mixture incubated at 37°C overnight. Cells were washed the following day, the relevant concentration of chemokine added back to all cultures, and incubated at 37°C for 5 days.

Construction of a Library of hMIP-1␣ Variant Proteins-105
single amino acid substituted variants of hMIP-1␣ were generated by oligonucleotide site-directed mutagenesis (Fig. 1A). All mutations were verified at the M13, yeast expression vector, and postexpression stages by DNA recovery and sequencing (data not shown). Initially, charged residues implicated in the aggregation process were mutated; however, the entire molecule was screened such that all hydrophilic residues were mutated to serine and all hydrophobic residues to alanine. Sixtyfive of the 69 hMIP-1␣ residues were substituted, and the four cysteine residues were unmodified to retain structural integrity. Expression of 75% of the mutants was comparable with that of hMIP-1␣, allowing purification of protein with substitutions at 55 of the 65 available residues.  (41). The number of alternative amino acid substitutions at each residue, which expressed well (Ն20% hMIP-1␣), is shown above the origin and those which expressed poorly (Յ20% hMIP-1␣) below the origin. Cysteine residues at amino acid positions 10, 11, 34, and 50 were not mutated to retain structural integrity. B and subsequent panels refer to amino acid residues at which the described properties have been identified are shown with a tall histogram and variants which were assayed for a property but did not meet the criteria (e.g. they were not disaggregated or they were less potent) are shown by short histograms. B, variants that expressed well and were disaggregated according to native polyacrylamide gel electrophoresis. Disaggregated variants migrated substantially further into the gel than hMIP-1␣, which remained near the well. C, variants that were disaggregated according to sedimentation equilibrium AUC analysis. Disaggregated variants possessed weight average molecular weights Յ 100,000 Da. D, disaggregated variants that retained full competitive receptor binding activity on FDCP-mix A4 cells. Fully active variants were defined as those with IC 50 values Յ 8 nM in this cellular assay. hMIP-1␣ Self-association-Substantial changes in the selfassociation of the hMIP-1␣ variant proteins were initially detected by native polyacrylamide gel electrophoresis. Of the 79 variants analyzed, 37, representing 28 residues, migrated more rapidly through the gel than hMIP-1␣ (Fig. 1B). Sedimentation equilibrium AUC analysis was used to provide a quantitative assessment of the effects of the substitutions on hMIP-1␣ selfassociation (36). Twenty-three of the 37 variants analyzed had substantially reduced (Ͻ100 kDa) weight average molecular weights than hMIP-1␣ (Ͼ100 kDa). These variants represented 16 positions in hMIP-1␣ at which substitution leads to significant disaggregation (Fig. 1C). Native gel electrophoresis appeared to enhance the effects of substitution on hMIP-1␣ variants, implying greater effects on self-association than observed by AUC analysis. The high glycine concentration in the gel may be responsible for the differences between the two assays. The AUC analysis is generally accepted to be a preferable method of analysis of self-association.
Identification of Active Disaggregated hMIP-1␣ Variants-The activity of the disaggregated hMIP-1␣ variants was assessed in a receptor binding assay using the murine cell line, FDCP-mix A4 (37,38), which binds and responds to hMIP-1␣. Nineteen variants, representing substitution at 16 residues, were assessed for their ability to bind the murine MIP-1␣ receptor expressed by these cells in a competitive receptor binding assay. The concentration of hMIP-1␣ or variants required to inhibit iodinated hMIP-1␣ binding by 50% (IC 50 ) was estimated. hMIP-1␣ had an IC 50 of 3.27 Ϯ 0.34 ng/ml (mean Ϯ S.E.; n ϭ 81). A range of hMIP-1␣ variant receptor binding activities was observed from wild-type to IC 50 values Ն 100 nM. For the purposes of screening, variants with mean IC 50 values Յ 8 nM were considered to be fully active as they fall within the 99.99% confidence interval for wild-type hMIP-1␣ activity (Fig. 1D). The four most active disaggregated variants representing two residues of hMIP-1␣ (Asp 26 and Glu 66 ) were essentially homogeneous tetramers at 0.5 mg/ml. Three alternative substitutions D26A, D26S, and D26Q gave rise to the same properties.

Substitution of Homologous Amino Acid Residues in hMIP-1␣, hMIP-1␤, and RANTES Produces Active Disaggregated
Variants-Alignment of the protein sequence of hMIP-1␣ with the structurally related chemokines hMIP-1␤ and RANTES reveals homologous charged amino acid residues that may play similar roles in the self-association pathways of these chemokines (Table I). Substitution of residues D27A and E67S in hMIP-1␤ and E26A and E66S in RANTES was performed to investigate the aggregation and biological properties of the variants. Aggregation of RANTES was so pronounced that at 0.5 mg/ml the estimate of weight average molecular weight was unreliable, and therefore a lower concentration of 0.1 mg/ml was used in these comparisons. AUC analysis indicated that residues 26 and 66, which are critical for hMIP-1␣ self-association, also play a key role in the self-association of hMIP-1␤ and RANTES (Fig. 2). Substitution of hMIP-1␣ at either resi- Comparison of the distribution of wild-type and disaggregated chemokine proteins in the ultracentrifuge cell at equilibrium. All samples were prepared at 0.1 mg/ml and centrifuged at 15,000 rpm. Absorbance was estimated at three wavelengths and the most complete data set used for analysis. The solid line depicts the best fit of the data assuming an ideal nonassociating protein. The straighter the distribution line and the more random the scatter of the residuals, the closer the protein solution is to a nonassociating ideal species. Curvature in the distribution data or the residuals indicates a self-associating protein species. Weight average molecular weights for the proteins are shown. due 26 or 66 resulted in the formation of an essentially nonassociating protein solution. Substitution at residue 67 in hMIP-1␤ was more effective than substitution at residue 27, although both variants continued to self-associate, reaching a lower weight average molecular weight than wild-type hMIP-1␤. Substitution at residue 66 in RANTES was more effective than that at residue 26, resulting in the formation of an essentially nonassociating RANTES solution. The lowest disaggregated chemokine weight average molecular weights were consistent with dimeric proteins at 0.1 mg/ml.
The biological activity of the disaggregated chemokines was assessed using human embryonic kidney (HEK) cells expressing either human CCR1 or CCR5 (19,20). Disaggregated chemokine variants retained their ability to bind CCR1 (Fig. 3A) and to induce signal transduction (Fig. 3B), indicating that the substitutions did not grossly affect their biological activities. hMIP-1␤ and its disaggregated mutants are not agonists of CCR1, but do signal effectively via CCR5. Furthermore, the human mononuclear cell chemotactic activity was not affected by disaggregating substitutions (Fig. 3C).
Structural Characterization of hMIP-1␣ and hMIP-1␣ D26A by NMR-The structures of hMIP-1␣ and hMIP-1␣ D26A were investigated by NMR. In order to assess whether gross structural changes accompanied the substitution of D26A, backbone 1 H and 15 N assignments were also completed for hMIP-1␣. The chemical shift changes between the wild-type and D26A forms are extremely small, being less than 0.1 ppm for ␣ or amide protons except adjacent to the site of the mutation. Apart from the mutated residue itself, the greatest shifts are 0.12 ppm for the amide proton of the subsequent residue Tyr 27 and for the residues across the ␤-sheet from residue 26, namely Phe 41 (␣ proton shift difference of 0.09 ppm) and Leu 42 (amide proton shift difference of 0.12 ppm). These chemical shift changes are much smaller than would be expected for a major structural disruption where shift changes of up to 2 ppm would be likely.
The full 1 H, 15 N, and 13 C heteronuclear assignment of hMIP-1␣ D26A was completed, which, in conjunction with heteronuclear edited NOESY spectra, allowed the calculation of an ensemble of solution state structures. The experimental data for the structure calculation included a total of 821 NOE constraints and 29 angle constraints. The NOE constraints break down into 306 intraresidual, 211 sequential, 98 medium range, and 172 long range monomer NOEs and 34 intermonomer NOEs (these figures count each NOE only once). A total of 30 structures were calculated, and the 10 with the lowest energy were selected to represent the ensemble of properly converged structures. The root mean square deviation for the overlay of individual monomers onto the average structure for residues 15-29 and 39 -63 is 0.7 Å for backbone atoms and 1.2 Å for all heavy atoms. The remaining regions of the protein showed a much higher root mean square deviation. 15 N relaxation data indicated that significant subnanosecond time scale motions occur for residues 1-4 and 67-69. The higher root mean square deviation for residues 5-15 and the loop from 30 -38 are for residues in the dimer interface and probably reflect the difficulty in properly defining the relative orientation of the two monomers in the dimer. This is reflected in the overlay of backbone heavy atoms of the second monomers in the ensemble of dimers of ϳ12 Å when only the first monomer in each dimer structure is overlaid. This introduces a consequent uncertainty in the dimer interface region. This variability in the dimer structure may represent true flexibility or may simply represent a limitation of current NMR methodology. That such uncertainty is reasonable is supported by a comparison of the two published RANTES structures (6,7). Comparison of the monomers from these structures shows a similar discrepancy between the average structures that is similar to that found in our ensemble of structures. In particular, the structures disagree on the positions of the N-terminal region and the 30-s loop. The relative orientations of the monomers in the dimer structures reported for RANTES show a similar corresponding discrepancy. In MCP-1, a similar variability in dimer structure has been reported between dimers from two crystal forms crystallized from the same drop (39). This indicates that in MCP-1 the dimer relationship is not rigidly fixed. The difficulty of properly defining the dimer led us to disregard NOEs that could not be clearly ascribed to either inter-or intramonomer contacts in order not to falsely constrain the ensemble of structures.
The structures show that the monomer structures of hMIP-1␣ D26A (and of hMIP-1␣ on the basis on the small chemical shift differences) are similar to those of hMIP-1␤, RANTES, and MCP-1. An example overlay of the backbone atoms of hMIP-1␣ and hMIP-1␤ is shown in Fig. 4A.
A Mechanism for CC Chemokine Disaggregation-Mutagenesis studies have identified two acidic residues at positions 26/27 and 66/67 of hMIP-1␣, hMIP-1␤, and RANTES that are critical for self-association, and yet mutation of these residues does not affect the biological activities of these molecules. Substitution at residue 66/67 is more effective at disaggregating the chemokines than are changes at position 26/27. These key residues must be involved in interactions that stabilize the multimeric form. Analysis of the three-dimensional structures of D26A hMIP-1␣, hMIP-1␤, and RANTES is complicated by the necessity to generate structures under nonaggregating acidic conditions, which abolish any interactions that the 26/27 and 66/67 residues may normally have. Although the structures show residues 26/27 and 66/67 to be close together in the monomer, it is not obvious how they interact, if at all (Fig. 4B). We note that the amide proton of the intervening residue Tyr 27 shows protection in hMIP-1␣ D26A, even though it is apparently solvent exposed and not clearly involved in any hydrogen bonding. Similar protection was observed in RANTES, which was thought to arise from interaction with the side chain of Asp 26 (7). This cannot be the case with hMIP-1␣D26A, and we cannot currently offer an alternative explanation. The most likely interaction for these acidic residues would be with basic residues. Substitution of basic residues at hMIP-1␣ positions Lys 44 , Arg 45 , or Arg 47 effectively inhibited aggregation but also reduced biological activity (Fig. 1), such that we have been unable to identify direct contacts for residues 26/27 and 66/67. The interaction of dimers, in the formation of tetramers, may bring acidic and basic regions together to stabilize the multimer. Charge removal or reversal would destabilize the tet- ramer and result in disaggregation. Mutagenesis of mMIP-1␣, substituting up to three C-terminal acidic residues, also resulted in substantial disaggregation implicating the C-terminal ␣-helix in multimer formation (18). It is possible that the position and/or flexibility of the C-terminal ␣-helix is critical to self-association and that substitution at residues 26/27 or 66/67 subtly alters its position or flexibility, thereby decreasing the stability of multimers.
Comparison of the HIV-1 Inhibitory Properties of Aggregating and Nonaggregating Chemokines-SF-2 (T-tropic, CXCR4 utilizing) and SF-162 (M-tropic, CCR5 utilizing) viruses were used to evaluate the specificity of chemokine-mediated inhibition of PBMC infection (Fig. 5A). HIV-1 infection was quantified by measuring soluble core protein (p24) antigen production by enzyme-linked immunosorbent assay. Both wild-type and disaggregated chemokines (0.1 M) inhibited SF-162 infection of two different donor PBMC cultures (Fig. 5A). All wild-type and variant chemokines, at a concentration of 0.1 M, had no effect on the ability of SF-2 to infect PBMC cultures. Generally, disaggregated variants substituted at residue 26/27 were less efficient at neutralizing SF-162 infection than wild-type che-mokines or variants substituted at residue 66/67 (data not shown). Since chemokine aggregation is concentration-dependent, we evaluated the effect of higher concentrations of chemokines in the PBMC HIV-1 infection assay (Fig. 5B). Surprisingly, high concentrations (1.0 M and greater) of RANTES induced an increase in extracellular p24 antigen from both SF-2-and SF-162-infected cultures. Similar effects were not observed with cultures treated with either hMIP-1␣ or hMIP-1␤. The eight-fold increase in p24 antigen release was accompanied by a 10-fold increase in infectious virus titer, as determined by the titration of extracellular virus on U87.CD4 cells expressing either CCR5 or CXCR4 co-receptors (data not shown). It is striking that stimulation was not observed in cultures treated with the disaggregated variants of RANTES. DISCUSSION hMIP-1␣, hMIP-1␤, and RANTES are CC chemokines that tend to self-associate, forming high molecular mass aggregates. This self-association is a dynamic, reversible process, and it is generally accepted that chemokine concentrations in vivo may be too low to encourage such aggregation. However, there are natural circumstances that may cause high local concentrations of chemokines. These include platelet degranulation, in inflammatory disease (40), and accumulation on cell membranes via receptors or glycosaminoglycans. In addition, should chemokines be used clinically, their concentration will be high at the injection site, encouraging self-association and precipitation which may reduce tissue penetration and induce inflammation. We believe that CC chemokine self-association must be fully characterized and its in vivo relevance determined if the immunomodulatory properties of these chemokines are to be fully understood.
We have explored the molecular basis for chemokine aggregation, using hMIP-1␣ as a model system, and have also investigated the aggregation of hMIP-1␤ and RANTES. Extensive mutagenesis and biophysical characterization, comprising the production of 105 single amino acid substituted variants of hMIP-1␣ identified two residues, Asp 26 and Glu 66 , as the key elements to the self-association process. Identical or analogous residues were found in homologous positions in both hMIP-1␤ and RANTES, suggesting that their substitution might affect self-association. Substitution of Asp 27 and Glu 67 in hMIP-1␤ and Glu 26 and Glu 66 in RANTES had similar effects, substantially disaggregating the chemokines and resulting in the production of proteins with greatly improved solution properties. Our analysis suggests that a single charged residue at either 26/27 or 66/67 is insufficient to support extensive self-association and that two charged residues must be present. Data presented here and elsewhere demonstrate that both aggregating and disaggregated forms of hMIP-1␣ generally have equivalent biological potencies (18,22,38). Disaggregated hMIP-1␤ and RANTES variants retain the biological activities and receptor selectivities associated with the aggregating forms. We have therefore generated novel reagents to evaluate the role of hMIP-1␣, hMIP-1␤, and RANTES self-association in vitro and in vivo. Since the GPCR-mediated activities of these chemokines-receptor binding, signal transduction, and chemotaxis are unaffected by the substitutions which disaggregate the chemokines, self-association is probably irrelevant to these activities. If this proposition is true, any biological differences observed between the wild-type chemokines and their disaggregated variants are probably due to their differences in their self-association and probably act via non-GPCR-mediated events.
Using wild-type and disaggregated forms of hMIP-1␣, hMIP-1␤, and RANTES we have shown that the effects of chemokines on HIV-1 infection are concentration-dependent. At lower con- centrations (Յ100 nM) hMIP-1␣, hMIP-1␤, and RANTES inhibit infection of PBMC of SF-162 virus, but at higher concentrations (Ն1000 nM) RANTES enhances infection by both SF-2 and SF-162 viruses. Interestingly, disaggregated RANTES variants failed to enhance HIV-1 infection, instead acting as effective inhibitors of M-tropic HIV-1 infection at all concentrations tested. Since all of the other biological activities tested were normal, we conclude that RANTES aggregation is directly responsible for this stimulation of HIV-1 infection. We speculate that this stimulation is not mediated by GPCRs and currently consider interactions with glycosaminoglycans to be a potential mechanism of RANTES-mediated viral stimulation.
Our observations have important implications for the clinical use of chemokines. RANTES may be a key regulator of HIV-1 infection; it is the most potent chemokine inhibitor of M-tropic HIV-1 infection (32) but as we show here can also act as a stimulator, enhancing viral infection. Administration of RAN-TES to HIV-1 infected individuals could increase HIV-1 replication and thus accelerate disease progression. In contrast, administration of disaggregated RANTES is likely to suppress HIV infection of new target cells if effective concentrations can be achieved. Epidemiological analysis of correlations between HIV-1 viral load and RANTES expression or inflammatory disease may be informative. In conclusion, we have identified amino acid residues critical to CC chemokine self-association and have identified a biological consequence of RANTES aggregation. It is likely that there will also be activities associated with hMIP-1␣ and hMIP-1␤ aggregation that will be revealed only by use of the relevant assay system. Nonaggregating variants of hMIP-1␣, hMIP-1␤, and RANTES may be essential for the safe clinical evaluation of these chemokines. hMIP-1␣ D26A, also known as BB-10010, has proven to be safe and well tolerated in phase I and phase II clinical trials even after large doses (300 g/kg) have been administered by subcutaneous injection (25)(26)(27)38).