A Zn2+ Ion Links the Cytoplasmic Tail of CD4 and the N-terminal Region of Lck*

Lck is a lymphoid-specific, Src family protein-tyrosine kinase that is known to interact with the T-cell coreceptors, CD4 and CD8. This interaction, which is critical for proper T-cell function, is mediated by the N-terminal unique region of Lck and the C-terminal cytoplasmic tail of the coreceptors. A pair of cysteines on each molecule is essential for association, suggesting that CD4 or CD8 may interact with Lck by jointly coordinating a metal ion. We describe here experiments in which a maltose-binding protein fusion protein bearing the CD4 tail has been coexpressed inEscherichia coli with an N-terminal fragment of Lck. The proteins associate in the expressing cells, forming a complex that can be affinity-purified. Formation of this complex, like the in vivo interaction, depends upon the two pairs of cysteines. Biochemical and biophysical experiments show that the complex dissociates in the presence of EDTA and that it contains a single Zn2+ ion. These results are consistent with the proposal that Lck and CD4 associate by thiol-mediated co-coordination of zinc.

Signal transduction through the T-cell antigen receptor activates a number of cytoplasmic protein-tyrosine kinases (PTKs), 1 including Lck, Fyn, Itk, and ZAP-70 (see Refs. 1-3 for review). Activation of Lck (p56 lck ) is one of the earliest and most central events in this signaling cascade. Lck is a Src family PTK found primarily in T-cells, and it is recruited to the activated T-cell receptor via its association with the cytoplasmic domain of the CD4 or CD8 coreceptors (4 -9). Genetic and mutational analyses demonstrate that Lck plays an indispensable role in antigen-dependent T-cell activation (10 -13) and in T-cell development in the thymus (14 -16). Lck-deficient T-cells fail to respond to antigen stimulation, and lose cytotoxic potential (11,17,18).
Like other Src family PTKs, Lck contains regulatory SH3 and SH2 domains, followed C-terminally by a conserved tyrosine kinase domain (19). Each family member has a unique N-terminal domain, which lacks the homology prevalent in the rest of the protein. The role of the unique domain remains unclear for most proteins in the superfamily, but in the case of Lck it has been shown to mediate association with the T-cell surface coreceptors CD4 and CD8 (20 -24). This association is essential for normal T-cell function (8). Structure-function analysis of deletion mutants in cell culture has shown that the cytoplasmic tail of either CD4 or CD8␣ (Lck binds the ␣ variant only) and the first 32 residues of the Lck unique region are sufficient for in vivo association (21). The interaction mediated by these regions appears to be quite specific. When the extracellular and transmembrane domains of platelet-derived growth factor receptor are fused to the cytoplasmic domain of CD4, the fusion protein complexes with Lck and transduces signals through it (25). Similarly, when the regulatory and kinase domains of p60 src are joined to the unique region of Lck, CD4 activation produces Src kinase activity (20). Analysis of point mutations has demonstrated that the coreceptor-Lck interaction is mediated by cysteine residues. More specifically, the sequence CXCP in the cytoplasmic tails of both CD4 and CD8␣ appears to interact with the sequence CXXC in the N-terminal domain of Lck. When these cysteines are mutated to alanine, serine, or even histidine, the complex, which is detectable by immunoprecipitation in wild type cells, does not form (20,22). This evidence has been corroborated by antibody induced capping and immunofluorescence microscopy of the CD4⅐Lck and CD8⅐Lck complexes in vivo (26,27). Lck appears to be the only Src-related PTK that utilizes cysteines to bind surface receptors.
There are at least three known cysteine-mediated receptor/ kinase interactions, and all of them involve Lck. CD4 and CD8␣ represent two of these. In addition, 4 -1BB, a T-cellspecific antigen whose structure resembles members of the nerve growth factor superfamily, also bears a CXCP motif and interacts with Lck (28). The dependence of these non-covalent interactions on four cysteines suggests that receptor and kinase may be held together by a metal ion coordinated by thiol groups from both molecules. This notion is supported by the observation that 1,10-ortho-phenanthroline, a metal chelator, disrupts in vitro reconstitution of a CD4⅐Lck complex from detergent lysates (20). Metal-mediated association has been previously demonstrated with the human immunodeficiency virus Tat protein, which has been shown to form zinc-and cadmium-linked homodimers (29). A direct demonstration of the participation of a metal ion has hitherto been lacking, however. This paper describes experiments in which fragments of the two proteins have been coexpressed as distinct fusion proteins in bacteria and their complex purified. The 1:1 complex of MBP-CD4-(396 -433) and Lck-(7-226) that we can isolate is cysteine-dependent and EDTA-sensitive, and it contains one equivalent of Zn 2ϩ . These results are consistent with the * 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. ¶ ¶ To whom correspondence should be addressed. E-mail: harrison@crystal.harvard.edu. 1 The abbreviations used are: PTK, protein-tyrosine kinase; MBP, maltose-binding protein; ICP, inductively coupled plasma atomic absorption spectroscopy; PCR, polymerase chain reaction; ␤-ME, ␤-mercaptoethanol; PAGE, polyacrylamide gel electrophoresis; VSV, vesicular stomatitis virus. model that the association is mediated by a co-coordinated zinc ion.

EXPERIMENTAL PROCEDURES
Cloning of CD4 and Lck Constructs-In order to coexpress two proteins in bacteria from two different plasmid vectors, it is necessary that those plasmids bear different origins of replication. A coexpression plasmid was constructed using pACYC184 (New England Biolabs), a low copy number cloning vector. The T7 expression cassette from a pRSET vector (Invitrogen) was amplified by the polymerase chain reaction (PCR), and the resultant fragment ligated into pACYC184. The cloning process replaced more than half of the pACYC184 tetracycline resistance gene, but left the chloramphenicol resistance gene intact. The resultant vector (pACYC-T7) was used for Lck expression. The LckN32 construct was made by PCR amplification of the appropriate DNA fragment, followed by ligation into pACYC-T7. DNA required for the Lck32 construct was excised from large quantities of pRSET vector bearing the Lck32 DNA. The fragment was then ligated into pACYC-T7. DNA fragments coding for different CD4 constructs were amplified by PCR using appropriate primers. Cys to Ala mutations were introduced by long 3Ј primers, which hybridized over the codons to be mutated. The various CD4 tail fragments were ligated into pMAL-c2, an MBP fusion protein expression vector (New England Biolabs).
Coexpression of CD4 and Lck-pMAL-c2 and pACYC-T7 vectors bearing CD4 and Lck constructs were used to transform Escherichia coli BL21. Cells were grown in the presence of 150 g/ml ampicillin and 75 g/ml chloramphenicol to select for cells bearing both plasmids. Expression was induced with 0.5 mM isopropyl-1-thio-␤-D-galactopyranoside during log phase growth. Suitable induction of the MBP-CD4 fusion proteins and the Lck 32 fragment was usually achieved in 2 h. The CD4 fusions express as arrays of bands between 45 and 48 kDa, indicating either proteolytic cleavage or premature termination. Expression of the N32 fragment was low and relatively non-inducible, but the yield was adequate for the studies described here.
Phosphotyrosine Affinity Purification-Isopropyl-1-thio-␤-D-galactopyranoside-induced cells were lysed by sonication in a standard physiological buffer consisting of 50 mM Tris, pH 7.5, 200 mM NaCl, 0.2% ␤-ME, and 2 mM benzamidine (the MBP-CD4 fusions are slowly degraded in the absence of benzamidine). The lysate was subjected to ultracentrifugation at 40 -60 K for at least 2 h to divide the soluble and insoluble components. The supernatant from this step was loaded onto a phosphotyrosine affinity column that had been washed at high and low pH, and then equilibrated in 50 mM Tris, pH 7.5, 200 mM NaCl, 0.2% ␤-ME, 2 mM benzamidine (hereafter called standard buffer). After loading, the column was washed with 10 column volumes of standard buffer, and then eluted with standard buffer plus 100 mM phenylphosphate. Yields generally range between 3 and 10 mg of complex/liter of induced E. coli.
Gel Filtration-Samples (0.5 ml) of CD4L⅐LckN32 complex were applied to a Superdex 200 sizing column (Amersham Pharmacia Biotech) that had been equilibrated in at least 1.5 column volumes of standard buffer. Gel filtration experiments were performed using the Pharmacia FPLC system. Elution profiles were monitored by A 254 and plotted; 0.5-ml or 1.0-ml factions were collected.
Native PAGE Experiments-Samples of purified CD4L⅐LckN32 complex were incubated in buffers containing the desired reagents (EDTA, phenanthroline, dithiothreitol, etc.). All solutions were buffered at pH 7.5, and contained 200 mM NaCl, at least 0.2% ␤-ME and 20% glycerol. Samples were incubated at room temperature for at least 1 h, separated by electrophoresis on a 6% native gel at pH 8.8, and visualized by Coomassie staining.
Inductively Coupled Plasma (ICP) Spectroscopy and Zinc Quantitation-ICP spectroscopic analysis was performed by Robert Foster (Thermo Jarrell Ash Corp.). Five samples were analyzed. Two contained CD4L⅐LckN32 complex from two separate bacterial preparations. Two contained blanks taken from the same preparations (buffer that eluted from the Superdex 200 column just before the complex), and the last contained ϳ0.5 mg/ml MBP. The concentration (in parts per billion) of 25 different metal atoms was determined for each sample. The samples containing CD4L⅐LckN32 complex were then subjected to amino acid analysis to determine their exact protein concentration. The analysis was performed at the HHMI Biopolymers facility at Harvard Medical School.

RESULTS AND DISCUSSION
Various portions of CD4 and Lck were coexpressed in bacteria. Fragments encompassing the cytoplasmic tail of CD4 were expressed as fusion proteins with MBP, chosen because it is monomeric. Five CD4 fusion proteins were expressed, two with wild type CD4 sequences and three with mutations (Fig. 1A). The larger of the wild type constructs (CD4L) include almost the entire C-terminal tail (residues 396 -433); the smaller (CD4S) was seven residues shorter, beginning after the Nterminal stretch of arginines. In the three mutant constructs, alanines were substituted for Cys-420 and Cys-422, the cysteines shown previously to be critical for association (20,22). The double mutant (CD4m1ϩ2) and both single mutants (CD4m1, CD4m2) were made. All constructs were well expressed. They appeared as ϳ48-kDa bands on SDS gels (Fig. 2, indicated bands), but the presence of fragments shorter than the full-length induced band suggested some degree of proteolytic degradation or premature termination. Coexpression of the MBP CD4 fusion proteins with Lck had no effect on this distribution of bands.
Two Lck fragments were coexpressed with the MBP-CD4 fusion proteins, LckN32 and Lck32 (Fig. 1B). LckN32 (residues 7-226) includes the unique region, the SH3 domain, and the The cysteines shown previously to be required for interaction with p56 lck (C420 and C422) are shown in bold (20,22). In the three CD4m constructs, these cysteines are mutated to alanines as indicated. B, Lck constructs. The full-length PTK is shown as a reference. The cysteines required for interaction with CD4 (C20 and C23) are shown in bold (20,22). The cloning fidelity of all constructs, with the exceptions of CD4S and Lck32, has been confirmed by DNA sequencing. Figure is not drawn to scale. SH2 domain of Lck, while Lck32 (residues 53-226) bears the SH3 and SH2 domains, but lacks the unique region (Fig. 2,  indicated bands). The presence of an SH2 domain in both fragments provides for simple, one step purification by phosphotyrosine affinity chromatography. The Lck32 fragment, like the CD4m fusions, was used in coexpression experiments as a control for any nonspecific interaction.
The expressed CD4 fusion proteins and Lck fragments are all soluble, fractionating in the supernatant portion of the cell lysate after ultracentrifugation. As an assay for CD4⅐Lck association, soluble fractions from all coexpression strains (CD4L⅐LckN32, CD4S⅐LckN32, CD4m1⅐LckN32, CD4m2⅐ LckN32, CD4m1ϩ2⅐LckN32, CD4L⅐Lck32) were subjected to phosphotyrosine affinity chromatography. If the two proteins are associated, they will both appear to bind to the column and will elute together. Conversely, if they are not associated, the CD4 fusion protein will not be retained and will not coelute with Lck. The results from these experiments are shown in Fig.  2. Significant coelution was observed only when CD4L was coexpressed with LckN32 ( Fig. 2A). We concluded that this was the result of protein-protein association, as CD4L was not retained by phosphotyrosine columns when expressed by itself (data not shown). Only full-length CD4L (the highest induced band) associates with LckN32, consistent with the assumption that the other induced bands are CD4 tail truncations of some sort. When LckN32 and MBP-CD4L are separately expressed and purified, they co-purify on phosphotyrosine-Sepharose when mixed in the presence, but not absence of zinc (data not shown).
All three of the cysteine mutant fusions failed to coelute with LckN32 (Fig. 2, D-F), and CD4L did not coelute with Lck32 (Fig. 2C). These results show that the interaction observed in the CD4L⅐LckN32 coexpression experiments depend upon the N-terminal unique region of Lck and upon cysteines 420 and FIG. 3. Gel filtration of the CD4L⅐LckN32 complex. A 254 elution profiles are shown along with Coomassiestained gels that encompass the peaks of interest. CD4L and LckN32 are indicated in both gels. The sample loaded in each experiment (L) and a set of molecular weight standards (M) are also shown. A, concentrated (ϳ20 mg/ml) CD4L⅐LckN32 complex taken directly from phosphotyrosine elution and applied to a Superdex 200 sizing column. The two primary peaks contain CD4L⅐LckN32 complex and unbound LckN32, respectively. The peak at 8.5 ml, which is not shown in the gel, represents aggregates of higher molecular weight protein. B, gel-filtered CD4L⅐LckN32 complex reapplied to the Superdex 200 column at low concentration (ϳ1 mg/ml). The primary peak contains CD4L⅐LckN32 complex, with a small shoulder containing unbound LckN32. The molecular mass of the CD4L⅐LckN32 complex was approximated using the following standards: dimeric transferrin receptor (140 kDa, 13.2 ml), bovine serum albumin (66 kDa, 14.6 ml), and ovalbumin (44 kDa, 15.6 ml).

FIG. 2. CD4⅐Lck copurification experiments.
Coelution from phosphotyrosine affinity columns was used to assay for protein:protein interaction. Samples from the crude soluble extract (S), the column flow-through (ft), the column wash, and the elution were subjected to SDS-PAGE along with known molecular weight standards (M). A, CD4L⅐LckN32; B, CD4S⅐LckN32; C, CD4L⅐Lck32; D, CD4m1⅐LckN32; E, CD4m2⅐LckN32; F, CD4m1ϩ2⅐LckN32. Bands representing the CD4 and Lck fusion proteins involved in each experiment are indicated. The high molecular weight bands seen in most of the phosphotyrosine elutions shown are bacterial proteins that purify by ion exchange. These proteins can be separated from CD4L⅐LckN32 complex by subsequent gel filtration. 422 in the CD4 tail. These are the same as the requirements demonstrated in vivo (20 -22). These data suggest that the interaction between CD4L and LckN32 indeed corresponds to the native CD4⅐Lck association, and that it is not an artifact of expression in E. coli.
The CD4S also copurified with LckN32 (Fig. 2B), but not as strongly. This observation suggests that the N-terminal arginine-rich segment lacking in CD4S but present in CD4L (residues 396 -402) may be necessary for proper association. Indeed, a somewhat larger deletion, encompassing residues 395-412 in murine CD4 (397-414 in humans) abolishes interaction with Lck in cell culture (20), consistent with the importance of the membrane-proximal region for stable interaction.
Large quantities of CD4L⅐LckN32 complex were purified using phosphotyrosine affinity chromatography (see also "Experimental Procedures") and concentrated to between 20 and 30 mg/ml. When CD4L⅐LckN32 complex prepared in this manner was subjected to gel filtration on a Superdex 200 sizing column (Amersham Pharmacia Biotech), the sample eluted in two primary peaks (Fig. 3A): a large peak at 13.8 ml, followed by a smaller peak at 16 ml (the peak observed at 8.5 ml is aggregated protein and contains neither CD4L nor LckN32). SDS-PAGE of this elution pattern (Fig. 3A) revealed that the first peak contained CD4L and LckN32 in roughly 1:1 stoichiometry. The smaller second peak contained unbound, monomeric LckN32. When the gel filtered complex was diluted and reapplied to the sizing column, it eluted in a single peak at 14.1 ml, indicating that the association is relatively stable (Fig. 3B). Comparison of elution volumes of the complex with those of known standards sets its molecular mass at ϳ90 kDa, 17 kDa higher than the expected mass of a 1:1 CD4L⅐LckN32 complex.
This effective mass could reflect either an extended structure or some tendency to form higher aggregates.
CD4L⅐LckN32 complex purified by phosphotyrosine affinity and gel filtration yielded a single major band when subjected to native PAGE (Fig. 4, lane 2). The band was identified as a 1:1 CD4L⅐LckN32 complex by experiments in which it was excised from the gel and analyzed by SDS-PAGE (data not shown). Preincubation with 0.5 mM EDTA led to complete dissociation of the complex and appearance of two new bands (Fig. 4, lane  3). SDS-PAGE analysis identifies the higher mobility band as unbound LckN32 and the lower as CD4L (data not shown). By contrast, neither 1% ␤-ME nor 25 mM dithiothreitol affects the complex (lanes 4 and 5, respectively). The zinc chelator 1,10ortho-phenanthroline also disrupts the CD4L⅐LckN32 association, but much less efficiently than does EDTA (approximately 50% dissociation at 14 mM 1,10-ortho-phenanthroline, data not shown). These variations in chelator effectiveness probably reflect the structural intricacies of the interaction itself.
We used ICP atomic absorption spectroscopy to detect the presence of metal ions in the CD4⅐LckN32 complex. Two samples of complex from separate bacterial preparations were purified by phosphotyrosine affinity chromatography and gel filtration. ICP spectroscopy was then used to determine the concentration of 25 separate metal ions relative to blanks taken from the same purifications. Zinc was the only metal to appear at levels above background in the CD4L⅐LckN32 samples (Table I). The protein concentration in both samples was determined by absolute amino acid analysis, and the number of zinc atoms per complex was calculated to be 0.90 for sample 1 and 1.05 for sample 2. As a control, a sample of MBP lacking a CD4 cytoplasmic tail was also subjected to ICP analysis, which gave only background levels of zinc (data not shown). Taken together with the published mutational studies and with our native gel experiments, these compositional data strongly support a model in which CD4 and Lck associate by cysteinemediated co-coordination of a zinc ion (Fig. 5). Zinc is a common structural element in small domains of intracellular proteins  The ICP-determined zinc concentration (in ppb) is displayed along with the background zinc level for that sample. Background levels of zinc were determined using sample buffer blanks taken from the same purifications. The protein concentration of each sample was determined by amino acid analysis. The micromolar concentrations of both protein and zinc were calculated assuming a molecular weight of 65.38 amu for zinc and 72.5 ϫ 10 3 amu for the protein (where amu is atomic mass units), which was taken to be 100% CD4L ⅐ LckN32 in a 1:1 stoichiometric complex. The density of the sample buffer (50 mM Tris, pH 7.5, 200 mM NaCl, 0.2% ␤-ME, 2 mM benzamidine) was assumed to be 1.0 g/ml.
(e.g. zinc fingers), but this role as a cross-linker represents a novel type of interaction.
As much as 90% of Lck is associated with CD4 in CD4 ϩ CD8 Ϫ cells (30). CD4 and Lck interact when expressed in non-lymphoid cell lines as well (20,21), indicating that T-cell-specific factors are not required in order for the complex to form. Other experiments have demonstrated that myristylation and membrane localization of Lck is not strictly required for interaction with CD4 (22). Furthermore, radioactive pulse-chase studies have indicated that the proteins may associate while CD4 is being processed in the endoplasmic reticulum, long before the coreceptor is transported to the cell membrane (21). These data suggest that the CD4-p56 lck interaction is both kinetically and thermodynamically favorable under physiological conditions. The results presented here support this hypothesis. CD4L and LckN32 apparently associate in bacterial cells, in the absence of myristylation, T-cell-specific proteins, and eukaryotic organelles. Moreover, gel filtration analysis of the purified complex indicates that there is little, if any, reversible equilibrium between bound and unbound states.
We can compare the CD4L⅐LckN32 association with what is known about the CD8␣-p56 lck association. The cytoplasmic tails of both CD4 and CD8␣ are shown in Fig. 6. The tail of CD8␣ is 10 residues shorter than that of CD4, and the CXCP motif is much closer to the membrane (8 residues away as opposed to 25). When the sequence TCQCPHR from CD4 and the sequence VCKCPR from CD8␣ were fused to the cytoplasmic tail of the viral transmembrane protein VSV-G (fusion proteins G4 and G8 in Fig. 6), the G8 fusion interacted with Lck, but the G4 fusion did not (22). In another experiment, a murine CD4 mutant (CD4-⌬412) was constructed in which residues 395-412 (397-414 in humans) were deleted (20). This deletion decreased the distance between the membrane and the CXCP motif to seven residues, just as in wild type CD8␣. The mutant also failed to interact with Lck. Thus, CD4 seems to have additional sequence requirements beyond the CXCP motif, whereas CD8␣ does not.
This conclusion is consistent with the coexpression and copurification data presented here. The CXCP motif is not enough to promote strong CD4⅐LckN32 association in bacteria; the arginine-rich segment (residues 396 -402), is also required. Further experiments will be required to determine if this sequence is functionally involved in the interaction, or if it is simply acting as a spacer to reduce the steric hindrance imposed by MBP. That the G4 fusion and the CD4-⌬412 mutant, which lack the arginine-rich stretch, both fail to associate with Lck suggests that this segment may indeed be significant (20,22). CD8␣ also bears an N-terminal arginine-rich sequence (residues 189 -193), but mutational studies in cell culture have shown that it is not required for association. FIG. 6. Comparison of the cytoplasmic regions of CD4 and CD8␣, and their interactions with Lck. A deletion of 18 residues in the arginine-rich region of the CD4 tail disrupts its interaction with Lck (20), and a VSV-G fusion protein with the seven-residue CD4 sequence TC-QCPHR does not interact with Lck (22). In contrast, a VSV-G fusion protein with corresponding region of CD8␣ does bind Lck (22). See "Results and Discussion" for further discussion.