Use of a general purpose mammalian expression vector for studying intracellular protein targeting: identification of critical residues in the nuclear lamin A/C nuclear localization signal

Proper cellular function depends on targeting proteins to specific subcellular locations (reviewed in von Heijne, 1990). In most cases, a short sequence of amino acids serves as a signal to direct a protein within the cell. Amino acid sequences that target proteins to the nucleus (Kalderon et al., 1984b; reviewed by Chelsky et al., 1989), the cytosolic side of the endoplasmic reticulum (Frangioni et al., 1992), Golgi apparatus (Swift and Machamer, 1991), lysosomes (Kornfeld, 1990) and mitochondria (Hurt et al., 1985), have been described. To demonstrate that a particular amino acid sequence directs the intracellular targeting of a protein, one must prove that it is both necessary and sufficient. A sequence is considered necessary if its removal or mutation disrupts proper targeting. To be considered sufficient, a sequence must be capable of redirecting a heterologous protein to the subcellular compartment of interest.

Two basic strategies have been employed to prove sufficiency: microinjection of a carrier protein covalently coupled to a peptide containing the putative targeting sequence (Goldfarb et al., 1986; Lanford et al., 1986); and transfection of an expression vector containing the putative targeting sequence fused in-frame to a heterologous protein (Kalderon et al., 1984b). A commonly employed heterologous protein for the latter strategy is chicken muscle pyruvate kinase (CMPK; Lonberg and Gilbert, 1983). This completely soluble, normally cytosolic protein is ideal for such studies, and a number of mammalian expression vectors containing it have been constructed (Kalderon et al., 1984b; Roberts et al., 1987; Dingwall et al., 1988; Gao and Knipe, 1992).

Unfortunately, none of these vectors offers both flexible cloning and versatile promoter control. To circumvent these problems, we have engineered the mammalian expression vector p3PK. This vector contains cloning sites for construction of N- and/or C-terminal fusions of putative targeting sequences to CMPK, and permits subcloning in cassette fashion to expression vectors controlled by different promoters. Since it is often desirable to use an identified targeting sequence as an affinity matrix for purifying anti-bodies and/or binding proteins, p3PK has also been designed such that targeting sequences fused to the C-terminus of CMPK can be rapidly subcloned into the pGEX bacterial protein purification system (Smith and Johnson, 1988). We demonstrate the versatility of p3PK by targeting CMPK to various organelles, and by using it to study the sequence requirements of the human nuclear lamin A/C nuclear localization signal.

Restriction endonucleases and other molecular biology reagents were purchased from Gibco-BRL. All other reagents were purchased from Sigma. Oligonucleotides were synthesized on a Milligen Cyclone Plus DNA synthesizer.

Construction of p3PK

The vector p3PK was designed to express amino acids 17 to 476 of chicken muscle pyruvate kinase (CMPK; Lonberg and Gilbert, 1983) under control of the SV40 promoter. It was constructed by digesting the plasmid pJ3Ω (Morgenstern and Land, 1990; ATCC catalog no. 37719) with HindIII and BamHI. A three-way ligation was then performed with the digested plasmid, a double-stranded oligonucleotide (d.s. oligo) adapter, and the BglII/BamHI fragment from the plasmid PK10/8 (Gao and Knipe, 1992). The d.s. oligo adapter contained a HindIII site, a consensus sequence for translation initiation (Kozak/Start site; Kozak, 1986), and a BglII site. It was formed by annealing two overlapping oligonucleotides:

5′ AGCTTACCATGGGTA 3′

3′ ATGGTACCCATCTAG 5′

The plasmid PK10/8 was a generous gift from Dr Min Gao (Bristol-Myers Squibb) and Dr David M. Knipe (Harvard Medical School). The BglII/BamHI fragment from this plasmid codes for amino acids 17-476 of chicken muscle pyruvate kinase (CMPK; Lonberg and Gilbert, 1983).

An in-frame stop codon for CMPK was added to the recombinant plasmid by digesting with BamHI and EcoRI, and ligating an overlapping d.s. oligo adapter containing a BamHI site, an inframe stop codon, an XbaI site for screening recombinants, and an EcoRI site:

5′ GATCCTGATGATCTAGAG 3′

3′ GACTACTAGATCTCTTAA 5′

The resulting plasmid contained two BglII sites, one of which would preclude cloning of N-terminal fusions. To remove the unwanted site, the plasmid was cut with limiting concentrations of BglII, treated with the Klenow fragment of DNA polymerase, and ligated back to itself. All constructions were verified by dideoxynucleotide fluorescence sequencing using an Applied Biosystems Model 373A fluorescence DNA sequencer.

The plasmid p3PK/MtLS was designed to contain the first twelve amino acids of the presequence of yeast cytochrome c oxidase subunit IV (Hurt et al., 1985) fused to the N-terminus of CMPK. It was constructed by digesting p3PK with HindIII and BglII, and ligating an overlapping d.s. oligo adapter containing (5′ to 3′): a HindIII site, an EcoRI site for screening recombinants, a Kozak/Start site, the putative targeting sequence, and a BglII site. The two oligonucleotides used to form the adapter were:

5′ AGCTTGAATTCACCATGCTTTCACTACGTCAATCTATAAGATTTTTCAAGA 3′

3′ ACTTAAGTGGTACGAAAGTGATGCAGTTAGATATTCTAAAAAGTTCTCTAG 5′

The plasmids p3PK/NL, p3PK/K₁→L₁ and p3PK/K ₂→L₂ contained wild type and mutant forms of the putative nuclear lamin A/C nuclear localization signal (NLS) fused to the C-terminus of CMPK. They were constructed by digesting p3PK with BamHI and EcoRI and ligating an overlapping d.s. oligo adapter containing (5′ to 3′): a BamHI site, the NLS (nucleotides 1456-1479 of human nuclear lamin A; McKeon et al., 1986), an in-frame stop codon, a HindIII site for screening recombinants, and an EcoRI site. Plasmid p3PK/NL codes for the wild type NLS and used the following overlapping d.s. oligos:

5′ GATCCACCAAAAAGCGCAAACTGGAGTCCTGAAAGCTTG 3′

3′ GTGGTTTTTCGCGTTTGACCTCAGGACTTTCGAACTTAA 5′

Plasmid p3PK/K₁→L₁, coding for a substitution in the nuclear lamin NLS (see below) used:

5′ GATCCACCCTGAAGCGCAAACTGGAGTCCTGAAAGCTTG 3′

3′ GTGGGACTTCGCGTTTGACCTCAGGACTTTCGAACTTAA 5′

Plasmid p3PK/K₂→L₂, coding for a substitution in the nuclear lamin NLS (see below) used:

5′ GATCCACCAAACTTCGCAAACTGGAGTCCTGAAAGCTTG 3′

3′ GTGGTTTGAAGCGTTTGACCTCAGGACTTTCGAACTTAA 5′

Cell culture of HeLa cells was carried out as previously described (Frangioni et al., 1992). DNA for transfection was prepared by alkaline lysis and CsCl banding as described (Ausubel et al., 1987). Twenty-four hours prior to transfection, 1.5×10⁵ cells in 2 ml of medium were plated onto sterile coverslips in 35 mm plastic Petri dishes. Cells were transfected using the modified calcium phosphate precipitation method (Chen and Okayama, 1987). Briefly, 4 μg plasmid DNA (per 35 mm dish) was mixed in 100 μl of 0.25 M CaCl₂. Then 100 μl of 2× BBS (50 mM N-,N-bis(2-hydroxyethyl)-2-amino-ethanesulfonic acid, pH 6.95, 280 mM NaCl, 1.5 mM Na₂HPO₄) was added to the mixture over 30 seconds, and the precipitate allowed to form for 20 minutes at room temperature. The total solution (200 μl) was applied directly to the cells which were incubated for 16 hours at 37°C and 3% CO₂, washed 3 times, re-fed with fresh medium, and allowed to incubate for an additional 32 hours at 37°C and 10% CO₂, before fixation as described below.

Whole serum from rabbits immunized with chicken muscle pyruvate kinase (Sigma, catalog no. P-6406) was the generous gift of Dr Morris Birnbaum (Harvard Medical School). Antibodies specific for CMPK were affinity-purified using a glutathione S-transferase (GST)/CMPK fusion protein covalently coupled to Affi-Gel 15 (Bio-Rad). The details of the purification and Affi-Gel coupling of GST/CMPK are described elsewhere (Frangioni and Neel, 1993). An 8 ml sample of whole rabbit serum was immunodepleted of GST reactive antibodies by incubation with an Affi-Gel 15 GST column (4 mg GST protein, 0.5 ml bed volume) for 1 hour at 4°C, with end-over-end rotation of the column. The flow-through was collected into a GST/CMPK column (1 mg protein, 0.5 ml bed volume) and incubated for 1 hour at 4°C. Bound proteins were washed with 40 bed volumes of PBS (8.4 mM Na₂HPO₄, 1.9 mM NaH₂PO₄, pH 7.4, 150 mM NaCl), 10 bed volumes of PBS adjusted to 500 mM NaCl, 20 bed volumes of PBS, 10 bed volumes of 0.1% NP-40 in PBS, and finally 30 bed volumes of PBS. Specific antibody was eluted with 100 mM glycine (pH 2.5), and 250 μl fractions were collected into tubes containing 1/10th volume 1 M Tris, pH 8.5. Pooled IgG fractions were analyzed by SDS-PAGE and supplemented with 100 μg/ml BSA and 0.2% azide, and stored at 4°C. Affi-Gel columns were stripped with 100 mM glycine, pH 2.2, re-equilibrated in PBS and stored at 4°C using 0.2% azide as a preservative.

Cells were washed gently 3 times with PBS at 37°C, fixed with 2% paraformaldehyde in PBS (pH 7.4) for 10 minutes at 37°C, washed once with PBS, quenched for 5 minutes at room temperature (RT) with 50 mM glycine in PBS (pH 7.4), washed once with PBS, and permeabilized with 0.1% NP-40 in PBS. All sub-sequent incubations and washes were carried out in 0.1% NP-40/PBS to reduce non-specific background. Blocking was achieved by incubation for 30 minutes at RT with 0.3% bovine serum albumin/0.1% NP-40/PBS. Cells were incubated with primary antibodies for 60 min at RT, washed 4 times with 0.1% NP-40/PBS, incubated with secondary antibodies for 45 min at RT, washed 3 times with 0.1% NP-40/PBS, washed twice with PBS, and mounted on glass slides with Mowiol (Calbiochem) containing 2.5% DABCO (Kodak) as described (Harlow and Lane, 1988). Affinity-purified anti-CMPK antibodies were used at a concentration of 1 μg/ml. Anti-mitochondria antibody mAb1273 (Chemicon, Inc.) was used at a dilution of 1:200. Secondary antibodies, FITC-conjugated goat anti-mouse (Tago) or TRITC-conjugated goat anti-rabbit (Tago), were used at a dilution of 1:350. Cells were photographed on an Olympus BH-2 microscope with Ektar 125 color film (Kodak) as described (Frangioni et al., 1992).

The p3PK vector, shown in Fig. 1, was constructed from the plasmids pJ3Ω and PK10/8 as described in Materials and Methods. This 4.9 kb mammalian expression vector expresses amino acids 17 to 476 of chicken muscle pyruvate kinase (CMPK) under the control of the SV40 early promoter. An SV40 polyadenylation site within the parent vector pJ3Ω provides the signal for polyadenylation. Replication of the plasmid in bacteria is mediated by a pBR origin, and ampicillin resistance is conferred by the Amp^r gene. Replication of p3PK in SV40 large T-containing mammalian cells is mediated by the SV40 origin of replication.

Short oligonucleotide adapters were inserted at each end of the CMPK cDNA (Fig. 1) to create multiple cloning sites. These adapters also provide a translation start signal (Kozak/Start site; Kozak, 1986) at the N-terminus of the CMPK cDNA, and a translation stop site at its C-terminus. The additional amino acids generated by these adapters have no effect on the expected subcellular localization of CMPK (see below). The multiple cloning sites can be used to generate fusion genes of a putative targeting sequence with the N- and/or C-terminus of CMPK. Although three restriction sites (XbaI, EcoRI, and ClaI) are available for use as the 3′ cloning site, we strongly recommend that the EcoRI site be used whenever possible in order to preserve the cassette features of p3PK (see below). Fig. 1 also shows the relative position of a unique SalI restriction site (amino acids 215-216 of CMPK), which can be used to insert targeting sequences into the middle of the CMPK molecule (discussed below).

CMPK/targeting sequence (TS) fusion genes constructed with p3PK can be subcloned in one step into several other useful vectors. For example, Fig. 2A demonstrates how different combinations of fusion genes can be subcloned into any of the pJxΩ (Morgenstern and Land, 1990) family of mammalian expression vectors. The pJxΩ vectors (ATCC catalog nos 37719-37724) differ only in the promoter/ enhancer combination controlling expression of the gene of interest (Fig. 2A, inset). To subclone from p3PK into a pJxΩ vector, the CMPK/TS fusion gene is excised from p3PK as a single unit by digestion with HindIII and EcoRI and ligated to a similarly digested pJxΩ family member. This swapping procedure can be accomplished for TS fused at either the N- and/or C-terminus of CMPK. We use the simple nomenclature of pxPK where x is the pJxΩ vector into which the p3PK fusion gene has been cloned (e.g. CMPK alone swapped into pJ4Ω creates p4PK, CMPK/TS swapped into pJ5eΩ creates p5ePK/TS, etc.).

C-terminal CMPK fusions can be directly subcloned into the pGEX-1λT and pGEX-2T bacterial fusion protein expression vectors (Smith and Johnson, 1988). As shown in Fig. 2B, the TS sequence alone can be cloned directly into the pGEX vectors at the same time it is cloned into p3PK, by digesting the pGEX plasmid with BamHI and EcoRI. Alternatively, the entire CMPK/TS fusion gene can be cloned into the pGEX plasmid by digesting the p3PK construct with BglII and EcoRI, and ligating the excised fusion gene to pGEX digested with BamHI and EcoRI. Sub-cloning from p3PK into pGEX retains the thrombin proteolytic site between GST and the insert. This allow the TS to be cleaved away from GST and purified separately.

To test the utility of p3PK in studying intracellular protein targeting, we asked whether a sequence previously shown to be a mitochondrial import signal in yeast could function in mammalian cells. DNA coding for the twelve amino acid pre-sequence from yeast cytochrome c oxidase subunit IV was cloned into p3PK as described in Materials and Methods. The resultant plasmid, which encodes a yeast mitochondrial import signal (MLSLRQSIRFFK) fused to the N-terminus of CMPK, was named p3PK/MtLS. HeLa cells, transiently transfected with p3PK or p3PK/MtLS, were fixed 48 hours post-transfection and processed for indirect immunofluorescence as described in Materials and Methods. Cells overexpressing CMPK were easily identified from untransfected, non-overexpressing cells due to the high signal to background ratio. Fig. 3 (column 1, row 1) displays the typical immunofluorescence pattern seen when CMPK alone was overexpressed in HeLa cells. Comparison of the fluorescence micrograph to the phase contrast image of the same field (Fig. 3, column 1, row 3) revealed that the signal was diffusely distributed throughout the cytoplasm and excluded the nucleus. However, when the putative twelve amino acid mitochondrial localization signal fused to the N-terminus of CMPK was overexpressed in HeLa cells, CMPK was seen to have a restricted subcellular localization (Fig. 3, column 2, row 1). The staining pattern excluded the nucleus, and within the cytoplasm had a tubular and punctate appearance consistent with localization to mitochondria. Fig. 3 (column 2, row 2) displays the same field stained with a mitochondria-specific monoclonal antibody and confirms localization of CMPK to the mitochondria. Control cells were transfected with p3PK/MtLS and incubated with each primary antibody separately, and both secondary antibodies together. Under these conditions, there was no detectable cross-reactivity of the secondary antibodies (data not shown).

A putative 8 amino acid nuclear localization signal from human nuclear lamins A and C was cloned into p3PK as described in Materials and Methods. The resultant vector, containing the wild type sequence TKKRKLES fused to the C-terminus of CMPK, was designated p3PK/NL. HeLa cells transiently transfected with p3PK/NL were fixed at 48 hours post-transfection and processed for indirect immuno-fluorescence as described in Materials and Methods. Fig. 3 (column 3, row 1) displays the resulting immunofluorescence pattern. By comparison of this fluorescence micrograph to the phase contrast image of the same field (Fig. 3, column 3, row 3), it was evident that CMPK localized exclusively to the nucleus. Only in extremely high level overexpressors was any CMPK signal detectable in the cytoplasm (data not shown).

Based on a previously proposed consensus sequence for nuclear localization signals (Chelsky et al., 1989), single point mutations were made in presumed critical residues of the nuclear lamin A/C nuclear localization signal (see below for discussion). One mutation substituted the first lysine with a leucine residue, changing the wild type sequence from TKKRKLES to TLKRKLES. The plasmid containing this mutant NLS fused to the C-terminus of CMPK was constructed as described in Materials and Methods and was labeled p3PK/K₁→L₁. As shown in Fig. 4 (column 2, row 1), this mutation did not prevent accumulation of CMPK in the nucleus, although a very slight additional cytoplasmic staining was seen in some cells. For comparison, the staining pattern seen with the wild type NLS/CMPK fusion is shown (Fig. 4, column 1, row 1). A second mutant NLS, in which the second lysine was substituted with leucine (TKLRKLES), was cloned into p3PK to generate the plasmid p3PK/K₂→L₂. When this construct was overexpressed in HeLa cells (Fig. 4, column 3, row 1), CMPK no longer localized to the nucleus, and exhibited a staining pattern identical to that seen with CMPK alone (compare to Fig. 3, column 1, row 1).

Using the mammalian expression vector p3PK, putative targeting sequences can be cloned as in-frame fusions to the N- and/or C-termini of the normally cytosolic protein chicken muscle pyruvate kinase (CMPK). There are two general strategies for constructing the desired fusion gene: PCR; and overlapping double-stranded oligonucleotide (d.s. oligo) adapters. The use of PCR in generating CMPK fusion genes, flanked by appropriate restriction sites, has been detailed previously (Frangioni et al., 1992). In general, PCR is the method of choice to clone sequences longer than 25 amino acids, while d.s. oligo adapters offer rapid cloning of shorter sequences. In designing the insert, one must be sure to add a HindIII site, a Kozak/start sequence, and a BglII site for N-terminal fusions, and a BamHI site, a stop codon, and an EcoRI site for C-terminal fusions (see Fig. 1). For PCR, these sequences are engineered into the primers. For both PCR and d.s. oligos, it is helpful to include an additional restriction site (e.g. EcoRI for N-terminal fusions, HindIII for C-terminal fusions) to assist with the screening of recombinants (see above for examples). Without such a screening site, it may be difficult to resolve short inserts.

Although not used in this paper, p3PK contains a unique SalI site which permits cloning of targeting sequences into the middle of the CMPK molecule. The SalI site, corresponding to amino acids 215-216 of CMPK, is contained within a slightly hydrophobic region of the molecule (data not shown). Negative results obtained using internal fusions, or for that matter N- or C-terminal fusions, must be interpreted with caution since flanking amino acids can affect the functioning of a targeting sequence (Roberts et al., 1987; Gao and Knipe, 1992).

The mammalian expression vector p3PK has been engineered to permit one step subcloning into the pJxΩ (Morgenstern and Land, 1990) family of expression vectors, thus offering a total of six different promoter/enhancer combinations for controlling CMPK expression. Expression levels of CMPK from p3PK (SV40 promoter) may not be high enough for immunodetection in some cell types. More often, though, overexpression of a fusion protein will saturate the subcellular compartment of interest resulting in ambiguous results. Expression from the vectors pJ5Ω and pJ5eΩ is under control of the glucocorticoid-inducible mouse mammary tumor virus (MMTV) promoter. pJ5eΩ offers slightly higher induction than pJ5Ω since it contains, in addition, a murine sarcoma virus (MSV) enhancer. By swapping the fusion gene of interest from p3PK into pJ5Ω (or pJ5eΩ), protein expression levels can be modulated by glucocorticoid treatment. Although saturation was not encountered in this study, we expect that it will be a common problem which can potentially be solved by proper promoter choice. When interpreting results though, one must also consider that immunodetection requires a threshold concentration of CMPK molecules. Although we have shown that CMPK can be used to visualize mitochondria (Fig. 3, column 2), the cytosolic face of the endoplasmic reticulum (Frangioni et al., 1992), and the nucleus (Fig. 3, column 3), saturation of other subcellular compartments may occur before the critical CMPK concentration is reached.

p3PK has also been designed to permit one step sub-cloning into the pGEX bacterial fusion protein vectors (Smith and Johnson, 1988). The pGEX plasmids express protein sequences of interest as C-terminal fusions to glutathione S-transferase (GST), which can be purified from crude bacterial lysates by glutathione agarose chromatography (Smith and Johnson, 1988; Frangioni and Neel, 1993). The GST fusion proteins can then be used to purify antibodies against the protein sequence of interest, and/or for probing mammalian cellular lysates for targeting sequence binding proteins (Kaelin et al., 1991). A GST fusion protein containing just the targeting sequence (TS) has the advantages of being structurally compact and potentially soluble. A GST fusion protein containing the entire CMPK/TS fusion gene has the advantage that the TS is expressed in the same context (CMPK) as that which functioned intracellularly. A significant disadvantage, however, is that GST/CMPK fusion proteins are extremely insoluble when expressed in bac-teria and require special conditions for solubilization and purification (Frangioni and Neel, 1993). For studies utilizing GST fusions to isolate target sequence binding proteins, a non-functioning point mutant (e.g. K₂→L₂ mutation, see also Kaelin et al., 1991) is helpful in identifying proteins that bind to the targeting sequence but which do not serve a targeting function.

In this study, we have shown the general usefulness of p3PK by redirecting CMPK to various intracellular compartments. The first twelve amino acids of the yeast cytochrome c oxidase subunit IV gene have previously been shown to be sufficient for import of a heterologous protein (dihydrofolate reductase) into the yeast mitochondrial matrix (Hurt et al., 1985). Our data suggest that this pre-sequence can, at the very least, be recognized by the translocation machinery of mammalian mitochondria, and cause co-localization of CMPK with mitochondria. However, it remains to be seen whether the yeast sequence is sufficient to direct translocation of CMPK across the outer and inner membranes of mammalian mitochondria.

We also targeted CMPK to the nucleus by constructing a C-terminal fusion with a putative 8 amino acid nuclear localization signal (NLS) from human nuclear lamins A/C. This sequence (TKKRKLES) contains a high degree of homology to the SV40 NLS (Table 1), and deletion of this region prevents nuclear localization of the lamins (Loewinger and McKeon, 1988). We provide the first direct proof that this sequence is sufficient for nuclear localization of a heterologous fusion protein. Moreover, a longer sequence from nuclear lamins A/C (SVTKKRKLE), when conjugated to human serum albumin, has recently been shown to promote nuclear import in an assay used to identify cytosolic factors involved in this process (Moore and Blobel, 1992).

Table 1 displays a comparison of the wild-type human nuclear lamin A/C nuclear localization signal with that of SV40 large T. Previous work has shown that substitution of lysine 128 of the SV40 large T NLS with non-basic, non-hydrophobic amino acids abolishes nuclear localization (Lanford et al., 1986; Kalderon et al., 1984a; Lanford et al., 1988). A previous analysis of published NLS sequences concluded that nuclear localization signals conform to the consensus sequence K R/K X R/K, where R/K specifies either R or K at that position, and X is either K, R, P, V, or A (Chelsky et al., 1989; Table 1). However, we found indirect evidence in the literature that substitution of a hydrophobic amino acid at the critical first lysine of the sequence KKRK might permit at least partial functioning of a nuclear localization signal (Loewinger and McKeon, 1988).

To test the hypothesis that a hydrophobic substitution of the first lysine would retain CMPK nuclear targeting, we fused the mutant sequence (TLKRKLES; Table 1) to its C-terminus. As shown above (Fig. 4, column 2), this mutant sequence was virtually indistinguishable from the wild type sequence in terms of its nuclear targeting. Changing the second lysine to leucine (TKLRKLES) resulted in complete loss of nuclear localization (Fig. 4 and Table 1). It will now be possible to test the hypothesis that an alternative minimal nuclear localization sequence has the form XKRK where X is a hydrophobic amino acid. Due to the importance of protein context on the functioning of NLSs (see above and Roberts et al., 1987; Gao and Knipe, 1992), results such as those obtained using isolated targeting sequences must ultimately be confirmed by mutation of the whole parent molecule.

In this paper, we have described strategies that should facilitate the identification of both amino acid residues critical for targeting and proteins that interact with targeting sequences. As more investigators use p3PK in their studies, we hope that a collection of p3PK vectors that target CMPK to various intracellular compartments will be formed. These vectors, all expressing the same heterologous protein (CMPK), will then form a resource for the characterization of new putative targeting sequences.

We thank Dr Min Gao (Bristol-Myers Squibb) and Dr David M. Knipe (Harvard Medical School) for their generous gift of plasmid PK10/8 and for many helpful discussions. We thank Dr Morris Birnbaum for the generous gift of anti-CMPK whole rabbit serum, and Dr Lan Bo Chen (Dana-Farber Cancer Institute) for mAb1273. We thank Drs Brian Burke, Frank McKeon and Morris Birnbaum (Harvard Medical School) for critical reading of this manuscript. We thank Ms Maureen Magane and Ms Celia Mokalled for administrative assistance. This work was funded by NIH Grant no. R01-CA-49152 (BGN) and a grant from the Rowland Foundation (JVF). JVF is a Markey Fellow in Developmental Biology at Harvard Medical School.

The vector p3PK has been deposited with the ATCC, and is available as catalog no. 77314.