From a FLOXIL3 gene trap integration library consisting of approximately 2 × 10⁶ unique proviral integrations [9], we have isolated 125 individual clones that converted to factor independence upon IL-3 withdrawal. Cellular sequences adjacent to U3Cre integration sites (gene trap sequence tags, GTSTs) were obtained from 102 clones by inverse PCR or 5'RACE using Cre-specific primers. Ninety clones yielded informative GTSTs (>30 nucleotides), of which 15 (17%) belonged to known genes, 10 (11%) matched single ESTs or full cDNAs with unknown function and 65 (72%) had no match within the public databases (Figure 2). Interestingly, the frequency of unknown genes recovered with this method was significantly higher than that obtained with a standard gene trap protocol and a similar gene trap vector. For example, only 782 (46%) of 1,687 GTSTs recovered from embryonic stem (ES) cells expressing a U3β geo gene trap virus were novel [7]. This difference is consistent with the ability of the Cre/loxP approach to enrich for transiently expressed genes that are difficult to isolate by standard methodology and therefore underrepresented in the nucleotide databases.

Figure 3a shows the structure of the proviral integration sites in the 15 previously characterized genes. Like earlier studies involving retroviral gene trap vectors with a reporter gene in the U3 region of the long terminal repeat (LTR), most integrations occurred in or near 5' exons, although integrations further downstream were observed occasionally. To confirm transcriptional induction by IL-3 withdrawal, polyadenylated RNAs were isolated from FDCP1 cells after various intervals of factor deprivation and hybridized on northern blots to gene-specific probes. Figure 3b shows several examples of transient gene inductions during ongoing cell death.

As summarized in Table 1, a significant number of the trapped known genes encode proteins involved in cell growth and survival. For example, the guanine nucleotide exchange factors (GEFs) for the Rho-family GTPases NET1 and h-PEM2 (KIAA0424) [10,11] belong to a large family of proteins that regulate cell growth and can transform cells in culture [12]. Transformation is dependent on the highly conserved dbl homology (DH) domain and in most cases requires amino-terminal truncation of the protein [13,14]. Interestingly, a novel splice variant of NET1, NET1A, which was isolated in these experiments, encodes an amino-terminally truncated protein that is transforming in its native form (Table 1). Thus, when overexpressed in NIH3T3 fibroblasts, NET1A but not NET1 produces transformed foci with typical rho morphology (F.W. and H.v.M., unpublished data). In addition to the DH domain, Rho-GEFs have a pleckstrin homology domain that connects the Rho signal transduction pathway to the phosphatidylinositol-3-phosphate kinase (PI3 kinase) signal transduction pathway [14]. The latter is responsible for mediating survival signals, particularly in hematopoietic cells, and is directly activated by IL-3 [15]. Surprisingly, two members of this pathway, the phosphatidylinositol 4/5 phosphate kinases (PIP5 kinases) type Iβ (which corresponds to human PIP5 kinase type Iα) and type Iγ, were induced by IL-3 withdrawal (Table 1). PIP5 kinase converts the lipid phosphatidylinositol-4-phosphate to phosphatidylinositol-4,5-phosphate (PIP2), which enhances cell survival by at least three mechanisms. First, PIP2 directly inhibits apoptosis by inactivating caspases 8, 9 and 3. Moreover, overexpression of PIP5 kinase type Iα in cultured cells prevents apoptosis induced by caspase 9 or tumor necrosis factor-α (TNF-α) [16]. Second, PIP2 is phosphorylated by Pl3 kinase to give phosphatidylinositol 3,4,5 phosphate (PIP3), which activates the survival kinase Akt. Akt prevents cell death by phosphorylating and inactivating the proapoptotic proteins BAD, caspase 9 and the forkhead family transcription factor FKHRL1, which induces the expression of Fas ligand [17,18,19,20]. Third, PIP5 kinase activates ATP-dependent potassium channels and thereby helps to maintain the polarity of the cell membrane [21]. As shown previously, apoptotic signals can cause membrane depolarization, which leads to cell death by directly inhibiting potassium channels [22].

Another gene with a putative survival function encodes the cell-cycle-regulated protein p38-2G4 (EBP1). Although less well characterized than the above proteins, P38-G4 is absent from Go cells and may stimulate cell-cycle progression [23,24].

Another class of survival genes inhibits apoptosis by interfering with Fas signaling. It includes the IL-12 receptor, which inhibits Fas-mediated apoptosis in certain types of T lymphocytes [25] and the transcription factor YB-1, which represses Fas gene expression [26]. Interestingly, YB-1 also activates the expression of the multidrug resistance gene (MDR1) whose product, the P glycoprotein, is an efflux pump that eliminates a variety of toxins from the cell, including proapoptotic anticancer drugs [27].

Similarly cytoprotective is VMAT2, the protein product of the vesicular monoamine transporter gene, which moves cytoplasmic monoamine transmitters into secretory vesicles and prevents cell death by sequestering proapoptotic molecules such as MPP⁺ and histamine [28,29].

Finally, IL-3 withdrawal induced transcription of several DNA repair genes. These were the mouse homologs of the bacterial alkB gene and of the yeast rec8 and rad50 genes. While AlkB protects DNA from damage by alkylating agents [30], Rec8 and Rad50 are DNA double-strand break repair proteins [31,32,33]. DNA double-strand breaks typically develop at the conclusion of any apoptotic process.

Only a few genes upregulated by IL-3 withdrawal (4 out of 15) were not directly related to cell death or survival and their induction may reflect some secondary events occurring during apoptosis (Table 1). However, the low frequency with which such genes were recovered underscores the specificity of the gene trap approach.

To further study transcriptional regulation during growth factor starvation, we analyzed the expression patterns of a large number of previously characterized genes in FDCP1 cells undergoing apoptosis by IL-3 withdrawal using Atlas mouse cDNA arrays (Clontech). Atlas arrays contain cDNA probes for 588 genes involved in development, oncogenesis, DNA repair, tumor suppression and apoptosis.

Figure 4a shows that of the 588 genes displayed on the arrays, only 12 were differentially expressed after factor deprivation. Differential expression of these genes was reproducible in independent array experiments and could be confirmed by northern blotting (Figure 4c).

Similar to the trapping results, several genes with survival functions were upregulated in the absence of IL-3. Among these were the survival kinase Akt, the c-kit receptor tyrosine kinase, the potent inhibitor of apoptosis FlipL, and the IL-3-receptor itself [34,35,36].

Moreover, survival functions seemed to be strengthened by the transient downregulation of the proapoptotic activin receptor [37,38]. In confirmation of previous results, however, the expression of several survival genes appeared IL-3 dependent. Thus, transcripts for antiapoptotic Pim-1, c-Myc and Bcl-XL proteins were repressed in the absence of IL-3 [39]. Finally, only two genes with arguably proapoptotic functions (that is, JunD and Chop10) were upregulated in these experiments (Figure 4c) [40].

Taken together, the gene trap and cDNA array results suggested that the transcriptional changes at the onset of apoptosis should favor cell survival rather than cell death.

The activation of protective mechanisms during the early stages of programmed cell death suggested that a transient exposure to an apoptotic stimulus might be favorable for cell survival. Accordingly, a transient factor deprivation should, by upregulating anti-apoptotic gene expression, reduce the FDCP1 cell's need for IL-3. To test this, we evaluated the ability of IL-3-deprived and non-deprived cells to form colonies in agar cultures that contained suboptimal amounts of IL-3. Figure 5a shows that cells pre-exposed to factor deprivation generated significantly more colonies than their non-exposed counterparts, indicating that survival mechanisms were induced by factor deprivation. Whereas growth stimulation was highest in cultures containing IL-3 at concentrations that normally fail to support colony formation, it gradually subsided with prolonged IL-3 deprivation.

In a second approach, we directly tested whether cells pre-exposed to factor deprivation develop increased apoptotic tolerance to subsequent factor starvation. To this end, IL-3 was initially withdrawn from FDCP1 cells for 1 or 2 hours. Subsequently, IL-3 was re-added for 4 hours only to be removed again for 6, 9 and 12 hours. At these time points cells were stained with annexin and apoptosis was quantified by flow cytometry. Figure 5b shows that pretreated cells died significantly more slowly that their non-pretreated counterparts, indicating that apoptotic prestimulation increases resistance to factor deprivation. Like the clonal survival assays, the cell protective effect was highest after a prestimulation of 2 hours.

On the basis of these results, we conclude that cell protective mechanisms are activated following growth factor withdrawal and transiently prevail prior to irreversible commitment to death.

Inverse PCR from genomic DNAs was performed using the Cre-specific primers described previously [9] and a combination of the blunt end restriction enzymes SspI and HincII. 5'RACE was performed with 1 μg of total RNA using the 5'RACE kit from Gibco-BRL and the manufacturer's instructions. The specific Cre reverse primers were as follows:5'-TGCGAACCTCATCACTCGTTG-3',5'-CATGTCCATCAGGTTCTTGCG-3' (nested). Amplification reactions were performed in a Perkin Elmer thermocycler and cloned into the p-GEMT-vector (Promega) as described previously [9]. Inserts were sequenced using an ABI 310 Genetic Analyzer (Perkin-Elmer).

FDCP1 cells were propagated at concentrations of 2 × 10⁵ cells/ml in Dulbecco's modified Eagle's medium (DMEM; Gibco), supplemented with 10% (v/v) fetal bovine serum (Boehringer-Mannheim) and 5 ng/ml recombinant mouse IL-3 (Peprotech) unless indicated otherwise. Agar cultures were an equal volume mixture of double-strength DMEM supplemented with 40% (v/v) fetal bovine serum and 0.6% (w/v) Bacto-agar (Difco) in double distilled water as previously described [9].

Apoptosis was measured in a FACScan flow cytometer after staining the cells with annexin using the Annexin-V-FLUOS detection kit (Roche) and the manufacturer's instructions.

Poly(A)⁺ RNA samples from FDCP-1 cells were reverse transcribed in the presence of ³²P-labeled dATP and hybridized to Atlas Mouse cDNA Expression Array (Clontech) according to the manufacturer's instructions. Filters were scanned with a PhosphoImager (Molecular Dynamics) and analyzed with the AtlasImageTM 1.5 software (Clontech). For northern blots, 2.5 μg of poly(A)⁺ was fractionated in 1% formaldehyde-agarose gels, transferred onto Hybond N membranes (Amersham) and hybridized to specific ³²P-dCTP-labeled probes produced by random priming (Rediprime, Amersham). Blots were exposed to Kodak-Biomax autoradiography film.

Cellular sequences adjacent to proviral integrations (gene trap sequence tags; GTSTs) were searched in the NCBI/NIH genomic databases [43] using the BlastN algorithm.