Isolation of differentially expressed human cDNA clones: similarities between mouse and human embryonal carcinoma cell differentiation

Mouse embryonal carcinoma (EC) cells are the stem cells of teratocarcinomas (Stevens, 1970) and have been used extensively as models for early mouse embryonic development (Martin, 1980; Hogan et al. 1983). Although analogous human cell lines do exist (reviewed Andrews et al. 1983), not until recently have the conditions to induce extensive somatic differentiation been defined. In addition, whereas mouse EC cells are regarded to some extent as developmentally similar to cells of the primitive ectoderm (Brinster, 1974; Evans et al. 1979), the developmental stage to which human EC cells correspond has been difficult to determine (Andrews et al. 1983). NT2D1 is a human EC cell line capable of extensive differentiation in response to retinoic acid (RA), producing several novel cell types (Andrews, 1984; Gonczol et al. 1984; Skowronski & Singer, 1985; Hauser et al. 1985; La Femina & Hayward, 1986; Lee & Andrews, 1986; Fenderson et al. 1987). However, except for neurones (Lee & Andrews, 1986), these other cells have not been characterized.

For this study, a cDNA library was constructed from NT2D1 cells that had been induced to differentiate in response to RA. (NT2D1 cells treated with 10 ⁻⁵ M-RA for 7 days are referred to here as ANT2D1 cells). Sequences corresponding to mRNAs that increase in abundance following RA-induced differentiation of NT2D1 cells were isolated by differential cDNA hybridization (St. John & Davis, 1979; Williams & Lloyd, 1979). The differentiation of NT2D1 cells is poorly characterized as compared to mouse EC models, for example F9 (reviewed by Hogan et al. 1983). It would therefore be expected that a comparison to previously characterized mouse EC models would yield information, defining similarities, if any, and differences between mouse and human EC cell differentiation. Such a comparison would apply the premise that homologous sequences similarly regulated between mouse and human during differentiation relate to common pathways of differentiation and control in mammals (see review of Gurdon, 1987). Whereas sequences found to be unique to human EC cell differentiation may be the result of sequence divergence or point to developmental differences between mice and humans.

To identify cDNA recombinants that are both conserved in sequence and similarly regulated following mouse and human EC cell differentiation, the approach of differential cDNA hybridization was again applied. The mouse model used to compare mouse and human EC differentiation is the well-characterized F9 system (reviewed by Hogan et al. 1983). The mouse EC cell F9 can be induced to differentiate in vitro either to cells resembling visceral endoderm (F9VE; Hogan et al. 1981) or parietal endoderm (F9PE; Strickland et al. 1980). It has been suggested, however, on the basis of detection of epidermal growth factor, transferring (Adamson & Hogan, 1984) and higher than expected levels of homeobox-containing mRNAs (Colberg-Poley et al. 1985) in F9 compared with other mouse EC cell lines that F9 may be predetermined. For this reason, the mouse EC cell line PCC41 was also included in the differential screenings.

By comparing differentiation in both mouse and human EC cells, it should be possible to isolate sequences that are similar between species in both sequence and temporal expression. Such sequences may be developmentally important in mammalian embryogenesis (Gurdon, 1987); also by the study of human EC cells, insights, particularly into early human development, may be gained.

NT2D1 was cloned by Andrews (1984) from Tera2 (Fogh & Trempe, 1975). F9 (Bemstine et al. 1973) and PCC4azal (Jakob et al. 1973) are mouse EC cell lines originally derived from the same transplantable teratocarcinoma, OTT6050 (Stevens, 1970). PCC41RA (Benham et al. 1983) is a ouabain-resistant, RA-adapted mouse EC cell line, subclone of PCC4azal.

All cell lines were maintained at 37 °C in Dulbecco’s modified Eagle’s medium (high glucose formulation), supplemented with 10% fetal calf serum. Cells were subcultured at, or near, confluence with 0 ·25 % (w/v) trypsin, 0-2 mm-EDTA every 2 to 3 days. NT2D1 cells were subcultured to a minimum cell density of 7 × 10⁴ cells cm². F9 cells were grown on tissue culture flasks treated with 0 ·1% gelatin.

NT2D1 cells were seeded at 7 ×10⁴ cells cm² in the presence of 10 ⁻⁵ M-RA (all-trans retinoic acid, dissolved in DMSO at 10 ⁻²M). After 6 days, the medium was replaced, maintaining the same regime. On the 7th day, cultures were harvested for RNA isolation and FACS analysis. NT2D1 cultures so treated are referred to here as ΔNT2D1.

F9 cells were differentiated to parietal endoderm-like cells as described by Strickland et al. (1980). Monolayer cultures of F9 were treated with RA at 10 ⁻⁷ M plus dibutyryl cyclic AMP at 10 ⁻³M. This regime was maintained for 7 days, after which the cells were harvested for RNA isolation. F9 cultures so treated are referred to as F9PE.

For F9 visceral endoderm-like differentiation, the procedure of Hogan et al. (1981) was used. Monolayer cultures of F9 cells were lightly ‘trypsinized’ so as to cause the detachment of cells as small clumps. These clumps were plated out onto bacteriological grade plates in the presence of 5 × 10 ⁻⁸ M-RA. The ‘embryoid bodies’ that formed were maintained in this regime for 15 days before being harvested for RNA isolation. These cells are referred to as F9VE.

GTC27 is a human EC cell line, GTC72 is a human cell line resembling visceral endoderm, GTC44 is a human cell line resembling parietal endoderm (Pera et al. 1987); cells were provided by Dr Martin Pera.

Cells were prepared and labelled for cell surface antigens FACS analysis as described in Andrews et al. (1981). X63 (Kohler & Milstein, 1975) was used as a negative control monoclonal antibody. For internal antigen staining (antibodies TROMA1 and TROMA3), single-cell suspensions were permeabilized and fixed in ice-cold 1:1 (v/v) ethanol: water. After washing in phosphate-buffered saline, fixed permeabilized cells were treated as for surface antigen staining. SSEA1 was detected using the monoclonal 480 (Solter & Knowles, 1978), SSEA3 was detected using the monoclonal 631 (Shevinsky et al. 1982).

RNA was prepared using the guanidinium thiocyanate method of Chirgwin et al. (1979). Enrichment for the poly(A)⁺ fraction was by a single round of oligo d(T)-cellulose chromatography (Maniatis et al. 1982).

Northern blot analysis used 1 μg of glyoxalated poly(A)⁺ RNA (Thomas, 1980) electrophoresed through 1·2% agarose gels and transferred to Gene Screen Plus (NEN) membrane. Prehybridization and hybridization was in 50 % (v/v) formamide, 1m-NaCl, 10% (w/v) dextran sulphate and 1 % (w/v) SDS at 42°C. After 18 h hybridization, filters were washed at 50°C in 0·1×SSC, 0·1% SDS for 40min. Autoradiography was at −70 °C with preflashed XAR5 (Eastman Kodak) film between Du Pont Lightening Plus intensifying screens (Laskey & Mills, 1975) for 3–30 h.

For subsequent reprobing, Northern blot filters were stripped of previous probes by incubating at 95 °C in 0·1×SSC, 1% SDS for 20min.

³²P-labelled DNA probes were prepared using the random oligo-primer method of Feinberg & Vogelstein (1984) to a specific activity of >10⁸ cts min⁻¹μg⁻¹.

A cDNA library of >10⁷ recombinants was constructed from poly(A)⁺ RNA derived from ANT2D1 described as in Wiles et al. (1988), using the EcoRI site of the lambda insertion vector NM1149 (Murray, 1983). The ³²P-cDNA probes were prepared from oligo d(T) primed poly(A)⁺ RNA, as described previously (Wiles et al. 1988) to a specific activity of >10⁸ctsmin⁻¹/μg⁻¹, with an average length of 400 bp. 5000 recombinants of the ANT2D1 cDNA library (unamplified) were screened using four replicate filters (22 × 22cm sheets of Hybond-N, Amersham) prepared as described by Benton & Davis (1977). Duplicate filters were probed with ³²P-cDNA (3–6×10⁶ ctsmin⁻¹ ml⁻¹) derived from NT2D1 and ANT2D1 under conditions described in Wiles et al. (1988). After 18 h, filters were washed in 0·l×SSC, 0·1 % SDS at 65°C for 1 h. Autoradiography was as described for Northerns, but with exposures of 1, 3 and 5 days.

For subsequent rescreening, the replicate filters were stripped of previous probes by incubating in 0·4 m-NaOH at 45°C for 30min. Filters were sequentially rescreened in duplicate, with ³²P-cDNA (3–6×10⁶ctsmin⁻¹ ml⁻¹) derived from F9, PCC41RA, F9PE and F9VE and washed in 1 × SSC, 0·1 % SDS at 65 °C for 1 h. Autoradiography was as described for Northerns with exposures of 1–3 days. A differential signal was scored when ‘filters’ showed in duplicate a markedly stronger signal with ³²P-cDNA derived from differentiated EC cells compared with undifferentiated.

After treating NT2D1 cells with 10⁻⁵ M-RA for 7 days (ΔNT2D1 cells), cell surface and internal antigens were examined by FACS analysis (data summarized in Table 1). Before RA treatment, NT2D1 expresses high levels of the cell surface antigen SSEA3 (Shevinsky et al. 1982), high levels of a cytokeratin antigen recognized by TROMA1 (Kemler et al. 1981), low levels of the cytokeratin antigen recognized by TROMA3 (Brulet et al. 1980) and low levels of the cell surface antigen SSEA1 (Solter & Knowles, 1978). Following RA induction of NT2D1, cell surface antigen SSEA3 decreases, while SSEA1 expression increases; these changes are consistent with Andrews (1984). Cytokeratin expression was also found to change, with a decrease in TROMA1 antigen and an increase in TROMA3 antigen. The TROMA3 antigen may correspond to a 40x10³Afr keratin polypeptide that Damjanov et al. (1984) showed increased following RA treatment of human EC NTera2 cl.4D3 cells (both NTera2 cl.4D3 and NT2D1 are derived from the same parental line, NTera2). Together with these phenotypic changes, there is also a marked change in morphology. All these changes are consistent with RA-induced differentiation of NT2D1 cells.

A λcDNA library was constructed from poly(A)⁺ RNA isolated from ANT2D1 cells (Wiles et al. 1988). 5000 ANT2D1 cDNA library recombinants were screened for clones corresponding to mRNAs that are more abundant following RA-induced NT2D1 differentiation. By comparisons of matched autoradiographs, it was found that 56(1·12 %) ANT2D1 cDNA recombinants detected a marked increase in signal with ³²P-cDNA derived from ANT2D1 as compared with NT2D1 (see Fig. 1 for examples). For brevity recombinants that behave in this way are referred to as being regulated.

Recombinants homologous to mouse sequences were identified by repeated hybridizations using the same filters. Each filter was rehybridized sequentially with P-cDNAs derived from PCC41RA, F9, F9VE and F9PE. Washing criterion was relaxed to allow detection of hybrids with greater than 77 % homology (see Anderson & Young, 1985); examples of this screen are shown in Fig. 1. Comparison of signals obtained with ³²P-cDNAs derived from mouse EC cells (PCC41RA and/or F9) with those of differentiated F9 cells (F9VE and/or F9PE) show 60 (1·2 %) of the ANT2D1 recombinants are homologous to mRNA species that are more abundant following differentiation of F9 as compared with undifferentiated mouse EC cells. Of these homologous regulated recombinants, 15 (0·3 %) were also common to those regulated following NT2D1 differentiation. Of recombinants regulated in the human system but not in the mouse, about two thirds gave a signal with mouse-derived ³²P-cDNAs but were not detected as being regulated following F9 differentiation. A summary of these results is given in Table 2.

Six recombinants giving strong differential signals when probed with ³²P-cDNAs derived from ANT2D1 as compared with NT2D1 were isolated. These recombinants are referred to as SO2, SO5, SO6, SO7, SO9 and SO13. Upon EcoRI cleavage, recombinants SO5 and SO13 each released two inserts. These double inserts are the result of either ligation of unrelated cDNAs during library construction or the presence of intrinsic EcoRI sites within the cDNA molecules (note, the cDNA library was constructed with EcoRI methylase-treated cDNA). The larger of each insert is suffixed with an ‘A’, the smaller with a ‘B’. These data are summarized in Table 3. Inserts were cross hybridized with each other and homology was detected between inserts SO2 and SO5A; homology was also detected between inserts SO7 and SO13B.

Northern blots of poly(A)⁺ RNA derived from NT2D1 and ANT2D1 were probed with each of the cDNA inserts (examples shown Fig. 2, tracks 1 and 2). A summary of the Northern blot analysis together with the result of densitométrie scans for the NT2D1 and ΔNT2D1 autoradiographs are given in Table 3. It should be noted that, because of the kinetics of Northern blot hybridization, increases in abundance of mRNAs as estimated by densitométrie scans of Northern autoradiographs are probably underestimates (see review of Anderson & Young, 1985). In addition, it is assumed that β-actin abundance is similar between cells, not varying significantly following differentiation. With the exception of insert SO5B, all clones detect mRNAs that increase in abundance following RA-induced NT2D1 differentiation. The increase in abundance ranges from 4-fold with SO13A, to at least 75-fold with SO9 and SO5A.

cDNA inserts were hybridized to mouse poly(A)⁺ RNAs derived from the mouse cell lines PCC41RA, F9, F9VE and F9PE (Fig. 2B,C,E and F, tracks 3, 4, 5 and 6; data summarized in Table 3). Three different NT2Dl-derived clones detect homologous sequences in mouse poly(A)⁺ RNAs (SO2/SO5A, SO13A and SO7/SO13B), while no signal was detected with clones SO6 and SO9.

SO clones that detect homologous mouse sequences all detect an increase in mRNA abundance following F9 differentiation. The cDNA clones SO7/SO13B (Fig. 2F, tracks 4, 5 and 6) and SO13A (Fig. 2B, tracks 4, 5 and 6) detect mouse mRNAs that are more abundant following F9 differentiation; the signal from SO13A being more intense in F9PE compared with F9VE poly(A)⁺ RNAs. SO5A/SO2 detect an mRNA only in F9VE (Fig. 2C, tracks 4, 5 and 6), with no signal being detected in PPC41RA, F9 or F9PE poly(A)⁺ RNAs.

SO5A was singled out for further investigation as it detected a large increase in steady-state mRNA abundance following NT2D1 differentiation and similarly with F9 differentiation to F9VE but not to F9PE. Fig. 3 panel A shows the result of probing total RNAs derived from NT2D1 cells subjected to 10⁻⁵ M-RA over time with SO5A. The abundance of mRNAs homologous to SO5A increases markedly after 2 to 3 days exposure to RA and remains constant over 7 days (compared to the abundance of β-actin and Class-I HLA mRNAs).

Direct analysis of SO5A expression in human embryos was not possible. However, Pera et al. (1987) have isolated human cell lines that have characteristics of visceral endoderm (GTC72) and parietal endoderm (GTC44). When Northern blots of total RNA derived from these cells were probed with SO5A a signal was detected only in the parietal-endoderm-like line, GTC44 (Fig. 3 panel B). This is in apparent contrast to the mouse F9 model where expression was detected only in F9VE (visceral-endoderm-like).

During a screen of sequences known to be regulated in mouse, it was noted that SPARC (cDNA clone pPE30, Mason et al. 1986) detects on Northern analysis of NT2Dl-derived poly(A)⁺ RNAs a major band at 2·2 kb plus a minor band at ∼2·7kb. This signal was stronger in ANT2D1 than NT2D1 poly(A)⁺ RNAs. To check if any of the SO clones were homologous to SPARC, inserts were hybridized with the SPARC cDNA probe pPE30. Homology was detected between pPE30 and cDNA clone SO13A. When used as a probe on Northern blots of poly(A)⁺ RNAs (Fig. 2B), SO13A detects a 2·2kb signal following RA-induced differentiation of F9 to F9VE and F9PE; the stronger signal being seen in F9PE. This is in agreement with the result observed by Mason et al. 1986 with the mouse SPARC. However, the insert SO13A is ∼3 kb. A blunt-end ligation of unrelated cDNAs during library construction may account for this, or the cloning of a possible precursor. The latter possibility is favoured, as following longer exposure of Northern blots of poly(A)⁺ RNA from NT2D1 and ANT2D1, probed with SO13A, additional bands at 2·7 kb and ∼3·3kb are detected. Swaroop et al. (1988) also detected a 2·7 kb band in human-derived RNAs, this reflects the use of another polyadenylation site.

Common pathways of mammahan differentiation would be expected to elicit coordinated expression of genes that are homologous between species (Gurdon, 1987). The identification of cDNA recombinants that are homologous and similarly regulated in both human and mouse is therefore not surprising. This data is, however, in contrast to previous work where it has generally been differences between mouse and human EC cells, not their similarities, that have been most evident (e.g. Andrews et al. 1983).

It should be remembered that with the differentiation of F9 and NT2D1 EC cells it is not just their EC phenotype that is alike, but also the reagent used to induce this differentiation; i.e. retinoic acid. Therefore, it is possible that sequences regulated in common in these two model systems are the direct result of RA treatment and not differentiation per se. An example of this has been observed by Deschamps et al. (1987) who described the ‘inappropriate’ expression of a mouse homeobox-containing gene FI24.1. They showed that in RA-treated mouse EC cells expression of H24.1 appears to be correlated to the presence of RA, as H24.1 was not detected by Northern blot analysis of RNA derived from cultures that had been differentiated spontaneously. However, this result does not preclude appropriate gene expression under the influence of RA. For example, with the differentiation of F9 to PE-like and VE-like cells, changes in vitro appear to correlate with in vivo changes; laminin, collagen IV, tissue plasminogen activator and SPARC are detected in F9PE and mouse parietal endoderm; in addition, alphafetoprotein, transferrin, SPARC and cytokeratins have been detected in both F9VE and normal mouse visceral endoderm (reviewed by Hogan et al. 1981; see also Mason et al. 1986). In the case of RA-induced differentiation of NT2D1, inappropriate expression of genes due to RA cannot account for all the differential signals detected. For instance, increases in mRNA homologous to SO5A would have been detected in both F9VE and F9PE, not just F9VE. Also upon differential cDNA hybridization, not all the recombinants recognized as regulated with RA-induced NT2D1 differentiation were similarly regulated following F9 differentiation. These differences were not generally due to a lack of homology between mouse and human sequences, as the majority of these recombinants gave signals but showed no detectable change following F9 differentiation. Furthermore mRNA homologues to SO5A were detected in CTG44, a cell line that was not exposed to RA.

Five different clones were isolated that detected sequences more abundant in ΔNT2D1 than in NT2D1 poly(A)⁺ RNAs. When these cDNA clones were used as probes on Northern blots of poly(A)⁺RNA derived from PCC41RA, F9, F9VE and F9PE, two general patterns were found; (i) clones that do not detect homologous mouse sequences (SO6 and SO9), and (ii) those that do detect homologous mouse sequences (SO2/SO5A, SO7/SO13B and SO 13A). The presence of the former class may indicate that some abundance changes in mRNA species are specific under conditions used here to human EC cell differentiation in vitro. This may be a reflection of the heterogeneous nature of NT2D1 RA-induced differentiation, or suggest that human EC cells represent a different embryonic state than mouse EC cells and thus exhibit different developmental capabilities (Andrews et al. 1983). The presence of the latter class of homologous regulated clones shows, however, that there are similarities in gene expression between mouse and human EC cell differentiation. Further work needs to be done to establish the relevance of these two classes of clones.

Clone SO5A detects by Northern blot analysis, an mRNA that is 75-fold more abundant in ANT2D1 as compared to NT2D1. With the mouse F9 model system, a mouse homologue to SO5A is detected in RNAs derived from F9VE but not in F9PE. This specificity may give a clue as to the phenotype/s induced following NT2D1 differentiation. It also provides a further marker for F9VE differentiation and thus possibly for normal mouse visceral endoderm. The hybridization of SO5A to RNA derived from GTC44, a human PE-like cell line is surprising in the light of the apparent specificity seen with F9VE. This result may reflect basic differences between mouse and human extraembryonic tissues, or possibly the lack of characterization of the cell lines involved. Interestingly, when mouse SPARC (M. F. Pera and R. Krumlauf, personal communication) or the putative human homologue (SO13A; data not shown) are hybridized to GTC44 and GTC72 RNAs, steady-state expression is only marginally higher in the human PE-like cell line GTC44. This is again in contrast to mouse where high levels of SPARC mRNAs are detected in PE and PE-like cells (Mason et al. 1986). Further experiments, including DNA sequencing and the comparing of expression of these clones in mouse VE and PE with human embryonic tissues may shed light on these differences.

One clone, SO13A, has been tentatively identified by hybridization as a human homologue of the mouse gene SPARC. Mason et al. (1986) found that SPARC mRNA is induced to moderate levels following F9 differentiation to F9VE and to high levels following differentiation to F9PE in vitro. The induction of this gene in vitro reflects the differentiation of mouse primitive endoderm to visceral endoderm and parietal endoderm in vivo (Mason et al. 1986). The identification of a known mouse gene which is regulated both in vitro and in vivo during the differentiation of primitive endoderm indicates that cross species cDNA differential hybridization may assist in the isolation of homologous sequences involved in early embryogenesis. Also, it can be argued that genes that are conserved in sequence and in temporal and spatial expression are likely to be developmentally important (Gurdon, 1987). Future experiments could make effective use of this using additional differential cDNA screenings, the ³²P-cDNAs being derived from a number of different characterized cells and tissues (of the same, or different, species); thus allowing one to survey a large number of cDNA recombinants simultaneously. The autoradiographs produced would lead to the composition of a form of ‘pictorial ROT curve’ for each tissue/phenotype. By such repeated cDNA differential hybridizations it should be possible to construct a molecular phenotype of NT2D1 differentiation.

From the data presented, it can be surmized that RA-induced differentiation of F9 and NT2D1 share some elements in common. As F9 differentiation in vitro also shares features with mouse development in vivo, genes that are homologous and similarly regulated between NT2D1 and F9 may be developmentally important. Differences, however, are also highlighted and these may point to marked differences between mouse and human early development. Further experiments are required to correlate the changes observed with NT2D1 differentiation with the in vivo situation; for example, using abortus material and human VE-like and PE-like cell fines. However, the isolation of these clones represent some of the first reagents that can be used for the study of human developmental molecular biology.

I thank Dr Peter N. Goodfellow in whose laboratory this work was done; also Drs B. L. M. Hogan and E. F. Wagner for helpful discussions and assistance. Cell lines and RNA preparation for Fig. 3 was done in collaboration with Drs M. F. Pera and S. Cooper. Monoclonal antibodies TROMA1 and TROMA3 were provided by Dr R. Kemler. I must also thank my colleagues B. Williams, B. Pym, C. Pritchard, G. Knott, P. J. Goodfellow, S. Carson and G. Banting for their patience and help.