The appearance of aneuploidy in human cancers has been linked to defects in several processes1,2. Many studies suggest that aneuploidy arises from defects in the conserved spindle checkpoint that normally governs progression through mitosis4,5. In response to lack of tension or improper microtubule attachment at the kinetochores, a group of sensor proteins (Bub3, Bub1, BubR1, Mps1 and Mad2) releases a diffusible signal that inhibits the ubiquitin ligase activity of the anaphase promoting complex (APC) or cyclosome6. APC/cyclosome function is required for sister chromatid separation and cytokinesis7. Although mitotic checkpoint defects are often observed in cancer cells challenged with microtubule poisons, and inactivation of checkpoint components produces aneuploidy in yeast and mammalian cells8,9,10, loss-of-function mutations in spindle checkpoint genes are rarely observed in human tumours11,12.

Adenovirus E1A is a potent viral oncoprotein that acts, in part, by inactivating the Rb gene product and deregulating the E2F transcription factors3. In performing a series of microarray experiments, we observed that E1A significantly increased the expression of MAD2 (Z.N., V.M. and S.W.L., unpublished observations), a crucial component of the spindle checkpoint that associates with the APC/cyclosome, and prevents its activation7,13. Northern and western blot analyses of MAD2 transcript and protein confirmed its upregulation by E1A in mouse embryo fibroblasts (MEFs) (Fig. 1a, compare lanes 1 and 2 or 3 and 4, respectively). Similarly, MEFs isolated from Rb-/- mice displayed increased Mad2 levels compared to Rb+/+ controls (Fig. 1a, compare lanes 5 and 6). IMR90 human fibroblasts expressing either E1A (Fig. 1b, compare lanes 1 and 2) or E2F-1 (Fig. 1b, compare lanes 1 and 3) also expressed elevated Mad2 levels, implying that deregulation of the Rb/E2F pathway induces Mad2 through a conserved mechanism.

Figure 1: Mad2 is an E2F target.
figure 1

a, MAD2 mRNA (lane 1 and 2) and protein (lane 3 and 4) were assessed in vector (V) or E1A-expressing MEFs by northern blotting and immunoblotting, respectively. p53 induction indicates E1A activity and α-tubulin (α-tub), β-actin or 28S RNA serve as loading controls. Mad2 protein levels were also compared in wild-type (Rb+/+) and Rb-/- MEFs (lanes 5 and 6). b, Immunoblotting of the indicated proteins in IMR90 cells infected with vector- (V), E1A- (E) or E2F-1-expressing adenoviral constructs. Cyclin A (Cyc A) is an established E2F target. c, Chromatin immunoprecipitation15 using an E2F-1 specific antibody in vector- (V) and E1A (E)-expressing IMR90 cells. Duplex PCR amplifications were performed using primers surrounding the putative E2F sites in the Mad2 promoter or the known sites in the cyclin A promoter, along with primers capable of amplifying Cyclin A genomic sequence (STS) lacking E2F binding sites (see Supplementary Fig. S1a). Input corresponds to PCR reactions containing 0.5% of total chromatin used in immunoprecipitation reactions. d, Relative luciferase signal in U-2OS cells co-transfected with empty vector (LPC) or E2F-1 (LPC–E2F) and a luciferase reporter with (MAD2Luc) or without (Luc) Mad2 promoter sequences. E2F-1 plasmid concentration (triangle). The Casp7luc reporter is the positive control. Values represent the mean ± s.e.m. (n = 3). e, Immunoblotting for Mad2 in control (C) or serum-starved IMR90 cells (t = 0), and at various times after serum addition. The percentages of cells in G2/M phase at the indicated time (shown below). f, Immunoblotting for Mad2 in IMR90 cells synchronized in S phase and released from a double-thymidine blockade. Histone H3-P is shown as a marker of mitosis. g, Mad2 expression by LSC in NIH-3T3 cells, where the gate represents the threshold considered positive for Mad2 based on the equivalent isotype control (inset). Cell populations of 2c and 4c DNA content are indicated.

The above results are consistent with the possibility that MAD2 is an E2F target. Concordantly, previous global chromatin immunoprecipitation/microarray studies suggested that E2Fs can bind the MAD2 promoter in normal fibroblasts14. Indeed, analysis of the MAD2 genomic sequence showed several putative E2F-binding sites in the MAD2 promoter (see Supplementary Fig. S1), and subsequent chromatin immunoprecipitation (ChIP) analysis confirmed in vivo binding of E2F-1 to their predicted sites in E1A-expressing cells (Fig. 1c, lane 2). Moreover, a MAD2 genomic fragment containing the E2F sites (Supplementary Fig. S1) conferred E2F responsiveness to a luciferase reporter following transient transfection into IMR90 (data not shown) or U-2OS cells (Fig. 1d) in a manner comparable to the caspase-7 promoter, an established E2F target15. Therefore, MAD2 is a physiological transcriptional target of E2F.

Many E2F targets are cell-cycle regulated3. Therefore, we examined the cell cycle distribution of Mad2 expression in normal synchronously and asynchronously cycling cells. The level of Mad2 was undetectable in quiescent IMR90 cells (Fig. 1e, t = 0) and increased after cells entered S phase following serum addition, reaching maximum levels in G2/M, albeit slightly after cyclin A. The increase in Mad2 was confined largely to mitosis, as assessed by co-expression with phosphorylated histone H3 following release from a double-thymidine-induced S-phase arrest (Fig. 1f), and by laser scanning cytometry (LSC) analysis of Mad2 expression in unsynchronized NIH-3T3 fibroblasts (Fig. 1g). Presumably, Mad2 expression is regulated during the normal cell cycle as part of a mechanism that coordinates the expression of spindle checkpoint proteins with the onset of G2/M.

E2F target genes are often constitutively expressed in cells with Rb pathway defects. Concordantly, Mad2 was expressed at aberrantly high levels throughout the cell cycle in IMR90 cells expressing E1A (Fig. 2a), which deregulates endogenous E2F, and in cells where Rb was suppressed by stable RNA interference (RNAi) (shRb16, Fig. 2a). Of note, Mad2 levels in E1A- and shRb-expressing IMR90 cells were similar to those produced by enforced expression of MAD2 (Fig. 2b, c). As in normal fibroblasts, Mad2 was confined to G2/M in the Rb-functional T24 bladder carcinoma17 and HCT116 colon carcinoma lines (Fig. 2d). However, in bladder and colon carcinoma lines with inactive Rb, such as HT1197 (ref. 17) and SW480 (ref. 18), respectively, Mad2 was expressed at high levels throughout the cell cycle (Fig. 2d). Thus, cells with Rb defects often aberrantly express Mad2.

Figure 2: Deregulating the Rb/E2F pathway produces aberrant Mad2 expression.
figure 2

a, Biparametric analysis of Mad2 expression and DNA content by LSC in IMR90 fibroblasts expressing a vector control (Vector), E1A, shRb or MAD2. b, c, Western blot analysis of Mad2 in the cells described in part a. Ran and α-tubulin (α-tub) protein levels serve as loading controls. d, LSC analysis of Mad2 expression and DNA content in T24, HCT116, HT1197 and SW480 tumour derived lines. Numbers indicate the total percentage of Mad2 positive cells.

Because the Rb pathway is frequently disrupted during tumorigenesis, we asked whether Mad2 is overexpressed in human tumour specimens. Retinoblastomas, which invariably lose Rb function and deregulate E2F, expressed high levels of Mad2 relative to normal retina (Fig. 3a, upper panels). Similarly, Mad2 was significantly increased in many bladder carcinomas when compared to normal urothelium (data not shown), particularly in the more advanced tumours, where Rb inactivation is common19 (Fig. 3a, middle panels). The elevated Mad2 levels observed in human tumours were not merely due to the increased proliferation of cancer cells, as Mad2 was not always detected in highly proliferating normal and tumour samples (see Supplementary Fig. S2).

Figure 3: Abnormal Mad2 expression in human tumours.
figure 3

a, Immunohistochemistry for Mad2. Top: nearly undetectable Mad2 expression in normal retina (left) compared to high levels in retinoblastoma (right). Middle: an example of a Mad2-negative transitional cell carcinoma of the bladder (TCC, left) compared to a Mad2-positive invasive TCC (right). Bottom: a Mad2-negative (left) and positive (right) neuroblastoma. b, Comparison of E2F-1 and MAD2 transcript levels in a cohort of 106 neuroblastic tumours and 12 neuroblastoma cell lines20. The inset represents the logarithm of MAD2 mRNA expression (X axis) versus E2F-1 (Y axis), indicating a near-linear correlation. c, Kaplan–Meier survival curves stratified by MAD2 microarray-based mRNA levels, with prognostic significance by the log-rank test using median values as cut points. Neuroblastoma cases with low (black curve) and high (blue curve) MAD2 mRNA levels are shown.

A clear link between deregulated E2F activity and aberrant Mad2 expression was observed in neuroblastoma. Here, MAD2 messenger RNA levels correlated tightly with E2F1 mRNA expression (an E2F member but also a prototypical E2F target gene), as shown by analysis of a microarray database containing information on 106 neuroblastoma tumours and 12 neuroblastoma cell lines20 (Fig. 3b; Kendall tau correlation coefficient: 0.568; P < 0.0001) and validated by immunohistochemistry (Fig. 3a, lower panels). Moreover, aberrant Mad2 expression was associated with MYCN (a proto-oncogene of the myelocytomatosis (MYC)-box genes) overexpression and amplification (which can drive E2F constitutive activation21) and poor patient prognosis (Fig. 3c). In fact, high Mad2 expression was more tightly associated with survival than MYCN amplification (P = 0.001 for MAD2; P = 0.0215 for MYCN). Interestingly, MAD2 also clusters with tens of genes capable of predicting disease outcome in breast cancer patients22. Therefore, aberrant Mad2 expression is linked to Rb pathway defects in human tumours, where it can be an effective indicator of patient prognosis.

Most advanced human cancers are aneuploid, and loss of Rb function has been associated with mitotic defects23. Hence, it is paradoxical that cells with defective Rb express higher levels of a protein that should facilitate proper checkpoint responses. We hypothesized that misexpressed or elevated Mad2 might compromise mitotic events, predisposing cells to genomic instability. Therefore, we examined the DNA content of IMR90 fibroblasts expressing E1A, shRb or Mad2, all of which expressed similarly high levels of Mad2 throughout the cell cycle (see Fig. 2). Biparametric LSC analysis comparing BrdU incorporation and DNA content demonstrated that E1A-, shRB- and MAD2-transduced IMR90 or NIH-3T3 cell populations accumulated cells with greater than 4c DNA content during the course of cell division (Fig. 4a). These effects were exacerbated by treatment with nocodozole, which acutely engages the spindle checkpoint (data not shown).

Figure 4: Cells with defective Rb or elevated Mad2 display chromosomal instability.
figure 4

a, Immunocytochemical detection of BrdU versus DNA content by LSC in asynchronous cultures of IMR90 (upper panels) or NIH-3T3 (lower panels) infected with E1A, shRB or MAD2. b, FISH analysis on IMR90 cells, with centromeric probes for chromosomes 1 and 17 and DAPI counterstaining. c, Quantification of ploidy alterations derived from E1A, shRb or MAD2 expression. FISH in IMR90 and HCT116 transduced cells (upper table). Shown are percentages of cells with abnormal number of positive signals for one or both probes. Percentage of infected cells exhibiting tetraploidy in metaphase spreads (lower table).

Analogous results were obtained using fluorescent in situ hybridization (FISH) to track specific chromosomes. The E1A-, shRB- and MAD2-expressing populations each accumulated cells with aberrant chromosome numbers (Fig. 4b, c). In contrast, IMR90 cells expressing a control vector retained normal chromosome content upon serial passaging (Fig. 4a–c, see Vector or V) and responded normally to nocodazole (data not shown). Chromosome counting of metaphase spreads confirmed that IMR90 and HCT116 cells with defective Rb or elevated Mad2 are frequently aneuploid (Fig. 4c). No centrosome amplification was detected in E1A- or MAD2-transduced fibroblasts by γ-tubulin (see Supplementary Fig. S3b) or β-tubulin immunofluorescence staining (spindle shape analysis; data not shown), suggesting that the observed mitotic defects are not due to the effects of cyclin E on centrosome duplication24,25. However, as cyclin E is an E2F target, centrosome amplification may exacerbate the chromosomal instability associated with Rb loss. Nevertheless, misexpression of Mad2 is sufficient to produce this phenotype.

We next examined mitotic progression of individual cells in E1A-, shRb-, MAD2- or vector-transduced populations co-expressing green fluorescent protein (GFP)-tagged histone (H2B–GFP), by phase contrast and fluorescence videomicroscopy. Whereas normal NIH-3T3 cells underwent an orderly and rapid mitosis, cells expressing E1A or shRb took an unusually long time to finish the process, displaying marked difficulties in completing cytokinesis (Fig. 5a–c; Supplementary Fig. S4a and S5, compare video A with C and D) and frequent defects in chromosome segregation (Fig. 5c; Supplementary Fig. S5, compare video B with G). Cells expressing Mad2 alone displayed similar defects (Fig. 5c; Supplementary Fig. S5, compare video A with E and F), although the chromosome segregation abnormalities were more pronounced, perhaps owing to the higher Mad2 levels that these populations expressed (Supplementary Fig. S5, compare video B with H). Similar phenotypes were also observed in E1A-, shRb- and MAD2-transduced HCT116 (Supplementary Fig. S5, videos I and J). In each cell population, the frequency of aberrant mitoses decayed after serial passaging, perhaps owing to lethality of cells with aberrant chromosome content and/or to cellular adaptation to elevated Mad2. Nevertheless, consistent with a prolonged mitosis, cells aberrantly expressing Mad2 displayed a delay in the degradation of securin and cyclin B (Supplementary Fig. S4b, c), whose sequential ubiquitination by the APC/cyclosome is inhibited by Mad27,13 and required for chromosome segregation and cytokinesis, respectively6,7,26.

Figure 5: Aberrant Mad2 expression contributes to the mitotic defects associated with Rb loss.
figure 5

a, Average time needed to complete mitosis in NIH-3T3 cells expressing control vector, shRb, E1A or MAD2 assessed from prometaphase (disruption of the nuclear envelope and chromatin condensation) to completed cytokinesis and reattachment of daughter cells. n, number of cells analysed. P values determined by Student's t-test. b, Time course of a normal mitotic division occurring in vector-transduced NIH-3T3 cells (v). Illustrative examples of prolonged mitoses occurring in cells expressing E1A, shRb or MAD2 (E1A- and MAD2-transduced NIH-3T3 cells shown). c, A fluorescence videomicroscopy series of H2B–GFP transfected cells showing normal chromosome segregation (vector-transduced NIH-3T3, v) and representative examples of the slow and abnormal chromosomal excision patterns observed in cells expressing shRb, E1A or MAD2 (shRb and MAD2 shown). d, Immunoblotting for Mad2 in HCT116 cells expressing E2F-, E1A- or Vector together with either a MAD2 hairpin (shMAD2) or a hairpin control (sh-control) as indicated. e, f, Cytogenetic analyses by karyotyping (e) or FISH (f) of HCT116 cells co-expressing E2F or E1A with a MAD2 shRNA (two sequence-independent hairpins, shMAD2A and shMAD2B) or a shRNA control. ≥200 interphase nuclei were analysed by FISH. Values were compared for statistical significance using the exact test of Fisher: P = 0.015 and P < 0.0001, comparing E2F-transduced cells coinfected with shMAD2A or shRNA control, at 0 and 12 h nocodazole (Noc.); P = 0.006 and P = 0.0007, for shMAD2B versus shRNA control, at 0 and 12 h respectively.

Complete downregulation of Mad2 in cells with intact Rb produces aneuploidy after a premature anaphase27. However, if Mad2 upregulation contributes to chromosome instability, we reasoned that a partial suppression of Mad2 in Rb defective cells would have the opposite effect—perhaps reducing the occurrence of aneuploid cells. Therefore, we used stable RNAi to partially suppress Mad2 in control, E2F- or E1A-transduced HCT116 cells. Consistent with a previous report27, stable expression of two independent short-hairpin RNAs (shMAD2) in cells with low Mad2 levels further reduced Mad2 expression and increased the percentage of aneuploid cells (data not shown). In contrast, co-expression of shMAD2 with E2F or E1A significantly reduced Mad2 to nearly normal levels (Fig. 5d, compare lanes 3 with 7 or 2 with 8) and reduced the fraction of aneuploid cells (Fig. 5e, f) without affecting E1A or E2F-1 levels (Supplementary Fig. S4d). Cells coexpressing shMAD2 and either E2F-1 or E1A were proliferating and showed similar S phase content compared to cells harbouring a control shRNA (data not shown). Therefore, although other Rb targets may contribute to chromosome instability, aberrant Mad2 activity is important for this process.

We suggest that aberrant expression of Mad2 arising as a result of Rb pathway defects can produce a hyperactive spindle checkpoint, thereby altering the sequence of mitotic events and the accuracy of chromosome transmission. Persistent inactivation of the APC/cyclosome, mediated by Mad2 signalling, delays the degradation of securin and cyclin B. Moreover, overexpression of securin (Pds1, PTTG)—observed in many different tumour types—produces aneuploidy and displays transforming activity in vivo and in vitro28. Interestingly, the behaviour of cells with aberrant Mad2 expression is reminiscent of cells treated for prolonged periods with nocodozole—following spindle checkpoint activation, these cells eventually adapt and enter the next cell cycle without a proper mitosis29. Together, our results demonstrate that deregulated Mad2 expression contributes to mitotic alterations and chromosome instability in the context of an altered Rb/E2F pathway.

Our data illustrate how aneuploidy in human cancers can arise as a byproduct of the Rb pathway defects that accompany tumorigenesis. During normal cell proliferation, the cell cycle is hardwired to the spindle checkpoint through E2F, presumably to ensure that Mad2 expression is tightly controlled. As a consequence, mutations that disrupt Rb function produce deregulated E2F activity leading to inappropriate proliferation, but also aberrant Mad2 expression and, eventually, aneuploidy. Hence, our results provide one explanation for why loss-of-function mutations in spindle checkpoint genes are uncommon in human cancers11,12. Our model also implies that aneuploidy is not necessarily a selected trait of cancer cells but can be an inadvertent consequence of defects that uncouple mitotic control from normal cell cycle progression. Nevertheless, this byproduct of Rb loss might subsequently increase the mutation rate, thereby fuelling further tumour evolution.


Cells, gene transfer and drug treatment

Human tumour cell lines (T24, HT1197, HCT116, SW480 and U-2OS), normal human fibroblasts (IMR90) and murine immortal fibroblasts (NIH-3T3) were obtained from the ATCC (American type culture collection, Rockville). Primary MEFs were isolated and cultured as described15. Cells were infected with high-titre recombinant retroviruses expressing E1A (LPC–12S), E2F-1 (LPC–E2F1), shRb (MSCV–shRB)16, shMAD2 (MSCV–shMAD2A or B)27, shmurMAD2 (shRNA control, MSCV–shmurMAD2)27 or MAD2 cDNA (Babe–MAD2) and selected with puromycin (2 µg ml-1)15. Adenoviruses E1A, E2F-1 and GFP control were used at a multiplicity of infection of 100 PFU cell-1 as described15. Nocodazole (Sigma) was used at 200 nM and 400 nM for normal fibroblast or tumour cell lines, respectively.

Protein and mRNA expression

Protein expression was assessed by immunoblotting using 30 µg of total cell lysate15. Blots were probed with antibodies directed against hsMad2 (clone 48, BD Biosciences), E1A (M58, BD Biosciences), p53 (CM5, Novocastra), E2F-1 (KH95, Santa Cruz), Cyclin A (BF683, Santa Cruz), Ran (sc-1156, Santa Cruz), Pds1 (ab-1, DCS-280, Neomarkers), cyclin B1 (05-373, Upstate), phospho-histone3 (06-570, Upstate), APC2 (ab-1, Neomarkers) or α-tubulin (B-5-1-2, Sigma). Anti-mouse, anti-rabbit (Amersham) or anti-goat horseradish peroxidase (HRP) (Sigma) were used as secondary antibodies and proteins were visualized using an ECL detection system (Amersham). mRNA levels were assessed by northern blotting using 10 µg of total RNA15.


Cells grown as monolayers on microscope-chamber slides were washed, fixed and incubated with the respective antibodies30: hsmad2 (BD Biosciences); alpha-tubulin FITC conjugated (clone DM 1A) and γ-tubulin (Sigma); and anti-Rb (Pharmingen, BD Laboratories). Mouse IgG served as an isotype negative control. Cells were washed and incubated with FITC-conjugated rabbit anti-mouse F(ab′)2 or anti-rabbit (DAKO), then washed and incubated with 5 µg ml-1 of propidium iodide (PI; Molecular Probes) and 0.1% RNase A (Sigma) in PBS before measurement by LSC (CompuCyte) as indicated30. A minimum of 3,000 cells were analysed per slide as previously described30.

Cell cycle analysis after low-serum synchronization of IMR90 cells was performed as previously described15. Mitotic index was quantified by measuring MPM2 expression (anti-MPM2, Upstate Biotechnology) versus DNA content (PI) by FACS27. For BrdU incorporation assays, slide-grown cells were incubated in 10 µM BrdU (Sigma) for 1 h 30 min (human cell lines) or 3 h (normal fibroblasts) immediately before collection. Cells were fixed with 80% ethanol overnight and incubated with FITC-conjugated anti-BrdU antibody after denaturing the DNA, following the manufacturer's instructions (Pharmingen, BD Laboratories). BrdU incorporation was measured by LSC, with the equivalent FITC-conjugated isotypic control as a baseline.

Chromosome analyses

Bicolour FISH was performed using probes for chromosomes 1 and 17 centromere-specific alphoid region (Vysis)20. The number of hybridization signals for each probe was assessed in a minimum of 200 interphase nuclei with strong and well-delineated contours. Chromosome analysis of cells in metaphase was performed as described27.


Sections were deparaffinized, treated with 1% H2O2, immersed in boiling 10 mM citrate buffer for 15 min and incubated in 10% normal horse serum for 30 min at room temperature. Anti-hsmad2 (Transduction Laboratories, BD Laboratories; dilution 1:100) and anti-human Ki-67 (MIB-1, DAKO; dilution 1:1000) antibodies were used. Samples were then incubated with biotinylated anti-mouse IgG at 1:500 dilution (Vector Laboratories) followed by avidin–biotin peroxidase complexes (1:25; Vector Laboratories) for 30 min. Diaminobenzidine was used as the chromogen, and hematoxylin as the nuclear counterstain.

Human tumour samples and gene expression analysis

Twelve cases of retinoblastoma and 95 bladder tumours on a tissue array were analysed by immunohistochemistry. In addition, 12 neuroblastoma cell lines and 106 neuroblastic tumours were studied20. Total RNA was extracted and cDNA was labelled and hybridized to Affymetrix human U95 oligonucleotide arrays as described20. Comparisons of E2F-1, MAD2 and MYCN expression levels were performed by Kendall's tau correlation test. Kaplan–Meier survival curves were generated by using the SPSS program (SPSS). E2F-1, MYCN and MAD2 expression were tested for prognostic significance (P < 0.05) by the log-rank test using median values as cut points. Protein expression was validated on a tissue microarray containing 50 neuroblastoma cases.