Loss of Tumor Suppressor STAG2 Promotes Telomere Recombination and Extends the Replicative Lifespan of Normal Human Cells

Telomeres, the specialized structures at chromosome ends, are composed of TTAGGG repeats and the shelterin protein complex (1). Because of the end replication problem and nucleolytic processing that occurs with each cell division, telomeres shorten to a limited threshold that signals checkpoint-dependent entry into senescence, a state of permanent growth arrest (2). However, if checkpoint function is compromised cells will continue to proliferate. This continued proliferation leads to crisis and cell death, unless cells can counteract the progressive loss of telomeric DNA. Eighty-five percent of human cancers achieve this by upregulating telomerase (3, 4). The remaining 15% of cancers activate alternative lengthening of telomeres (ALT; ref. 5), a recombination-based mechanism marked by high rates of telomere sister chromatid exchange (T-SCE; refs. 6, 7). ALT cells exhibit defective (persistent) sister telomere cohesion into mitosis that contributes to the high level of T-SCE (8).

Sister chromatid cohesion is established in S phase during DNA replication to keep sisters in proximity for recombination and repair (9). Cohesion is removed in mitosis in a two-step process. During G₂ and early mitosis cohesin is removed from telomeres and arms by the prophase pathway (10). A small amount of cohesin is protected from removal and remains at centromeres holding sister chromatids together (against the spindle forces) until the metaphase to anaphase transition. Centromere cohesion is essential for the faithful distribution of sister chromatids and defects can led to chromosomal missegregation and aneuploidy (11). Cohesion is mediated by the cohesin ring, a tripartite structure composed of SMC1, SMC3, and SCC1, and a peripheral SA subunit found as two isoforms SA1 and SA2, which are required for telomere and centromere cohesion, respectively (12, 13). SA1 is distinguished by a unique N-terminal 72 amino acid domain that contains a DNA-binding AT-hook motif and binds the shelterin subunit TRF1 (14), which facilitates its association with telomeric DNA (15).

The gene encoding SA2 (STAG2) is frequently mutated in human cancer, whereas mutation of the gene encoding SA1 (STAG1) is rare (16, 17). STAG2 mutations are most common in bladder cancer, but are also found in Ewing sarcoma, melanoma, glioblastoma, and other cancers. In fact, STAG2 is one of only 12 genes found to be significantly mutated in four or more cancer types (18). Approximately 85% of STAG2 mutations are truncating and often result in loss of expression, indicating STAG2 as a tumor suppressor gene (16). However, it is not known how loss of SA2 promotes tumorigenesis. The initial report identifying STAG2 mutations in cancer showed (using isogenic human cultured cell systems) that STAG2 mutations can lead to aneuploidy (17). However, subsequent studies on naturally occurring tumors showed limited correlation between STAG2 mutations and aneuploidy (19). Here we set out to determine how STAG2 tumors maintain sister chromatid cohesion and how STAG2 inactivation contributes to tumorigenesis.

VM-CUB-3 (20), SK-ES-1, SK-NEP-1, TC-32, H4, 42MGBA, 42MGB STAG2 knock-in, and HCT116 STAG2 knockout (17) were obtained from Dr. Todd Waldman, Georgetown Medical School (Washington, DC) in 2015. UM-UC-3 (20), SK-N-MC (21), and U138MG (17) were obtained from ATCC in 2014. LOX IMVI (17) was obtained from Frederick National Laboratory in 2014. HCT116, HEK293T, and BJ were obtained from ATCC. SuperHeLa (22) was obtained from Dr. Joachim Lingner, EPFL Lausanne (Lausanne, Switzerland) in 2006. HeLa1.2.11 (23) and HTC75 (24) were obtained from Dr. Titia de Lange, Rockefeller University (New York City, NY), in 1999 and tested for mycoplasma (Invitrogen Testing Kit). Cells were store in liquid nitrogen, thawed, and passaged for a few population doublings prior to use. Cells were grown under standard conditions. Where indicated cells were grown in the presence or absence of the telomerase inhibitor BIBR 1532 (Selleckchem) at a final concentration of 20 μmol/L. Cells were passaged twice every 7 days and BIBR 1532 was freshly added at each passage.

Cells were fixed and processed as described previously (25). Briefly, cells were isolated by mitotic shake-off, fixed twice in methanol:acetic acid (3:1) for 15 minutes, cytospun (Shandon Cytospin) at 2,000 rpm for 2 minutes onto slides, rehydrated in 2× SSC at 37°C for 2 minutes, and dehydrated in an ethanol series of 70%, 80%, and 95% for 2 minutes each. Cells were denatured at 75°C for 2 minutes and hybridized overnight at 37°C with subtelomeric FITC-conjugated (16ptelo, 13qtelo, or 4ptelo) and TRITC-conjugated 13q14.3 deletion (arm) and 10 cen (centromere) probes from Cytocell. Cells were washed in 0.4× SSC at 72°C for 2 minutes, and in 2× SSC with 0.05% Tween 20 at room temperature for 30 seconds. DNA was stained with 0.2 μg/mL DAPI. BJ cells were incubated in 30 ng/mL nocodazole (Sigma) for 16 hours prior to shake-off. Mitotic cells were scored as having telomeres cohered if 50% or more of their loci appeared as singlets, that is, one out of two or two out of three. Quantification of FISH analyses is the average of two independent experiments (n = approximately 50 cells each; range 33–78 cells) ± SEM.

Chromosome orientation FISH (CO-FISH) was performed as described previously (26). Briefly, cells were treated with 10 μmol/L BrdUrd:BrdC (3:1) for 24 hours prior to harvest and colcemide (0.5 μg/mL final concentration) was added 8 hours prior to harvest. Cells were harvested by trypsinization, hypotonically swollen in 10 mmol/L Tris, pH 7.4, 10 mmol/L NaCl, and 5 mmol/L MgCl₂ for 10 minutes at 37°C, and fixed twice for 15 minutes in methanol/acetic acid (3:1). Metaphase spreads were prepared by dropping fixed cells on coverslips followed by centrifugation at 1,000 rpm for 10 seconds in an Eppendorf 5810R centrifuge. Coverslips were air-dried overnight and were rehydrated in PBS for 5 minutes, treated with RNase A (0.5 μg/mL in PBS) for 10 minutes at 37°C, stained with Hoechst 33258 (0.5 μg/mL in 2× SSC) for 10 minutes at room temperature and exposed to 365-nm UV light (Stratalinker 1800 UV irradiator) for 45 minutes. The BrdUrd/dC substituted DNA strand was digested with Exonuclease III (10 U/mL) for 30 minutes at 37°C. The slides were dehydrated through an ethanol series (75%, 95%, and 100%) and hybridized with TAMRA-OO-(TTAGGG)₃ PNA probe in hybridization solution for 2 hours at room temperature. The slides were washed for few seconds with 70% formamide/10 mmol/L Tris-HCl pH 7.2, 0.1% BSA, and incubated with FITC-OO-(CCCTAA)₃ PNA probe in hybridization solution for 2 hours at room temperature. The slides were washed twice for 30 minutes with 70% formamide, 10 mmol/L Tris, pH 7.5, and 0.1% BSA, and were washed three times for 5 minutes in 0.1 M Tris, pH 7.5, 0.15 M NaCl, and 0.08% Tween 20. After the washes, cells were dehydrated in an ethanol series of 70%, 95%, and 100%, and DNA was counterstained with 4,6-diamino-2-phenylindole (DAPI) 0.5 μg/mL. CO-FISH on HCT116 WT or SA2KO cells was performed as described above, but the spreads were hybridized with only one probe Cy3-conjugated (CCCTAA)₃ PNA probe (PNAbio).

Cells were resuspended in four volumes of TNE buffer [10 mmol/L Tris (pH 7.8), 1% Nonidet P-40, 0.15 M NaCl, 1 mmol/L EDTA, and 2.5% protease inhibitor cocktail (Sigma)] and incubated for 1 hour on ice. Suspensions were pelleted at 8,000 x g for 15 minutes. Equal amounts of supernatant proteins (determined by Bio-Rad protein assay) were fractionated by SDS-PAGE and analyzed by immunoblotting.

Immunoblots were incubated separately with the following primary antibodies: goat anti-SA1 C-term A300-157A (0.5 μg/mL; Bethyl Laboratories, Inc.); goat anti-SA1 N-term A300-156A (0.5 μg/mL; Bethyl Laboratories, Inc.); rabbit anti-SA1 172 (0.5 μg/mL; ref. 8); rabbit anti-SA2 N-term A300-580A (0.20 μg/mL; Bethyl Laboratories, Inc.); goat anti-SA2 C-term A300-158A (0.20 μg/mL; Bethyl Laboratories, Inc.); rabbit anti-SMC3 – ChIP grade ab9263 (0.5 μg/mL; Abcam); rabbit anti-SMC1 A300-055A-T (1.0 μg/mL; Bethyl Laboratories, Inc.); rabbit anti-Rad21 A300-080A (0.5 μg/mL; Bethyl Laboratories, Inc.); rabbit anti-ATRX sc15408 (0.5 μg/mL; Santa Cruz); mouse anti-DAXX VMA00318 (1:1,000; Bio-Rad); and mouse anti-α-tubulin (1:10,000; Sigma Aldrich, T5768), followed by horseradish peroxidase-conjugated donkey anti-rabbit or anti-mouse IgG (Amersham) or donkey ant-goat (Bethyl Laboratories, Inc.; 1:2,500). Bound antibody was detected with Super Signal West Pico (Thermo Scientific).

Cells were lysed as above and supernatants precleared with protein G-Sepharose rotating at 4°C for 30 minutes. Nonspecific protein aggregates were removed by centrifugation and the supernatant was used for immunoprecipitation analysis or fractionated directly on SDS-PAGE. For immunoprecipitation of endogenous SMC3 from HCT116 (WT or SA2KO and HeLaI.2.11 (shvec or shSA2), cells lysis was performed in 0.7 mL (per one 10 cm dish) of TNE buffer. Equal amounts of supernatant proteins (determined by Bio-Rad protein assay) were used as starting material. Ten micrograms of the supernatant was used as input and the rest was incubated with a 3 μg of SMC3 antibody or rabbit IgG for 3 hours. Protein G beads were added for 2 hours and washed five times with 1 mL TNE buffer. Samples were fractionated by SDS-PAGE and analyzed by immunoblotting.

Cells were fixed in 2% paraformaldehyde in PBS for 10 minutes at room temperature, permeabilized in 0.5% NP-40/PBS for 10 minutes at room temperature, blocked in 1% BSA/PBS, and incubated with mouse anti-γH2AX #05-636 (0.2 μg/mL; Millipore) and rabbit anti-53BP1 NB 100-304 (0.1 μg/mL; Bovus Biologicals) or mouse anti-PML sc966 (2.0 μg/mL; Santa Cruz) and rabbit anti-TRF1 415 serum (1:1,000). For staining RAD51 foci, cells were permeabilized in Triton X-100 buffer (0.5% Triton X-100, 20 mmol/L Hepes-KOH at pH 7.9, 50 mmol/L NaCl, 3 mmol/L MgCl₂, 300 mmol/L sucrose) for 5 minutes at room temperature, fixed in 3% paraformaldehyde (in PBS, 2% sucrose) for 10 minutes at room temperature, permeabilized in Triton X-100 buffer for 10 minutes room temperature, blocked in 1% BSA/PBS, and incubated with rabbit anti-RAD51 (4 μg/mL; Santa Cruz Biotechnology, 8349) for 2 hours. Primary antibodies were detected with FITC-conjugated or TRITC-conjugated donkey anti-rabbit or anti-mouse antibodies (1:100; Jackson Laboratories). DNA was stained with DAPI (0.2 μg/mL).

To detect RAD51 foci colocalization with telomeres, immunofluorescence for RAD51 was performed as described above. After the incubation with secondary antibody, the samples were fixed with 2% paraformaldehyde in PBS for 10 minutes at room temperature, then washed six times with PBS and dehydrated in 70%, 95%, and 100% ethanol for 5 min each. Samples were air dried and then denatured at 80°C for 3 minutes in hybridization mix [0.5 μmol/L of a Cy3-conjugated (CCCTAA)₃ PNA probe (PNAbio) in 10 mmol/L NaHPO₄, 10 mmol/L NaCl, 20 mmol/L Tris pH 7.5, 70% formamide, 1× Denhardts, 0.1 mg/mL tRNA, 0.1 mg/mL herring sperm DNA]. Cells were hybridized for 2 hours at room temperature, then washed three times for 10 minutes with 70% formamide, 10 mmol/L Tris, pH 7.2. Cells were washed three times with PBS and DNA was stained with DAPI (0.2 μg/mL).

Genomic DNA was isolated from STAG2 and control tumor cell lines and digested with HinfI, Alu1, MboI, and RsaI. Approximately 3 μg of the digested DNA was fractionated on 1% agarose gels using pulsed-field gel electrophoresis. Telomeres were detected by hybridization to a ³²P end-labeled (TTAGGG)₄ oligonucleotide probe as described (25). The same procedure was used for the BIBR 1532-treated cell lines. For BJ cells, DNA was isolated and digested as described above. DNA was fractionated on a 0.8% agarose gels. G-overhang was detected under native conditions by incubating the gel with a ³²P end-labeled (CCCTAA)₃ oligonucleotide probe overnight at 43°C. The gel was denatured and reprobed overnight at 55°C to detect the total G-strand telomeric DNA. The mean telomere length was determined using Telometric (Fox Chase Cancer Center).

The C-circle assay was performed as described (27). Genomic DNA was prepared and digested as described above and its concentration determined using a Nanodrop spectrophotometer. Fifty nanograms of DNA was combined with 0.2 mg/mL BSA, 0.1% Tween, 1 mmol/L each dATP, dGTP, and dTTP, 1 × φ29 Buffer, and 7.5 U φ29 DNA polymerase (NEB) in 20 μL final volume and incubated at 30°C for 8 hours, then 65°C for 20 minutes. The reaction products were diluted with 60 μL 2× SCC and dot blotted onto a 2× SCC-soaked Amersham Hybond-XL membrane (GE Healthcare). Membranes were UV-cross-linked and hybridized with end-labeled ³²P-(CCCTAA)₃ probe. The membranes were washed (2 × 5 min with 2× SCC at room temperature), exposed to PhosphoImager screens, and scanned using a Typhoon Phosphoimager. To detect the genomic DNA, the membrane was stripped and denatured using a denaturation buffer (1.5 M NaCl, 0.5 M NaOH) for 2 × 10 minutes in a 42°C shaker and neutralized for 10 minutes at room temperature in neutralization buffer (3 M NaCl, 0.5 M Tris-HCl, pH 7.0). The membrane was reprobed with ³²P end-labeled α-satellite probe (5′-ATGTGTGCATTCAACTCACAGAGTTGAAC-3′; ref. 28).

The TRAP assay was performed as described previously (4). To generate cell extracts, 1 × 10⁶ cells were washed with ice-cold PBS, resuspended in 200 μL CHAPS buffer, incubated on ice for 30 minutes, and then centrifuged at 12,000 x g for 20 minutes at 4°C. The assay was performed with 2,500, 5,000, and 10,000 cell equivalents for tumor cell lines and 5,000 and 10,000 cell equivalents for single cell clones. As a control, extracts were heat-inactivated at 80°C for 10 minutes. Point-one microgram of TS primer was incubated with the cell extract for 25 minutes at 30°C, followed by PCR with the reverse (CX) primer. PCR products were fractionated on 10% acrylamide gels and visualized with SYBR Green (1:10,000; Thermo Fisher Scientific).

For the SA β-galactosidase assay (29), cells were fixed in 2% formaldehyde and 0.2% glutaraldehyde in PBS for 5 minutes, washed three times in PBS, and stained for 4 hours at 37°C in staining solution (1 mg/mL X-gal, 150 mmol/L NaCl, 2 mmol/L MgCl₂, 5 mmol/L K3Fe[CN]6, 5 mmol/L K4Fe[CN]6, and 40 mmol/L NaPi, pH 6.0).

Images were acquired using a microscope (Axioplan 2; Carl Zeiss, Inc.) with a Plan Apochrome 63× NA 1.4 oil immersion lens (Carl Zeiss, Inc.) and a digital camera (C4742-95; Hamamatsu Photonics). Images were acquired and processed using Openlab software (Perkin Elmer). For chromosome specific FISH, if foci fell in more than one optical plane of focus, multiple planes were merged using Openlab software.

Statistical analysis was performed using Prism 7 software. Data are shown as mean ± SEM. Student unpaired t test was applied. P < 0.05 values were considered significant: *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.0001; ****, P ≤ 0.0001; ns, not significant.

Previous examination of STAG2 tumor cell lines by chromosome spread analysis showed defects in centromere cohesion (17). Despite defective centromere cohesion, STAG2 tumors do not associate with aneuploidy, suggesting that other mechanisms might be acting to hold sister chromatids together. We sought to determine the status of cohesion along chromosome arms and telomeres. Because arm/telomere cohesion is not maintained during chromosome spread preparation, we measured sister chromatid cohesion using chromosome specific FISH (25, 30). In this procedure mitotic cells are isolated by “shake-off” from asynchronous cultures, fixed immediately to preserve cohesion, and analyzed with fluorescent chromosome-specific probes.

We obtained a panel of 10 STAG2 tumor lines that were either reported (or predicted) to show loss of SA2 expression across a range of tumor types (Supplementary Fig. S1A; refs. 17, 20, 21). Immunoblot analysis with antibodies directed against the N- or C-terminal domains of SA2 showed complete loss of SA2 protein in all 10 STAG2 lines, whereas SA1 exhibited a range of expression levels from one-tenth to two-fold, the levels found in control HeLaI.2.11 cells (a clonal HeLa line with long telomeres; ref. 23; Fig. 1A and Supplementary Fig. S1B). Analysis of centromere cohesion using a 10 cen probe in two STAG2 tumor lines (LOX–IMVI and UM-UC-3) revealed (as expected) premature loss of centromere cohesion (doublets in mitosis) that was rescued by transient expression of SA2 (singlets in mitosis; Fig. 1B–D). FISH analysis with a dual probe against the 13q subtelomere and arm revealed (unexpectedly) persistent cohesion (singlets in mitosis) at telomeres and arms that was rescued by transient expression of SA2 (doublets in mitosis; Fig. 1E and F). We extended the analysis to additional STAG2 tumor lines using a 16p subtelomere probe. Although non-STAG2 tumor lines (HTC75 and HeLa) show limited persistent cohesion (10% of mitotic cells), STAG2 tumor lines show excessive persistent cohesion (50%–60% of mitotic cells; Fig. 1G and H). Together, these data indicate an abnormal pattern of sister chromatid cohesion in mitotic STAG2 tumor cells, where centromeres are prematurely separated, but telomeres and arms are persistently cohered (Fig. 1I).

Because STAG2 tumors contain a number of additional mutations (16), we sought to determine if loss of STAG2 was (in and of itself) sufficient to induce persistent telomere cohesion. We used lentiviral infection to generate HeLa.I.2.11 cell lines stably expressing two different SA2 shRNAs (SA2-3 and SA2-4) and found that SA2 depletion-induced persistent telomere cohesion (Fig. 1J–L). We next wanted to determine if increased cohesion was due to a change in the level or association of cohesin subunits. Immunoblot analysis showed that the levels of SMC1, SMC3, and SCC1 were unaffected by depletion of SA2, whereas SA1 was slightly increased (Fig. 1M). Immunoprecipitation analysis showed that the ring was intact; SMC1, SCC1, and SA1 coimmunoprecipitated with SMC3 (Fig. 1M), consistent with previous results in STAG2 tumor cells (31). We obtained similar results in a HCT116 SA2 KO cell line (ref. 17 and Fig. 1N). Thus, despite normal levels of cohesin subunits and formation of the cohesin ring, loss of SA2 induces defective cohesion.

SA2 is the major SA subunit in somatic cells. There is 12- to 15-fold more SA2-cohesin than SA1-cohesin in HeLa cells (32). We considered that upon loss of SA2, even if there is a two-fold increase in SA1, which was the maximum that we observed in our panel of STAG2 tumor lines (see Fig. 1A and Supplementary Fig. S1B), most of the cohesin rings would be without an SA subunit. We reasoned that the tripartite ring alone may be unable to establish or maintain centromere cohesion and at the same time could contribute to persistent cohesion at telomeres and arms. We thus asked if SA2 rescue of defective cohesion depends on its ability to bind the cohesin ring. We introduced wild-type SA2 or SA2.D793K (a ring binding-deficient point mutant; ref. 33) into HTC116 STAG2 KO cells. We found that wild-type SA2, but not SA2.D793K, rescued the defective telomere (Fig. 2A–C) and centromere (Fig. 2D and E) cohesion. Finally, consistent with a role for the ring in persistent cohesion, we found that depletion of the SCC1 ring subunit rescued persistent telomere cohesion in HCT116 STAG2 KO cells (Fig. 2F–H) and it rescued persistent arm/telomere cohesion in LOX-IMVI and UM-UC-3 STAG2 tumor cells (Fig. 2I–K).

We considered that keeping sister telomeres in close proximity through G₂ into mitosis might promote increased sister telomere recombination, as it does in ALT cells (8). To measure the rate of T-SCE in STAG2 tumor cells we performed CO-FISH analysis (26). For controls we measured the rate in non-ALT super-telomerase HeLa cells (a HeLa line overexpressing telomerase that has very long telomeres; ref. 22) and observed a low rate of T-SCE and in the ALT cell line U2OS and observed a high rate of T-SCE (Fig. 3A and B). STAG2 tumor cells (SK-N-MC, U138MG, LOX-IMVI, and 42MGBA) all exhibited high rates of T-SCE (Fig. 3A and B). Knock-in of STAG2 in 42MGBA STAG2 tumor cells (17) reduced T-SCE levels (Fig. 3B), suggesting that the high rate was due to loss of STAG2. To determine if depletion of SA2 in non-STAG2 tumor cells was sufficient to induce T-SCE, we performed CO-FISH on HeLaI.2.11 cell lines stably expressing STAG2 shRNAs and observed a significant increase in T-SCE in SA2-depleted cells (Fig. 3C and D). Similarly, knockout of SA2 in HCT116 cells led to increased levels of T-SCE (Supplementary Fig. S2A and S2B). Together, these data suggest that persistent cohesion promotes T-SCE. To further test this, we asked if the increased T-SCE in HCT116 SA2 KO cells was dependent on the cohesin ring. As shown in Supplementary Fig. S2C and S2D, depletion of the ring subunit SCC1 rescued the T-SCE in SA2 KO cells. Finally, SA2 KO did not induce a genome wide increase in SCE as determined by Giemsa staining (Supplementary Fig. S2E and S2F).

ALT cells exhibit high rates of T-SCE and have long heterogeneous telomeres. To analyze the telomeres in STAG2 tumor lines, we performed telomere restriction fragment analysis (Fig. 3E). Telomerase positive non-ALT cell lines typically have short (∼4–7 kb) homogeneous telomeres, as shown for HEK293T and HTC75 cells. STAG2 tumor cell lines showed a range of telomere lengths. Four lines had relatively long telomeres; SK-N-MC, U138MG, and 42MGBA were between 12 and 18 kb and LOX-IMVI ranged from 14 kb to extremely long, similar to the U2OS ALT line. Telomeres of the remaining STAG2 lines were between 4 and 6 kb, similar to telomerase positive tumor cells. We observed no correlation between the ploidy level and telomere length in STAG2 tumor cell lines (Supplementary Fig. S2G).

Because STAG2 tumor cells exhibited some features of ALT: high levels of T-SCE and in some cases long telomeres, we examined other properties of ALT in our STAG2 tumor lines. Hallmark features of ALT cells include the loss of ATRX and DAXX expression (34, 35), the presence of ALT-associated promyelocytic leukemia (PML) bodies (APB; ref. 36) and partially single-stranded telomeric extrachromosomal (CCCTAA) DNA circles (C-circles; ref. 27), and the absence of telomerase activity (37). We measured ATRX and DAXX levels in our panel of 10 STAG2 cell lines using immunoblot analysis and found that most expressed ATRX and DAXX (Fig. 4A). We detected little or no C-circles in most STAG2 cell lines, compared to robust levels for ALT cells (Fig. 4B). One exception was LOX-IMVI cells, which showed similar C-circle levels as ALT cells. This could be due to the exceptionally long telomeres in that cell line (see Fig. 3E). To determine the frequency of APBs, we performed immunofluorescence analysis and measured colocalization of the shelterin subunit TRF1 with PML. As shown in Fig. 4C and D, APBs were minimally or not at all detected in STAG2 tumor lines compared to U2OS. Even when APBs were detected in STAG2 cells, they were much smaller than those in the U2OS ALT cells (see Fig. 4C). Finally, measurement of telomerase activity using the telomere repeat amplification protocol (TRAP) indicated high levels of telomerase activity in all STAG2 tumor lines tested (UM-UC-3, SK-ES-1, UM138G, SK-N-MC, and LOX-IMVI) similar to telomerase positive HT1080 tumor cells (Fig. 4E). To rule out the possibility that telomerase activity resulted from a mixed population of cells in the tumor, we isolated single cell clones from LOX-IMVI cells and performed TRAP analysis. As shown in Fig. 4F, each clone displayed telomerase activity. Taken together, these data indicate that despite high levels of T-SCE and longer than average telomeres, STAG2 tumors are not ALT.

Our studies indicate that STAG2 tumor cells have two mechanism of telomere maintenance available to them: recombination and telomerase. We cannot inhibit recombination during long-term growth, because it inhibits cell growth, but we can inhibit telomerase using the nontoxic selective small molecule inhibitor BIBR 1532 (38, 39). Previous studies showed that continuous treatment of HT1080 cells with 10 μmol/L BIBR had no effect on proliferation for 150 population doublings (PD) and led to telomere shortening (38). We used the TRAP assay to confirm that BIBR 1532 inhibited telomerase in HT1080 cells and in our STAG2 tumor lines in a dose-dependent manner (Fig. 5A). We passaged cells for 120 to 160 PD in the absence or presence of 20 μmol/L BIBR 1532; cells were passaged twice every 7 days and BIBR 1532 was freshly added at each passage. We then used telomere restriction fragment analysis to measure telomere length. As shown in Fig. 5B, as expected, telomeres in HT1080 cells shorten. Measurement of the mean telomere lengths (Fig. 5C) indicates that telomeres shorten at a rate of 21 bp/PD (Fig. 5D), similar to previous studies (38). By contrast, the STAG2 cell lines exhibited minimal or no shortening with rates ranging from 6 bp/PD (UM-UC-3) to 0 bp/PD (U138MG; Fig. 5B–D). Together, these data indicate that the STAG2 cells (unlike typical telomerase positive tumor cells) do not rely exclusively on telomerase for telomere length maintenance.

We considered that having two pathways of telomere maintenance might provide cells with a growth advantage. We did not detect an obvious effect on cell growth in telomerase-positive HeLa cells depleted of SA2 or in HCT116 STAG2 KO cells. But what about normal (telomerase negative) human cells? Considering that continued passage of normal human cells leads to replicative senescence due to telomere shortening, we reasoned that SA2-depletion (in this case) might provide a growth advantage by promoting recombination and extending critically short telomeres. We infected BJ fibroblasts at an early PD (PD 28) with lentiviruses expressing two different SA2 shRNAs (SA2-3 and SA2-4) to generate SA2-depleted BJ cell lines (BJ-2). Immunoblot analysis confirmed depletion of SA2 (Supplementary Fig. S3A) and FISH analysis showed induction of persistent telomere cohesion (Supplementary Fig. S3B and S3C). We performed CO-FISH analysis and observed a small, but significant increase in T-SCE in SA2-depleted BJ cells (Supplementary Fig. S3D). As another way to measure recombination, we determined the frequency of RAD51 foci, a marker of homologous recombination that appears in G₂ phase of the cell cycle. As shown in Fig. 6A and B, we observed a significant increase in RAD51 foci upon depletion of SA2. To determine if these foci coincided with telomeres, we performed dual staining for RAD51 with telomeres using TTAGGG-PNA FISH. As shown in Fig. 6C and D, the RAD51 foci were frequently at or near telomeres in SA2-depleted cells, indicating increased recombination at telomeres.

To determine the impact of SA2 depletion on growth, the BJ-2 cell lines were carried for multiple population doublings. As shown in Fig. 6E, initially at early PDs (from PD 30 to 45; 0 to 60 days in culture), both vector control and SA2-depleted cells grew at the same rate. However, at late PDs (after PD 45; after 60 days in culture), the growth rate for control, but not for SA2-depleted cells, slowed. SA2-depleted cells continued to grow at the same rate for several PDs before slowing. Ultimately, SA2-depletion enabled BJ cells to grow for an additional seven to eight cell divisions up to PD 59 to 60 compared to PD 52 for vector control cells (Fig. 6E). Aging human cells accumulate DNA damage that signals senescence (40). We thus asked if the DNA damage signal was attenuated in SA2-depleted cells. We measured the frequency of DNA damage foci at early (day 25) and late (day 84) PD. As shown in Fig. 6F and G, we observed a significant decrease in the levels of γH2AX/p53BP1 DNA damage foci in SA2-depleted late PD cells. To measure senescence, we subjected the cells to the senescence associated ß-gal assay (29) at early (day 25) and late (day 84) PD. As shown in Fig. 6H and I, we observed a significant decrease in ß-gal positive cells in SA2-depleted late PD cells. We independently generated a second set of SA2-depleted BJ cell lines (BJ-1) and confirmed that SA2-depletion extended lifespan, reduced DNA damage foci, and delayed induction of senescence (Supplementary Fig. S3E–S3G).

Our results, showing that SA2-depleted cells have reduced DNA damage signaling and delayed senescence at late PDs, suggested the possibility that telomere shortening was attenuated in these cells. We used telomere restriction fragment analysis under denaturing conditions using a CCCATT telomere probe to analyze telomere length in late PD BJ-2 cells (day 74) and observed longer telomeres in the SA2-depleted cell lines than the vector control (Fig. 6J, right). Telomere shortening in primary cells results from the end replication problem and nucleolytic processing of the 5′ C-strand to generate the single-stranded 3′ G-strand overhang (41). Reduced telomere shortening could result from reduced C-strand degradation. However, probing of the native telomeric DNA (prior to denaturing) showed that the amount of G-strand overhang was not diminished in SA2-depleted cells (Fig. 6J, left). We repeated these analyses in late PD BJ-2 cells (day 66) and obtained similar results (Supplementary Fig. S3H).

Finally, we measured the impact of SA2-depletion in BJ cells on aneuploidy. We performed dual 10 and 6 cen FISH and scored monosomy and trisomy to determine aneuploidy for each chromosome at early and late PD. As shown in Supplementary Fig. S3I and S3J, we observed a small but significant increase in aneuploidy in SA2-depleted cells compared to the vector control at early PD. At late PD, both vector and SA2-depleted cells showed an increase, consistent with a general increase in aneuploidy in aging cells (42). These results are consistent with previous studies showing that loss of STAG2 in human cultured systems can lead to aneuploidy (17, 20).

We have found that STAG2 tumor cells exhibit a dramatic change in the pattern of sister chromatid cohesion in mitosis. Normally, cohesin is removed from telomeres and arms by the prophase pathway, but remains at centromeres until the metaphase to anaphase transition. In STAG2 tumors, cohesion is maintained at telomeres and arms, but is lost from centromeres. We show that this pattern can be rescued by reintroducing SA2 and further that rescue depends on SA2 binding to the cohesin ring. Although it is not surprising that SA2 loss leads to a centromere cohesion defect, the observed persistent arm and telomere cohesion was unexpected. Normally, SA2-cohesin is the predominant species; it is 12- to 15-fold more abundant than SA1-cohesin (32). Thus, loss of SA2, without sufficient compensation by SA1 would mean that the majority of cohesin rings would lack an SA subunit. Indeed, in STAG2 tumors SA2 is lost, SA1 varies only slightly (up or down), and the ring subunit levels remain the same. Thus, the bulk of cohesin in STAG2 tumors is comprised of just the tripartite ring. One possibility is that rings lacking a SA subunit are not removed efficiently from arms and telomeres by the prophase pathway. The resulting persistent arm/telomere cohesion would keep sisters cohered in mitosis, thereby compensating for the lack of centromere cohesion.

We found that STAG2 tumors undergo high rates of sister telomere recombination, a feature that it shares with ALT cells. In addition, some STAG2 tumors show long heterogeneous telomeres like ALT cells. However, STAG2 tumors do not exhibit most of the hallmarks of ALT cells including APBs, C-circles, and loss of ATRX. In addition, STAG2 tumors are telomerase positive. Telomerase positive tumors generally rely on telomerase for cell growth and exhibit telomere shortening upon inhibition of telomerase (43). However, STAG2 tumors exhibited minimal telomere loss during long-term growth in the presence of a telomerase inhibitor. The capacity for telomere recombination may permit STAG2 tumor cells to maintain their telomeres when telomerase is inhibited. Such a property would be an important consideration for anticancer therapy. Telomerase is an attractive target for cancer therapeutics under active investigation (44). Based on our studies, STAG2 tumors would not be good candidates for telomerase anticancer therapy. Along these lines, it will be important to determine if other (non-STAG2) classes of telomerase positive tumors have a capacity for recombination.

Our data are consistent with the notion that persistent telomere cohesion promotes T-SCE. We showed that depletion of SA2 alone was sufficient to induce T-SCE in non-STAG2 tumor cells, that reintroduction of STAG2 into STAG2 tumor cells rescued T-SCE, and that depletion of the ring subunit SCC1 rescued T-SCE in SA2 KO cells. How might persistent telomere cohesion promote recombination? Studies in mouse suggest that telomeres are highly susceptible to homologous recombination and that this is normally repressed by shelterin (45). Here we suggest that timely resolution of telomere cohesion contributes to repression of recombination and that the persistent cohesion at telomeres (induced by SA2 loss) drives the increase in T-SCE. Homologous recombination between sister telomeres can only occur in a small window of the cell cycle: between post-replication in S phase and separation in mitosis. Resolution of telomere cohesion normally occurs in late S/G₂ and is complete by prophase (14). By contrast, in STAG2-depleted cells telomeres remain cohered throughout G₂ and into mitosis. This several fold increase in the window of time that sisters are cohered could underlie the increase in T-SCE.

Mutational inactivation of STAG2 is an early event in tumorigenesis (16), whereas telomerase activation is late (44). Thus, it was pertinent to determine the impact of SA2 depletion on telomerase negative normal human cells. We observed persistent telomere cohesion and increased recombination at telomeres. We did not detect an effect on cell growth, at least at early PDs. However, at later PDs upon senescence onset, the advantage of SA2-depletion was striking. Cells showed reduced DNA damage signaling and diminished telomere shortening that resulted in delayed senescence. How might increased sister chromatid cohesion and recombination slow the rate of telomere attrition? As telomeres shorten in the absence of telomerase, the shortest telomeres activate a DNA damage response (40). Short telomeres may be subject to increased processing and resection and/or incomplete replication, which would lead to length asymmetry of sister chromatids (46). Here, a recombination-based mechanism using the sister chromatid as a copy template could allow lengthening of a shortened strand, delaying the DNA damage signal and extending lifespan. Ultimately, the telomere reserve is depleted and cells senesce, but not before the critical period prior to senescence is extended. In normal human cells, telomere shortening serves to limit the number of cell divisions to prevent genomic instability and cancer (47). A recent report indicates that a majority of mutations in cancer arise from replication errors and further, that stem cells with longer replicative lifespan have a higher number of mutations and increased incidence of cancer (48). Lifespan extension of normal human cells by inactivation of STAG2 could promote tumorigenesis by extending the time period during which tumor-driving mutations can occur.

No potential conflicts of interest were disclosed.

Conception and design: Z. Daniloski, S. Smith

Development of methodology: Z. Daniloski, S. Smith

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Z. Daniloski, S. Smith

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Z. Daniloski, S. Smith

Writing, review, and/or revision of the manuscript: Z. Daniloski, S. Smith

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Z. Daniloski, S. Smith

Study supervision: S. Smith

We thank members of the Smith Lab for critical reading of the manuscript and helpful discussion and Todd Waldman for generously providing STAG2 tumor and HCT116 KO cell lines.

This work was supported by the National Cancer Institute of the NIH under award number R01CA200751 to S. Smith and the Department of Defense PRCRP Horizon Award No. W81XWH-16-1-0590 to Z. Daniloski.

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.