Selected Alkylating Agents Can Overcome Drug Tolerance of G₀-like Tumor Cells and Eradicate BRCA1-Deficient Mammary Tumors in Mice

Although immunotherapy and more personalized treatment (1) have led to important advances in the treatment of cancer, chemotherapy using alkylating agents remains a cornerstone in the treatment of many tumors (2, 3). The most important cellular target of alkylating agents is DNA, in which bifunctional alkylators induce interstrand and/or intrastrand crosslinks (4). It is usually assumed that the interstrand crosslinks are the lethal lesion, as they block DNA copying and are difficult to repair. The basic explanation for the efficacy of alkylating agents has long been the fact that tumor cells replicate rapidly (5). In recent years, we have learned that DNA-crosslinking agents preferentially kill tumors that show defects in the DNA damage response (4, 6, 7). In particular, tumors that are defective in homology-directed DNA repair due to the lack of BRCA1 or BRCA2 function appear to benefit from intensive chemotherapy with alkylating agents, as shown for patients with HER2-negative, high-risk breast cancers that show BRCA1/2-like signatures (8, 9). Despite this high sensitivity, some residual cancer cells persist and lead to tumor recurrence and treatment failure. The precise mechanisms underlying this residual disease are poorly understood (10).

The current choice of DNA crosslinkers to treat breast cancer is empiric, because these have been selected long before any genetic analysis of tumor subgroups was possible. Some chemotherapy regimens using alkylating agents have been developed clinically based on the concept of “more is better.” Agents that mainly caused bone marrow toxicity were favored, as their dose could be substantially escalated using peripheral blood progenitor cell transplantation (PBPC-Tx). As the nature and extent of DNA crosslinks differs between alkylating agents (3, 4), it is conceivable that tumors that cannot employ homologous recombination (HR) for DNA repair show differential responses to various DNA crosslinkers.

We have previously shown that mammary tumors generated in the K14cre;Brca1^F/F;p53^F/F (KB1P) mouse model for hereditary breast cancer are highly sensitive to cisplatin (11) and carboplatin, as well as to the PARP inhibitor olaparib given as single agent or in combination with platinum drugs (12). Despite this high sensitivity, tumors were usually not eradicated, and we could also not achieve this goal by more frequent dosing of cisplatin (13), or with other cytotoxic drugs with a different site of action, such as docetaxel or doxorubicin (11, 14). The drug-tolerant “remnants” surviving cisplatin treatment were neither enriched in tumor-initiating cells, nor in biochemical defense mechanisms (13). We therefore hypothesized that the drug-tolerant residual cells causing tumor relapse were stalled in the cell cycle, and we have now identified a drug-tolerant tumor cell population in a G₀-like cell-cycle stage.

An important question is whether these drug-tolerant cells can be eradicated at all by drug treatment. When cell lines are tested for drug response in clonogenic assays, there are usually colonies coming up after treatment with doses above 10 times the IC₅₀, indicating that there are cells in the population that can survive drug doses that cannot be reached in mice or patients. To test this in our model, we have investigated a range of alkylating agents. Whereas we obtained cures in about half of our mice with melphalan, we were able to completely eradicate the tumors with nimustine, an alkylating agent previously identified in an in vitro compound screen to preferentially kill Brca2^−/− mammary tumor cells (15). Eradication required the full MTD; at half the MTD, all tumors relapsed. This shows that the drug-tolerant G₀-like cells can be eliminated, if sufficient damage is inflicted.

Brca1^−/−;p53^−/−mouse mammary tumors were generated and transplanted into syngeneic mice as described previously (11, 16). In this study, we used KB1P mice, which were backcrossed to an FVB/N background (17). Tumor-bearing mice were treated as indicated in the different experiments. Melphalan (Alkeran, GlaxoSmithKline) and docetaxel (Taxotere, Sanofi-Aventis) were reconstituted immediately before administration, according to the manufacturer's protocol. Thiotepa (Ledertepa, Sigma) and bendamustine hydrochloride (Sigma) were dissolved with physiologic salt solution (stock solution of 10 mg/mL) just before injection. Nimustine hydrochloride (1 g; Sigma) was dissolved in 6.5 mL DMSO and just before injection, diluted 25-fold with physiologic salt solution. Cisplatin and carboplatin solutions ready for injection were obtained from Mayne Pharma. 4-hydroxy-cyclophosphamide (4-OH-CP) was purchased from Dr. Ulf Niemeyer (IIT GmbH, University of Bielefeld, Bielefeld, Germany). For temozolomide (Temodal), a 5 mg/mL dosing solution was prepared using corn oil, and the freshly prepared solution was sonicated for 10 to 15 minutes before oral gavage. Tumor volume was calculated using caliper measurements (v = length × width² × 0.5). Total white blood cell numbers were measured on a hematology analyzer (Becton Dickinson). All experimental procedures were approved by the Animal Ethics Committee of the Netherlands Cancer Institute (Amsterdam, the Netherlands).

The following antibodies were used for IHC staining of tumors: the in-house–produced NKI-A59 antibody for the detection of cisplatin DNA adducts (as described previously; ref. 18), anti-Ki67 from Abcam (ab15580, 1:3,000), anti-γH2AX from Cell Signaling Technology (#2577, 1:50), and also anti–cleaved caspase-3 from Cell Signaling Technology (#9661, 1:400). The terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) assay was performed using the ApopTag Peroxidase In Situ Apoptosis Detection Kit (Chemicon International, S7100). Antigen retrieval was performed by boiling the formalin (10%)-fixed tissue samples for 15 minutes in citrate buffer (pH 6.0). Slides were incubated at 4°C overnight with the primary antibodies, followed by incubation with a biotinylated goat anti-rabbit secondary antibody (DakoCytomation, # E043201, 1:800 in 1% PBSA) for 30 minutes at room temperature. For detection, we used a standard StreptABC amplified staining procedure with DAB (DakoCytomation, # K037711) and hematoxylin counterstaining. Positive and negative (only secondary antibody) controls were included for each slide and staining procedure. Positively labeled cells were counted in the tumor sections in 10 fields of 650 × 650 μm.

Two individual GFP-labeled Brca1^−/−;p53^−/− mammary tumors were orthotopically transplanted into 20 wild-type FVB/N mice each. After transplantation, 30 mg 5-iodo-2-deoxyuridine (IUdR) per kg was injected intraperitoneally (i.p.) daily for about 3 weeks until tumors reached the size of about 20 mm³. Four days after the last IUdR injection, tumors were treated with 100 μL saline intravenously (i.v.) or with 6 mg cisplatin per kg i.v. Tissue specimens were stained with DAPI, and IUdR-positive cells were detected by immunofluorescence. Tumors were analyzed by immunofluorescence prior to and 8 days posttreatment.

Two individual GFP-labeled Brca1^−/−;p53^−/− mammary tumors were orthotopically transplanted into wild-type FVB/N mice. A total of 160 mg 5-ethynyl-2′-deoxyuridine (EdU) per kg was injected i.p. 10 days after saline or cisplatin treatment. Two hours after EdU injection, tumors were harvested and stained with DAPI, and EdU-positive cells were detected by immunofluorescence. For double labeling with IUdR and EdU, animals were injected with IUdR as described above, and EdU was injected 2 hours before saline or cisplatin treatment. Tumors were analyzed by immunofluorescence prior to and 8 days posttreatment.

The Brca1^−/−;p53^−/−cell lines KB1P-B11 and KB1P-G3 were derived from a genetically engineered mouse model for BRCA1-mutated breast cancer as described previously (17). The cell lines were last authenticated by genotyping of the specific Brca1 and p53 mutations in January 2017. The KB1P-B11 cell line was transduced with the vectors CSII-EF-MCS expressing the fluorescent ubiquitination-based cell-cycle indicator (FUCCI) mKO2-hCdt1(30/120) or mAG-hGem(1/110) (19) and the packaging and envelope plasmids pCMV-R 8.2, pMDG and pRSV-Rev. After selection using zeocin, the cells were subcloned and sorted for green and red fluorescence subsequently to produce a cell line stably expressing the FUCCI system (KB1P-B11-FUCCI). Cell lines were tested every 6 months for mycoplasma contamination using MycoFluor Mycoplasma Detection Kit (Thermo Fisher Scientific).

KB1P-B11 cells were plated and treated 24 hours later with 0.7 μmol/L cisplatin for 24 hours, and colony formation was tracked using brightfield settings on the Zeiss Axiovert 200 M fluorescence/live cell imaging system. KB1P-B11-FUCCI cells were treated with 5- or 10-fold IC₅₀ carboplatin (IC₅₀ = 610 nmol/L) or nimustine (IC₅₀ = 750 nmol/L) 2 days after seeding. The IC₅₀ was determined using a classical clonogenic assay. Cell cycle–specific response of the KB1P-B11-FUCCI cells was analyzed by live cell imaging for 72 hours.

To identify Ki67-negative cells, KB1P-B11-FUCCI cells were grown to 100% confluency. The cells were fixed and permeabilized using the BD Cytofix/Cytoperm Kit (cat. no. 554722) and stained for Ki67 (Ki67-BV421, Clone 16A8, BioLegend, 0.2 mg/mL) and DNA content (Draq5 BioLegend, 5 nmol/L). The cell-cycle stage of Ki67-negative KB1P-B11-FUCCI cells was analyzed using the BD LSRII Flow Cytometer (Ki67-BV421 excitation: 407 nm, filter: 450/50; mAG: excitation: 488 nm, filter: 525/50; mKO2: excitation: 561 nm, filter 585/15; Draq5: excitation: 640 nm, filter: 710/50).

To sort cells for clonogenic assays, KB1P-B11 and KB1P-G3 cells were labeled with Hoechst 33342 (Sigma), the cell-cycle stage was analyzed, and cells were sorted for 2n and 4n content using FACS (BD FACSAria, blue/violet laser 375 nm). The cell-cycle stage of KB1P-B11-FUCCI cells was analyzed using a BD FACSAria III (mAG: excitation: 488 nm, filter: 530/30; mKO2: excitation: 561 nm, filter 582/15) and cells in different cell-cycle stages (nonfluorescent (mAG⁻; mKO2⁻), red (mAG⁻; mKO2⁺), double positive (mAG⁺; mKO2⁺), and green (mAG⁺; mKO2⁻) were sorted.

IC₅₀ of KB1P-B11 and KB1P-G3 was determined in a short-term clonogenic assay: Cells were treated for 24 hours with different drug concentrations and allowed to grow for 6 days. Cells were fixed with 4% paraformaldehyde and stained with 1% Crystal Violet, and the drug concentration at which only half of the colonies were growing out compared with untreated cells was calculated. KB1P-B11-FUCCI cells were treated with cisplatin (0.7 μmol/L), carboplatin (5 μmol/L), or nimustine (4.2 and 5 μmol/L) for 24 hours 1 day after seeding or left untreated. Nonfluorescent, red, green, and double-positive cells were sorted into FCS and seeded at equal numbers [5,000 cells (untreated), 100,000 cells (treated)]. Cells were fixed after 6 (untreated) or 9 (treated) days, and colony formation (at least 20 cells) was quantified. For long-term clonogenic assay, KB1P-B11 or KB1P-G3 was treated with 5-fold the IC₅₀ of nimustine or carboplatin for 24 hours and was allowed to grow for 3 weeks.

We have previously shown that the KB1P tumors, which contain a large irreversible deletion of the Brca1 gene, do not acquire resistance to the MTD of cisplatin (11, 20). Despite this inability of tumors to acquire resistance, animals were usually not cured and tumors eventually relapsed, even after several treatment cycles (Fig. 1A). Consistent with the absence of acquired drug resistance, we did not observe a shortening in the time interval between tumor regrowth after 5 repeated cisplatin treatments for 12 individual KB1P tumors (Fig. 1B). Intensifying the drug treatment did not prevent relapse. Although an additional cisplatin MTD treatment on day 14 significantly increased the time until relapse, the tumors still regrew (13). On the basis of the time until relapse for the day 0 and day 0 + 14 treatments (Fig. 1C), we estimated the surviving fraction of tumor cells to be <10⁻⁸ (tumor of 200 mm³ contains about 10⁸ cells) after applying two cisplatin doses on day 0 + 14 (surviving fraction = 2^{−growth delay/doubling time}, tumor doubling time: about 3 days). If the doubling time is constant and cells are randomly hit by cisplatin, a third cisplatin dose on day 28 (after the animals have recovered from the previous cisplatin doses) should definitely result in tumor eradication. It did not however (Fig. 1C). Spreading the additional cisplatin dose over several days would be expected to help eradicate tumor cells, which are by chance not in the right cell-cycle stage when cisplatin is dosed, but this treatment regime did not increase tumor eradication either (Supplementary Fig. S1). To test whether drug-tolerant tumor cells accumulate in residual tumors, we compared the cisplatin response of small drug-naïve tumors (about 20 mm³) with the response of remnants of the same size 14 days after surviving the initial cisplatin treatment. As shown in Fig. 1D, tumor remnants relapsed earlier than the naïve tumors (P = 0.006). This difference is even more significant given the fact that the residual tumors contain an increased amount of stroma and about a third less tumor cells, as we have shown previously (13). In contrast to the small tumor remnants, previous treatment does not influence the response of the relapsed tumors (Fig. 1A and B), suggesting that the drug-tolerant cells only represent a small fraction of the latter. Together, our data show that the tumor remnants remaining after the initial cisplatin treatment are enriched in cells that are drug tolerant, but have not acquired stable drug resistance.

We recently showed that residual tumors are not enriched in tumor-initiating cells, which strongly suggests that the tumor-initiating cells of our model do not have increased biochemical defense mechanisms against cisplatin (13). The residual tumor cells are also accessible to drug: We found a homogeneous distribution of the platinum–DNA adducts throughout the tumor (Supplementary Fig. S2A). Ten days after cisplatin treatment, several residual tumor cells appeared to have removed platinum–DNA adducts, but upon re-treatment, the adducts reappeared. We infer from these results that the residual tumor cells of our model are accessible to the drug.

Another plausible mechanism for tumor cells to escape drug toxicity is to stop dividing, and to enter transiently into a quiescent program. This hypothesis is supported by the substantial decrease in Ki67-positive cells in the residual tumors (Fig. 2A). To verify this result, we investigated EdU uptake into the tumor remnants. To distinguish tumor from stromal cells, we introduced the GFP marker into our model (Supplementary Fig. S2B; ref. 21). In drug-naïve tumors, we found that many tumor cells were easily labeled with EdU. In contrast, the residual GFP-positive tumor islands after cisplatin treatment hardly incorporated EdU (Supplementary Fig. S3A). Hence, residual tumor cells dropped out of cycle.

We next investigated whether slowly cycling cells are already present in drug-naïve tumors, or whether they are exclusively induced by drug treatment. For this purpose, we performed a label-retaining assay of two different GFP-labeled KB1P donor tumors using IUdR (Fig. 2B). Daily injection of IUdR after orthotopic transplantation successfully labeled most tumor cells once a palpable tumor (about 3–5 mm in diameter) was detected (“2 hours after the last IUdR injection”). At this size, the IUdR injections were stopped, and 4 days later, most tumor cells had lost the label (“4 days after the last IUdR injection”). We then compared the effect of saline versus cisplatin treatment on the tumors 12 days after the last IUdR injection. Whereas labeled cells were rare (<1%) in the saline-treated tumors (“12 days after the last IUdR injection; 8 days after saline”), the IUdR-retaining cells were enriched (about 25%–35%) in the remnants of cisplatin-treated tumors (“12 days after the last IUdR injection; 8 days after cisplatin”). This shows that slowly cycling cells are present within our KB1P tumor model, and these cells have an advantage in surviving the cisplatin treatment onslaught. In addition, we investigated cells that went through the S-phase just before cisplatin or saline treatment. To mark those cells, we injected EdU into mice 2 hours before cisplatin or saline treatment (Supplementary Fig. S3B–S3D). As expected, IUdR-labeled cells did not pick up EdU (Supplementary Fig. S3C). Moreover, in contrast to the quiescent IUdR-retaining cells, the number of proliferating EdU-positive cells clearly decreased after cisplatin treatment (Supplementary Fig. S3C and S3D), showing that these cells are more vulnerable.

In our model, we observed the presence of giant and multinucleated tumor cells in the cisplatin-treated tumor remnants (Fig. 3A). We could even reproduce this phenotype in vitro in Brca1^−/−;p53^−/− cell lines (KB1P-B11 and KB1P-G3) that we derived from a spontaneous KB1P tumor. These cell lines highly mimic the spontaneous tumor in their genomic profile (Supplementary Fig. S4A), and they mimic the morphology and drug response of KB1P tumors if grafted back into mice (17). We found many giant and multinucleated KB1P-B11 cells within a few days after treatment with 0.7 μmol/L cisplatin. In tumors or colonies that relapsed 3 to 4 weeks later, those cells were largely absent (Fig. 3A). The formation of multinucleated cells and subsequent “giant cell death” in the G1-phase of the cell cycle has been described to result from DNA interstrand crosslinks (22). Using time-lapse video microscopy, we found that the multinucleated cells after cisplatin treatment are doomed and eventually undergo crosslink-induced giant cell death (Fig. 3B).

To determine from which cells relapsing colonies evolve, we backtracked colonies formed after cisplatin treatment by time-lapse video microscopy. When KB1P-B11 cells were treated with 0.7 μmol/L cisplatin for 24 hours, we found that colonies were always derived from small cells with a single nucleus (an example is shown in Fig. 4A). To determine the cell-cycle stage from which the single-nucleated cells are able to repopulate colonies, we labeled cells with Hoechst. Whereas equal numbers of untreated cells (unsorted, 2n, 4n, or 2n + 4n) resulted in the same number of colonies (Fig. 4B), only cells with a 2n DNA content, hence cells in G₀ or G₁, were able to form new colonies after cisplatin treatment (Fig. 4C).

To further characterize the cell-cycle stage from which Brca1^−/−;p53^−/− mammary tumors arise after cisplatin treatment, we introduced the fluorescent ubiquitination-based cell-cycle indicator (FUCCI) construct (19) in KB1P-B11 cells. The cells express the fluorescent proteins mKO2-hCdt1(30/120) in G₁, mAG-hGem(1/110) in S-G2-M, both fluorescent proteins in G₁–S transition, and there is a nonfluorescent stage after mitosis of the cells. As we observed an enrichment of Ki67-negative cells in the nonfluorescent pre-G₁ population (Fig. 5A), we called them G₀-like. In contrast to the cells used by Tomura and colleagues (23), we could not identify an increase of cells with high intensities of mKO2 as potential G₀ cells when cells were grown to confluency or treated with drugs. This difference may be due to the use of a different promoter for the FUCCI system and the different genetic background in our model. We then analyzed the colony formation capacity (experimental layout in Supplementary Fig. S4B) of the different cell-cycle stages after treatment with DNA-crosslinking agents. The cell-cycle stage did not influence the colony formation capacity of untreated cells. If treated with cisplatin or carboplatin, the G₀-like cells had a much higher colony formation capacity than cells in G₁, G₁–S transition, or cells in S-G2-M-phase (Fig. 5B). The same pattern was observed with other drug concentrations than those used in Fig. 5B and C (data not shown). Thus, residual G₀-like cells are drug tolerant to platinum drugs.

Evers and colleagues recently found in an in vitro screen that BRCA2/p53–deficient cells are hypersensitive to the alkylator nimustine (a chloroethylnitrosourea; ref. 15). We therefore tested whether the platinum drug-tolerant G₀-like KB1P cells respond differently to nimustine treatment. Intriguingly, nimustine treatment resulted in a different cell cycle–dependent sensitivity pattern: The selective drug tolerance of G₀-like cells is not detectable when the cells are treated with nimustine at concentrations that resulted in about 100 colonies (Fig. 5C; Supplementary Fig. S4B).

As time-lapse video microscopy revealed that multinucleation in response to DNA-crosslinking agents kills the BRCA1/p53–deficient tumor cells, we analyzed whether nimustine causes more multinucleated cells than platinum drugs. For this purpose, we first determined the carboplatin and nimustine IC₅₀ in a short-term (1 week) clonogenic assay (Supplementary Fig. S5A). Interestingly, filming of the KB1P-B11-FUCCI cells showed that 10-fold nimustine IC₅₀ caused more multinucleation than the corresponding carboplatin concentration (Fig. 5D, left). Moreover, the difference was most evident when multinucleation of G₀-like cells was compared after nimustine or carboplatin treatment (Fig. 5D, right). Most of the multinucleated cells were able to complete mitosis and underwent cell lysis in G₁ (Supplementary Fig. S5B). Consistent with the inability of multinucleated cells to form colonies, we observed in long-term (3 weeks) clonogenic assays that even at 5-fold IC₅₀ concentrations, some colonies grew back when cells were treated with carboplatin, but fewer colonies grew back with the equivalent dose of nimustine (Fig. 6A).

Together, these data strongly suggest that in contrast to platinum drug-induced DNA crosslinks, the increased amount of nimustine-induced interstrand crosslinks (24, 25) cannot be removed effectively in the G₀-like cells. This results in multinucleated cells, which subsequently die.

On the basis of these in vitro results, we tested nimustine efficacy in Brca1^−/−;p53^−/− (KB1P) or p53^−/− (KP) tumors. As Evers and colleagues (15) also identified the bifunctional alkylator melphalan as an efficient compound for the treatment of BRCA2/p53–deficient cells, we examined this agent in parallel in our in vivo experiments. Moreover, we tested the nimustine analogue, nitrogen mustard bendamustine, which has recently reemerged as chemotherapeutic treatment. Although KP tumors were not particularly sensitive to the MTD of melphalan (10 mg/kg i.p.), nimustine (30 mg/kg i.p.), or bendamustine (40 mg/kg i.p.) (Supplementary Fig. S6A and S6B), all KB1P tumors in the nimustine-treated, half of the melphalan-treated, and one third of the bendamustine-treated animals appeared to be eradicated (Fig. 6B, top). Dose matters however. When we lowered the nimustine dose to 75% of the MTD, half of the tumors relapsed, and all tumors regrew when 50% of the MTD was injected (Fig. 6B, bottom). In contrast, we could not eradicate the KB1P tumors using the monofunctional alkylator temozolomide (Supplementary Fig. S6C and S6D), and most KB1P tumors relapsed after high dose monotherapy or combination therapy with carboplatin, thiotepa, and cyclophosphamide (CTC) (Supplementary Fig. S7A; refs. 8, 26). Unfortunately, the gut toxicity of the cocktail of alkylating agents CTC turned out to be the limiting factor in mice, and we could not escalate the dose above MTD levels by mimicking bone marrow reconstitution (Supplementary Figs. S7B–S7E and S8A). Despite the lack of eradication, relapsing tumors were still sensitive to repeated treatments of monotherapy or combination therapy with CTC (Supplementary Fig. S8B). This pattern is similar to what we found in response to cisplatin (Fig. 1A) or carboplatin (12). It supports the notion that in the absence of functional BRCA1, resistance to DNA-crosslinking agents given at MTD levels does not evolve in our model (27).

To further explore the differential sensitivity of KB1P tumors to carboplatin or nimustine, we investigated the drug-induced damage in situ. IHC analysis revealed that 72 hours after treatment, more DNA damage foci (determined by γH2AX positivity), apoptotic cells (measured by cleaved caspase-3 positivity) and DNA breaks (measured by TUNEL) were present in nimustine-treated tumors (Fig. 6C) compared with carboplatin-treated tumors. This difference between nimustine- and carboplatin-treated tumors was not detectable 24 hours after treatment, indicating that DNA damage caused by nimustine is not resolved as efficiently as damage caused by platinum drugs.

To investigate whether carboplatin is more effective when the drug-tolerant G₀-like cells are depleted, we pretreated animals with docetaxel for 24 hours. This resulted in an enrichment of cells arrested in the M-phase (Fig. 6D). When the docetaxel-pretreated animals were dosed with carboplatin, we observed that 6 of 9 animals were cured, and their survival was significantly increased compared with carboplatin treatment alone (Fig. 6D). This suggests that reducing the G₀-like fraction of tumor cells is a useful therapeutic strategy to achieve tumor eradication.

Using a mouse model for BRCA1-deficient breast cancer, we identified a slowly cycling drug-tolerant population of tumor cells, which survives the platinum drug therapy and causes tumor relapse. We show that these residual tumor cells are in a G₀-like cell-cycle stage and can be effectively targeted with the chloroethylating agent nimustine, resulting in disease eradication.

A major difference in DNA damage between platinum drugs and nimustine is the type of DNA crosslinks caused. DNA-crosslinking agents bind to nucleotides either on the same DNA strand or on complementary DNA strands causing intrastrand and interstrand crosslinks, respectively. Platinum drugs, such as cisplatin or carboplatin, mainly cause intrastrand crosslinks (28), whereas nimustine mainly causes interstrand crosslinks (24, 25).

By measuring DNA damage foci, apoptotic cells and DNA breaks 24 and 72 hours after treatment, we have shown that DNA damage caused by nimustine is not resolved as efficiently as damage caused by platinum drugs in BRCA1-deficient tumors. The increased sensitivity of G₀-like cells to nimustine may be explained by the impaired DNA damage repair of the nimustine-induced interstrand crosslinks. Platinum-induced intrastrand crosslinks can be removed by nucleotide excision repair (NER) during the G0-G1-phase of the cell cycle independent of a sister chromatid as template DNA (29). Platinum adducts linking two neighboring nucleotides on the same strand bend the DNA significantly, which might facilitate the recognition and repair of crosslink sites (30). In contrast, only a subset of interstrand crosslinks is repaired by NER. Unresolved interstrand crosslinks lead to stalled replication forks in the S-phase, which are stabilized and repaired by the Fanconi anemia pathway and HR. BRCA1 is important in further processing of the defect by HR (31, 32). In BRCA1-deficient tumor cells, interstrand crosslink repair might therefore be impaired. Unresolved interstrand crosslinks caused by nimustine may lead to stalled replication forks, multinucleation, and cell death, whereas the intrastrand crosslinks caused by platinum drugs may be resolved by NER before the cells enter the S-phase.

Characterization of drug-tolerant tumor cells using cell lines derived from the BRCA1-deficient mouse model showed that relapsing colonies after cisplatin treatment are derived from single-nucleated 2n cells. In the Brca1^−/−;p53^−/− model, the multinucleated cells observed after treatment with DNA-crosslinking agents in vivo and in vitro lacked clonogenicity and eventually died. Using FUCCI-expressing Brca1^−/−;p53^−/− cells, we found that most cells do not die from mitotic catastrophe, but exit mitosis without proper segregation of sister chromatids. This event is described as mitotic slippage (33) resulting in multinucleated cells. The multinucleated cells in G₁ can still enter S-phase due to the defective G₁–S checkpoint in p53-deficient cells (34). Nevertheless, the multinucleated cells die after several cell cycles, and interestingly, most of the cells die in G₁ and not in S-phase, as described in other p53-deficient cells (33). Cell death in G₁ must be executed via a p53-independent pathway in the Brca1^−/−;p53^−/− model (35). These results differ from those obtained by Puig and colleagues (36). In their rat model, tumor cells can escape cisplatin-induced cell death through DNA endoreduplication and reversible polyploidy. In contrast, our results are more compatible with the findings of Osawa and colleagues (22), who observed giant cell death of multinucleated cells following treatment with DNA-crosslinking agents. A difference to the rat model is that the relapsing tumors in our genetically engineered model do not show increased resistance to cisplatin treatment. Because of the large intragenic deletion of the Brca1 gene, no restoration of BRCA1 function is possible, a resistance mechanism against platinum drugs that was described earlier (37). This lack of acquired resistance we also observed in response to other frequently used alkylating agents, including cyclophosphamide and thiotepa. Unresolved interstrand crosslinks might therefore explain the formation of more multinucleated cells and cell death after nimustine treatment compared with platinum drug treatment in the Brca1^−/−;p53^−/− cells.

The most likely explanation for the platinum drug tolerance of G₀-like cells is that they have more time to repair DNA crosslinks before entry into S-phase than G₁ cells. We do not think that stabilization of stalled replication forks plays a role, even though this stabilization is known to preserve genomic integrity (38, 39) and can contribute to acquired PARPi resistance in BRCA2-deficient cells (40). In contrast, the BRCA1-deficient cells in our model do not acquire resistance to the DNA crosslinking agents. In particular, the nimustine-induced interstrand crosslinks cannot be resolved by increased fork stabilization that is sufficient to overcome PARPi-induced fork stalling.

There is also no evidence that G₀-like cells are more proficient in repairing the DNA damage. RNA sequencing analysis of the G₀-like cells did not show increased expression of NER-related genes compared with G₁ cells (data not shown). Nevertheless, further characterizations of the drug-tolerant G₀-like cells may be helpful to find drug targets to specifically deplete G₀-like cells before treatment with platinum drugs.

The G₀-like drug-tolerant cells in our BRCA1-deficient mouse model do not appear to be identical to the subpopulation of reversibly drug-tolerant cells identified by Settleman and colleagues in tumor cell lines treated with tyrosine kinase inhibitors (TKI). These cells have been called “drug-tolerant persisters” (DTP) to emphasize the analogy with drug-resistant “persisters” arising in bacterial cultures treated with antibiotics (reviewed by Harms and colleagues, 2016; ref. 41), even though this analogy is not persuasive (42). DTPs are reported to be stalled in G₁, but G₀ was not separately checked. A subfraction of DTPs can give rise to “drug-tolerant expanded persisters” that can multiply in drug indefinitely (43). Various independent mechanisms appear to contribute to the formation of DTPs (43–45). Resistance of DTPs is not limited to TKIs, as some cross-resistance was observed to cisplatin (43). Although superficially similar to DTPs, at the RNA level, the drug-tolerant, G₀-like cells that we find in our mouse tumors lack the characteristic features reported for DTPs, such as activated IGF-1 receptor signaling (43), elevated aldehyde dehydrogenase levels (44), or high peroxiredoxin 6 (data not shown; ref. 45). Moreover, the G₀-like cells that we find in vivo in our model are not very resistant, as shown by treatment with the appropriate alkylating agent nimustine, in contrast to the DTPs that are more than 100-fold resistant to TKIs. Although we also find rare colonies in vitro after treating our cells with 5-fold the IC₅₀ of nimustine, we think that these rare surviving cells are probably eliminated in vivo by the intact host immune system.

The finding that even at 5-fold, the IC₅₀ of cisplatin or nimustine some colonies grow back in vitro raised the question whether these drug-tolerant cells can be eradicated at all by drug treatment. The fact that we were able to eradicate the Brca1^−/−;p53^−/− tumors with nimustine in vivo is consistent with the results that high-dose chemotherapy using alkylating agents can cure patients with high-risk BRCA1/2-like breast cancer (8, 9). Although not possible in our model, upfront mutations that restore BRCA function may be present in drug-naïve human tumors and counteract tumor eradication. The fact that tumor eradication is achieved in patients with high- risk BRCA1/2-like breast cancer (8, 9), however, supports the notion that this is not a general clinical hurdle. Eradication of the mouse Brca1^−/−;p53^−/− tumors with nimustine is highly dose dependent, and this strongly suggests that the amount of damage caused is crucial for the successful eradication of residual tumor cells. In our model, we did not succeed in testing the standard high-dose chemotherapy used in patients due to the gut toxicity of the DNA-crosslinking agents in mice. Nevertheless, we infer from our data that the high-dose chemotherapy in humans also results in an increased number of interstrand crosslinks of transiently dormant G₀-like cells. When these cells reenter the cell cycle with persisting crosslinks, damage cannot be properly repaired in the absence of BRCA1/2 function, and the cells become multinucleated and eventually die. Future studies aimed at identifying specific markers for these G₀-like cells are crucial to understand their role in drug tolerance in human cancer.

No potential conflicts of interest were disclosed.

Conception and design: M. Pajic, J. Jonkers, P. Borst, S. Rottenberg

Development of methodology: M. Pajic, S. Blatter, A. Küçükosmanoğlu, S. Rottenberg

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Pajic, S. Blatter, C. Guyader, M. Gonggrijp, A. Küçükosmanoğlu, W. Sol, R. Drost, J. Jonkers, S. Rottenberg

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Pajic, S. Blatter, P. Borst, S. Rottenberg

Writing, review, and/or revision of the manuscript: M. Pajic, S. Blatter, C. Guyader, M. Gonggrijp, J. Jonkers, P. Borst, S. Rottenberg

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Blatter, A. Kersbergen, S. Rottenberg

Study supervision: P. Borst, S. Rottenberg

We thank the late Adrian Begg (the Netherlands Cancer Institute) for his help with calculating the surviving fraction and Ben Floot (the Netherlands Cancer Institute) for providing the NKI-A59 antibody. We are also grateful to Sjoerd Rodenhuis and Sabine Linn (Antoni van Leeuwenhoek Hospital) for suggesting clinically relevant alkylating agents/combinations. Rob Wolthuis (VU University Medical Center, Amsterdam) was very helpful in discussing cell cycle–related questions. Moreover, we wish to thank Sjoerd Rodenhuis, Nora Gerhards, and Paola Francica for critical reading of the manuscript.

This work was supported by grants from the Dutch Cancer Society (2009-4303), the Netherlands Organization for Scientific Research (NWO-VIDI-91711302), the European Research Council (ERC-CoG-681572), the Swiss National Science Foundation (project grant 310030_156869), and the Swiss Cancer Research foundation (MD-PhD-3446-01-2014 to S. Blatter).

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.