CHIP buffers heterogeneous Bcl-2 expression levels to prevent augmentation of anticancer drug-resistant cell population

Abstract

Many types of cancer display heterogeneity in various features, including gene expression and malignant potential. This heterogeneity is associated with drug resistance and cancer progression. Recent studies have shown that the expression of a major protein quality control ubiquitin ligase, carboxyl terminus of Hsc70-interacting protein (CHIP), is negatively correlated with breast cancer clinicopathological stages and poor overall survival. Here we show that CHIP acts as a capacitor of heterogeneous Bcl-2 expression levels and prevents an increase in the anticancer drug-resistant population in breast cancer cells. CHIP knockdown in breast cancer cells increased variation in Bcl-2 expression levels, an antiapoptotic protein, among the cells. Our results also showed that CHIP knockdown increased the proportion of anticancer drug-resistant cells. These findings suggest that CHIP buffers variation in gene expression levels, affecting resistance to anticancer drugs. In single-cell clones derived from breast cancer cell lines, CHIP knockdown did not alter the variation in Bcl-2 expression levels and the proportion of anticancer drug-resistant cells. In contrast, when clonal cells were treated with a mutagen, the variation in Bcl-2 expression levels and proportion of anticancer drug-resistant cells were altered by CHIP knockdown. These results suggest that CHIP masks genetic variations to suppress heterogeneous Bcl-2 expression levels and prevents augmentation of the anticancer drug-resistant population of breast cancer cells. Because genetic variation is a major driver of heterogeneity, our results suggest that the degree of heterogeneity in expression levels is decided by a balance between genetic variation and the buffering capacity of CHIP.

Introduction

Breast cancer is the most common cancer among women.¹ There have been many advances in the understanding of the molecular mechanisms underlying breast cancer and the development of novel therapies. However, it remains the most frequent cause of cancer-related death in women.¹ One difficulty in breast cancer therapy is cell-to-cell heterogeneity in cancer, such as in gene expression and malignant potential.² Cancer heterogeneity is a precondition for the evolution of cancer and engenders resistance to anticancer drugs.^{3, 4}

Recent studies have indicated that anticancer drug-resistant cells are derived from emerging clones within primary tumors.^{5, 6} Anticancer drug resistance of cancer cells can occur via several mechanisms, including evasion of apoptosis, increased drug efflux and drug inactivation,⁷ and corresponds closely to the prognosis of breast cancer patients.⁵ Treatment failure in over 90% of metastatic cancer patients is believed to be caused by anticancer drug resistance.⁸ As remarked above, anticancer drug resistance is thought to occur in emerging clones caused by cancer heterogeneity. Accordingly, understanding the molecular mechanisms underlying cancer heterogeneity may pave the way to overcome anticancer drug resistance and significantly improve the survival of cancer patients.

Many types of anticancer drugs activate the apoptotic pathway.⁷ Consistently with this observation, the antiapoptotic protein, Bcl-2, is reported to cause anticancer drug resistance in breast cancer cells.^{9, 10, 11, 12} Bcl-2 is one of the Bcl-2 family proteins and suppresses the apoptotic pathway by inhibiting the loss of mitochondrial membrane potential and the release of cytochrome c.^{13, 14} Bcl-2 expression is clinically associated with both the loss of apoptosis and the presence of lymph node metastases in tumor samples from ductal breast carcinomas,¹⁵ implying that Bcl-2 expression is involved in cancer progression.

We previously reported a novel mechanism for the suppression of breast cancer malignancy by the ubiquitin E3 ligase, carboxyl terminus of Hsc70-interacting protein (CHIP).¹⁶ CHIP has been previously identified as a tetratricopeptide repeat domain-containing protein that interacts with heat shock proteins (HSPs).¹⁷ CHIP acts as a co-chaperone with HSPs for the quality control of other proteins and as a ubiquitin ligase for target proteins to induce proteosomal degradation.^{18, 19, 20} In our previous study, CHIP suppressed the tumorigenic and metastatic potential of breast cancer cells by inhibiting the oncogenic pathway.¹⁶ CHIP expression levels were negatively correlated with breast cancer clinicopathological stages.¹⁶ Moreover, recent clinical studies suggest that lower levels of CHIP are predictive of poorer overall survival in breast cancer patients.²¹ However, it has been unclear how CHIP prevents breast cancer progression.

In the present study, we showed that CHIP knockdown enhances heterogeneous Bcl-2 expression levels among breast cancer cells, depending on their genetic variation. Consequently, CHIP knockdown increases the proportion of anticancer drug-resistant cells. Our findings suggest that CHIP acts as a capacitor of heterogeneous Bcl-2 expression levels to suppress the malignant progression of breast cancer.

Results and discussion

CHIP expression was significantly associated with prognosis of breast cancer patients.²¹ However, its underlying molecular mechanisms remain to be elucidated. Accordingly, to investigate the underlying mechanism by which CHIP regulates breast cancer prognosis, CHIP expression levels in breast cancer cell lines were reduced using shRNA (shCHIP; Figure 1a and Supplementary Figure 1). The cells were treated with the anticancer drug cisplatin²² to evaluate their sensitivity to the drug. In the absence of cisplatin, the number of shCHIP cell colonies was comparable to that of cells treated with control shRNA (shCtrl; Figure 1b). However, in the presence of cisplatin, the number of shCHIP cell colonies was much higher than that of the control cells (Figure 1b), suggesting that CHIP knockdown reduced cells’ sensitivity to cisplatin.

Figure 1

CHIP knockdown increases the proportion of anticancer drug-resistant cells. (a) CHIP protein levels in MCF-7 or T47D cells after the introduction of shRNA against LacZ (shCtrl) and CHIP (shCHIP). MCF-7 and T47D cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and penicillin–streptomycin mixed solution (Nacalai Tesque). A RNA interference experiment was performed as described in Kajiro et al.,¹⁶ with minor modification. We used a retroviral expression system; GP2-293 cells (Clontech) were co-transfected with the pVSV-G vector and the pSINsi-hU6 (Takara) vector containing either the CHIP or LacZ (control) target sequence. MCF-7 and T47D cells were incubated with the retroviral supernatant in the presence of 8 μg/ml polybrene. Twenty-four hours after infection, the viral supernatant was replaced with fresh DMEM containing 10% FBS. The infected cells were selected with 1 mg/ml G418. The target sequences were 5′-gcacgacaagtacatggcgga-3′ for CHIP and 5′-gctacacaaatcagcgatt-3′ for LacZ. Immunoblotting was performed as described in Kajiro et al.,¹⁶ with minor modification. Cells were lysed in buffer containing 50 mM Tris (pH 7.5), 150 mM NaCl, 0.5% Triton-X-100 and 0.5 mM EDTA. Mouse monoclonal antibodies against β-Actin (1:1000; Santa Cruz, clone C4) and rat monoclonal antibodies against human CHIP (1:250; Green Space Biomed, Japan) were used for immunoblotting. (b) Colony-formation capability of MCF-7 or T47D shCtrl and shCHIP cells with cisplatin treatment. Colonies were stained with crystal violet and images of MCF-7 colonies are shown (left panel). The number of colonies for these conditions is shown in the right panel. MCF-7 cells were seeded at 6.0 × 10³ cells per 10-cm dish. After 24 h, the cells were treated with 5 μM cisplatin (LKT Laboratories). The cells were then cultured for 10 days. Surviving colonies were stained with 2% crystal violet (Nacalai Tesque) in methanol and counted. T47D cells were seeded at 8.0 × 10³ cells per 10-cm dish. After 48 h, the cells were treated with 5 μM cisplatin and cultured for 2 weeks. Surviving colonies were stained with 2% crystal violet in methanol and counted. (c) Cell viability of shCtrl and cisplatin-resistant (cisR-) shCHIP cells treated with the indicated anticancer drugs. cisR-shCHIP clones (Cl) were isolated from among shCHIP cells survived under cisplatin treatment condition as described in Figure 1b. shCtrl cells without cisplatin treatment were used as a control. shCtrl and cisR-shCHIP MCF-7 and T47D cells were seeded at 1.0 × 10⁵ cells/ml in 12-well culture plate. The cells were treated with 75 μM cisplatin, 100 nM paclitaxel (Wako), 25 μM etoposide (Calbiochem) or 750 nM epirubicin (Pfizer) for MCF-7 cells, and 75 μM cisplatin, 750 nM paclitaxel, 2.5 μM etoposide and 250 nM epirubicin for T47D cells. After 48 h, surviving cells were stained with 0.4% trypan blue and counted with Countess (Invitrogen). Error bars represent s.d.; n=3 for b; n⩾3 for c. *P<0.05; **P<0.01, Student’s t-test.

Three clones (cisR-shCHIP) from cisplatin-resistant shCHIP cells (Supplementary Figures 1 and 2) were isolated and their sensitivity to other anticancer drugs was evaluated. The number of viable cells in cisR-shCHIP clones was higher than that of control cells after exposure to these drugs (Figure 1c and Supplementary Figure 3a).

Bcl-2 expression levels were measured in cisR-shCHIP cells, considering that Bcl-2 has been reported to participate in resistance to anticancer drugs.^{9, 10, 11, 12} As expected, the Bcl-2 mRNA and protein levels in cisR-shCHIP cells were higher than those in control cells (Figure 2a and Supplementary Figure 2 and 4). To confirm that Bcl-2 was responsible for anticancer drug resistance in cisR-shCHIP cells, Bcl-2 expression was depleted in cisR-shCHIP cells using shRNA (cisR-shCHIP/Bcl-2; Figure 2b). Bcl-2 downregulation resulted in reduced numbers of viable cisR-shCHIP cells in the presence of anticancer drugs (Figure 2c and Supplementary Figure 3b).

Figure 2

CHIP knockdown reduces breast cancer cells’ Bcl-2-dependent sensitivity to anticancer drugs. (a) Bcl-2 mRNA levels in shCtrl and cisR-shCHIP cells. Real-time reverse transcription PCR (RT–PCR) was performed as described in Kajiro et al.,¹⁶ with minor modification. Cells were lysed in 1 ml of Sepasol-RNA I Super G (Nacalai Tesque) and total RNA was extracted according to the instruction manual and treated with DNase (Promega) for 30 min at 37 °C. cDNA was synthesized from 1 μg total RNA using RevatraAce reverse transcriptase (Toyobo) and random primers. The cDNA was amplified by real-time RT–PCR using Thermal Cycler Dice TP800 (Takara) and SYBR Premix Ex Taq (Takara). Samples were normalized by PPIA mRNA levels. The primers for real-time PCR are as follows: 5′-ctgggaatcgatctggaaatcc-3′ and 5′-tgcataaggcaacgatcccatc-3′ for Bcl-2, 5′-acgtggtataaaaggggcgggag-3′ and 5′-tcaccaccctgacacataaaccctg-3′ for PPIA. (b) CHIP and Bcl-2 protein levels in shCtrl, shBcl-2, cisR-shCHIP and cisR-shCHIP/Bcl-2 MCF-7 cells. RNA interference was performed as described in Figure 1a with pSUPER-retro (OligoEngine) vector containing either the Bcl-2 or luciferase (control) target sequence. shCtrl and cisR-shCHIP MCF-7 cells were infected with the retroviral supernatant. The infected cells were selected with 1 μg/ml puromycin. The target sequences were 5′-tggatgactgagtacctgaac-3′ for Bcl-2 and 5′-gaagctgcgcggtggtgttgt-3′ for luciferase. Immunoblotting was performed as described in Figure 1a. A mouse monoclonal antibody specific for Bcl-2 (1:1000; BD Biosciences, clone 7) was used in this experiment. (c) Viable cell numbers among shCtrl, shBcl-2, cisR-shCHIP and cisR-shCHIP/Bcl-2 cells treated with anticancer drugs. The cells were treated with the indicated anticancer drugs and surviving cells were counted as described in Figure 1c. (d) Tumor weights in mice with shCtrl, shBcl-2, cisR-shCHIP and cisR-shCHIP/Bcl-2 cell-derived xenografts and injected with cisplatin. Representative tumors are shown in the left panel. BALB/cAJcl-nu/nu female mice 4-week-old were purchased from CLEA Japan. MCF-7 cells were trypsinized and resuspended in Matrigel (BD Biosciences) at 1.0 × 10⁸ cells/ml. The mice were given bilateral subcutaneous injections of 1.0 × 10⁷ cells (0.1 ml) and kept in a pathogen-free environment. When tumors reached ~100 mm³, the mice were treated with cisplatin (4 mg/kg) or phosphate-buffered saline (PBS; Nissui) as a vehicle every 2 weeks by i.p. injection. After 53 days, all mice were killed and tumor tissues were collected and weighed. All animal experiments were in accordance with institutional guidelines. Scale bar=1 cm. Error bars represent s.d.; n=3 for a; n⩾3 for c and d. *P<0.05; **P<0.01; NS, not significant, Student’s t-test.

Mouse xenograft models were used to assess the involvement of CHIP and Bcl-2 in tumor formation (Figure 2d). In mice with shCtrl and shBcl-2 cell xenografts, tumor weights were reduced after cisplatin treatment. In mice with cisR-shCHIP cell xenografts, tumor weights were increased and were not markedly reduced by cisplatin treatment. In contrast, tumor weights in mice with cisR-shCHIP/Bcl-2 xenografts were significantly reduced by cisplatin treatment (Figure 2d). Taken together, these results indicate that the population of anticancer drug-resistant cells that increased owing to CHIP knockdown was dependent on Bcl-2 expression.

CHIP was added back to cisR-shCHIP clones to investigate the mechanism of increased Bcl-2 expression; however, increased Bcl-2 expression levels were not reduced by CHIP re-expression (Figure 3a). These results indicate that Bcl-2 expression in cisR-shCHIP clones was irreversibly altered. In addition, transient CHIP knockdown by siRNA for 48 h did not change the expression levels of Bcl-2 (Supplementary Figure 5). This finding, together with Figure 3a, indicates that CHIP does not directly change the expression levels of Bcl-2 by protein degradation or transcriptional regulation. It is thus possible that cisplatin treatment resulted in the selection of high Bcl-2-expressing clones from among shCHIP cells.

Figure 3

CHIP knockdown enhances variation in Bcl-2 expression level and contributes to anticancer drug resistance through the selection of Bcl-2^high-expressing cells. (a) Bcl-2 protein levels in cisR-shCHIP cells in which CHIP expression was recovered. For overexpression of CHIP, we used the ViraPower Adenoviral Expression System (Invitrogen), according to the manufacturer’s protocol. To generate CHIP adenoviral supernatants, 293A cells were co-transfected with pAd/CMV/V5-DEST vector containing either the full-length of CHIP (Ad-CHIP) or GFP (Ad-GFP). MCF-7 cells were incubated with each adenoviral supernatant. Protein was extracted 48 h after adenoviral infection. Immunoblotting was performed as described in Figures 1a and 2b. (b) Bcl-2 mRNA levels in MCF-7, T47D shCtrl and shCHIP clones (left panels). We isolated shCtrl or shCHIP clones from MCF-7 or T47D cells in normal culture condition as described in Figure 1a and performed real-time RT–PCR as described in Figure 2a. Box plots of Bcl-2 mRNA levels in shCtrl and shCHIP clones are shown (right panels). (c) Bcl-2 protein expression patterns in MCF-7 shCtrl and shCHIP cells. The cells were permeabilized and fixed with FACS Permeabilizing Solution 2 (BD Biosciences) for 10 min at room temperature (RT). The cells were washed with 0.5% BSA in PBS and stained by FITC-conjugated anti-Bcl-2 antibody (BD Biosciences, clone 100) for 30 min in the dark at RT. The cells were evaluated for Bcl-2 expression using FACSAria (BD Biosciences). Collected events were analyzed using FlowJo (Tree Star). (d) Bcl-2 protein expression patterns in MCF-7 shCHIP cells with and without cisplatin treatment. shCHIP cells were treated with cisplatin for 9 days as described in Figure 1b, and FACS analysis was performed as described in Figure 3c. P-values compared with control were shown in b and c, statistical F-test.

To avoid the possible effects of cisplatin selection, shCtrl and shCHIP clones were isolated from MCF-7 and T47D cell lines without cisplatin treatment (Supplementary Figure 1). Bcl-2 and CHIP expression levels were evaluated for each clone. There was considerable variation in the Bcl-2 mRNA levels among shCHIP clones (Figure 3b, left). A box plot analysis of these data indicated that the divergence of Bcl-2 expression from mean levels was significantly higher for shCHIP clones than for shCtrl clones (Figure 3b, right; F-test, P=2.80 × 10⁻¹² for MCF-7 and P=7.28 × 10⁻⁶ for T47D). To exclude the possibility that the difference in Bcl-2 expression levels among shCHIP clones was due to CHIP knockdown efficiency, we evaluated the correlation between CHIP and Bcl-2 expression levels, but found no significant correlation (Supplementary Figure 6; Spearman correlation analysis, r_s=0.02, P=0.45 for MCF-7 and r_s=0.22, P=0.095 for T47D).

Variation in Bcl-2 protein levels was assessed by fluorescence-activated cell sorting among cells expressing shCtrl and shCHIP without cisplatin treatment. Compared with shCtrl cells, variation in Bcl-2 protein levels was also increased among shCHIP cells (Figure 3c; F-test, P=5.70 × 10⁻¹⁴⁷). Taken together, these results indicate that CHIP suppresses variations in Bcl-2 expression levels in breast cancer cell lines.

The effect of cisplatin treatment on Bcl-2 expression profiles in shCHIP cells was also evaluated. In the cells that survived cisplatin treatment, a Bcl-2^low-expressing population was decreased, whereas a Bcl-2^high-expressing population was increased (Figure 3d). The results indicate that these variations in Bcl-2 expression among shCHIP cells contributed to drug resistance through the selection of Bcl-2^high-expressing cells.

Next, we addressed why the variation in Bcl-2 levels increased following CHIP knockdown. It has been reported that CHIP regulates cellular protein quality control in cooperation with HSPs, such as HSP90 and HSP70.^{18, 19, 20, 23} In Drosophila melanogaster and Arabidopsis thaliana, HSP90 masks heritable variation, such as genetic and epigenetic variation, during developmental processes.^{24, 25, 26} The MCF-7 and T47D cell lines were derived from uncloned human breast cancer cells.^{27, 28, 29} It is thus possible that CHIP knockdown resulted in the expression of cryptic genetic and/or epigenetic variation among these cells. To test this hypothesis, we isolated single-cell clones (sc-MCF-7) from the MCF-7 cell line and used the clones to generate sc-MCF-7 shCtrl and shCHIP clones (Supplementary Figure 1 and Figure 4a). In contrast to uncloned MCF-7 cells (Figure 3b), the variation in Bcl-2 mRNA levels among sc-MCF-7 shCHIP clones was not significantly different from that among sc-MCF-7 shCtrl clones (Figure 4b; F-test, P=0.52 for sc-MCF-7 #1 and P=0.83 for sc-MCF-7 #2). CHIP knockdown did not affect the variation in Bcl-2 protein levels among sc-MCF-7 cells (Figure 4c). In addition, the colony-formation capability of sc-MCF-7 cells was not significantly increased owing to CHIP knockdown in the presence of cisplatin (Figure 4d). These results indicate that Bcl-2 expression and cisplatin resistance were not altered by CHIP knockdown in cells with the same genetic and epigenetic background.

Figure 4

CHIP buffers variation in Bcl-2 expression level caused by genetic background. (a) CHIP protein levels in single-cell clones (#1 or #2) from MCF-7 cells (sc-MCF-7) after the introduction of shRNA against LacZ (shCtrl) and CHIP (shCHIP). We isolated two clones from MCF-7 cells (#1 and #2), and generated shCtrl or shCHIP cells from the clonal cells (sc-MCF-7 shCtrl or shCHIP cells). RNA interference was performed as described in Figure 1a. We performed immunoblotting using the cells as described in Figure 1a. (b) Bcl-2 mRNA levels in shCtrl and shCHIP clones derived from sc-MCF-7 cells. We isolated clones from sc-MCF-7 shCtrl or shCHIP cells and evaluated the Bcl-2 mRNA levels of the clones as described in Figure 3b. (c) Bcl-2 protein expression patterns in sc-MCF-7 shCtrl and shCHIP cells. sc-MCF-7 shCtrl and shCHIP cells were evaluated for Bcl-2 expression using FACSAria as described in Figure 3c. (d) Colony formation rate of sc-MCF-7 shCtrl and shCHIP cells with cisplatin treatment. A cisplatin-treated colony-formation experiment was performed as described in Figure 1b. (e) Bcl-2 expression patterns in shCtrl and shCHIP cells derived from sc-MCF-7 #1 with or without ENU treatment. sc-MCF-7 #1 shCtrl and shCHIP cells were treated with 1 μg/ml N-ethyl-N-nitrosourea (ENU; Sigma-Aldrich) once weekly. After the treatment, the cells were evaluated for Bcl-2 expression using FACSAria as described in Figure 3c. (f) Bcl-2 expression patterns in ENU-treated sc-MCF-7 #1 after the introduction of siRNA for luciferase (siCtrl) or CHIP (siCHIP). sc-MCF-7 #1 cells were treated with 1 μg/ml ENU once a week three times. The ENU-treated cells were treated with siRNA for luciferase (control; Invitrogen) or CHIP using Lipofectamine RNAiMAX (Invitrogen) according to the instruction manual. siRNA was reintroduced 5 days after the first introduction. The target sequences were 5′-ggcaatcgtctgttcgtgggccgaa-3′ for siCHIP#1 and 5′-ccagcgctcttcgaatcgcgaagaa-3′ for siCHIP#2. Zero, 3, 7, and 10 days following siRNA treatment, cells were analyzed for Bcl-2 expression using FACSAria as described in Figure 3c. (g) Scatter plots of genome-wide gene expression profiles of parental versus clonal shCtrl and shCHIP cells. Numbers in parentheses indicate numbers of probe sets for which expression levels differed by⩾2 between clonal and parental cells. Total RNA was extracted with a FastPure RNA Kit (Takara). Purified RNA (100 ng) was used to synthesize labeled aRNA and aRNA was fragmented using a GeneChip 3′IVT Expression Kit (Affymetrix), following the manufacturer’s protocol. Using 12.5 μg fragmented labeled aRNA, a hybridization mix was prepared according to the protocol and hybridized to U133 Plus 2.0 GeneChips (Affymetrix). The chips were then washed in the GeneChip Fluidics Station (Affymetrix) and finally scanned using the GeneChip Scanner 3000 (Affymetrix). Raw data were obtained using GCOS software version 1.0 and initially analyzed using GeneSpring version 12.1 (Agilent Technologies). Data sets have been deposited in the Gene Expression Omnibus of NCBI under accession number GSE49349. (h) Number of probes for which expression levels differed by⩾2 between clonal and parental MCF-7 shCtrl and shCHIP cells. P-values compared with control were shown in b, c, e and f, statistical F-test. Error bars represent mean s.d.; n=3 for d; n⩾7 for h. *P<0.05; **P<0.01; NS, not significant, Student’s t-test.

To test whether genetic diversity was a cause of increased variation in Bcl-2 expression levels resulting from CHIP knockdown, sc-MCF-7 shCtrl and shCHIP cells were treated with the mutagen N-ethyl-N-nitrosourea (ENU).³⁰ For sc-MCF-7 shCHIP cells, the variation in Bcl-2 expression levels was significantly higher than that for sc-MCF-7 shCtrl cells, and this difference was dependent on the frequency of ENU treatment (Figure 4e). Then sc-MCF-7 cells were treated with ENU to change their genetic variation. After the treatment, CHIP expression in the cells was depleted using siRNA in the absence of ENU and the time-dependent changes of Bcl-2 expression were evaluated with fluorescence-activated cell sorting. CHIP knockdown proved to increase the variation in Bcl-2 protein levels in a time-dependent manner (Figure 4f). Taken together, these results suggest that a difference in genetic background causes increased variation in Bcl-2 expression in MCF-7 cells. ENU is originally used to induce mutations, but it is also possible that ENU indirectly affects epigenetic gene regulation. We can thus not exclude the effects of ENU on epigenetic background.

The expression levels of Bcl-2 are reportedly controlled by the regulation of mRNA stability in the 3′- and 5′-untranslated regions of Bcl-2 mRNA^{31, 32} or by the levels of transcription.^{33, 34, 35, 36, 37, 38, 39} In our data, the transcription levels of Bcl-2 were markedly increased in cisR-shCHIP cells (nuclear run-on assay; data not shown). We infer that the changes in transcriptional regulation may be one of the causes of the variation in Bcl-2 expression. Numerous genes are reported as transcriptional regulators of Bcl-2, including p53,³³ WT-1,³⁴ CREB,³⁵ NF-kB,³⁶ E2F-1,³⁷ AP-2³⁸ and BP1.³⁹ Numerous genetic mutations are accumulated in cancer cells and associated with gene expression, as has been reported in breast cancer studies.^{40, 41} We accordingly consider that the factor(s) with mutation that is buffered by CHIP and affects Bcl-2 expression may be among the regulator genes of Bcl-2. We have not yet identified these genes, and consider their identification a long-term goal.

To evaluate the effect of CHIP knockdown on the divergence of global gene expression levels from their mean levels, we then compared DNA microarray data from shCHIP and shCtrl clones with those in their parental cells. Scatter plots of these data showed that the difference between gene expression in clonal cells and that in parental cells was higher for shCHIP cells than that for shCtrl cells (Figure 4g and Supplementary Figure 7). The number of probes for which the expression levels differed by⩾2 between clonal cells and parental cells was significantly increased for shCHIP cells as compared with that for shCtrl cells (Figure 4h). These results indicated that CHIP knockdown affects variation in global gene expression.

Cancer heterogeneity can arise in multiple ways, such as by genetic and epigenetic variation or cancer stem cell differentiation or in response to extrinsic environmental differences.⁴² Among these ways, genetic variation is known to be a major driver of phenotypic heterogeneity among cancer cells.⁴³ Overall, our results suggest that CHIP buffers heterogeneous Bcl-2 expression among breast cancer cells, which is typically caused by genetic variation. This phenomenon is a cause of augmentation of anticancer drug-resistant cells (Supplementary Figure 8). Our results also suggest that the degree of heterogeneity in expression levels is decided by a balance between the genetic variation and buffering capacity of CHIP, with this balance linked to breast cancer progression.

Previous studies indicate that HSP90 buffers phenotypic heterogeneity in the developmental processes of D. melanogaster and A. thaliana.^{24, 25, 26} It has been reported that HSP90, which has an important function in the protein quality control system by refolding mis- or unfolded proteins,⁴⁴ masks heritable variation, such as genetic and epigenetic variation, and acts as a capacitor for phenotypic heterogeneity in developmental processes.^{24, 25, 26} Here we suggest that CHIP, which participates in the protein quality control system,¹⁹ buffers heterogeneous expression levels in a manner that is dependent on the genetic background of human breast cancer cells. Our findings show the possibility that the buffering capacity of protein quality control-related factors is involved in cancer heterogeneity.