The CRL7FBXW8 Complex Controls the Mammary Stem Cell Compartment through Regulation of NUMB Levels

Abstract

NUMB is a tumor suppressor gene that functions by inhibiting the action of the NOTCH proto-oncogene and enhancing the levels and activity of the tumor suppressor protein p53. In breast cancer (BC), NUMB loss of function (LOF), mediated by various molecular mechanisms, is a frequent and causal event. Herein, it is established that loss of NUMB protein, resulting from protein hyper-degradation, is the prevalent mechanism of NUMB LOF in BC. Through an RNAi-based screening, the CRL7^FBXW8 complex is identified as the E3 ligase complex responsible for NUMB hyper-degradation in BC. Genetic and pharmacological inhibition of CRL7^FBXW8 rescues the transformation-related phenotypes induced by NUMB LOF in BC cell lines and in patient-derived xenografts. These effects are directly dependent on the restoration of NUMB protein levels. Thus, enhanced CRL7^FBXW8 activity, through its interference with the tumor suppressor activity of NUMB, is a causal alteration in BC, suggesting it as a potential therapeutic target for precision medicine.

Keywords: CRL7FBXW8 complex; NUMB; breast cancer; cancer stem cells; proteasomal degradation.

Figure 1

Loss of NUMB expression at the protein level in BCs correlates with prognosis. a) Kaplan–Meier (KM) survival analysis of the IEO cohort. Time to death related to BC (DRBC) was determined in patients stratified by NUMB IHC status (deficient or proficient). Reliable IHC data were obtained for 808 patients. b) Univariate and multivariable analyses of the BCs analyzed in panel (a). HR, hazard ratio. c) Relative NUMB mRNA levels, determined by trimmed mean of M values (TMM) normalization of RNAseq data, in NUMB‐deficient and NUMB‐proficient BCs, defined by IHC, as described in panel (a). The analysis was performed on 716 patients for whom reliable IHC and RNAseq data were available. P‐value was calculated by the nonparametric Wilcoxon test using JMP. d) KM survival analysis of the 716 patients defined in panel (c), stratified by levels of NUMB mRNA expression determined by RNAseq. Left: BCs were classified as NUMB‐low or NUMB‐high relative to the average NUMB mRNA expression across the entire cohort. Right: BCs were stratified based on the quintiles of NUMB mRNA levels. The Cox proportional hazards model was used to calculate the hazard ratio and corresponding p‐values. e) Pie chart showing the relative contribution of different NUMB LOF mechanisms in BC. Data on NUMB protein levels, expression of Ex3‐containing NUMB isoforms, and NUMB hyper‐phosphorylation in the 890 case cohort were extracted from this paper and our previous publications.^{[ 1b,e ]} In 683 cases, data were retrievable for all three categories (missing samples were mainly due to loss of tissue‐microarray cores in the “NUMB hyper‐phosphorylation” category^{[ 1e ]}). Low mRNA expression of Ex3‐containing isoforms occurred in the presence (11% of cases) or absence (7% of cases) of decreased NUMB protein expression. The cumulative frequency of NUMB alterations was ≈65%. Other mixed alterations: low NUMB + NUMB hyper‐phosphorylation (2.6%), low Ex3‐containing isoforms + NUMB hyper‐phosphorylation (1.2%), all three alterations together (0.4%).

Figure 2

The CRL7^FBXW8 complex is responsible for NUMB degradation in a NUMB‐deficient BC cell line. a) MDA‐MB‐361 and MDA‐MB‐231 BC cell lines were treated for 24 h with bortezomib (BTZ, 20 nm), MLN4924 (MLN, 0.5 µm), or vehicle (Veh), and effects on NUMB levels were determined by immunoblotting (IB). p21 was used as a control for proteasome inhibition by BTZ. MLN efficacy was controlled by assessing the levels of NEDDylated CUL1 (upper band, red arrowhead) compared with non‐NEDDylated CUL1 (lower band, black arrowhead). Actin, loading control. Results are representative of three independent experiments. In this and all subsequent figures: molecular weigth (MW) markers for IB are shown in KDa. b) NUMB mRNA levels in MDA‐MB‐361 or MDA‐MB‐231 cells determined by RT‐qPCR (reverse transcription quantitative polymerase chain reaction). Left: basal expression (expressed as 2^{−Δ Ct} ); right: mRNA levels upon treatment with BTZ or MLN as in panel (a). In the right panel, the 2^{−ΔΔ Ct} method was used to assess relative NUMB expression, using Veh‐treated cells as the reference sample. Data are expressed as mean ± standard error (SE) (normalized to Veh in the right panel) from four independent experiments. No statistically significant differences were observed in the unpaired t‐test. c) Organotypic cultures of MDA‐MB‐361 or MDAMB‐231 cells were established in Matrigel and treated for ≈24 h with 20 nm BTZ, 0.5 µm MLN or Veh when organoids reached ≈40 µm in diameter. Organoids were fixed and analyzed for NUMB and p21 expressions by immunofluorescence (IF). Nuclei were counterstained with DAPI (4‐'6‐diamidino‐2‐phenylindole). Results are representative of two independent experiments. Bar = 100 µm. d) Transient knockdown (KD) of RBX1 or FBXW8 in MDA‐MB‐361 and MDA‐MB‐231 cells using different siRNA (small interfering RNA) oligos (#1 and #2). As negative controls cells were transfected with nontargeting control (Ctrl) siRNA, or mock‐transfected. BTZ and MLN treatments (as in panel (a)) were used as positive controls for NUMB restoration. IB was performed with the indicated Ab. The arrows in the CUL1 blot are as in panel (a). For FBXW8, the two bands represent different isoforms. Actin, loading control. e) Individual CRL7^FBXW8 complex components were transiently silenced in MDA‐MB‐361 cells. For CUL1 and CUL4A, the arrowheads point to NEDDylated (red) and non‐NEDDylated (black) forms. For FBXW8, green arrowheads point to the two isoforms. For CUL7, the black arrowhead points to the protein (the lower band visible in the blot is nonspecific). MDA‐MB‐231 cells were used as a reference for “normal” NUMB expression levels. f) FLAG‐tagged NUMB was transfected into HEK293 cells. Lysates were immunoprecipitated (IP) with anti‐FLAG, with or without FLAG peptide to displace the specific interactions, or with anti‐HA (human influenza hemagglutinin epitope) as an additional negative control. Interacting proteins were detected by IB. p53, a known NUMB interactor, served as positive control. For CUL1, arrowheads point to NEDDylated (red) and non‐NEDDylated (black) forms. Note that in HEK293 cells, FBXW8 migrates as a single band. Input, 15 µg; IP, 8 mg. Results are representative of three independent experiments.

Figure 3

NUMB polyubiquitination in BC cell lines requires the CRL7^FBXW8 complex. a) MDA‐MB‐361 (361) and MDA‐MB‐231 (231) cells were treated for 24 h with MLN4924 (MLN, 0.5 µm), or vehicle (Veh). Lysates were then immunoprecipitated (IP, Abs are shown on the top) and immunoblotted (IB, as shown on the right). Input, 10 µg; IP, 6 and 3 mg for MDA‐MB‐361 and MDA‐MB‐231, respectively. b) MDA‐MB‐361 and MDA‐MB‐231 BC cells were transiently silenced for RBX1 or FBXW8 or with a nontargeting control (Ctrl) siRNA. Lysates were IP (as shown on the top) and IB (as shown on the right). Input, 10 µg; IP, 6 and 3 mg for MDA‐MB‐361 and MDA‐MB‐231, respectively. c) MDA‐MB‐361 cells were stably transduced with the doxycycline (DOX)‐inducible small hairpin RNA (shRNA) constructs (pTRIPZ‐based vectors) targeting the indicated E3 ligases, and then treated with DOX. Left: IB was performed with the indicated Abs. MDA‐MB‐231 cells (231) were used as a NUMB‐proficient control. Vinculin, loading control. Results are representative of two independent experiments. Right: RT‐qPCR analysis of the expression levels of KD target genes (black bars) and NUMB (white bars), relative to the Ctrl KD (dashed red line = 1). Data are the mean ± standard deviation (SD) of technical triplicates.

Figure 4

The CRL7^FBXW8 complex modulates NUMB‐dependent biological activities. a) MDA‐MB‐361 and MDA‐MB‐231 cells were stably transduced with the indicated DOX‐inducible shRNA constructs (pTRIPZ‐based vectors) and treated or not with DOX. IB was performed with the indicated Ab. Red Fluorescent Protein (RFP) is a reporter expressed upon DOX stimulation, used as a control for shRNA induction efficiency. p21, positive control for RBX1 KD. Actin, loading control. The MDA‐MB‐361 and MDA‐MB‐231 blots come from a single larger blot in which irrelevant lanes were spliced out. The NUMB blots represent different exposures (shorter exposure for MDA‐MB‐231 because of high endogenous NUMB levels). b) Representative images of a 3D Matrigel assay performed with MDA‐MB‐361 (left) and MDA‐MB‐231 (right) cells, silenced with the indicated shRNAs. After fixation, organoids were stained with anti‐NUMB (pseudocolored in red) and DAPI (blue). RFP (pseudo‐colored in gray), reporter of DOX‐induced shRNA expression. Bar, 50 µm. In MDA‐MB‐361 cells, the reduction in organoid size following RBX1 or FBXW8 KD was consistent with the increased NUMB levels. Data are representative of two independent biological replicates. c) MDA‐MB‐361 or d) MDA‐MB‐231 cells stably transduced with the DOX‐inducible shRNA pTRIPZ constructs, as in panel (a), were analyzed for organoid‐forming efficiency (OFE) in Matrigel or spheroid‐forming efficiency (SFE) in methylcellulose, in the presence of DOX. e,f) MDA‐MB‐361 cells were stably silenced for NUMB with a DOX‐inducible shRNA construct (pTRIPZ‐vector) and further silenced for RBX1 or FBXW8 (SMARTvector particles) and analyzed for e) OFE) and f) SFE. Protein and mRNA levels are shown in Figure S4b,c (Supporting Information). c–f) Data are normalized to Ctrl in each experiment and expressed as mean ± SD. The number of biological replicates is shown (n = 3). Statistical analysis was performed with the unpaired t‐test, and significant differences are indicated by asterisks. In this and all subsequent figures, *, p < 0.05, **, p < 0.01, and ***, p < 0.001.

Figure 5

Tumorigenic potential of NUMB‐deficient BC cells is dependent on CRL7^FBXW8‐mediated NUMB degradation. a) Scheme of the BTZ (black circles) and MLN (red stars) in vivo treatment protocol. Cells were transplanted into the inguinal mammary fat pads of NSG mice. When xenografts reached palpable size (≈20 mm³) drugs were administered as indicated. b) Growth curves of MDA‐MB‐361 or MDAMB‐231 xenografts in NSG mice, mock‐treated (vehicle), or treated with BTZ (350 µg kg⁻¹, once a day) or MLN (30 mg kg⁻¹, twice daily). Data are expressed as mean ± SE. Statistical analysis was performed with the unpaired t‐test and significant differences between treated versus vehicle control are indicated; n, number of tumors per experimental group (in this and all subsequent figures). Inset images show typical final tumor sizes. Bar, 1 cm. c) Tumors, excised at the end of the experiments shown in panel (b), were IB as indicated. For CUL1, the arrowheads point to NEDDylated (red) and non‐NEDDylated (black) forms. d) Effects of in vitro BTZ (20 nm) and MLN (0.5 µm) treatments on the OFE (Matrigel) and SFE (methylcellulose) of MDA‐MB‐361 or MDA‐MB‐231 cells. Data are normalized to respective control and expressed as mean ± SD. Statistical analysis was performed with the unpaired t‐test. e) Scheme of the BTZ and DOX treatment protocol used for xenografts of MDA‐MB‐361 cells stably expressing the DOX‐inducible shRNA‐NUMB construct. BTZ was administered as indicated by red arrows. DOX was administered via food pellets over the period indicated by the black arrow. Representative animals were sacrificed at days 1 and 7 to assess NUMB levels in xenografts by IHC. f) Growth curves of MDA‐MB‐361‐inducible‐shNUMB xenografts treated with BTZ or vehicle in the presence and absence of DOX. Data are expressed as mean ± SE. Statistical analysis was performed with unpaired t‐test, and significant differences between BTZ/+DOX versus BTZ/‐DOX samples are indicated. g) Analysis of NUMB expression by IHC on FFPE sections of xenografts as in panel (f), at days 1 and 7 of the experiment (see also panel (e)). Bar = 100 µm.

Figure 6

The CRL7^FBXW8 complex is responsible for the tumorigenicity of NUMB‐deficient BC cells. a) Scheme of the treatment protocol of xenografts derived from MDA‐MB‐361 or MDA‐MB‐231 cells transduced with DOX‐inducible constructs expressing shRBX1, shFBXW8, or control shRNA (shCtrl). DOX was administered via food pellets over the indicated period. Two groups of xenografted mice were treated in parallel with BTZ or vehicle (red arrows) as controls. b,c) Growth of b) MDA‐MB‐361 or c) MDAMB‐MB‐231 xenografts treated as outlined in panel (a). Data are expressed as mean ± SE. Statistical analysis was performed with unpaired t‐test. Significant differences between DOX‐/BTZ‐treated versus control samples are indicated. n, number of tumors per experimental group. d) Representative final xenograft sizes from experiments in panels (b) and (c). Bar, 1 cm. e) Xenografts excised at the end of the experiments in panels (b) and (c) were analyzed by IB as indicated. RFP is a reporter of DOX‐induced shRNA expression used to assess the efficacy of DOX treatment.

Figure 7

Targeting of the ubiquitin‐proteasome system has antitumoral effects in NUMB‐deficient PDXs. a) NUMB, hormone receptor (ER, estrogen receptor; PgR, progesterone receptor) and HER2 status in the selected PDXs are shown. Def, deficient; Pro, proficient, Neg, negative. b) IHC staining of NUMB expression in the primary tumors from which the PDXs were established. Bar, 100 µm. c) NUMB mRNA levels in the indicated PDXs relative to the reference cell line MDA‐MB‐231 (dashed line). Results are expressed as mean ± SD, normalized to MDA‐MB‐231 (2^{−ΔΔ Ct} method). d) SFE in methylcellulose of PDX‐derived primary cells treated or not at plating with BTZ (20 nm) or MLN (0.5 µm). Data are normalized to corresponding vehicle (Veh) control and expressed as mean ± SD (n = 3). Statistical analysis was performed with the unpaired t‐test. Significant differences are indicated. e,f) Growth of PDXs in NSG mice treated with BTZ or vehicle alone following the treatment protocol in Figure 5a. Data are expressed as mean ± SE. Statistical analysis was performed with unpaired t‐test. Significant differences are indicated. The insets show typical final tumor sizes. Bars, 1 cm.

Figure 8

CRL7^FBXW8‐mediated NUMB degradation is essential for maintenance of the CSC compartment in clinical NUMB‐deficient BCs. a) SFE and b) OFE of PDX‐derived primary cells transduced with the indicated pTRIPZ lentiviral vectors encoding DOX‐inducible shRNA constructs, in the presence of DOX. Data are normalized to the corresponding Ctrl‐shRNA sample and expressed as mean ± SD (n = 3). Statistical analysis was performed with the unpaired t‐test. Significant differences are indicated. c) NUMB‐deficient T2 BC primary cells transduced as in panel (a) were transplanted in NSG mice. Once outgrowths were palpable; mice were administered DOX as shown in Figure 6a; and tumor growth monitored. Data are expressed as mean ± SE. Statistical analysis was performed with the unpaired t‐test. Significant differences between shRBX1 or shFBXW8 versus shCtrl samples are indicated. d) Tumors at the end of the experiment in panel (c) were stained by IHC to detect NUMB levels. Bar, 100 µm. e) NUMB‐deficient T2 BC primary cells transduced as in panel (a) were assessed for frequency of tumor‐initiating cells (TICs) in a limiting‐dilution xenograft assay. The ELDA software (https://bioinf.wehi.edu.au/software/elda/) was used for statistical analysis.^{[ 32 ]} The frequency of TICs, with 95% confidence intervals in parentheses, and the pairwise test for differences in SC frequency versus control, are shown. f–h) The same set of experiments reported in panels (c)–(e) was replicated with the NUMB‐proficient TA primary BC culture transduced as indicated in panel (a). g) Bars, 100 µm.