CDK12 promotes tumorigenesis but induces vulnerability to therapies inhibiting folate one-carbon metabolism in breast cancer

Abstract

Cyclin-dependent kinase 12 (CDK12) overexpression is implicated in breast cancer, but whether it has a primary or only a cooperative tumorigenic role is unclear. Here, we show that transgenic CDK12 overexpression in the mouse mammary gland per se is sufficient to drive the emergence of multiple and multifocal tumors, while, in cooperation with known oncogenes, it promotes earlier tumor onset and metastasis. Integrative transcriptomic, metabolomic and functional data reveal that hyperactivation of the serine-glycine-one-carbon network is a metabolic hallmark inherent to CDK12-induced tumorigenesis. Consistently, in retrospective patient cohort studies and in patient-derived xenografts, CDK12-overexpressing breast tumors show positive response to methotrexate-based chemotherapy targeting CDK12-induced metabolic alterations, while being intrinsically refractory to other types of chemotherapy. In a retrospective analysis of hormone receptor-negative and lymph node-positive breast cancer patients randomized in an adjuvant phase III trial to 1-year low-dose metronomic methotrexate-based chemotherapy or no maintenance chemotherapy, a high CDK12 status predicts a dramatic reduction in distant metastasis rate in the chemotherapy-treated vs. not-treated arm. Thus, by coupling tumor progression with metabolic reprogramming, CDK12 creates an actionable vulnerability for breast cancer therapy and might represent a suitable companion biomarker for targeted antimetabolite therapies in human breast cancers.

Fig. 1. CDK12 overexpression enhances spontaneous and carcinogen-induced breast tumorigenesis.

Top, Representative mammary gland histology of age-matched adult (a) and old (b) female nulliparous CDK12 knock-in (CDK12-KI) vs. wild-type (WT) FVB mice; upper/middle images, hematoxylin and eosin (H&E); lower images, IHC with an anti-human CDK12 specific antibody. Bars (a): upper images, 2 mm; middle/lower images, 50 µm; Bars (b): upper images, 4 mm; middle/lower images, 100 µm. Bottom, Quantitation of the experiment performed by scoring the highest degree lesion observed in each mouse. HYP, hyperplasia; HAN, hyperplastic alveolar nodule; DYS, dysplasia; Adv. HAN, advanced HAN; pre-malig., pre-malignant; IC, infiltrating carcinoma. P, P-values, Pearson’s test; n, number of mice in each group. c Frequency (%) of breast cancers (BC) in old CDK12-KI (n = 23) vs. WT (n = 26) mice as in (b). The presence of multiple or multifocal breast cancers is expressed as a percentage of total positive mice. P, P-values, two-tailed Fisher’s test. d Top, Typical 3D-Matrigel outgrowth morphology from primary mammary epithelial cells (MECs) isolated from age-matched adult CDK12-KI vs. WT female mice. In immunofluorescence (IF) images, DAPI nuclear staining is shown in blue. Bottom, total number (Rel. Acini No.) and percentage of filled and disorganized acini from CDK12-KI vs. WT MECs. Data are means ± SD. **P = 0.001; ***P < 0.001, two-sided unpaired t-test (n = 5). e Left, Kaplan–Meier analysis of tumor-free survival of CDK12-KI (n = 11) vs. WT (n = 8) mice treated with DMBA (1 mg/dose/week) for 6 weeks. Sacrifice time at the maximum tolerable tumor burden. P, P-value, log-rank test. Right, Representative histology of DMBA-treated CDK12-KI mice showing an infiltrating carcinoma (IC) vs. WT female mice showing normal/hyperplastic alveoli. Bars, upper panels: right, 1 mm, left 3 mm; lower panels: right, 150 µm; left, 100 µm. f Distribution of mammary lesions detected at sacrifice time in CDK12-KI (n = 10) vs. WT (n = 8) mice treated as in (e). Data are as in (a, b). P, P-value for incidence of IC + pre-malignant lesions in CDK12-KI vs. WT, two-tailed Fisher’s test. g Number of tumors in individual CDK12-KI vs. WT mice treated as in (e). *P = 0.0286, two-sided unpaired t-test. Source data are provided as Source Data file.

Fig. 2. CDK12 overexpression increases tumor onset and multiplicity and metastatic spreading in PyMT oncogene-driven mammary tumorigenesis.

a Kaplan–Meier analysis of tumor-free survival of female (left) and male (right) PyMT/CDK12-KI vs. PyMT/WT control mice. All mice are heterozygous for the Cre recombinase transgene. Mice were sacrificed when the maximum tolerable tumor burden was reached (~1.2 cm). The number of mice in the different groups are indicated. Comparison of survival curves, log-rank test. P = 0.0031, in Females; P < 0.0001, in Males. b Distribution of the number of mammary tumors/mouse detected in individual PyMT/CDK12-KI (n = 16) vs. PyMT/WT (n = 16) mice at a fixed sacrifice time (~5 weeks old). ***P < 0.001, two-sided unpaired t-test. c Relative average tumor volume in PyMT/CDK12-KI (n = 12) vs. PyMT/WT (n = 12) mice at the time of sacrifice. Data are means ± SD. ***P < 0.001, unpaired t-test. d Representative images of primary mammary tumors (left panels) and synchronous lung metastases (Lung, right panels) in PyMT/CDK12-KI and PyMT/WT mice. Left panels, inserts show a magnification of the tumor areas. Right panels, arrowheads indicate metastatic lesions; boxed areas are magnified on the right. H&E, hematoxylin and eosin staining; CDK12, anti-CDK12 IHC staining. Bars in primary mammary tumor panels, 2 mm; bars in Lung panels: left images, 2 mm; middle/right images: top, 150 µm (PyMT/WT); bottom, 300 µm (PyMT/CDK12-KI). e Left, Quantitative analysis of the incidence of lung metastases in PyMT/CDK12-KI vs. PyMT/WT mice. The number of mice in each group is indicated. P, P-value, two-tailed Fisher’s test. Right, quantification of metastases in lung sections from PyMT/CDK12-KI (n = 11) vs. PyMT/WT mice (n = 10). *P = 0.018, two-sided unpaired t-test. Source data are provided as Source Data file.

Fig. 3. CDK12 overexpression induces aberrant functional phenotypes in normal mammary MCF10A cells.

a Immunoblot analysis of CDK12 expression in MCF10A cells stably transfected with CDK12 (CDK12-OE) or empty vector control (EV), and in CDK12-amplified BT474 breast cancer cells. Vinculin (Vinc.), loading control. b Growth kinetics of CDK12-OE vs. EV MCF10A cells in 2-D culture. Data are means ± SEM (n = 3). **P = 0.002, ***P < 0.001, two-sided unpaired t-test. c Typical morphological appearance of 3D-acini generated by CDK12-OE vs. EV MCF10A cells cultivated in Matrigel for 18 days. Left panels in each condition are magnified on the right to better show the hollowed vs. filled/disorganized/branched structures of EV vs. CDK12-OE MCF10A cells, respectively. Bar, 100 μm. d Quantitation of the experiment in (c). Data are means ± SD (n = 5). ***P < 0.001, two-sided unpaired t-test.

Fig. 4. CDK12 overexpression induces aberrant metabolic phenotypes in normal mammary MCF10A cells.

a KEGG enriched pathways of differentially upregulated genes in RNAseq (n = 2) of CDK12-OE vs. EV MCF10A cells. Metabolic pathways are in red. b Hierarchical clustering heatmap of the entire set of metabolites differentially expressed in the steady-state profile of CDK12-OE vs. EV MCF10A cells by t-test analysis (see abbreviations in Supplementary Table 1). Three independent samples performed in triplicate for each condition were analyzed by LC-MS. The color code scale indicates the normalized metabolite abundance (ranging from −2 up to 2). The clusters containing EV and CDK12-OE MCF10A cells are highlighted in green and red, respectively. c Pathway impact analysis of steady-state metabolite profiling as in (b). Color intensity (white-to-red) and size of each circle reflects increasing statistical significance, based on the P-value [-log(P)] from the pathway enrichment analysis (y-axis) and the pathway impact value derived from the pathway topology analysis (x-axis), respectively. Gluconeo., gluconeogenesis; TCA, tricarboxylic acid; hypotau., hypotaurine; GSH, Glutatione. d Schematic of the metabolic pathways upregulated in CDK12-OE vs. EV MCF10A cells showing derivation and contribution of carbon atoms in intermediate metabolites traced with ¹³C-Glucose. Key genes involved in these pathways are also indicated. Red circles, ¹³C-labeled carbon atoms. Solid arrows represent direct metabolic reactions, and dashed arrows represent multiple reactions and indirect connections between two metabolites. Outputs of the different metabolic routes are colored in red. SSP, serine synthesis pathway. Met. Salv., methionine salvage. e Isotopolog abundance of serine (Ser), glycine (Gly), 5,10 Me-THF, 5,10-MTHF in CDK12-OE vs. EV MCF10A cells grown in medium containing ¹³C-Glucose (17.5 mM). Data are the mean ± SD (n = 3), represented as percentage isotopomer labeling relative to control condition. ***P < 0.001 vs. EV, two-sided unpaired t-test. m + n: mass of the isotopomer + n, where n represents the number of heavy carbons (¹³C). Source data are provided as Source Data file.

Fig. 5. CDK12 overexpression induces metabolic dependency on SGOC network pathways in MCF10A cells.

a Three-day proliferation of CDK12-OE and EV MCF10A cells treated with: (i) varying glucose concentrations; (ii) 5 mM 2-deoxyglucose (2DG); (iii) +/− 0.4 mM serine. Data are the mean ± SD (n = 3), expressed as relative to day 0 in each condition (=1, dashed line). Statistical significance was evaluated for each cell line relative to controls (25 mM, −2DG, + Ser) in each condition, and for control-treated CDK12-OE vs. EV cells in the same experimental condition. **P = 0.002; ***P < 0.001, two-sided unpaired t-test. b Three-day proliferation of CDK12-OE and EV MCF10A cells silenced for PSAT1 (PSAT1-KD), MTHFD1 (MTHFD1-KD) and CDK12 (CDK12-KD) or treated with THZ531 (100 nM) and MTX (1μM). Data are the mean ± SD, expressed as relative to day 0 in each condition (=1, dashed line). Statistical significance, two-sided unpaired t-test (n = 3): (i) relative to Ctr-KD in each condition; (ii) relative to CDK12-OE cells silenced for CDK12 (CDK12-KD), for CDK12-OE cells treated with THZ531, silenced for CDK12 and concomitantly treated with THZ531, or treated with MTX. **P = 0.002; ***P < 0.001; n.s., not significant. c RT-qPCR analysis of knockdown (KD) efficiency in the cells described in (b). Data are the mean ± SEM relative to control-silenced cells (Ctr-KD = 1, indicated by horizontal dashed line) for each cell line (n = 1, performed in technical triplicates). d RT-qPCR analysis of the indicated SGOC enzymes in THZ531-treated (100 nM for 48 h) CDK12-OE vs. EV MCF10A cells. Data are the mean ± SEM relative to mRNA levels in control-treated (Veh.) cells (one representative experiment, out of two). e Acini formation in 3D-Matrigel by CDK12-OE vs. EV MCF10A cells treated with THZ531 (100 nM), methotrexate (MTX, 1 μM), or control-treated (Veh.). Left, data are the mean ± SD, expressed as relative to EV in vehicle-treated condition. Right, data are the mean ± SD and are expressed as percentage of disorganized acini (%) vs. total acini number. Statistical significance, two-sided unpaired t-test (n = 3) relative to vehicle in each condition, and relative to EV in the same experimental condition for vehicle-treated CDK12-OE cells. Source data are provided as Source Data file.

Fig. 6. CDK12 overexpression induces metabolic dependency on the SGOC network in PyMT/CDK12.

a Left, GSEA of RNA-seq data (n = 2) from PyMT-CDK12 vs. PyMT-WT tumor cells, obtained from a pool of 3 different tumors for each strain, showing significant association of CDK12 overexpression with enrichment of gene sets involved in glycolysis (Reactome), glycine-serine-threonine metabolism (Gly-Ser-Thr metabolism, KEGG), one-carbon metabolism (one-carbon pool by folate, KEGG). Right, RT-qPCR analysis of PSAT1 and MTHFD1 in PyMT-CDK12 vs. PyMT-WT cells. mRNA levels are reported relative to control in each condition ± SEM (n = 1, out of two independent experiments). b Three-day proliferation of PyMT-CDK12 vs. PyMT-WT MECs silenced for PSAT1 (PSAT1-KD), MTHFD1 (MTHFD1-KD) and CDK12 (CDK12-KD). Data are expressed as relative to day 0 in each condition (=1, dashed line) and are the mean ± SD (n = 3). Statistical significance, two-sided unpaired t-test, relative to Ctr-KD in each condition, and relative to WT in the same experimental condition for Ctr-KD PyMT-CDK12 cells. ***P < 0.001 relative to matching control. c RT-qPCR showing efficiency of PSAT1, MTHFD1 and CDK12 knockdown in PyMT/CDK12 and PyMT/WT MECs. Data are expressed as relative to control-silenced cells (Ctr-KD = 1, indicated by horizontal dashed line) for each condition (one experiment performed in technical triplicates).

Fig. 7. CDK12 overexpression induces metabolic vulnerability to methotrexate (MTX) treatment in PyMT/CDK12 mice.

a PyMT/CDK12 vs. PyMT/WT xenograft response to MTX (10 mg/kg) or control vehicle (Veh.) (n ≥ 8 tumors/condition). Data, expressed as relative to day 1 of treatment, are the mean ± SEM. *P = 0.019 (8 days); P = 0.013 (12 days) relative to matching control, two-sided unpaired t-test (Scatter plots and fitting curves in Source data). b Number (left) and total area (middle) of synchronous lung metastases detected at the endpoint in the experiment in (a). Data are the mean ± SD; *P = 0.019; ***P < 0.001, two-sided unpaired t-test, relative to vehicle in each condition, and relative to WT in the same experimental condition for vehicle-treated PyMT-CDK12 mice. *P = 0.019; ***P < 0.001. Left, n = lung sections, WT: Veh. n = 8, MTX n = 6; CDK12: Veh. n = 8, MTX n = 6. Middle, n = lung sections, WT: Veh. n = 14, MTX n = 11; CDK12: Veh. n = 12, MTX n = 13. Right, Histology showing the lung metastatic burden for each condition. Inserts, magnification of typical metastatic lesions in dashed boxed. Bars, 3 mm. c Three-days in vitro growth of PyMT/CDK12 vs. PyMT/WT tumor cells treated with paclitaxel (PTX, 20 nM), or MTX (1 μM) vs. vehicle. Data, expressed as to relative 72 h of treatment of each cell line, are the mean ± SD (n = 3). ***P < 0.001 vs. matching condition, two-sided unpaired t-test. d PyMT/CDK12 vs. PyMT/WT xenograft response to first-line PTX (5 mg/kg) and MTX (10 mg/kg) (n ≥ 8 tumors/condition). Data, expressed as relative to day 1 of treatment, are the mean ± SEM. *P < 0.05; **P < 0.01 relative to matching vehicle-treated controls, two-sided unpaired t-test (Scatter plots and fitting curves in Source data). e Representative IHC images for KI67 in PyMT/CDK12-KI vs. PyMT/WT tumor xenografts. The percentage of KI67-positive cells is indicated. Bars, 150 μm. f Growth response in PTX-resistant PyMT/CDK12 xenografts from (d), randomized to second-line MTX (10 mg/kg) or no treatment (Veh.). Data, expressed as relative to day 1 of treatment (n ≥ 4 tumors/condition), are the mean ± SEM. *P = 0.020; **, P = 0.003 relative to vehicle-treated tumors, two-sided unpaired t-test (Scatter plots and fitting curves in Source data). Source data are provided as Source Data file.

Fig. 8. CDK12 overexpression is a prognostic and therapeutic biomarker in breast cancer patients and predicts response to antimetabolite therapy targeting the SGOC network.

a Pearson correlation coefficient analysis of the METABRIC breast cancer patient cohort showing positive correlation of a SGOC network gene signature (SGOC_CDK12 average z-score) with CDK12 transcript levels (CDK12 z-score) in all breast cancers and in the indicated molecular subtypes. Derivation of the SGOC_CDK12 signature is described in Supplementary Fig. 4. The number of patients in each group is indicated. R, correlation coefficient value; P, P-value. Padj, FDR adjusted P value for the 3 subgroups tested. b Cumulative incidence of distant metastasis in CDK12^HIGH, CDK12^LOW and CDK12^NULL HER2-negative (HER2-neg) breast cancer patients in response to chemotherapy (CT) with taxane/anthracycline (Tax/AC)- vs. methotrexate-based (CMF, cyclophosphamide, MTX, fluorouracil). The 10-year cumulative distant metastasis rate (%) and the patient number of each class are indicated. HR, multivariable hazard ratio adjusted for standard clinical parameters (see Methods). P, P-value, stratified Wald test. c Cumulative distant metastasis in CDK12^HIGH and CDK12^LOW hormone-negative, lymph node-positive patients randomized to metronomic cyclophosphamide+methotrexate (CM) chemotherapy vs. no chemotherapy (control, Ctr). Patients no.: CDK12^HIGH = 31, CDK12^LOW = 73 in the control group; CDK12^HIGH = 35, CDK12^LOW = 74 in the CM therapy group. HR, hazard ratio (see also Supplementary Fig. 5c for cohort description and multivariable analyses). P, P-value, stratified likelihood-ratio test. d Cumulative incidence of deaths in CDK12^HIGH vs. CDK12^LOW neoadjuvant triple-negative breast cancer (TNBC) patients treated with second-line CMF after failure of first-line Tax/AC chemotherapy. HR, multivariable hazard ratio for CDK12^HIGH vs. CDK12^LOW patients. The 10-year cumulative death rate (%) for each category is indicated. Patient no., CDK12^HIGH = 82, CDK12^LOW = 43, see also Supplementary Fig. 5e). HR, multivariable hazard ratio; P, P-value, stratified Wald test. Source data are provided as Source Data file.

Fig. 9. Alterations of SGOC metabolism are relevant to real life CDK12^HIGH human breast cancers.

a Two CDK12^HIGH (Pt#1 and #2) and two CDK12^LOW (Pt#3 and #4) breast cancer patient-derived xenografts (PDXs) were silenced with a lentiviral CDK12 shRNA or with a control shRNA (Ctr), and injected orthotopically into the mammary fat pads of NGS mice. Tumor volume was measured at the endpoint (4–6 weeks post-injection). Bars represent the mean ± SD (6 mice per experimental group). ***P < 0.001 vs. Ctr-KD cells, two-sided unpaired t-test. b Kaplan–Meier survival analysis of mice injected intracardiacally with CDK12^HIGH Pt#1 and Pt#2, or CDK12^LOW Pt#3 PDX cells treated as in (a). Individual mice were sacrificed when signs of distress or severe debilitation became manifest. Number of mice/experimental group: PT#1, 14 for Ctr-KD and 13 for CDK12-KD; Pt#2, 23 for Ctr-KD and 12 for CDK12-KD; Pt#3, 14 for Ctr-KD and 14 for CDK12-KD. P, P-values, by log-rank test. c RT-qPCR analysis of the indicated SGOC genes in CDK12^HIGH PDX cells (Pt#1 and Pt#2) silenced or not for CDK12 as in (a). Data are expressed as relative to control shRNA (Ctr-KD = 1) for each condition. Bars represent the means ± SEM (n = 5, Pt#1; n = 3, Pt#2). *P < 0.05; **P < 0.01; ***P < 0.001, vs. Ctr-KD, two-sided unpaired t-test. d Relative abundance of the indicated folate cycle metabolites in CDK12^HIGH PDX-derived cells control- (Ctr-KD) or CDK12-silenced (CDK12-KD), identified by LC-MS metabolomics. Data are expressed as relative to an internal standard (reserpine) and are means ± SD (n = 3). **P = 0.001; ***P < 0.001, vs. Ctr-KD, two-sided unpaired t-test. e Three-day in vitro proliferation of CDK12^HIGH (Pt #1 and #2) vs. CDK12^LOW (Pt #3) cells, control silenced (Ctr-KD) or silenced for CDK12 (CDK12-KD), PSAT1 (PSAT1-KD) and MTHFD1 (MTHFD1-KD), or cultured in the presence of THZ531 (100 nM). Data are expressed as relative to day 3 in each condition and are the mean ± SD (n = 3). ***P < 0.001 relative to Ctr-KD in each condition, two-sided unpaired t-test. Source data are provided as Source Data file.

Fig. 10. Preclinical validation of the therapeutic actionability of SGOC metabolism alterations in CDK12-overexpressing human breast cancers.

a Three-days in vitro growth of CDK12^HIGH (Patient #1, #2 and #5) vs. CDK12^LOW (Patient #3 and #4) PDX cells control silenced (Ctr-KD) or silenced for CDK12 (CDK12-KD), in response to treatment with paclitaxel (PTX, 20 nM), methotrexate (MTX, 1 μM), PTX plus CDK12-KD, and vehicle (Veh), as a control. Data are expressed as relative to day 3 in each condition and are the mean ± SD (n = 3). *P = 0.035, ***P < 0.001, relative to vehicle in each condition, two-sided unpaired t-test. b In vivo growth kinetics of CDK12^HIGH (Pt#1 and #2) and CDK12^LOW (Pt#3) PDXs transplanted in the mammary fat pads of NSG mice administered with vehicle control (Veh.) or methotrexate (MTX, 10 mg/kg) (n ≥ 4 tumors per experimental group). Data are the mean ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001, relative to vehicle-treated mice, two-sided unpaired t-test. c Three-days in vitro growth of cells derived from taxanes/anthracycline chemotherapy-resistant (CT-R) CDK12^HIGH (Patients #A, #B and #C) and CDK12^LOW (Patients #D and #E) PDXs, in response to treatment with paclitaxel (PTX, 20 nM), methotrexate (MTX, 1 μM) or vehicle control (Veh.) for 72 h. Data are expressed as relative to day 3 in each condition and are the mean ± SD (n = 3). ***P < 0.001 relative to vehicle in each condition, two-sided unpaired t-test. d In vivo growth kinetics of two tumor xenografts derived from taxane/antracycline-resistant CDK12^HIGH PDX cells (CT-R) (Patient #A and #B) transplanted in NSG mice administered with vehicle control (Veh) or methotrexate (MTX, 10 mg/kg) (n ≥ 8 tumors/per experimental group). Data are the mean ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001, relative to matching controls, two-sided unpaired t-test. Source data are provided as Source Data file.