An intermediate Rb–E2F activity state safeguards proliferation commitment

Cell culture

All experiments were performed using MCF-10A human mammary epithelial cells (American Type Culture Collection (ATCC), CRL-10317, RRID: CVCL_0598) unless otherwise noted. MCF-10A cells were cultured in DMEM/F12 growth medium with HEPES (Gibco, 11039047), supplemented with 5% horse serum (Gibco, 16050122), 20 ng ml^–1 EGF (PeproTech, AF-100-15), 0.5 μg ml^–1 hydrocortisone (Sigma, H0888), 100 ng ml^–1 cholera toxin (Sigma, C8052) and 10 μg ml^–1 insulin (Sigma, I1882). Cells were passaged using trypsin-EDTA (0.05%, Gibco, 25300054), and trypsin was neutralized in DMEM/F12 supplemented with 20% horse serum. RPE-1 human retinal pigment epithelial cells (ATCC, CRL-4000, RRID: CVCL_4388) were cultured in DMEM/F12 with HEPES supplemented with 10% FBS (Sigma, TMS-013-B) and 0.01 mg ml^–1 hygromycin B (Invivogen, ant-hg-1). BJ-5ta human foreskin fibroblast cells (ATCC, CRL-4001, RRID: CVCL_6573) were cultured in DMEM growth medium (Gibco, 11995065) supplemented with 20% Medium 199 (Thermo Fisher, 11150059), 10% FBS and 0.01 mg ml^–1 hygromycin B. U2OS human osteosarcoma epithelial cells (ATCC, HTB-96, RRID: CVCL_0042) and Lenti-X 293T human embryonic kidney cells (Takara Bio, 632180, RRID: CVCL_4401) were cultured in DMEM growth medium with 10% FBS. For MCF-10A serum starvation, cells were cultured in starvation medium (growth medium without horse serum, EGF and insulin, but instead supplemented with 0.3% BSA) after two washes of starvation medium. For RPE-1 and BJ-5ta serum starvation, cells were cultured in starvation medium (growth medium without FBS but instead supplemented with 0.3% BSA) after two washes of starvation medium. For mitogen release, starvation medium was exchanged with starvation medium supplemented with EGF or growth medium. Cells were cultured at 37 °C and 5% CO₂. For microscopy experiments, 96-well glass-bottomed plates (Cellvis, P96-1.5H-N) were coated with collagen (Advanced Biomatrix, 5005-B, 30–60 μg ml^–1 for at least 1 h) for all cell lines (except RPE-1) or with bovine plasma fibronectin (Sigma-Aldrich, F1141, 10-20 µg ml^–1 for at least 1 h) for RPE-1 and cells were seeded into wells at least the night before performing experiments.

Cell line generation

All constructs were introduced into cells using third-generation lentiviral transduction^47,48. In brief, lentivirus was produced in HEK-293T cells co-transfected with packaging plasmids pMDLg/pRRE (Addgene, 12251), pRSV-rev (Addgene, 12253) and pCMV-VSV-G (Addgene, 8454) together with the lentiviral plasmid with Lipofectamine 2000 (Thermo, 11668019). At 72 h after transfection, virus was collected from the supernatant, filtered through a 0.22 μm filter (Millipore, SCGP00525) and concentrated using 100 kDa centrifugal filters (Millipore, UFC910024). Virus was then transduced into cells in growth medium. For constitutively expressed fluorescent constructs, positive fluorescent cells were sorted using a BD Influx cell sorter (performed at the Stanford Shared FACS Facility) or a BD Aria II cell sorter (performed at the Weill Cornell Medicine Shared FACS Facility). mVenus-positive MCF-10A cells were single-cell cloned to establish cell lines expressing the E2F reporter. Doxycycline-inducible constructs (TetOn in pCW backbone with puromycin selection marker) were selected with 1 μg ml^–1 puromycin until control cells died. TetOn cells were grown in the absence of doxycycline until the time of experiment. Doxycycline was added 5 h before release for induction. MCF-10A, RPE-1, BJ-5ta, U2OS cells were acquired directly from ATCC. Lenti-X 293T cells were acquired directly from Takara Bio. MCF-10A cells were validated by RNA-seq. All cell lines tested negative for mycoplasma contamination.

Plasmid generation

Plasmids generated in this study were assembled using Gibson assembly of PCR-amplified inserts and restriction-enzyme-digested plasmid backbones. Genomic DNA was extracted using a DNeasy Blood & Tissue kit (Qiagen, 69504) according to the manufacturer’s instructions. The nuclear marker H2B-miRFP670 was amplified from pLenti-PGK-CMV-H2B-miRFP670 (a gift from T. Kudo, M. Covert Laboratory, Stanford University) and inserted into the pLV backbone to generate pLV-EF1a-H2B-miRFP670. The CDK2 activity reporter was amplified from pCSII-hDHB(amino acids 994–1087)–mVenus¹³ and inserted into pLV backbone to generate pLV-EF1a-DHB–mTurquoise. The CRL4^Cdt2 reporter was amplified from tFucci(CA)2/pCSII-EF⁹ and inserted into the pLV backbone to generate pLV-mCherry–hCDT1(1–100)ΔCy. 12S-E1A was amplified from pBabe 12S E1A (Addgene, plasmid 18742) and inserted into the pCW backbone (derived from pCW-Cas9, Addgene, plasmid 50661) to generate pCW-12S-E1A-HA-puro. E2F1 was amplified from HA-E2F-1 wt-pRcCMV (Addgene, plasmid 21667) and inserted into the pCW backbone to generate pCW-HA-E2F1-puro. Cyclin E1, cyclin D1 and Rb were amplified out of cDNA from MCF-10A cells and inserted into the pCW backbone to generate pCW-CCNE1-HA-puro, pCW-CCND1-HA-puro and pCW-HA-hRB-WT-puro, respectively. Rb(ΔCDK) was amplified from pCMV HA hRb ΔCDK (Addgene, plasmid 58906) and inserted into the pCW backbone to generate pCW-HA-hRb-ΔCDK-puro. Rb T373A, S608A and S612A mutations were introduced by PCR on pCW-HA-hRb-WT-puro to generate pCW-HA-hRb-WT-T373A-puro, pCW-HA-hRb-WT-T373A/S608A-puro and pCW-HA-hRb-WT-T373A/S608A/S612A-puro, respectively. Rb T373 and S608 reverse mutations were introduced by PCR on pCW-HA-hRb-ΔCDK-puro to generate pCW-HA-hRb-ΔCDK-T373-puro and pCW-HA-hRb-ΔCDK-T373/S608-puro, respectively. Rb T373D/S608D/S612D mutations were introduced by PCR on pCW-HA-hRb-ΔCDK-puro to generate pCW-HA-hRb-ΔCDK-T373D/S608D/S612D-puro.

E2F reporter development

Extended Data Fig. 1a explains the workflow for development of the E2F reporter. To optimize the promoter for the E2F transcriptional activity reporter, we selected 11 candidate genes that were upregulated by growth medium (containing serum, EGF and insulin) and downregulated by the CDK4/6 inhibitor (thus specific to the CDK–Rb–E2F pathway) in the RNA sequencing experiments (Extended Data Fig. 1b). We further validated the selected 11 genes by RT–qPCR for their responsiveness to growth medium and the CDK4/6 inhibitor (Extended Data Fig. 1c,d). We next selected promoter regions from each gene based on a genome database, Ensembl (https://www.ensembl.org), and inserted the promoter regions into the prototype reporter construct such that the promoter drives mVenus expression. We then evaluated the sensitivity and specificity of the reporter by live-cell imaging (Extended Data Fig. 1e). The fold change in DMSO indicated the sensitivity to growth medium, whereas the fold change in the CDK4/6 inhibitor indicated the specificity to the CDK–Rb–E2F pathway. Among the top three constructs (FAM111B, DTL and CDC6) for these criteria, we decided to use the CDC6 promoter because it is free of cytotoxicity and has no background signals in quiescence.

Candidate constructs for an E2F reporter consisted of a E2F target gene promoter, mVenus, NLS (SV40), PEST (mouse ornithine decarboxylase) and a polyA signal (bGH). The PEST sequence was added to achieve a rapid turnover rate of the reporter by targeting the mVenus protein for degradation. The maturation half-time of mVenus is reported as 4.1 min^49,50. The degradation half-life of eGFP with the PEST sequence used in the reporter is reported as 1–2 h⁵¹. E2F target gene promoter regions were either obtained as synthesized oligonucleotides (gBlocks) or amplified out of genomic DNA from MCF-10A cells. The following human E2F target gene promoter regions were used for the reporters (relative position from the transcription start site and the accession numbers in NCBI): CDC6 (−1057 to 239, NM_001254.4); FAM111B (−545 to 1765, NM_198947.4); DTL (−1948 to 143, NM_016448.4); MCM10 (−435 to 49, NM_018518.5); RRM2 (−3317 to −8, NM_001034.4); E2F1 (short) (−123 to 59, NM_005225.3) (the same promoter sequence as the previous E2F reporter¹⁰); CDC45 (−457 to 49, NM_003504.5); EXO1 (−1004 to 1771, NM_130398.4); E2F1 (long) (−720 to 1681, NM_005225.3); CLSPN (−446 to 263, NM_022111.4); CDT1 (−1208 to 532, NM_030928.4); and TCF19 (−1589 to 6, NM_007109.3).

As predicted, the mVenus signal increased as cells transitioned into S phase and then decreased^52,53 (Fig. 1c,d and Extended Data Fig. 1f), and the mVenus signal amplitude correlated in single cells with the mRNA abundance of known E2F targets (Extended Data Fig. 2f,g; nonlinearity in the mRNA puncta area can be explained by the overlap of mRNA puncta, Extended Data Fig. 2h–k). Moreover, the mVenus signal was upregulated by knockdown of the E2F suppressor Rb (Extended Data Fig. 1g), overexpression of adenovirus E1A (which inactivates Rb) (Extended Data Fig. 1h) and induced expression of E2F1 (Extended Data Fig. 1i,j) or cyclin D1 (cyclinD1–CDK4/6 phosphorylates Rb) (Fig. 1e and Extended Data Fig. 1k). All these results provided validation that the reporter measures E2F transcriptional activity. The E2F reporter monitors the activity of both activating and repressing E2Fs and therefore can be used as a global E2F transcriptional activity reporter (Extended Data Fig. 2a–e). We note that a previous E2F reporter, which is based on the E2F1 promoter, is regulated by both E2F and MYC^17,54, but the E2F reporter developed here does not respond to MYC induction (Extended Data Fig. 1l).

The E2F reporter and other reporter constructs used in the study are available through the non-profit organization Addgene (plasmids 212665-212675) (https://www.addgene.org/Tobias_Meyer/).

siRNA transfection

MCF-10A cells were transfected with siRNA using DharmaFECT 1 (Dharmacon, T-2001-03) according to the manufacturer’s protocol using 20 nM siRNA and 1:500 diluted DharmaFECT 1 final concentration. Cells were incubated for 4–6 h in serum starvation medium containing transfection mixture, followed by a medium change. Pools of four siRNA oligonucleotides (ON-TARGETplus, Dharmacon) were used for siControl, siRB1, siE2F1, siE2F2, siE2F3 and siE2F7. For RB1 knockdown while overexpressing doxycycline-inducible Rb constructs, oligonucleotides that target the RB1 3′ untranslated region was selected to avoid knockdown the constructs.

Chemicals

The following stock solutions of drugs were dissolved in DMSO (Sigma, D2650 or Santa Cruz, sc-358801): doxycycline hyclate (Sigma, D9891); the CDK4/6 inhibitor palbociclib (Selleck Chemicals, S1116); the CDK2 inhibitor PF-07104091 (ChemieTek, CT-PF0710); the CDK1 inhibitor RO-3306 (Cayman Chemical, 15149); the EGFR inhibitor gefitinib (Selleck Chemicals, S1025); and the PP1 and PP2A inhibitor calyculin A (Santa Cruz, sc-24000). Neocarzinostatin (Sigma, N9162) was dissolved in a solution containing 20 mM MES buffer, pH 5.5.

RT–qPCR

RNA was extracted from cells using QIAshredder (Qiagen, 79656) and RNeasy Mini kits (Qiagen, 74106). cDNA was generated by reverse transcription using RevertAid reverse transcriptase (Thermo, EP0442) and Oligo(dT)18 primer (Thermo, SO132) or Random Hexamer primer (Thermo, SO142) according to the manufacturer’s instructions. qPCR was performed using iTaq Universal SYBR Green supermix (Bio-Rad, 1725122) on a LightCycler 480 II (Roche) according to the manufacturer’s instructions.

RNA fluorescent in situ hybridization

Cells were fixed in 4% paraformaldehyde in PBS for 10 min at room temperature followed by a PBS wash. Cells were permeabilized in 0.2% Triton X-100 in PBS for 15 min followed by a PBS wash. RNA fluorescent in situ hybridization (FISH) was carried out using a ViewRNA ISH cell assay (Thermo, QVC0001) according to the manufacturer’s instructions. Cells were washed in PBS and incubated with 1 μg ml^–1 Hoechst 33342 (Invitrogen, H3570) in PBS for 10 min, followed by a final PBS wash before imaging. The following hybridization probes were used: E2F1 (Thermo, VA1-12108-VC); E2F2 (Thermo, VA4-16111-VC); E2F3 (Thermo, VA6-16112-VC); E2F7 (Thermo, VA6-3183369-VC); CCNE2 (Thermo, VA1-3005686-VC); and CDC6 (Thermo, VA4-3084153-VC).

Western blotting

Cells were grown in 60 mm dishes. At the time of lysis, cells were washed in PBS, lysed in 2× Laemmli sample buffer with 200 mM DTT, passed through a 25 G needle 10 times and heated at 90 °C for 4 min. Samples were then separated by SDS–PAGE using 7.5% Mini-Protean TGX gels (Bio-Rad, 4561026) or 4–20% Mini-Protean TGX gels (Bio-Rad, 4561096) in Tris–glycine–SDS running buffer, followed by wet transfer onto Immun-Blot Low Fluorescence PVDF/filter paper (Bio-Rad, 1620261) in Tris–glycine buffer + 20% methanol. Membranes were washed in TBST (25 mM Tris, pH 7.5, 150 mM NaCl and 0.05% Tween 20), blocked for 1 h in 5% milk + 0.01% NaN₃ in TBST and incubated overnight with primary antibodies in 5% BSA + 0.01% NaN₃ in TBST. For fluorescence detection, membranes were incubated with the secondary antibodies goat anti-rabbit IgG (H+L) secondary antibody Alexa Fluor 680 (1:20,000, Thermo, A-21109) and goat anti-mouse IgG polyclonal antibody IRDye 800CW (1:20,000, Li-Cor, 925-32210). For chemiluminescence detection, membranes were incubated with a secondary HRP-linked goat anti-rabbit IgG secondary antibody (1:2,000, Cell Signaling Technology, 7074) and developed with SuperSignal West Pico PLUS chemiluminescent substrate (Thermo, 34580). Membranes were imaged using an Odyssey Infrared imaging system (Li-Cor). The following primary antibodies were used: mouse anti-Rb antibody (1:1,000; Cell Signaling Technology, 9309); rabbit anti-phospho-Rb (T252) antibody (1:1,000; Abcam, ab184797); rabbit anti-phospho-Rb (T356) antibody (1:1,000; Abcam, ab76298); rabbit anti-phospho-Rb (T373) antibody (1:1,000; Abcam, ab52975); rabbit anti-phospho-Rb (S608) antibody (1:1,000; Cell Signaling Technology, 2181); rabbit anti-phospho-Rb (S780) antibody (1:250; Thermo, 701272); rabbit anti-phospho-Rb (S788) antibody (1:250; Abcam, ab277775); rabbit anti-phospho-Rb (S795) antibody (1:1,000; Cell Signaling Technology, 9301); rabbit anti-phospho-Rb (S807/S811) antibody (1:20,000; Cell Signaling Technology, 8516); rabbit anti-phospho-Rb (T826) antibody (1:1,000, Abcam, ab133446); rabbit anti-EGFR antibody (1:1,000, Cell Signaling Technology, 4267); and rabbit anti-phospho-EGFR (Y1045) antibody (1:1,000; Cell Signaling Technology, 2237).

Immunofluorescence

General protocol

Cells were fixed in 4% paraformaldehyde in PBS for 10 min at room temperature followed by a PBS wash. If cells expressed fluorescent proteins that spectrally overlapped with the fluorophores used in later steps, the fluorescent proteins were chemically bleached⁵⁵ in 3% H₂O₂ + 20 mM HCl in PBS for 1 h and washed in PBS. Cells were permeabilized and blocked in blocking buffer (10% FBS, 1% BSA, 0.1% Triton X-100 and 0.01% NaN₃ in PBS) for 1 h. Cells were then incubated with primary antibodies overnight in blocking buffer at 4 °C, washed in PBS and incubated with secondary antibodies in blocking buffer for 1 h at room temperature. Cells were washed in PBS and incubated with 1 μg ml^–1 Hoechst 33342 (Invitrogen, H3570) in PBS for 10 min, followed by a final PBS wash before imaging.

Iterative immunofluorescence

If simultaneously staining for targets with antibodies of the same species, the iterative indirect immunofluorescence imaging (4i) method³¹ was used to sequentially image multiple antibodies. In brief, the first round of imaging was identical to the general immunofluorescence protocol, with the exception that cells after the post-Hoechst PBS wash were washed in ddH₂O and then placed in imaging buffer (700 mM N-acetyl cysteine in ddH₂O, pH 7.4, Sigma, A7250). Cells were imaged and then washed in ddH₂O. The previous-round antibodies were eluted by 3 × 10-min incubations in elution buffer, which consisted of 0.5 M glycine (Sigma), 3 M urea (Sigma), 3 M guanidinium chloride (Sigma) and 70 mM TCEP-HCl (Goldbio, TCEP50) in ddH₂O, pH 2.5, followed by a PBS wash. Cells were then checked under a fluorescence microscope to ensure proper elution. Cells were then blocked with blocking buffer, consisting of 1% BSA and 150 mM maleimide (dissolved immediately before use, Sigma, 129585) in PBS for 1 h and then washed in PBS, followed by primary antibody incubation, and the subsequent steps were performed the same as in the first round and repeated as needed.

Pre-extraction for chromatin-bound protein

If chromatin-bound proteins were being stained, soluble proteins were extracted from cells. Immediately before fixation, medium was aspirated off from cells and the plate was placed on ice. Cells were incubated for 5 min in ice-cold pre-extraction buffer, consisting of 0.2% Triton X-100 (Sigma, X100) and 1× Halt Protease Inhibitor cocktail (Thermo, 78429) in PBS. After pre-extraction, 8% paraformaldehyde in H₂O was directly added to wells 1:1 with wide-orifice tips to minimize cell detachment, and cells were fixed for 1 h at room temperature, after which the sample was treated using the general staining protocol.

EdU incorporation and labelling

If measuring EdU incorporation, cells were pulsed with 10 μM EdU (Cayman Chemical, 20518) in growth medium for 15 min before fixation. EdU is incorporated throughout the EdU pulse, such that incorporated EdU reflects the average rate of DNA synthesis over the length of the pulse. Thus, 15 min of a short EdU pulse is more reflective of the instantaneous DNA synthesis rate compared with a longer pulse such as 1 h. After blocking cells (before primary antibodies), cells were washed once with PBS and then a click reaction was performed in 2 mM CuSO₄, 20 mg ml^–1 sodium ascorbate in TBS (Tris 50 mM and NaCl 150 mM pH 8.3) with 3 μM AFDye 568 picolyl azide (Click Chemistry Tools, 1292) for 30 min, followed by a PBS wash.

Antibodies

The following primary antibodies were used: mouse anti-Rb antibody (1:250; BD, 554136); rabbit anti-phospho-Rb (T373) antibody (1:100 to 1:1,000; Abcam, ab52975); rabbit anti-phospho-Rb (S608) antibody (1:100 to 1:250; Cell Signaling Technology, 2181); rabbit anti-phospho-Rb (S780) antibody (1:100 to 1:250, Thermo, 701272); rabbit anti-phospho-Rb (S807/S811) antibody (1:2,500; Cell Signaling Technology, 8516); rabbit anti-phospho-Rb (T826) antibody (1:100; Abcam, ab133446); rabbit anti-p21 antibody (1:2,500; Cell Signaling Technology, 2947); mouse anti-cyclin E antibody (1:400; Santa Cruz, sc-247); rabbit anti-c-Myc antibody (1:800; Cell Signaling Technology, 5605); rabbit anti-53BP1 antibody (1:500; Cell Signaling Technology, 4937); rabbit anti-HA tag antibody (1:1,000; Cell Signaling Technology, 3724); and mouse anti-HA tag antibody (1:1,000; BioLegend, 901501). The following secondary antibodies were used: goat anti-rabbit IgG Alexa Fluor 514 (1:2,000; Thermo, A-31558) and goat anti-mouse IgG Alexa Fluor 647 (1:2,000; Thermo, A-21235).

Microscopy

For automated epifluorescence microscopy, cells were imaged using a Ti2-E inverted microscope (Nikon) or ImageXpress Micro XLS microscope (Molecular Devices). For imaging on the Ti2-E, multichannel fluorescent images were taken using a 89903-ET-BV421/BV480/AF488/AF568/AF647 Quinta Band set (Chroma Technology) with an LED light source (Lumencor Spectra X) and an ORCA-Flash4.0 V3 sCMOS camera (Hamamatsu). A ×20 (Nikon CFI Plan Apo Lambda, 0.75 NA) objective was used to acquire images. For imaging on the ImageXpress, images were taken with appropriate single-band filter sets with a white-light source, using a ×20 (Nikon CFI Plan Apo Lambda, 0.75 NA) and Zyla 4.2 sCMOS camera (Andor). All images were acquired in 16-bit mode with 2 × 2 or 4 × 4 binning, and acquisition settings were chosen to not saturate the signal.

For live-cell time-lapse imaging, 96-well plates were imaged within an enclosed 37 °C, 5% CO₂ environmental chamber in 200 μl of medium. Around 4–9 sites were imaged in each well (with the number of wells imaged varying depending on the experiment) every 12 min. Light exposure to cells was limited by using the minimum exposure necessary to maintain an acceptable signal-to-noise ratio on a per-channel basis, and total light exposure was always limited to below 200 ms per site for each time point. When performing the live-cell imaging followed by fixed-cell imaging, cells were immediately taken off the microscope following the final time point and fixed. When matching fixed-cell imaging back to either live-cell imaging or previous rounds of fixed-cell imaging, the plate position (which can shift slightly when replacing the plate on the microscope) was aligned to approximately the same location and further aligned computationally during image analysis.

Image analysis

Image analysis was performed using a custom Matlab pipeline as previously described^8,14. In brief, images were first flatfield corrected (illumination bias determined by pooling background areas from multiple wells from the same imaging session), and then the background was subtracted locally. Cells were segmented for their nuclei based on either Hoechst staining or H2B signal. For the E2F and CRL4^Cdt2 reporters, the mean signal within the nucleus was then calculated. CDK2 activity was calculated by taking the ratio between the median cytoplasmic intensity and the mean nuclear intensity. The cytoplasm was sampled by expanding a ring outside the nucleus (with inner radius of 0.65 μm and outer radius of 3.25 μm) without overlapping with the cytoplasm from a neighbouring cell. For RNA FISH measurements, cells were segmented for their whole-cell regions by spatially approximating an area encompassing the nucleus and reaching as far as 15.6 μm outside the nuclear mask without overlapping other cell regions. A foreground mask of FISH puncta was generated by top hat-filtering the raw image with a circular kernel radius of 1.3 μm and thresholding on absolute intensity. RNA puncta count was calculated as the number of foreground pixels within a given whole-cell region. The image processing pipeline and code used to generate all figures in this study have been deposited into GitHub (https://github.com/MeyerLab/image-analysis-konagaya-2022).

The categorization of MCF-10A cells into S enter, E2F reverse and undecided was based on the following criteria: S enter cells are cells that entered S phase, which is detected by the CRL4^Cdt2 reporter signal; E2F reverse cells are cells with E2F activation, without S phase entry, and with an E2F activity decrease of more than half from the peak; undecided cells are cells with E2F activation, but without S phase entry or a decrease in E2F activity.

Protein structural modelling

Structures were modelled using ColabFold⁵⁶, a simplified AlphaFold2 algorithm^30,57, without templates (https://colab.research.google.com/github/sokrypton/ColabFold/blob/main/beta/AlphaFAlp2_advanced.ipynb). Multiple sequence alignments (MSAs) were generated using MMseqs2 and unpaired (generates separate MSAs for each protein). The Rb–E2F1–DP1 complex was modelled as a heterotrimer with a 1:1:1 stoichiometric ratio. To reduce memory requirements, only a subset of the MSA was used as input to the model by subsampling the MSA to a maximum of 512 cluster centres and 1,024 extra sequences. Relaxation of the predicted structures using amber force fields was disabled because it barely moves the main-chain structure. Turbo setting (compiles once, swaps parameters and adjusts the maximum MSA) was used to speed-up and reduce memory requirements. The number of random seeds = 1, the number of ensembles = 1 and the threshold for tolerance = 0. Training setting (which activates the stochastic part of the model) was disabled. The predicted template modelling score was used to rank and assess the confidence of the predicted protein–protein interaction. Five models were computed through three recycles, and the one with the highest predicted template modelling score was visualized using PyMOL. The Rb–E2F1–DP1 complex predicted in the study is available through ModelArchive (https://modelarchive.org/doi/10.5452/ma-jcq2m).

PSP analysis

PSP plots were used to show single-cell correlations of Rb phosphorylation between two different sites. Each phosphorylation signal was normalized by the total Rb antibody signal in the same cell, and each axis was adjusted to the average phosphorylation signal in S phase of 1 (when Rb is hyperphosphorylated). The colour bar indicates the relative cell population density. A red line shows fitting with a preferential relative phosphorylation or dephosphorylation rate between the two sites (PSP_coeff).

We constructed a kinetic phosphorylation–dephosphorylation model to ask whether the observed differences in dephosphorylation kinetics between T373 and S807/S811 (Fig. 5a) could explain the preferential phosphorylation at T373 over S807/S811 (Fig. 3c). Phosphorylated fractions at one site (x; such as T373) and another reference site (y; such as S807/S811) are described as follows:

where α₁ and indicate α₂ the phosphorylation rate for x and y, respectively; and β₁ and β₂ indicate the dephosphorylation rate for x and y, respectively. At equilibrium,

We assumed that the Rb-targeting kinases (CDK4/6 and CDK2) do not have selectivity to T373 over S807/S811 because the convex relationship in the PSP plot does not change with treatment with the CDK4/6 inhibitor or the CDK2 inhibitor (DMSO: Fig. 5k, CDK4/6i, Fig. 3c, the top left panel, CDK2i, and Extended Data Fig. 10b). With this assumption,

Thus,

where \({{\rm{PSP}}}_{{\rm{coeff}}}=\frac{{\beta }_{2}}{{\beta }_{1}}\) indicates the relative dephosphorylation rate between two sites. We used the final equation above to fit the measured phosphorylation fractions (0.1 ≤ x ≤ 0.9 and 0.1 ≤ y ≤ 0.9) in the PSP plots and obtained PSP_coeff values.

ChIP–seq analysis

The ChIP–seq signals and peaks of E2F proteins on the CDC6 promoter were downloaded from ENCODE⁵⁸ (https://www.encodeproject.org/). All the experiments were performed in duplicate and with an irreproducible discovery rate cut-off of 0.05. The following target E2F proteins and cell lines were used in the ChIP–seq analysis (the accession numbers of the fold change over control and irreproducible discovery rate thresholded-peaks in ENCODE): E2F1 in HepG2 (ENCFF846JMO.bigWig; ENCFF919WXY.bigBed); E2F1 in MCF-7 (ENCFF858GLM.bigWig; ENCFF692OYJ.bigBed); E2F2 in HepG2 (ENCFF826PYA.bigWig; ENCFF629CDJ.bigBed); E2F3 in K562 (ENCFF838PBU.bigWig; ENCFF922ILX.bigBed); E2F4 in HepG2 (ENCFF491MUP.bigWig; ENCFF311TOD.bigBed); E2F4 in MCF-7 (ENCFF232RTG.bigWig; ENCFF249IZG.bigBed); E2F5 in HepG2 (ENCFF518RZY.bigWig; ENCFF582JSQ.bigBed); E2F6 in A549 (ENCFF190MLK.bigWig; ENCFF550XVR.bigBed); E2F7 in K562 (ENCFF979YTG.bigWig; ENCFF212JSU.bigBed); E2F8 in HepG2 (ENCFF805BWM.bigWig; ENCFF320WJO.bigBed); and E2F8 in MCF-7 (ENCFF898YRB.bigWig; ENCFF641CMX.bigBed).

MSA analyses

The amino acid sequences of human Rb orthologues were downloaded from the NCBI database (https://www.ncbi.nlm.nih.gov/). Rb orthologues used in the MSA analysis (accession numbers in NCBI): Homo sapiens (NP_000312.2), Mus musculus (NP_033055.2), Gallus gallus (NP_989750.2), Xenopus tropicalis (NP_001269454.1), Danio rerio (NP_001071248.1) and Strongylocentrotus purpuratus (XP_030838139.1). Full-length protein sequences were aligned using Clustal Omega (https://www.ebi.ac.uk/Tools/msa/clustalo/) with the default parameters as follows: output guide tree = true, output distance matrix = false, dealign input sequences = false, mBed-like clustering guide tree = true, mBed-like clustering iteration = true, number of iterations = 0, maximum guide tree iterations = –1, maximum HMM iterations = –1, output alignment format = clustal_num, output order = aligned, sequence type = protein

Statistical analysis

Statistical analyses were performed using two-sided, two-sample t-test or one-way ANOVA and Scheffé’s post hoc comparison. Quantifications are represented as the mean ± s.e. or median ± 25th and 75th percentiles as specified in the figure legends. P values are as follows: *P < 0.05, **P < 0.01 and ***P < 0.001. Further statistical details of experiments are reported in the figure legends. No statistical methods were used to predetermine sample size. The experiments were not randomized and investigators were not blinded to allocation during experiments and outcome assessment. 

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.