Cell-Cycle Regulation Accounts for Variability in Ki-67 Expression Levels

Ki-67 is a nuclear protein expressed in all proliferating vertebrate cells, and it is a widely used biomarker to estimate the proportion of dividing cells to grade tumors. Ki-67 expression might have prognostic value, such as in the IHC4+C score in breast cancer (1). However, inconsistency in assessments of Ki-67 labeling index hinders its use to stratify patients for therapy (2). This variability might contribute to inconsistency regarding the prognostic value of Ki-67 labeling index in a given cancer type, for example, triple-negative breast cancer (TNBC; refs. 3, 4). Therefore, it is critical to define what constitutes Ki-67–positive expression and what the clinical significance of different Ki-67 levels is. This requires a better understanding of the control and functional significance of Ki-67 expression.

In cultured cells, Ki-67 levels are highest in G₂ phase and mitosis (5). In HL60 cells, Ki-67 protein was reported highly unstable throughout the cell cycle (6), but this has not been confirmed in other cell lines. Furthermore, Ki-67 transcriptional control is poorly understood. The MKI67 promoter of the gene encoding Ki-67 is GC-rich, contains Sp1-binding sites, but lacks a TATA box (7, 8). In primary fibroblasts, it is bound by E2F proteins (9), and Ki-67 mRNA accumulates upon E2F overexpression (10). E2F-dependent transcription, which is required for S-phase onset, is repressed by RB family proteins, whose phosphorylation by cyclin-dependent kinases (CDK) promotes cell-cycle progression. Although RB expression is lost in many cancers, it is not clear whether this leads to upregulation of Ki-67. It is also not known whether Ki-67 is frequently over- or underexpressed in cancers, for example, due to copy number variation, translational regulation, or mutation of sites affecting protein stability or promoter activity.

We, and others, recently showed that Ki-67 is not required for proliferation of mammalian cells in culture (11–13). Furthermore, mice with a disrupted Mki67 gene were healthy and fertile, despite minimal Ki-67 expression (11). Conversely, mice lacking the Fzr1 gene maintained Ki-67 in differentiated, nonproliferating tissues (11). Thus, Ki-67 expression can be uncoupled from, and is not required for, cell proliferation. This raises the possibility that Ki-67 staining in cancers might not always reflect cell proliferation.

Like Ki-67, CDK4 and CDK6, which trigger RB phosphorylation, are not essential for cell proliferation in most cell types in mice (14). Nevertheless, CDK4 and D-type cyclins are required for certain cancers, including breast cancers (15). CDK4/CDK6 inhibition with PD0332991 (palbociclib) has shown major benefits in breast cancer clinical trials, leading to its approval in certain clinical settings (16), and it is currently in trials for a variety of other cancers. Preclinical models using tumor explants suggest that RB-positivity can predict responses to palbociclib (17). However, in these experiments, palbociclib effects were determined by Ki-67 expression itself. Yet Ki-67 expression may be directly promoted by CDK4/CDK6–dependent RB phosphorylation. Thus, upon CDK4/CDK6 inhibition, it remains possible that cells might continue to proliferate without Ki-67 expression. It is therefore essential to determine whether loss of Ki-67 after CDK4/CDK6 inhibition indeed reflects cell-cycle arrest by correlating with independent markers of cell proliferation.

In this study, we show that variability of Ki-67 levels is due to cell-cycle regulation of Ki-67 mRNA and protein in normal human cells, proliferating tissues in mice and human cancers. Furthermore, in cells that have recently exited the cell cycle, low-level Ki-67 persists. Ki-67 protein is degraded from mitosis to G₁, and G₁ arrest by CDK4/CDK6 inhibition causes loss of Ki-67 mRNA. Effects of palbociclib on Ki-67 expression always correlated with its effects on cell proliferation, including in vivo. These results have important implications for interpreting Ki-67 staining in cancer histopathology and for its use as a diagnostic marker.

All animal experiments were performed in accordance with international ethics standards and were subjected to approval by the Animal Experimentation Ethics Committee of Languedoc Roussillon.

The Apc^lox (Apc^tm2.1Cip) line (18) was provided by C. Perret (Cochin Institute, Paris, France). The Villin-Cre^ERT2 line (19) was provided by S. Robine (Curie Institute, Paris, France).

Cell lines were not authenticated but were mycoplasma-free (tested weekly using Mycoalert kit). Normal human diploid foreskin fibroblasts (HDF) were from frozen stocks provided by J. Piette (CRBM, Montpellier, France) in 2001. IMR-90, U2OS, HeLa, HCT-116, MCF7, MDA-MB-231, MDA-MB-468, CAL51, HBL100, and IMR90 were originally obtained from the ATCC from 2000 to 2010. IMR-90 fibroblasts expressing HPV-16 E7 oncogene were described previously (20). U2OS, HeLa HCT-116, MCF7, MDA-MB-231, CAL51, and HBL100 were grown in DMEM with 10% FBS. HDFs were grown in DMEM supplemented with 10% FCS and 2 mmol/L l-glutamine. Cells were grown under standard conditions at 37°C in a humidified incubator containing 5% CO2. IMR-90 WT and E7 were grown in 3% oxygen in DMEM/F12 medium supplemented with 10% FBS and 4 mmol/L l-glutamine.

Primary cells (HDF, BJ hTERT, IMR90) were treated with 1 μmol/L PD0332991 (Tocris), 100 μg/mL cycloheximide, 20 μmol/L MG132 (Tocris), 10 μg/mL bleomycin (Sigma-Aldrich), or 2 μg/mL ICRF-193 (Sigma-Aldrich).

HDF at 20% confluency were washed with PBS and incubated with medium supplemented with 0.1% FBS for 72 hours. Cell-cycle entry was stimulated by adding fresh medium with 10% FBS. Onset of S-phase was observed 16 hours later by EdU incorporation.

HDF at 25% confluency were incubated with medium supplemented with 2 mmol/L thymidine or 2 mmol/L hydroxyurea for 24 hours. Cells were released from G₁–S block by washing twice with fresh medium for 5 minutes.

Seven hours after release from G₁–S block, HDFs were incubated with 9 μmol/L RO3306 (Tocris) for 14 hours. Cells were released from G₂ block by washing twice with fresh medium for 5 minutes. Mitosis was observed around 1 hour after release.

Frozen washed cell pellets were lyzed directly in Laemmli buffer at 95°C (without β-mercaptoethanol and bromophenol blue) and sonicated on ice for 10 minutes in 30 seconds/30 seconds ON/OFF intervals. Protein concentrations were determined by BCA assay. Proteins were separated by SDS-PAGE (7.5% and 12.5% gels) at 35 mA in TGS buffer and transferred to Immobilon membranes with a semidry blotting apparatus.

Ten micrograms of total cell lysate in 5 μL was spotted onto nitrocellulose, blocked, and probed using standard immunoblotting procedues. Signals were quantified using PXi4 Imaging System (Syngene) and GeneTools analysis software.

Antibodies were: Ki-67: clone SP6 (Abcam), 35 (BD Biosciences), SolA15 (eBioscience); RB1 (BD Biosciences); cyclin A: 6E6 (Novocastra), H-432 (Santa Cruz Technology); cyclin E1: HE12 (Santa Cruz Biotechnology); cyclin D1: DSC6 (Cell Signaling Technology), EP272Y (Millipore); cyclin B1: GNS1 (Santa Cruz Biotechnology); CNDKN1A: C-19 (Santa Cruz Biotechnology); phospho-histone H3S10: 9701 (Cell Signaling Technology).

Purified RNA (1,000 ng) extracted by RNeasy Mini Kit (Qiagen) was reverse-transcribed using SuperScript II Reverse Transcriptase (Life Technologies) according to the manufacturer's instructions. qPCR was performed using LightCycler 480 SYBR Green I Master (Roche) and LightCycler 480 qPCR machine with conditions: 95°C 10 minutes, 40 cycles of 95°C 20 seconds, 58°C 20 seconds, and 72°C 20 seconds. Reaction specificity was verified by melt curve analysis. Each sample was performed in three replicates. qPCR primers used were:

• MKI67 (F: TGACCCTGATGAGAAAGCTCAA, R: CCCTGAGCAACACTGTCTTTT);

• CCNA2 (F: AGGAAAACTTCAGCTTGTGGG, R: CACAAACTCTGCTACTTCTGGG);

• CCNE1 (F: CCGGTATATGGCGACACAAG, R: ACATACGCAAACTGGTGCAA);

• PCNA (F: CCTGCTGGGATATTAGCTCCA, R: CAGCGGTAGGTGTCGAAGC);

• CCNB1 (F: TGTGTCAGGCTTTCTCTGATG, R: TTGGTCTGACTGCTTGCTCT);

• CCND1, (F: GCTGCGAAGTGGAAACCATC, R: CCTCCTTCTGCACACATTTGAA);

• B2M (F: GCGCTACTCTCTCTTTCTGG, R: AGAAAGACCAGTCCTTGCTGA);

• RPL19 (F: ATGCCGGAAAAACACCTTGG, R: GTGACCTTCTCTGGCATTCG).

Analysis of DNA replication in cells was achieved by treatment with 2 mmol/L 5-ethynyl-2′-deoxyuridine (EdU; Life Technologies) before fixation. Replicating cells were visualized following the protocol from Click-iT EdU AlexaFluor-488 Imaging Kit (Life Technologies). Beta-galactosidase staining was conducted using Senescence Detection Kit (ab65351) following the manufacturer's instructions.

Three million log-phase viable mouse pathogen-free (IMPACT1, Iddex) MDA-MB-231 and MDA-MB-468 cells in 0.2 mL, 50% v/v Matrigel (BD Biosciences) were injected subcutaneously into 6-week-old female athymic nu/nu mice (Envigo). Tumors were grown to an average volume of 150–200 mm³ measured by caliper using the formula “π/6 × S² (smaller radius) × L (larger radius)” before initiation of treatment. Mice were then randomized into treatment groups for each cell line: vehicle control or 150 mg/kg/day PD 0332991. PD 0332991 was orally administered (gavage) as a solution in 50 mmol/L sodium lactate, pH 4 for 5 consecutive days. Following this treatment, mice were euthanized by cervical dislocation, and tumors were excised by dissection. For each tumor sample, one part was fixed in 10% formalin overnight before transfer into 70% ethanol and another snap frozen in liquid nitrogen.

Mouse intestinal epithelium and tumors were processed for IHC as described previously (21).

We retrieved 585 transcriptomes of colorectal cancers (22) and used data as originally normalized. Spearman correlation was computed between MKI67 Affymetrix chip probes and all other gene probes over all samples simultaneously or for each sample cluster (cancer subtypes as defined in the original publication) separately. When multiple probes were available, the pair giving the highest absolute correlation (sum of absolute values over all clusters when computed for each cluster separately) was retained. Correlation P values were obtained with a Student t distribution with n–2 degrees of freedom (n = number of samples), and corrected for multiple hypotheses (Benjamini–Hochberg; ref. 23). Correlations with FDR<0.05 were considered. Breast tumor transcriptomes (1098 samples) were obtained from the Broad GDAC interface (http://gdac.broadinstitute.org/) to The Cancer Genome Atlas (TCGA) data. We used normalized RNA-seq data as provided and generated clustering of tumors with the Firehose online tool, which found 7 sample classes (Supplementary Fig. S1). Correlation computations were performed as above.

Ki-67 proteomics data (11) were filtered to obtain a Ki-67 interactome by retaining all bait proteins with Mascot identification score ratios in the top 5% with respect to negative controls (empty vectors or unrelated protein, TRIM39). We further removed all proteins present in more than 5% of the CRAPome (24). A total of 181 proteins passed this selection (Supplementary Table S1).

Interaction networks of Ki-67 highly correlated genes were obtained using pairwise interactions found in STRING database v10 (25), considering only top 10% STRING scores.

We first investigated whether Ki-67 expression in vivo varies in a cell-cycle–dependent manner by assessing Ki-67 staining on mouse intestinal epithelium. Cells in S-phase or G₂ were identified by injecting BrdUrd 2 hours prior to sacrifice. As expected, Ki-67 was absent in nonproliferating Paneth and goblet cells, whereas Ki-67 stained both BrdUrd-positive and negative nondifferentiated cells in the crypt (Fig. 1A). Ki-67 expression was higher in BrdUrd-positive cells (S-phase and G₂), and lowest in BrdUrd-negative cells, that are mostly G₁. Highest staining was observed in mitotic cells (Fig. 1A, yellow arrowheads).

We next asked whether Ki-67 expression was upregulated in tumors. We deleted exon 14 of one Apc allele in all intestinal epithelial cells using Villin-Cre^ERT2-mediated recombination. Adenoma formation then initiates through clonal growth from discrete epithelial cells that have lost the remaining Apc allele. Fig. 1B shows that Ki-67 levels were variable in both healthy tissue and in adenomas, similarly to proliferating cell nuclear antigen (PCNA; Fig. 1C), whose expression oscillates throughout the cell cycle. These results suggest that Ki-67 expression varies through the cell cycle in both normal tissues and in tumors.

We next looked at cell cycle exit in vivo. In the intestinal epithelium, actively proliferating crypt cells migrate upwards, exit the cell cycle, and differentiate before reaching villi. We compared Ki-67 IHC staining using different development times in serial sections of mouse intestinal epithelium. Using short exposures, Ki-67–positive staining can be seen only in cells within the proliferating crypt compartment, but longer exposures revealed low levels of Ki-67 in cells that have recently exited this compartment and started populating the adjacent villus base, where differentiating nonproliferating cells reside (Fig. 1D). Cells that have migrated up the villi have decreasing Ki-67 levels. Therefore, low Ki-67 levels reflect exit from the cell-cycle and terminal differentiation, but their assessment can be influenced by the staining protocol.

These results indicated that MKI67 levels are linked to the cell cycle in mice. To see whether the same is true in humans, we investigated coregulated gene expression using COXPRESdb (26). Gene ontology (GO) analysis (27) revealed that mechanisms involved in mitosis and the G₂–M transition constituted the 16 most enriched biological processes of the 100 most correlated genes (Supplementary Fig. S2A; Supplementary Table S2).

To investigate genes coregulated with MKI67 in human cancers, we first analyzed data from colorectal cancers, which have been extensively characterized at the molecular level (22). We assessed the proportion of cell-cycle genes among genes whose expression correlates with that of MKI67. Figure 2A shows that at a very high correlation coefficient (>0.6), around 80% of genes have a cell-cycle annotation. We next exploited STRING (25) to identify functional interactions. We superimposed high-confidence interactions determined from our own, high-confidence Ki-67 physical interaction data (11). Figure 2B shows that many proteins encoded by these MKI67 coregulated genes functionally interact in a cell-cycle network. Ki-67 interacts with both the mitotic and S-phase subnetworks. The remaining nodes also interact with the cell-cycle network and many have a metabolism annotation (which includes transcriptional regulation). Correlations with cell-cycle genes were maintained across colorectal cancer subtypes (Supplementary Fig. S2B). To see whether these results can be generalized to other cancers, we next interrogated TCGA with Ki-67 and searched for correlated expression across all breast cancer subtypes (28). Again, cell-cycle genes were predominant among the genes most correlated with MKI67 (Supplementary Fig. S2C and S2D), with the top hits being cyclins (CCNA2, CCNB1, CCNB2), CDK1, FOXM1 (a mitotic gene transcription factor; ref. 29), BUB1B and DLGAP5 (which controls mitotic spindle microtubule dynamics; Fig. 2C; ref. 30).

These results indicate that the main predictor of Ki-67 mRNA levels is the cell-cycle phase, with MKI67 most correlating with expression of genes involved in mitosis. We next experimentally analyzed Ki-67 and cyclin A2 mRNA and protein levels in a panel of human cancer cell lines: nontransformed human dermal fibroblasts (HDF) and their counterparts transformed with HPV-16 E7 oncogene, and cancer cells of different tissue origins and varying aggressiveness (U2OS, HeLa, HCT-116, MCF-7, HBL-100, CAL-51, and MDA-MB-231). As different human cancer cells express multiple smaller isoforms of Ki-67 (31), we quantified the total level of all Ki-67 species by dot-blotting. As controls, we used U2OS cells expressing nontargeting shRNA or Ki-67 shRNA (11), and HDF treated with the CDK4/CDK6 inhibitor PD0332991 (PD), which can induce senescence (32). As expected, Ki-67 shRNA or PD caused loss of both Ki-67 protein and mRNA. Ki-67 expression directly correlated with cyclin A2 expression at both protein (Fig. 2D) and mRNA (Fig. 2E) levels in all cell lines, including HDF treated with PD. Thus, Ki-67 levels are similar between cell types but depend on cell-cycle regulation.

To better understand cell-cycle variability of Ki-67 expression, we followed Ki-67 protein and mRNA levels in HDF cells synchronized throughout two cell cycles after release from quiescence. We used sequential block and release from CDK1 inhibition with RO-3306, which arrests cells at the G₂–M transition (33), and thymidine, which arrests cells in S-phase (Fig. 3A and B). Both Ki-67 and cyclin A2 mRNA levels oscillated, peaking at G₂ before decreasing in M-phase, as identified by phospho-histone H3 staining, dropping further in G₁ and rising again in the next cell cycle (Fig. 3C). Protein levels were also cyclic, peaking in mitosis (Fig. 3D). However, unlike mitotic cyclins, Ki-67 was not completely degraded during mitosis. We confirmed and extended these results to other cell lines by quantifying Ki-67 expression in asynchronous single HDF, HCT-116 or U2OS cells by immunofluorescence using markers of different cell-cycle stages (Supplementary Fig. S3A and S3B). Similar cell-cycle variation in Ki-67 expression level was found in a genome-scale proteomics and transcriptome analysis of a minimally perturbed cell cycle in human leukemic NB4 cells (34).

Next, we asked whether Ki-67 mRNA is sensitive to serum withdrawal after the restriction point, in cells released from a hydroxyurea-mediated S-phase block. Whereas D-type cyclin expression rapidly declined between S-phase and mitosis upon serum withdrawal, Ki-67 mRNA was stable, recapitulating cyclin A2 (Fig. 4A). Arresting cells in G₁ by CDK4/CDK6 inhibition with PD for 24 hours caused disappearance of Ki-67 mRNA and protein. Both Ki-67 loss and G₁ arrest were prevented by inactivating RB via expression of the HPV16 E7 oncogene (Fig. 4B–E). Ki-67 protein was degraded after release from the G₂–M block (Fig. 4F), rather than throughout the cell cycle as reported for HL60 cells (6). Ki-67 degradation at the mitosis/G₁ transition suggested that it involves the ubiquitin–proteasome system, in agreement with our previous demonstration (13) that Ki-67 downregulation is dependent on FZR1 (also known as CDH1), which activates the Anaphase-Promoting Complex (APC/C). Indeed, inhibiting the proteasome with MG132 prevented loss of Ki-67 protein upon PD treatment (Fig. 4G). Thus, CDK4/CDK6 inhibition eliminated Ki-67 expression by G₁ arrest, where Ki-67 mRNA expression is abolished and ubiquitin-mediated protein degradation occurs.

We next investigated how Ki-67 levels change in cells entering or leaving the cell cycle. First, HDFs were released from serum starvation, and DNA content and Ki-67 expression were determined over 30 hours (Fig. 5A). Very low levels of Ki-67 were detected in serum-starved cells, although cyclin A2 was absent and no cells were in S-phase. Whereas cyclin D1 levels rose rapidly, Ki-67 protein remained at a low level throughout the G₀–G₁ transition and rose upon entry into S-phase, when cyclin A2 became detectable. The major increase in Ki-67 expression occurred between S-phase and mitosis.

To verify whether persistent Ki-67 expression is a consistent feature of physiologically quiescent cells, we analyzed human umbilical cord T lymphocytes, which enter the cell cycle upon IL2 stimulation. Ki-67 was completely undetectable in nonstimulated T-lymphocytes. Again, Ki-67 appeared at a late stage, after 48-hour stimulation, coincident with cyclin A2 expression (Fig. 5B).

We reasoned that low level Ki-67 might persist in early stages of cell-cycle arrest and is gradually eliminated. To test this, we assessed Ki-67 expression by immunofluorescence in individual cells arrested using different approaches. We induced quiescence either by contact inhibition or serum starvation, or by DNA-damaging agents, ICRF-193 or bleomycin (Supplementary Fig. S4; ref. 20). Residual nuclear Ki-67 staining could clearly be seen in quiescent cells and was higher in contact-inhibited cells, which more readily enter the cell cycle when released, than serum-starved cells (Fig. 5C and D). However, Ki-67 was similar to background staining in cells with DNA damage (Fig. 5E and F). In all cases, cyclin A2 disappearance confirmed cell-cycle exit. Background Ki-67 levels could be detected 1 day after bleomycin treatment (Fig. 5F), but were essentially undetectable after 3 days and 7 days, by which time (7 days) senescent cells were readily visualized by β-galactosidase activity staining (Supplementary Fig. S5).

Taken together, these results show that high Ki-67 expression is a late marker of cell-cycle entry and its highest levels occur in G₂ and M-phase. Quiescent cells and cells entering the cell cycle have low Ki-67, and Ki-67 is undetectable in deeply quiescent or senescent cells.

Our results suggest that Ki-67 expression could be useful for assessing cellular responses to CDK4/CDK6 inhibition. However, as neither CDK4/CDK6 nor Ki-67 are essential for cell proliferation in all cells, it was important to verify whether Ki-67 expression always recapitulates cell proliferation status upon CDK4/CDK6 inhibition. We tested this using a panel of cancer cell lines treated with a range of concentrations of PD for 24 hours, and then exposed to 5-ethynyl-2-deoxyuridine (EdU) for a further 24 hours, to assess DNA replication. Samples were taken to measure Ki-67 protein and mRNA. This revealed that the cell-cycle responses to PD strictly correlated with the effects on Ki-67 mRNA and protein levels (Fig. 6A). In RB-positive cells, this was further correlated with loss of RB phosphorylation. PD most effectively prevented S-phase onset and downregulated Ki-67 in RB-positive MDA-MB-231 and MCF7 cells. HCT-116 cells, which are RB-positive, responded poorly to PD, possibly because they are also mutated for KRAS and PI3KCA (35). PD had no effect on low-RB–expressing HeLa cells.

To confirm the robustness of Ki-67- and cyclin A2-correlated expression upon PD treatment, we interrogated the Cancer Cell Line Encyclopedia (36). We also compared expression of Mki67 with mRNA encoding cyclins B1, D1, and D3. This showed an extremely tight correlation between MKI67 and CCNA2 (cyclin A2) in all cell lines at any concentration of PD, and a good, but slightly lower, correlation with CCNB1 (cyclin B1; Fig. 6B). Ki-67 and D-type cyclin expression was not correlated. This data also shows that PD sensitivity (inversely related to IC₅₀) and Ki-67 levels were not correlated. Thus, there is no general value of Ki-67 expression in predicting sensitivity to PD. We then extended this question to all drugs analyzed in the CCLE. Considering all drugs together, MKI67 expression showed no correlation with drug-sensitivity (IC₅₀; Supplementary Fig. S6A). However, MKI67 and CCNA2 showed a weak correlation with sensitivity to topoisomerase inhibitors irinotecan and topotecan (Supplementary Fig. S6B).

Finally, we performed an experiment in mice to see whether effects of CDK4/CDK6 inhibition on cell proliferation and Ki-67 expression correlate in tumors in vivo. We compared PD-sensitive and resistant cell lines, MDA-MB-231 and MDA-MB-468, respectively (37). We engrafted these lines subcutaneously into nude mice and allowed tumors to grow to 200 mm³ before treating mice with vehicle or PD for 5 days by oral administration. As expected from in vitro experiments, PD treatment arrested tumor growth from MDA-MB-231, but not MDA-MB-468 (Fig. 7A). This was reflected by strongly decreased IHC staining for cyclin A2, PCNA, and Ki-67 in MDA-MB-231, but no change in these markers in MDA-MB-468 (Fig. 7B). The number of cells scored positive for Ki-67 and PCNA was most similar in untreated samples, and higher than cyclin A. This is because Ki-67 and PCNA, although variable, are present throughout the cell cycle, whereas cyclin A is only present from S-phase to G₂. However, responses to PD in MDA-MB-231 were more complete for Ki-67 and cyclin A than for PCNA, where low level staining remained in a minority of treated cells. This might reflect a difference in half-life of PCNA compared with cyclin A and Ki-67, both of which are degraded by the APC/C in mitosis and G₁. Analysis by qRT-PCR confirmed that protein levels were recapitulated by mRNA levels, and also showed that cyclin D mRNA levels did not change (Fig. 7C). Thus, Ki-67 is a good marker for cell proliferation status in response to PD in vivo.

In light of the variability of the Ki-67 index in cancer biopsies and lack of consistent correlation with responses to therapy, it is important to understand how Ki-67 expression is controlled. We find that cell-cycle regulation accounts for variability in Ki-67 expression in primary cells and cancer cell lines as well as in tumors and human cancers. Thus, low and high level Ki-67 should be scored as positive to determine the Ki-67 labeling index. However, extremely low Ki-67 levels can be detected in quiescent cells by IHC upon long exposure. Unlike the situation in senescent cells, which have no Ki-67, such low levels of Ki-67 staining persist in cells that have recently stopped proliferating and entered quiescence.

Ostensibly, this is incompatible with the idea of Ki-67 as a specific marker for proliferating cells, but is consistent with a previous report that Ki-67 could be detected at sites of ribosomal RNA synthesis in quiescent cells (38). We speculate that a basal level of Ki-67 might be a marker for the recently described primed state for cell-cycle reentry termed G(Alert) (39). This basal level of Ki-67 in arrested cells contributes to the variability in assessments of Ki-67 staining index in cancers as cells might be variably classed as Ki-67–positive or negative. Basal Ki-67 expression might itself be a useful marker to identify cells within tumors that proliferate slowly or are quiescent, and thus are more resistant to chemotherapy or radiotherapy than proliferating cells (40). Such populations appear to be responsible for relapse after chemotherapy in colorectal cancer patients (41). Furthermore, in breast cancer, cells with low proliferation rates, and therefore low Ki-67 index, can sustain the tumor niche for highly proliferative clones, with which they remain in equilibrium (42). Quiescent cells would likely be undetectable upon standard IHC analysis, but our data suggest that they could be identified and distinguished from proliferating or senescent cells by more sensitive IHC analysis. Cells with such low levels of Ki-67 should be scored separately from cells with higher Ki-67 levels, which are proliferating, as they may have implications for prognosis of relapse.

We find that Ki-67 cell-cycle regulation relies on two opposing mechanisms dependent on conserved cell-cycle regulators: CDK4/CDK6 phosphorylates RB, allowing Ki-67 mRNA expression in G₁, and this is opposed by protein degradation in late mitosis and early G₁ by the ubiquitin–proteasome system. This corroborates our recent findings that Ki-67 protein expression is maintained in nonproliferating cells mutated for the Fzr1 gene, which encodes the CDH1 activator of the mitotic/G1 ubiquitin ligase, APC/C. Eliminating both RB and APC/C-CDH1 bypasses CDK4/CDK6 inhibition in breast cancer cells, and their combined gene knockout in nematodes circumvents the requirement for CDK4 (43). Thus, CDK4/CDK6 inhibition might both prevent Ki-67 transcription and promote its degradation. The mechanisms regulating Ki-67 expression link it to the cell cycle, resulting in maximal Ki-67 levels in mitosis and minimal Ki-67 levels in late G₁. In cancer cells, inhibition of entry into S-phase strictly correlates with downregulation of Ki-67. Although tumor explants with inactivated RB, which do not respond to PD0332991, have a higher initial Ki-67 index (17), we find that in CCLE data, Ki-67 expression does not generally correlate with PD0332991 sensitivity. However, we confirmed in vivo that PD0332991 treatment abrogates Ki-67 expression only when it abolishes cell proliferation. This provides a rationale for using Ki-67 expression as a biomarker to measure responses to PD0332991 or other CDK4/CDK6 inhibitors currently under development. Indeed, recent phase II trials with one such inhibitor, abemaciclib, found that it significantly reduced Ki-67 expression in patients with untreated early-stage breast cancer (44). Our data confirm that this reliably indicates reduced cell proliferation.

It has long been assumed that Ki-67 is essential for cell proliferation, and several previous studies have supported this notion (45–50). However, using mice mutant for Ki-67, we recently demonstrated that, rather than controlling cell proliferation directly, Ki-67 is required to organise heterochromatin in proliferating cells (11). Nevertheless, Ki-67 downregulation using oncolytic viruses armed with Ki-67 shRNA decreased tumor growth in xenograft experiments in immunodeficient mice (50). Taken together, this suggests that even if Ki-67 is not required for cell proliferation directly, it might promote tumorigenesis. Further analysis will be required to determine whether Ki-67 expression is required for tumorigenesis and its biochemical mechanisms of action.

Taken together, our results show that the average level of Ki-67 mRNA and protein in proliferating cells is similar and independent of cell type, and its levels in any one cell depend on the cell-cycle phase. In all circumstances examined, including CDK4/CDK6 inhibition, loss of Ki-67 reflected loss of cell proliferation. Thus, Ki-67 expression can be used as a biomarker for inhibition of cell proliferation by CDK4/CDK6 inhibitors, and probably any drug.

No potential conflicts of interest were disclosed.

Conception and design: M. Sobecki, K. Mrouj, L. Krasinska, V. Dulic, D. Fisher

Development of methodology: J. Colinge

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Sobecki, F. Gerbe, P. Jay, V. Dulic, D. Fisher

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Sobecki, J. Colinge, F. Gerbe, P. Jay, V. Dulic, D. Fisher

Writing, review, and/or revision of the manuscript: J. Colinge, L. Krasinska, V. Dulic, D. Fisher

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases):

Study supervision:

The authors thank the members of the Fisher laboratory for helpful discussions, technical staff of MRI imaging facility and IGMM and IGF mouse facilities, IRCM Cell culture unit of Montpellier SIRIC (INCa-DGOS-Inserm 6045) for cell lines, Karim Chebli and Susan Prieto for help with xenograft experiments, and Benedicte Lemmers and Susana Prieto for help with immunohistochemistry.

This work was supported by grants from the GEFLUC LR, Agence Nationale de la Recherche (ANR-09-BLAN-0252) and the Ligue Nationale contre le Cancer (EL2010.LNCC/DF and EL2013.LNCC/DF; all to D. Fisher). Further funding was provided by Worldwide Cancer Research (16-0006 to D. Fisher). M. Sobecki was supported by the Ligue Nationale Contre le Cancer and by the Fondation pour la Recherche Médicale. K. Mrouj was supported by the Ligue Nationale contre le Cancer. J. Colinges was supported by the Fondation ARC pour la Recherche sur le Cancer (PJA 20141201975).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.