The history and future of targeting cyclin-dependent kinases in cancer therapy

Integration of multiple signalling pathways through control of CDK4 and CDK6 activation

An understanding of the biology of CDKs is critical to deciphering the clinical results seen with CDK inhibitors, particularly in regard to determining biomarker and combination strategies. In most adult tissues, the majority of cells exit the cell cycle with diploid DNA content and are maintained in a quiescent G0 state. Tissue maintenance involves cues that physiologically induce cell cycle entry in a highly regulated manner. The mechanisms through which cells initiate entry into the cell cycle have been comprehensively described. Extracellular signals — including those activated by peptide growth factors (for example, RAS, mitogen-activated protein kinase (MAPK) and mammalian target of rapamycin (mTOR)) and nuclear receptors (for example, the oestrogen receptor (ER) in mammary epithelia) — converge on the cell cycle to drive progression from G0 or G1 phase into S phase through regulation of the metazoan-specific CDK4 or CDK6 complex^2,3,12,13. CDK4 and CDK6 emerged phylogenetically with the appearance of multicellular organisms, and are subjected to multiple levels of regulation to control the transition into S phase. CDK4 and CDK6 are structurally related proteins that harbour many biochemical and biological similarities, although most published studies have focused on CDK4 (REF. 14). CDK6 is particularly important in promoting the proliferation of haematological precursors^15,16.

The activity of CDK4 and CDK6 is primarily controlled by their association with D-type cyclins (that is, cyclin D1, cyclin D2 and cyclin D3)^17,18. Among these, cyclin D1 is the best characterized. The expression of cyclin D1 is characterized as a ‘delayed-early’ response to mitogenic signalling, and intricate promoter and enhancer interactions control its transcription¹⁹. Although less well studied, cyclin D3 conforms to a similar pattern as cyclin D1, whereas the regulation of cyclin D2 remains more enigmatic, although cyclin D2 also drives proliferation in certain contexts^20-24. The differential expression of paralogues of D-type cyclins is likely to reflect tissue-specific aspects of normal physiology, wherein different D-type cyclins are expressed to promote CDK4 or CDK6 activation^3,25.

In addition to the transcriptional regulation of CDK4 and CDK6, the stability, intracellular localization and association of cyclin D with CDK4 and CDK6 are tightly regulated (FIG. 2a). In particular, cyclin D1 is unstable and actively shuttles between the cytoplasm and the nucleus. Phosphorylation of threonine 286 on cyclin D1 actively promotes its export and ubiquitinmediated degradation^26,27. In contrast to other CDKs, for which cyclin association seems to occur relatively spontaneously, for CDK4 and CDK6 this process is regulated by multiple mechanisms²⁸. The inhibitor of CDK4 (INK4) proteins, which include p16^INK4A, p15^INK4B, p18^INK4C and p19^INK4D, represent CDK4- and CDK6-interacting proteins that seem to solely function as inhibitors of CDK4 and CDK6 (REFS 3,29,30). The INK4 proteins weaken the binding of D-type cyclins to CDK4 and CDK6, and also interact with the catalytic domains of CDK4 and CDK6 to potently suppress kinase activity^14,31,32. These proteins therefore negatively regulate CDK4 and CDK6 in response to stress conditions³³. For example, p16^INK4A is induced by multiple oncoproteins to counteract transformation. Moreover, under stress conditions associated with cellular ageing³⁴, overexpression of p16^INK4A results in a profound G1 arrest of the cell cycle. Similarly, p15^INK4B is induced by transforming growth factor-β-mediated suppression of epithelial cell proliferation³⁵. CDK4 and CDK6, similar to other CDK proteins, are also subjected to phosphoregulation^36,37. Thus, CDK4 and CDK6 serve as key nodes of integration downstream of multiple signalling pathways, in which their activation initiates progression into the cell cycle (FIG. 2a).

The association of D-type cyclins with CDK4 and CDK6 can induce kinase activity with a unique substrate spectrum compared with other CDKs³⁸. In particular, CDK4 and CDK6 have a specific preference for the phosphorylation of the tumour suppressor retinoblastoma protein (RB) and the related proteins p107 (also known as RBL1) and p130 (also known as RBL2)^39,40 (FIG. 2b). RB, the first tumour suppressor identified, has been extensively studied^41,42. The RB protein does not have catalytic activity but functions through the assembly of multiprotein complexes to control the cell cycle. In particular, RB can bind to the E2F transcription factors, recruit co-repressors and repress the transcription of target genes that are regulated by E2Fs^41,42 (FIG. 2b). The E2Fs regulate the expression of a set of genes involved in cell cycle control (for example, cyclin E (CCNE), CCNA and CCNB1), dNTP biosynthesis (for example, dihydrofolate reductase (DHFR), ribonucleotide reductase M1 (RRM1) and RRM2) and mitotic progression (for example, polo-like kinase 1 (PLK1), BUB1 mitotic checkpoint serine/threonine kinase (BUB1) and spindle checkpoint protein MAD2 (MAD2)). The phosphorylation of RB by CDK4 and CDK6 initiates an intricate process of phosphorylation-mediated disruption of RB function that releases E2F and initiates subsequent progression through the cell cycle (FIG. 2c). CDK4 and CDK6 also phosphorylate forkhead box protein M1 (FOXM1), leading to stabilization of FOXM1 as a further mediator in the expression of genes required for progression though mitosis³⁸ (FIG. 3).

Deregulation of the CDK4/6-RB-p16^INK4A pathway in cancer

The CDK4/6–RB axis is critical to cell cycle entry; therefore, it is unsurprising that the vast majority of cancers subvert this axis to promote proliferation^2,42,43 (FIG. 4). Most oncogenes promote the induction of p16^INK4A as an intrinsic break to deregulated proliferation^34,44-46. Overexpression of p16^INK4A ultimately engages RB to suppress growth and cell cycle progression, and promotes oncogene-induced senescence. Oncogene-induced senescence must be subverted to enable subsequent oncogenic proliferation, which occurs through two principal means in tumours: loss of p16^INK4A or loss of RB^29,47. Loss of p16^INK4A uncouples the oncogenic stress from the suppression of CDK4 or CDK6 activity, whereas loss of RB deregulates downstream signalling in the cell cycle. Consistent with this model, RB is required for the cell cycle arrest mediated by p16^INK4A (REFS 48,49). In addition, RB-negative tumours express super-physiological levels of p16^INK4A and are therefore insensitive to additional expression of p16^INK4A owing to the absence of RB²⁹.

A contrasting mechanism of deregulating the CDK4/6–RB axis is the direct oncogenic activation of CDK4 or CDK6 activity. Deregulated cyclin D1 protein expression, gene translocation and gene amplification are observed in many tumour types^50-54, and a plethora of functional data support the specific oncogenic activity of cyclin D1 (REFS 17,18,51). Furthermore, amplification of CDK4 and CDK6 is observed in several different types of cancer^55,56. Importantly, the distinct mechanisms of pathway dysregulation are mutually exclusive and are frequently tumour type-specific. For example, RB loss is a hallmark of small cell lung cancer, deregulation of cyclin D1 is common in breast cancer, and loss of p16^INK4A is particularly common in glioblastoma (FIG. 4).

Distal regulation of CDK2 and its deregulation in cancer

Although all CDKs have similarities, CDK2 is structurally and functionally related to CDK1 (REF. 3). CDK2 has a considerably broader substrate profile than CDK4 and CDK6, and it phosphorylates a large number of proteins involved in cell cycle progression (for example, p27^KIP1 and RB), DNA replication (for example, replication factors A and C), histone synthesis (for example, NPAT), centrosome duplication (for example, nucleophosmin (NPM)), among other processes^57-59 (FIG. 3). In vitro, CDK2 and its preferred E-type and A-type cyclin partners assemble spontaneously to form active kinase complexes^3,60. Much of the control over CDK2 involves the synthesis and availability of the cyclins, with RB and E2F regulating the abundance of CDK2, cyclin E1 and cyclin E2 transcripts and proteins^61-65. This process couples mitogen-mediated activation of CDK4 and CDK6 with the activation of CDK2 (REFS 66,67) (FIG. 2c). In contrast to CDK4 and CDK6, CDK2 is not regulated by INK4 proteins^30,68 but by the CDK-interacting protein/kinase inhibitory protein (CIP/KIP) class of CDK inhibitors, which bind to CDK2–cyclin complexes and render them inactive^60,69-71. For example, p21^CIP1 acts as a DNA damage checkpoint (it is a critical downstream target gene of p53 that inhibits DNA synthesis), whereas p27^KIP1 is responsive to mitogenic signalling as a further control on deregulated proliferation^71,72. Additionally, CDK2 is regulated by phosphorylation events⁷³. Therefore, multiple mechanisms in addition to the CDK4/6–RB axis can modulate the activity of CDK2 and subsequent DNA replication.

Recently, it has become clear that deregulation of CDK2 also occurs frequently in certain types of cancer⁷⁴. Cyclin E1 or cyclin E2 amplifications are key oncogenic events in multiple cancers, particularly in uterine and ovarian cancers^75-77 (FIG. 4). Ectopic expression of cyclin E bypasses the need for CDK4 or CDK6 activity to initiate the S phase^78-80, and it is therefore assumed that amplification of E-type cyclins may be oncogenic in a similar manner (that is, bypassing the physiological requirement for CDK4 or CDK6 activity to initiate expression of E-type cyclins). The CDK inhibitor p27^KIP1 is downregulated in many cancers, although the genetic loss of p27^KIP1 is fairly rare^81,82.

CDK1 is a key determinant of mitotic progression

CDK1 was cloned on the basis of complementarity to the cdc2 gene of Schizosaccharomyces pombe⁸³. The expression of CDK1 and associated cyclins (cyclin A2 and cyclin B1) is coordinated through the activity of E2F-assembled complexes^63,65. These include the canonical E2F and RB constituents, as well as the transcription factor FOXM1 and the DREAM (dimerization partner, RB-like, E2F and multivulval class B) complex, which are particularly relevant for the coordination of transcripts involved in mitotic progression^84-86. The cyclins that assemble with CDK1 are intrinsically unstable and are regulated by ubiquitin-mediated degradation mechanisms, and the assembly and localization of CDK1 complexes are regulated by multiple overlapping mechanisms^87-90.

Mouse knockout experiments have shown that CDK1 is required for mammalian cell proliferation; it is the only CDK that can initiate the onset of mitosis (that is, the M phase)⁹¹. Premature initiation of mitosis before completion of the S phase results in chromosomal shattering and cell death⁹². Multiple factors restrain the activity of CDK1 until DNA replication is complete and there is minimal DNA damage. This integration of DNA replication and CDK1 activity is mediated by checkpoint signalling kinases such as CHK1 and WEE1, which suppress the activity of CDK1 via inhibitory phosphorylation⁹³, as well as through the cell division cycle 25 (CDC25) family of phosphatases. At the onset of mitosis, activation of CDK1 occurs rapidly through a positive feedback loop whereby CDK1 phosphorylates and inactivates WEE1. CDK1 subsequently phosphorylates multiple substrates, leading to nuclear envelope breakdown, chromosome condensation and mitotic spindle assembly⁹⁴ (FIG. 3). The subsequent progression from metaphase to anaphase is controlled by the spindle assembly checkpoints, and progression through anaphase is dependent on the attenuation of CDK1 activity through the degradation of cyclin B1 by the anaphase-promoting complex^95,96.

Interestingly, in contrast to the genetic deregulation of the CDKs that coordinate the S phase, there is limited evidence to show that CDK1 activity is dysregulated by direct genetic alteration in tumorigenesis. Derangement of p53 signalling or of DNA damage checkpoints indirectly leads to the deregulation of CDK1 (REFS 97,98), and high cyclin B1 expression is generally associated with a more aggressive cancer phenotype^99,100. However, the requirement that CDK1 activity must be attenuated to exit mitosis and the lethal aspects of uncoordinated CDK1 activity are likely to limit its potential as a direct oncogenic driver.

The role of cell cycle-independent CDKs

In addition to the well-known CDKs involved in regulating the cell cycle, there is an equivalent number of CDKs involved in basal transcriptional regulation^3,10,11 (FIG. 3). In particular, CDK8 is part of the mediator complex that regulates a plethora of genes^101,102. CDK7 has a general role in the phosphorylation of the RNA polymerase II carboxyterminal domain that contributes to the initiation of transcription¹⁰³, and CDK9 also phosphorylates RNA polymerase II, thereby promoting elongation of transcription. Finally, CDK11 acts on the splicing machinery. In each of these contexts, the CDK activity is directed by specific cyclin interactions. Accumulating evidence suggests that these transcription-regulating CDKs may represent therapeutic targets in cancer.

In addition to the CDKs involved in transcriptional regulation, there remains a class of CDKs, including CDK3 and CDK5, for which the underlying functions are elusive. CDK3 was found to be intrinsically important for cell cycle control based on cell-based experiments that used a dominant-negative version of CDK3 (REF. 72) (FIG. 3). However, it was subsequently revealed that many mouse strains harbour an inactive CDK3, suggesting that its role in the cell cycle can be readily compensated¹⁰⁴. CDK5 was largely viewed as a neuronal kinase implicated in the protection of the nervous system from injury¹⁰⁵. However, recent work suggests that it harbours many functions similar to CDK4 and CDK2 in driving progression from G1–S and in RB phosphorylation in medullary thyroid cancer models¹⁰⁶. CDK5 might therefore have potential as a therapeutic target in this thyroid cancer subtype¹⁰⁶.

The first generation of CDK inhibitors

Over the past 20 years, several CDK inhibitors have been developed as potential cancer therapeutics and tested in numerous trials and in several tumour types (FIG. 5). The first-generation CDK inhibitors developed were relatively nonspecific and may therefore be referred to as ‘pan-CDK’ inhibitors (an example of which is flavopiridol (also known as alvocidib; developed by Sanofi-Aventis), although some compounds, such as olomucine (not commercially developed) and roscovitine (also known as seliciclib; developed by Cyclacel), have relatively low affinity for CDK4 and CDK6). As the field of CDK biology progressed in parallel with the development of these agents, our understanding of their targets and interpretation of their behaviour have also evolved. For example, it was initially believed that roscovitine was a relatively specific inhibitor of CDK1, CDK2 and CDK5; however, subsequent data demonstrated that it also inhibited transcription, probably through the inhibition of CDK7 and CDK9 (REF. 107).

Of these first-generation inhibitors, flavopiridol is the most extensively investigated CDK inhibitor so far, with >60 clinical trials carried out between 1998 and 2014 (see Supplementary information S1 (box)). Flavopiridol is a semi-synthetic flavonoid derived from rohitukine, a chromone alkaloid, and has been shown to inhibit CDK1, CDK2, CDK4, CDK6, CDK7 and CDK9 (REFS 108,109). Although flavopiridol can induce cell cycle arrest in G1 and G2 phases, in certain contexts it also induces a cytotoxic response, probably as a result of CDK7 and CDK9 inhibition that leads to suppression of transcription¹¹⁰. Although the underlying broad-spectrum nature of flavopiridol results in substantial in vitro activity, substantially less activity was observed in vivo¹¹⁰. Consequently, flavopiridol did not meet the initial high expectations for a CDK inhibitor, and low levels of clinical activity were seen in Phase II studies in several solid tumour types (see Supplementary information S1 (box)). There is evidence to indicate that flavopiridol may have clinical activity in haematological malignancies, such as chronic lymphocytic leukaemia (CLL) and mantle cell lymphoma^111,112, in which scheduling also seems to influence flavopiridol efficacy. For example, a relatively short infusion time (4 hours) resulted in response rates of 41% in 22 assessable patients with CLL^113,114. Patients with a high disease burden and high-risk genetic features achieved durable responses, and tumour lysis syndrome was reported in approximately 40% of patients with CLL treated with flavopiridol¹¹⁵. Despite these reports and extensive investment, no Phase III studies have emerged and drug development of flavopiridol was consequently discontinued in 2012.

In parallel with flavopiridol, roscovitine, a purine-based CDK inhibitor, was among the first agents to be evaluated in the clinic (see Supplementary information S1 (box)). Of the 56 patients treated in the Phase I setting, 1 patient achieved a partial response¹¹⁶. A subsequent blinded, randomized Phase II trial (APPRAISE) that compared roscovitine with the best supportive care was performed in patients with advanced non-small cell lung cancer. The APPRAISE study was terminated after 187 patients were enrolled; although results were never published, roscovitine did not seem to improve progression-free survival in this patient population (see Cyclacel press release). Currently, only a single trial is ongoing for roscovitine in Cushing disease.

Second-generation inhibitors of multiple CDKs

Following on from flavopiridol and roscovitine, other CDK inhibitors were developed with the aim of increasing selectivity for CDK1 and CDK2 and/or increasing overall potency (FIG. 5). Again, numerous CDK inhibitors seemed to be particularly promising in preclinical studies, but only a few progressed past Phase I clinical trials^117-120 (see Supplementary information S1 (box)).

Of these second-generation CDK inhibitors, dinaciclib (also known as MK-7965 and SCH727965; developed by Merck) has been most extensively studied in the clinic. Dinaciclib was specifically developed as a highly potent inhibitor of CDK1, CDK2, CDK5 and CDK9 (half-maximal inhibitory concentration (IC₅₀) values in the range of 1–4 nM), with less activity against CDK4, CDK6 and CDK7 (IC₅₀ values in the range of 60–100 nM). Compared to flavopiridol, dinaciclib exhibited superior activity with regard to suppression of RB phosphorylation in cell-based assays¹¹⁸. Moreover, dinaciclib inhibited cell cycle progression in >100 tumour cell lines of various tumour types and induced the regression of established solid tumours in a range of mouse models¹¹⁸. Initial results from Phase I studies were promising: dinaciclib induced stable disease in a range of malignancies and displayed tolerable toxicity¹²¹. However, results from recent randomized Phase II trials of dinaciclib in solid tumours have been disappointing. A randomized Phase II trial of dinaciclib versus the chemotherapeutic agent capecitabine in advanced breast cancer was stopped after 30 patients were enrolled because interim analysis showed that the time to disease progression was inferior with dinaciclib treatment¹²². In a study evaluating dinaciclib monotherapy in patients with non-small cell lung cancer, dinaciclib showed no activity in previously treated patients¹²³. In adult patients with advanced acute myeloid leukaemia (patients ≥60 years of age) or acute lymphoblastic leukaemia, no objective responses were observed with dinaciclib¹²⁴. However, in patients with relapsed multiple myeloma, preliminary encouraging single-agent activity was reported in a Phase I/II trial, with 2 patients of 27 achieving partial responses¹²⁵, and dinaciclib, similar to flavopiridol, demonstrated clinical activity in pretreated patients with CLL¹²⁶. Based on these data, a randomized Phase III study of dinaciclib in refractory CLL is ongoing. Therefore, dinaciclib may prove useful in the treatment of certain haematological malignancies, in which flavopiridol also had evidence of activity.

Other CDK inhibitors include AT7519 (developed by Astex), a pyrazole 3-carboxyamide compound that acts as an inhibitor of CDK1, CDK2, CDK4, CDK6 and CDK9. AT7519 has been evaluated in combination with bortezomib in a Phase II clinical trial enrolling patients with previously treated multiple myeloma. The combination was well tolerated, and more than one-third of patients achieved partial response¹²⁷. R547 (developed by Hoffman-La Roche) is an inhibitor of CDK1, CDK2 and CDK4 with less potency for CDK7, glycogen synthase kinase 3α (GSK3α) and GSK3β. R547 was tested in a Phase I study in 2007 as an intravenous weekly infusion¹²⁸. Although reported to have manageable side effects, there have not been further clinical trials with this compound. SNS-032 (also known as BMS-387032; developed by Bristol-Myers Squibb), which was initially described as having greater selectivity for CDK2 than CDK1 and CDK4, is now recognized to also target CDK7 and CDK9. Two Phase I clinical studies with SNS-032, one in 2010 in advanced lymphoid malignancies¹²⁹ and one in 2008 in selected advanced tumours¹³⁰, have been reported, but no further developments are apparent. The development of AZD5438 (developed by Astra Zeneca) — an orally bioavailable, potent inhibitor of CDK1, CDK2 and CDK9 with less selectivity for CDK5 and CDK6 (REF. 131) — was discontinued after it was reported to be intolerable when administered continuously in patients with advanced solid tumours¹³². AG-024322 (developed by Pfizer) — a potent inhibitor of CDK1, CDK2 and CDK4 — was also discontinued in 2007 after the Phase I study was terminated, as it failed to achieve an acceptable clinical end point (see Supplementary information S1 (box)).

Reasons for failure of CDK inhibitors with low selectivity

The general failure of non-selective CDK inhibitors in the clinic can be partly explained by at least three key underlying principles. First, there was a lack of clear understanding of the mechanism of action. For many of the CDK inhibitors with low specificity, there remains a lack of clarity with regard to which CDKs are actually being inhibited in vivo and therefore the corresponding mechanism that could underlie the therapeutic effect. For example, flavopiridol has been associated with diverse distal cellular effects, including cell cycle inhibition, transcriptional suppression, apoptosis, autophagy and endoplasmic reticulum stress^133-135. This lack of understanding confounds the ability to develop these agents as targeted therapies and to design effective combination strategies.

Second, there was a lack of appropriate patient selection. The vast majority of studies conducted with CDK inhibitors with low specificity were in unstratified patient cohorts. This is because there are essentially no biomarkers that may select for sensitive subpopulations for this class of inhibitors. The potential activity of both flavopiridol and dinaciclib in CLL and the rare ‘extraordinary’ responders suggest that there are molecular-based reasons that some tumours are vulnerable to such agents. Although the molecular underpinnings for these responses are unknown, it is tempting to speculate that the inhibition of CDKs that control transcription could underlie at least part of this activity.

Third, there is a lack of a therapeutic window. Many of these CDK inhibitors target several proteins that are critical to the proliferation (for example, CDK1) and survival (for example, CDK9) of normal cells. This limits the ability to achieve therapeutic levels of these drugs because of their intrinsic inability to discriminate between cancerous and healthy tissues. Consistent with this point, the toxicities associated with non-selective CDK inhibitors include diarrhoea, myelosuppression, anaemia and nausea^116,121,136.

A case for selectivity of CDK inhibitors

The ascribed weaknesses of pan-CDK inhibitors suggest that improved selectivity for certain CDKs is the key to the successful development of CDK inhibitors as therapeutic cancer agents. Selective inhibitors of CDK2 might provide the ability to target genetic and/or driver events in tumours driven by cyclin E amplification. Emerging data suggest that targeting CDK1 is toxic in certain contexts, and it may be challenging to achieve a therapeutic window. For example, synthetic lethal screens against KRAS mutations have indicated a potential sensitivity to CDK1 knockdown, although follow-up studies are required¹³⁷. Similarly, CDK1 or CDK9 inhibition is synthetically lethal with MYC^138,139. Pharmacologically, CDK1 inhibitors seem to potently cooperate with inhibitors of poly(ADP-ribose) polymerase¹⁴⁰ by compromising DNA repair pathways. An alternative strategy is to selectively target CDK7, CDK8 or CDK9, which are associated with basal transcription, because cancer cells may harbour unique vulnerabilities to selective suppression. CDK8 may function as an oncoprotein in colorectal cancer by regulating the transcription of β-catenin target genes¹⁴¹. A covalent inhibitor of CDK7 (THZ1), which is relatively specific for CDK7 compared with other CDKs, has shown activity in multiple cancer cell models¹⁴². Similarly, a specific inhibitor for CDK9 (CDKI-73) that exhibited activity in animal models of leukaemia was recently developed¹⁴³.

Rationale for targeting CDK4 and CDK6

Based on the myriad findings from mechanistic studies and studies of CDK4 and CDK6 deregulation in cancer, three important features and expectations arose for CDK4 and CDK6 inhibitors in the clinic. First, one would expect that a pure CDK4 or CDK6 inhibitor would elicit a single phenotype in tumours: cytostatic G0/G1 arrest. Second, this effect would be a direct reflection of the engagement of RB to suppress gene expression and proliferation. Third, such effects would be particularly actionable in tumours that exhibit deregulation of CDK4 and CDK6 activity as opposed to other CDKs. Initial data from mouse models seeded confusion as to whether CDK4 and CDK6 were therapeutic targets because many tissues in the mouse developed normally despite the absence of CDK4 and/or CDK6 (REFS 15,144) and in the absence of D-type cyclins¹⁴⁵. These data reflected substantial compensatory plasticity with other CDKs. Despite uncertainties arising from the mouse knockout models, it subsequently became clear that attenuation of CDK4 and CDK6 activity could prevent the development of specific mouse tumour types. For example, cyclin D1 is crucial for the development of mammary tumours driven by HER2 (also known as ERBB2)¹⁴⁶, and similar observations were obtained in NOTCH-driven T cell leukaemia mouse models¹⁴⁷ and cell lines^147,148.

Pharmacological approaches

The development of selective inhibitors of both CDK4 and CDK6 has markedly changed the perception of CDKs as therapeutic targets in cancer. Through a combination of chemical screening and optimization, it was found that pyrido[2,3-d]pyrimidin-7-one compounds with a 2-amino pyridine side chain at the C2 position act as CDK inhibitors with a high degree of selectivity for CDK4 and CDK6 relative to other CDKs¹⁴⁹ (FIG. 5). Subsequent optimization resulted in the compound PD-0332991 (also known as palbociclib; developed by Pfizer) that induced potent G1 arrest in cell culture and xenograft models^150,151. As anticipated from the basic biology of G1–S transition, the effects of palbociclib were dependent on the presence of a functional RB protein, thus demonstrating a degree of biological specificity that had not been previously described for CDK inhibitors^150,151. Subsequently, multiple independent groups have demonstrated that specific CDK4 and CDK6 inhibitors arrest the cell cycle through the downstream blockade of phosphorylation of RB, as well as the related p107 and p130 proteins. This blockade results in the loss of expression of S-phase cyclins, nucleotide biosynthesis, DNA replication machinery and mitotic regulatory genes^152,153. Dual CDK4 and CDK6 inhibitors have been shown to be active in multiple preclinical models, including xenografts, genetically engineered mouse models and primary human tumour explants^152,154-156 (TABLE 1). Parallel drug discovery efforts at Eli Lilly and Novartis resulted in the development of the drugs LY-2835219 (also known as abemaciclib) and LEE011, respectively^157-162. Each drug is structurally similar and chemically distinct from the less specific pan-CDK inhibitors (FIG. 5). The selectivity of all of these compounds is likely to reflect binding to the specialized ATP-binding pocket of CDK4 and CDK6 and specific interactions with residues in the ATP-binding cleft, although this has not been proven by structural analysis. In contrast to other CDK4- and CDK6-selective compounds, abemaciclib inhibits CDK9 in in vitro kinase assays, although there is no evidence of functional inhibition in cellular models¹⁵⁸.

Single-agent clinical outcomes

There are numerous emerging clinical studies with dual CDK4 and CDK6 inhibitors (TABLE 2). Results from a Phase I study with palbociclib monotherapy indicated promising clinical efficacy and a well-tolerated toxicity profile in patients with RB-positive advanced solid tumours and non-Hodgkin lymphoma. Of the 33 patients enrolled and treated with palbociclib (once daily for 14 days followed by 7 days off), 1 patient with testicular cancer achieved a partial response and 9 patients achieved stable disease¹⁶³. The anticipated cytostatic nature of dual CDK4 and CDK6 inhibitors resulted in prolonged stable disease duration of 18, 22 and 24 months in 3 men with growing teratoma syndrome with inoperable tumours¹⁶⁴. In patients with relapsed mantle cell lymphoma (which exhibits frequent cyclin D1 amplification) receiving palbociclib monotherapy, 5 of the 17 patients had >1 year progression-free survival, with 1 complete response and 2 partial responses¹⁶⁵. A Phase II study in liposarcoma, a disease with frequent CDK4 amplification, also reported favourable progression-free rates in patients with CDK4 amplification and RB expression^165,166. The recently reported Phase I study with LEE011 monotherapy in patients with advanced solid tumours demonstrated that LEE011 was well tolerated, with 2 confirmed partial responses and 40% of patients with stable disease¹⁶⁷. Similarly, abemaciclib exhibited single-agent activity associated with delayed disease progression and particularly robust activity in metastatic ER-positive breast cancer, although data are from a relatively small study with one group of patients¹⁶⁸. There are multiple ongoing Phase II trials evaluating dual CDK4 and CDK6 inhibitors as monotherapy in various tumour types (TABLE 2).

These early trials defined several key clinical hallmarks of inducing CDK4 and CDK6 inhibition in patients with cancer. Most importantly, it seems that neutropaenia is the principal dose-limiting toxicity of palbociclib and LEE011. Although neutropaenia is a common side effect of cytotoxic agents, the neutropaenia associated with palbociclib and LEE011 is distinct in that it is rapidly reversible, reflecting a cytostatic effect on neutrophil precursors in the bone marrow. Consequently, both palbociclib and LEE011 are dosed intermittently to accommodate a break for haematological recovery¹⁶⁷. Interestingly, abemaciclib exhibits more prominent gastrointestinal-associated toxicity, whereas neutropaenia is less evident, enabling continuous dosing. The reasons behind these observations and implications for future development remain unclear.

Hormone therapy combination strategies

Preclinical investigation suggested that dual CDK4 and CDK6 inhibitors positively interact with other therapeutic agents. In particular, synergy was observed when palbociclib was combined with hormone therapy in ER-positive breast cancer cell lines, although the observed effects can range from additive to synergistic depending on the model¹⁵⁵. Additionally, CDK4 and CDK6 inhibition has shown activity in multiple ER-positive breast cancer models that have acquired resistance to ER antagonists^155,169,170. Importantly, resistance to endocrine therapy is associated with the deregulation of proliferation-associated genes that are regulated by the CDK4/6–RB-E2F axis, suggesting a basis for cooperativity in the clinic^43,171. These observations triggered a series of randomized Phase II studies that have consequently transformed the CDK field.

The PALOMA-1 Phase II clinical trial randomized 165 women with advanced ER-positive breast cancer into two treatment groups: the aromatase inhibitor letrozole versus letrozole plus palbociclib. Data from this trial showed that the combination of palbociclib plus letrozole elicited significant improvement in median progression-free survival compared with letrozole alone (20.2 months compared with 10.2 months; hazard ratio (HR) = 0.488 (95% CI: 0.319–0.748) and one-sided p = 0.0004). The overall survival analysis after 61 deaths demonstrated a trend in favour of the letrozole plus palbociclib combination (37.5 months versus 33.3 months, respectively; HR = 0.813; p = 0.2105)^172-174. Consequently, palbociclib received Breakthrough Therapy designation by the US Food and Drug Administration in April 2013. ER-positive breast cancer is characterized by frequent dysregulation of CDK4 and CDK6 activity due to the overexpression and amplification of the gene encoding cyclin D1 (CCND1). Those cancers with amplification of CCND1 seemed to derive no greater benefit from palbociclib, an observation that is likely to reflect the central nature of cyclin D1 in promoting ER-positive breast cancer proliferation regardless of whether high CCND1 expression was due to amplification or other mechanisms.

As a follow-up to these findings, multiple Phase II and III trials of combination therapies were initiated. The combination therapies tested each include a dual CDK4 and CDK6 inhibitor (abemaciclib, LEE011 or palbociclib) and a hormone therapy (letrozole, anastrazole or fulvestrant) (TABLE 2).

CDK4 and CDK6 inhibitor biomarker strategies

Preclinical work has defined a series of biomarkers that may be used in the selection of tumours that may respond to dual CDK4 and CDK6 inhibitors. The most conservative and best supported of these markers is the direct assessment of the CDK4–RB–p16^INK4A pathway. Data from multiple groups have demonstrated that RB is necessary for the arrest induced by CDK4 and CDK6 inhibition^{152-157,175,176}, and loss of RB is therefore a marker of resistance to CDK4 and CDK6 inhibition. Loss of RB results in supra-physiological expression of p16^INK4A, which may also be a biomarker of resistance. For example, high levels of p16^INK4A are identified in malignancies caused by human papilloma virus, a virus that can inactivate RB in cervical and in head and neck cancers^29,177. Whether dual assessment of RB loss and induction of p16^INK4A expression is better than either biomarker alone is uncertain. There remains a considerable need to identify other predictive markers for tumours with CDK4 and/or CDK6 dependence or ‘addiction’ that can be selectively targeted. Examples of other predictive markers could be the amplification of cyclin D1 or CDK4 and CDK6, loss of p16^INK4A or other genetic alterations leading to the deregulation of CDK4 or CDK6 activity. This concept has been incorporated into ‘basket trial’ designs with palbociclib (LUNG-MAP) and LEE011 (SIGNATURE), in which patients with specific signature mutations that would be expected to deregulate CDK4 and CDK6 activity can be enrolled for treatment with these inhibitors.

The future of CDK4 and CDK6 inhibitors

Preclinical and clinical data suggest that dual CDK4 and CDK6 inhibitors could have broad-ranging efficacy in many cancer indications. Several questions have arisen from the published work regarding understanding which diseases would benefit the most from dual CDK4 and CDK6 inhibitors.

One important question is how to determine whether an RB-proficient tumour benefits from CDK4 and CDK6 inhibition. In some tumour types, CDK4 and CDK6 inhibition has a surprisingly modest clinical effect despite molecular alterations indicating a robust response^163,178. Some cancer types seem either to be innately resistant or to acquire rapid resistance to the effects of CDK4 and CDK6 inhibition. For example, CDK4 and CDK6 suppression seems to have little clinical effect in colorectal cancer, triple-negative breast cancer and melanomas. Therefore, such tumours would not benefit from monotherapy in the absence of potent combination strategies and robust predictive markers. In these cancers, other CDKs, particularly CDK2, are likely to compensate for selective CDK4 and CDK6 inhibition. However, the factors that determine whether other CDKs can compensate for CDK4 and CDK6 inhibition are poorly understood.

Another question is how to optimize schedules for treatment. CDK4 and CDK6 inhibition antagonizes the effect of cytotoxic chemotherapy and radiotherapy because the vast majority of cytotoxic chemotherapies require cells to be cycling^179-181. Despite this requirement, several studies are underway to evaluate scheduling with CDK4 and CDK6 inhibition, following the concept that release from CDK4 and CDK6 inhibition may synchronize cells and thereby sensitize cancer cells to a subsequent cytotoxic treatment, or may prevent ongoing proliferation or re-population of cancer cells between cytotoxic administrations¹⁸². Although such scheduling approaches have been shown to be potentially beneficial in preclinical models, translating this to the clinic, where proliferation rates of tumours are highly variable, will be challenging. A variation of this principle in patients with a known RB-inactivated cancer is the potential use of CDK4 and CDK6 inhibition to protect normal cells from the effect of chemotherapy or radiotherapy while rendering the tumour vulnerable¹⁷⁹.

Finally, it is important to determine ideal combinations. Considerable interest lies in the potential for combining CDK4 and CDK6 inhibitors with other targeted therapies. Substantial preclinical work has demonstrated that CDK4 and CDK6 inhibition may be synergistic with MEK inhibition in NRAS-positive melanoma¹⁸³. Similarly, in pancreatic cancer, CDK4 and CDK6 inhibition is synergistic with inhibitors of insulin-like growth factor 1 receptor and mTOR^184,185. Furthermore, in cancers with mutated phosphatidylinositol-4,5-bisphosphate 3-kinase (PI3K), catalytic subunit-α (PIK3CA), PI3K inhibitors may synergize with CDK4 and CDK6 inhibitors¹⁶⁰. Studies with LEE011 have incorporated this CDK4 and CDK6 inhibitor into the triplet combinations with a PIK3CA inhibitor (BYL719) and letrozole (ClinicalTrials.gov identifier: NCT01872260), and the mTOR inhibitor everolimus with exemestane in ER-positive breast cancers (NCT01857193). Recently released data from a study in melanoma suggest that such rational combinations are effective and that CDK4 and CDK6 inhibition could represent a preferred combination agent with a range of targeted agents¹⁸⁶.