The other side of the coin

Abstract

Although cancer-regulatory genes are dichotomized to oncogenes and tumor-suppressor gene s, in reality they can be oncogenic in one situation but tumor-suppressive in another. This dual-function nature, which sometimes hampers our understanding of tumor biology, has several manifestations: (1) Most canonically defined genes have multiple mRNAs, regulatory RNAs, protein isoforms, and posttranslational modifications; (2) Genes may interact at different levels, such as by forming chimeric RNAs or by forming different protein complexes; (3) Increased levels of tumor-suppressive genes in normal cells drive proliferation of cancer progenitor cells in the same organ or tissue by imposing compensatory proliferation pressure, which presents the dual-function nature as a cell–cell interaction. All these manifestations of dual functions can find examples in the genes along the CCND–CDK4/6–RB axis. The dual-function nature also underlies the heterogeneity of cancer cells. Gene-targeting chemotherapies, including that targets CDK4, are effective to some cancer cells but in the meantime may promote growth or progression of some others in the same patient. Redefining “gene” by considering each mRNA, regulatory RNA, protein isoform, and posttranslational modification from the same genomic locus as a “gene” may help in better understanding tumor biology and better selecting targets for different sub-populations of cancer cells in individual patients for personalized therapy.

Cancer regulatory genes are often dichotomized to oncogenes and tumor-suppressor gene s. Those that promote cell proliferation or migration and/or inhibit differentiation or cell death are considered oncogenes, which may be amplified in tumor cells with gain of function. Conversely, those that inhibit cell proliferation or migration and/or promote differentiation or cell death are regarded as tumor-suppressor gene s, which may be deleted from tumor cells with loss of function. Despite being oversimplified, this dichotomy has greatly helped in delineating molecular mechanisms behind tumor biology. However, in reality, virtually all cancer regulatory genes manifest oncogenic features in one situation while showing tumor-suppressive properties in another.¹ To provide examples, in this primer we review the dual functions of cyclin-dependent kinase 4 (CDK4) as well as its regulatory CCNDs (cyclins) and other proteins and analyze the meaning of the dual functions from the evolutionary point of view, as assigning multiple functions to one single gene can make the genome more efficient and thus smaller. Since CDK4 is a key target for cancer chemotherapies and some of its inhibitors such as PD0332991 are currently on clinical trials, the clinical relevance of this dual function nature of genes to the CDK4 inhibitors in particular and other targeted chemotherapies in general are reviewed and discussed as well.

There Is No Such Thing Called Tumor-suppressor gene s or Oncogenes

By the current estimation, the human genome encompasses about 3.0–3.2 billion bases but only slightly over 20 000 protein coding genes,^2,3 when those open reading frames smaller than 100 amino acids are regarded as non-coding and are thus excluded.⁴ Human exome, the part of the genome formed by all (roughly 180 000) exons that encode amino acids, constitutes about 1% of the whole genome, or about 30–32 megabases of DNA.^2,3,5-7 These figures together suggest that the majority of the human genome may not function via canonically defined genes, or, to keep the notion that the human genome functions via the “gene” that is the basic physical and functional unit of heredity, we need to redefine the “gene”, as pointed out by Gerstein et al.⁸

Most genes can be expressed to multiple RNA transcripts by using different transcription initiation sites or termination sites, and some of the transcripts are pre-mRNAs, while others may be precursors of regulatory RNAs. By alternative splicing, one pre-mRNA may produce not only different mature mRNAs but also different regulatory RNAs, such as long non-coding RNAs, microRNAs, and small interfering RNAs. One mature mRNA may yield different protein isoforms by using different translation start codons or stop codons. Each of the protein isoforms may undergo many posttranslational modifications, such as phosphorylation, methylation, and even splicing.^9,10

For many genes, each of their pre-mRNAs, mRNAs, regulatory RNAs, protein isoforms, and protein modifications may function differently or even oppositely. For instance, the long mRNA form of the Bcl-x gene, i.e., Bcl-xL, encodes an oncoprotein, whereas its short mRNA form, i.e., Bcl-xS, encodes a tumor suppressor.¹¹ In the FANCL gene, on the other hand, it is the long mRNA form that is tumor-suppressive, while the short form is oncogenic.¹² The titin gene expressed in muscle encodes the largest (4.2 MDa) and the third most abundant protein in the human, with over one million mRNA variants derived from different alternatives of splicing of its 363 exons.^13,14 At the DNA level, many genes can change their functions by mutating their sequence, as exemplified by p53, for which over 30 000 mutation types have been identified.¹⁵ While the wild-type (wt) p53 often manifests tumor-suppressive features, some of its normal mRNA variants¹⁶ or DNA mutants^17,18 are oncogenic. Actually, p53 was originally taken to be an oncogene, because its protein level is high in many cancers. Gene mutations can also change any of the above-described mechanisms for gene product multiplicity; for instance, mutations at splice sites can change the splice pattern.¹⁹ More complexly, nucleotide change can also occur at the mRNA level by RNA editing that, unlike DNA mutation, provides cells with a temporary and reversible mechanism for altering protein sequence.

In addition to the gene product multiplicity, interactions among genes also diversify genes’ functions at yet another level of complexity. The best-known manner of interaction is the formation of complexes among different proteins, as to be detailed later in this primer for the complexes containing CDK4. Chromosomal translocation, which results in a fusion gene and ensuing chimeric RNA that may or may not encode a fusion protein, can be regarded as another manifestation of gene–gene interaction. One less-discussed manner is that DNA mutation or RNA editing of a regulatory gene may affect products of its target gene via any of the above-described mechanisms. For example, mutation of some splisosome genes can change titin protein isoforms.²⁰ Formation of chimeric RNA at the RNA level,²¹ such as via trans-splicing, can also be regarded as a manner of gene–gene interaction. Actually, a new definition of “gene” is needed⁸ to cover such observations, such as that the 7 Msh4 mRNA variants in mouse involve sequences from 4 different chromosomes,²² and that there may be many chimeric RNAs in human formed between nuclear RNA and mitochondrial RNA.²³

Summarizing the above information with 2 examples of the extremes, i.e., that the titin gene is expressed to over a million mRNA variants, and that p53 has over 30 000 mutation types leads us to a conclusion that the whole genetic repertoire consists of not just 20 000 genes but 20 000 groups of genes, if each of the pre-mRNAs, mature mRNAs, regulatory RNAs, protein isoforms, and protein modifications is considered one individual “gene” with somewhat different functions. The real number of gene groups is actually much larger, because both strands of the DNA double helix in most loci, including unannotated ones, are transcribed.^24-26 Redefining “gene” in this way explains better the so-diversified human biology and related sociology relative to the canonical gene definition.

Each gene is just a tool used by the cell to elicit some specific function, and whether it should be oncogenic or tumor-suppressive is more up to the cell in a particular situation than to the gene per se, similar to a weapon that can be used by a person to defend or to attack. To use an analogy, an actor may play the hero in some movies but the villain in some others, depending on the script. Some actors almost always play the hero, making us forget that they can also do well as a negative role if required. Similarly, p53 usually acts as a tumor-suppressor gene , but, if the cell needs it, it can act as an oncogene as well, as described above. In conclusion, each gene is just an actor or actress whose role is assigned by the host cell to deal with a particular situation.

There is probably evolutionarily determined programming for which (canonically defined) genes use permanent DNA mutations or flexible RNA editing and for which other genes use the aforementioned means such as alternative splicing or protein modification to respond to environmental changes. Obviously, p53 is allowed to more often change its DNA sequence, whereas titin has more mRNA variants as alternatives. Actually, the United States adapts this elegant way to prevent any chance of administrative chaos by listing 16 persons on the presidential line of succession in the Constitution, with the vice president, the speaker of the House, the president pro tempore of the Senate, and the Secretary of State at the top of the list. Similarly, human cells reserve over 30 000 mutation types of p53 as backups to deal with different situations. Since about 12% single-nucleotide polymorphisms in the human genome are actually harmful,²⁷ some mutations may actually be more beneficial than the wt to the host.^27-29 An intriguing question is whether cells can convert some mutations back to wt if the situation returns to normal, since such “back mutation” or “reverse mutation” of inherited mutations has been well documented.^30-33 It deserves attention whether during sporadic carcinogenesis there is a short period of window for such reverse mutation or secondary mutations, both of which have been observed for the BRCA1 and BRCA2 genes after chemotherapy of some cancers,^34-36 leading to spontaneous abortion of the carcinogenesis. Some mutations of a gene occur more frequently than the others and are thus called “hot spots”, which are reminiscent of those at the top of the presidential line of succession as the backups for the “wild-type” (current) president of the USA. p53 knockout animals or cell lines are so often used in cancer research with a rationale that “p53 functions are often lost in cancer cells”. Unfortunately, this rationale is much closer to wrong than to right and has little human relevance, because cancer cells rarely lose the entire 19148-bp human p53 gene. Instead, the vast majority of cancer cells manifest one or some of the 30 000 alterations (including partial deletions) that in many cases lose some tumor-suppressive functions but gain some oncogenecity.

In a Tissue, Tumor Suppressors in Normal Cells Drive Proliferation of Pre-Malignant Cells

In an in vivo situation, cancer progenitor cells, e.g., initiated cells that are collectively dubbed hereinafter as pre-malignant cells regardless of how they have emerged, need to proliferate continuously in order to evolve toward a malignant state. Many, probably most, environmental tumor-promoting agents such as chemical carcinogens and physical (e.g., radiation) or biological (e.g., virus) factors that drive their proliferation are actually inhibitors of cell proliferation, as detailed before.³⁷ When insulted by these tumor-promoting agents, called tumor promoters hereafter, normal cells stop replicating, which is referred to as “mitoinhibition”, to avoid DNA damage and avoid passing damaged DNA to their progenies. This mitoinhibition is likely achieved by elevated expression of tumor-suppressor genes such as p53, probably in association with decreased expression of oncogenes (Fig. 1). However, the organ or tissue still has to regenerate a certain number of cells to balance the routine cell loss. This need for regeneration is actually enhanced by tumor promoters, not only because decreased expression of oncogenes often weakens cell survival, but more also because elevated expression of tumor-suppressor gene s often induces cell death, which is commonly considered by peers as apoptosis but by us as “stress-induced cell death (SICD)”, a programmed cellular death mechanism different from physiological apoptosis.^38,39 “Fortunately”, those pre-malignant cells that already exist in the organ or tissue are resistant to the effects of tumor promoters, i.e., being less mitoinhibited and having less SICD potential, probably due to inactivation of some critical tumor-suppressor gene s and/or activation of some critical oncogenes (Fig. 1). These cells with the “resistant phenotype”⁴⁰ are the only ones capable of proliferating, thus replicate robustly to compensate for the cell loss of the organ or tissue while evolving to a malignant state. Therefore, activation of tumor-suppressor genes in the normal cells provides an impetus for pre-malignant cells to proliferate. Because of this interaction between pre-malignant cells and their surrounding normal cells, tumor-suppressor genes are as important as oncogenes in driving carcinogenesis in vivo, which is another manifestation of genes’ dual-function nature at the tissue level. Knockout of a tumor-suppressor gene from the whole tissue or organ will weaken the mitoinhibition and decrease the SICD potential of the normal cells, thus decreasing the compensatory proliferation pressure onto the pre-malignant cells. This major but less often discussed mechanism of carcinogenesis is one of the major reasons why animals with a tumor-suppressor gene deleted, such as p53^−/− mice, do not develop cancer as efficiently as some peers expected and do not much resemble the human situation, since no tumor-suppressor gene is deleted from all cells of a human organ or tissue as a genetic basis for a sporadic cancer (germline mutation triggered carcinogenesis starts embryonically and thus is different). If one day it becomes applicable to knockout one gene from only one single cell of an organ or tissue with its surrounding cells intact, this “mitoinhibition-resistant phenotype” principle can be further tested, although, as reviewed before,³⁷ it has been tested intensively by Farber’s team^41-47 and many other researchers^48-50 since it was proposed by Haddow in 1938.⁴⁰

Figure 1. Depiction of how tumor-suppressor genes (TSG) promote carcinogenesis via the “mitoinhibition-resistant phenotype” principle, with liver carcinogenesis as an example. There already is a cell in the liver that has been initiated to undergo carcinogenic process, due to such as spontaneous or chemical-induced mutation of some critical gene(s). This so-called “initiated cell” may show decreased expression of TSG and/or increased expression of oncogenes (ONG). If the liver encounters a chronic insult from some tumor-promoting agent (promoter) such as a long-term exposure to a chemical carcinogen, normal hepatocytes will react by increasing their TSG expression and likely also decreasing their ONG expression, so as to arrest cell growth to prevent DNA damage or to repair damaged DNA. If the DNA damage is irreparable, the hepatocyte will die of SICD, which will further enhance the demand for regeneration. Partly because of its low TSG levels and/or high ONG levels, the initiated cell is resistant to the promoter, i.e., being much less growth-arrested and having a less potential to SICD, and thus proliferates robustly to compensate for the cell loss of the liver, eventually evolving to a cancer (Ca). Therefore, sporadic cancer only occurs in those organs or tissues that have a routine cell turnover, but not in those that can no longer regenerate such as neurons or heart muscle.

Many genetically engineered animal models of carcinogenesis, usually with 2 or more genes manipulated, such as in ras-and-myc double transgenic mice, can produce solid tumors at 100% incidence, i.e., in all animals.^51,52 However, although in these models all cells in the target organ or tissue bear the same genotype, usually only one to several cancer masses appear until the animal capitulates.^51,52 Since there are millions of target cells in the same organ or tissue, the carcinogenesis efficiency is less than one of a million. In other words, there are few, if any, genetically engineered animal models that can convert even a tiny portion of individual target cells into individual cancer masses, although at the animal level the penetrance is 100%. This situation, which to our knowledge has not been mentioned and addressed in the literature, has been puzzling us for a long time and may not be explained until we are able to manipulate genes in single cells of the target organ.

For Example, the CCND–CDK4/6–RB Axis Also Plays Tumor-Suppressive Roles

CDK4 or its close sibling CDK6 binds to 1 of the 3 CCND proteins, mainly CCND1, to form a holoenzyme, and p27 or p21 protein facilitates the CDK4/6-CCND assembly.⁵³ The holoenzymes phosphorylate RB1 or its sibling p107 or p130. Once phosphorylated, these RB proteins are inactivated and release E2F1 or other E2F proteins to regulate transcription of many proliferation regulatory genes, usually leading to acceleration of G₁ progression of the cell cycle. On the other hand, p16 and related proteins can bind to CDK4 and inhibit its kinase activity, thus collectively coined as inhibitors of CDK4 (INK4), although they also bind to CDK6. Many types of cancer manifest increased expression of CCND1 or CDK4 with high frequencies of gene amplification, which makes this CCND–CDK4/6–RB–E2F axis a key promoter of cancer formation and progression. Since the oncogenic property of this axis has been reviewed extensively, herein we just focus on its tumor-suppressive aspect, which, to our knowledge, has so far not yet been reviewed comprehensively.

It is often forgotten, and is thus worth mentioning, that CCND-CDK4/6 can drive cell proliferation but cannot render cells competent for proliferation, meaning that once a cell has fully differentiated and lost its replication capacity, CCND-CDK4/6 cannot make it proliferating. CDK4 is often used in combination with other genes to immortalize cells, as reviewed before,⁵⁴ but likely the cells need to be relatively young. Aged cells that can only undergo a few rounds of replication are probably hard, if not impossible, to immortalize by a CDK4-containing regimen.

As reviewed extensively before,⁵⁵ nearly 2 decades ago ectopic expression of CCND1 had already been shown to inhibit cell proliferation and induce apoptosis or senescence of several cell types^56-61 in addition to driving cell proliferation, although such ectopic expression usually results in only modest accumulation of the CCND1 protein.⁶² Transformation of human fibroblasts by SV40 coincides with a downregulation of CCND1,^63-65 whereas many cancer cell lines manifest a decreased CCND1 protein level compared with normal diploid fibroblasts,^64-69 although fibroblasts are not vigorous controls for epithelial-originated cancers. In transgenic mice, c-myc induces mammary tumors while it inhibits CCND1.⁷⁰ In the pioneer study by Land et al. in 1983, colonies of ras-transformed rat primary embryonic fibroblasts appear in the dish at very early time point, but the cells have not yet been immortalized and eventually die; however, co-transfection with c-myc can immortalize the ras-transformed cells.⁷¹ Since ras^72,73 induces whereas c-myc^37,54,55 inhibits CCND1 expression, decrease in the CCND1 may be a mechanism that prevents the death of ras-transformed cells and thus be a mechanism for the ras-myc collaboration in transformation.^54,55

In a cohort of breast cancer cases, low proliferation was observed in tumors with low p16, with high CCND1 but normal or high p16 expression, or without cell cycle defects.⁷⁴ A study of another cohort concludes that CCND1 upregulation is not associated with increased proliferative activity in the tumors, although loss of Rb1 expression is significantly associated with an increased proliferation of the tumor cells regardless of CCND1 abnormalities.⁷⁵ Oyama et al. also show that the vast majority of invasive lobular breast carcinomas manifest overexpression of CCND1 protein, but most CCND1-positive cells do not stain for Ki67 (a marker for proliferating cells), suggesting that CCND1 may contribute to the tumor progression via a mechanism irrelevant to its promotion of cell proliferation.⁷⁶ Indeed, CCND1 expression in human breast cancer does not seem to be clustered into the group of proliferation-related genes.⁷⁷ However, while the above-described studies, using cultured cells, animals, and human samples, seem to suggest a lower CCND1 level in transformed or malignant cells, it may sometimes be a misperception, because this “feature” may simply reflect a changed cell cycle distribution with decreased G₁ and increased S fractions. This is because in cycling cells CCND1 protein level starts to elevate in the G₂ phase of the cell cycle and remains high in the M phase and the G₁ phase of the next cycle, but it levels off in S phase to allow DNA synthesis.^72,73 Therefore, a smaller G₁ fraction and a larger S fraction seen in transformed or malignant cells that grow faster, relative to the normal counterparts, will collectively show a decreased CCND1 level (Fig. 2). This decrease may be discerned on immunoblots, but immunohistochemical staining intensity may still be stronger in individual malignant cells than in normal cells, according to our experience. Some proliferation markers such as Ki67 and proliferation cell nuclear antigen (PCNA) stain cells at the S and G₂–M phases and thus may or may not statistically differ from CCND1 staining, depending on how large the G₂–M portion is, which is usually the smallest wherein both CCND1 and proliferation markers are highly expressed (Fig. 2).

Figure 2. Illustration of the cell cycle and CCND1 relationship. In slowly growing tissues or cultures, G₁-phase cells constitute the largest portion, while the S-phase portion is much smaller, whereas the situation is the opposite in fast-growing tissues or cultures. Cells in the G₂/M phases usually constitute the smallest portion, although they may still be larger in the fast-growing than in the slow-growing tissues or cultures. CCND1 level is higher in G₂/M and G₁ but lower in S, and thus cannot sensitively reflect the proliferative rate.

CDK4, CDK6, CCND2, and CCND3 may manifest tumor-suppressive features as well in some situations. Transfection with a CDK4R24C or CDK6R31C mutant that cannot be inhibited by p16 may lead to cell death.⁷⁸ Even transfection with the wt CDK6 can cause cell death.⁷⁹ Overexpression of CDK6 and CCND1 causes apoptosis of chondrocytes without enhancing proliferation.⁸⁰ Ectopic expression of CDK6 inhibits cell proliferation, which is probably elicited via the p53 and RB pathways as p53 and p130 are increased.^79,81,82 Actually, overexpression of CDK6 or CCND3 in transgenic mice inhibits skin tumor development and, in the case of CCND3, the CCND3–CDK4 kinase activity may be involved.^83,84 CCND3 is required for differentiation of several tissues.^85-87 CCND2, including its splice variant CycD2SV,⁸⁸ is actually regarded by many peers as a tumor suppressor, because its expression level is decreased in many malignancies,^89-93 whereas restoration of its expression prevents prostate carcinogenesis.⁹⁴

p16 is commonly regarded as a tumor suppressor, but it sometimes manifests oncogenic properties. Its mRNA and protein levels are unexpectedly higher in basal cell carcinomas than in normal human skin.⁹⁵ A subset of breast, ovarian, and lung cancers manifests high levels of p16, which is usually associated with Rb1 mutation.^96-100 p16 is shown to be required for the survival of some cervical carcinoma cell lines in a CDK4/6-dependent manner.⁷⁸ Free p16 protein in the cytoplasm has been shown to promote migration of liver cancer cells in culture, which is somewhat opposite to the functions of the cytoplasmic CCND1 and CDK4.¹⁰¹

The canonical tumor suppressor RB1 sometimes helps cell proliferation or survival^102,103 and is overexpressed or amplified in some cancers,¹⁰⁴ whereas its loss may induce cell death.^103,105,106 More surprisingly, transgenic mice expressing a constitutively active Rb1 gene actually develop mammary gland cancer, which is thought to occur via the aforementioned “mitoinhibition” mechanism, i.e., the growth suppression of active RB1 protein imposes a selection pressure in favor of transforming mutations that accelerate cell proliferation.¹⁰⁷ Since RB1 regulates a plethora of genes that inhibit cell proliferation or promote apoptosis, besides those cell cycle-promoting genes such as CCNE1, CCNA2, and PCNA,¹⁰⁸ it remains possible that RB1 functions differently in normal cells and in cancer progenitor cells. RB1 also plays roles in replication of DNA, repair of DNA damage, and G₂/M progression of the cell cycle.¹⁰⁸ The Rb1 gene produces not only the wt, but also several N-terminally truncated proteins by alternative splicing,^109,110 by using downstream ATG start codons, or by caspase-cleavage.^111-117 There are at least 932 mutations identified for the Rb1,¹¹⁸ some of which may result in not only oncogenic proteins but also smaller protein isoforms.¹¹⁹ These smaller protein isoforms have different abilities in growth inhibition, with some N-terminal truncated ones being more potent, partly because the C-terminus contains a docking sequence that cannot be inactivated by CCND-CDK4/6.¹²⁰ However, chimerism of the wt with some smaller isoforms manifests decreased tumor suppression.¹²¹ Whether different mechanisms for generation of different RB1 protein isoforms are differentially used in cancer and non-cancer cells remains elusive.

Being the major target of RB1, E2F1 not only activates transcription of genes that drive cell proliferation but also induces cell apoptosis in some situations,^122-124 such as when RB1 is lacking.¹²⁵ E2F1 knockout mice develop spontaneous tumors in a number of tissues, demonstrating its tumor-suppressive property in vivo in some situations.¹²⁶

CCND1 Not Only Drives G₁ Progression, but Also Arrests It, Allowing DNA Repair

The functional duality of CCND1 may be attributed more to the cellular context than to the CCND1, per se, but until now it still baffles us as to what cellular properties direct the high-CCND1 cells to replicate, arrest proliferation, undergo apoptosis, or enter into senescence. Serving as a cellular sensor for extracellular growth stimuli, CCND1 is activated by Ras, ERK, and many growth stimuli-induced molecules at the G₂ phase and remains high until the next G₁ phase.^72,73 At the G₁–S boundary, CCND1 protein is quickly degraded to allow DNA synthesis (Fig. 2). Phosphorylation of CCND1 protein at T286 by glycogen synthase kinase 3β (GSK-3β) is required for CCND1 ubiquitination by 26S proteasome, and this degradation is enhanced by its binding to CDK4,^127,128 but T286 phosphorylation does not promote its disassociation from CDK4.¹²⁹ CCND1 that is unbound to CDK4 is more rapidly degraded in some,¹³⁰ but probably not all,¹²⁹ cells via ubiquitination without phosphorylation at T286.¹²⁹

If a cell encounters a genomic insult such as radiation that causes DNA damage, the cell arrests G₁ progression in part by rapidly degrading the CCND1 protein,^131,132 or, if it is too late for the cell to hold itself at G₁, it arrests S progression by sustaining a high CCND1 level.¹³³ Both G₁ and S arrests allow the cell to repair its DNA and to prevent passing the genetic change to the daughter cells. If the DNA damage is irreparable, the sustained CCND1 protein at the S phase will instead trigger more DNA damages and ensuing apoptosis^73,134 that is considered by us as SICD.^38,39 Although CCND1 is bound by CDK4 and p21 at the G₁, genomic stress only degrades CCND1, but not CDK4 and p21. In consequence, CDK4 is released to bind another CCND1 molecule and brings it for degradation while p21 is released to bind CDK2, thus sequestering CDK2 activity and further arresting G₁ progression.^131,132 It has been shown that CCND1 and CCNA2 undergo cleavage by caspases during irradiation-induced apoptosis of Xenopus embryo, and the cleaved CCND1, alone or in complex with CDK4/6, is much more potent in binding to p27, leading to reduced phosphorylation of RB1.¹³⁵

Irradiation-induced apoptosis has been shown to be more evident in CCND1 knockout (−/−) mouse fibroblasts (MEFs) than in the wt counterparts, suggesting that CCND1 confers protection against irradiation.¹³⁶ However, ectopic expression of CCND1 in several premalignant and malignant cell lines of breast origin also enhances irradiation-induced apoptosis.^137,138 This incongruity may in part be related to the method of irradiation, as Shimura et al. show that single irradiation downregulates CCND1 protein level, but fractionated irradiation causes CCND1 accumulation via DNA-PK/AKT-mediated inhibition of its proteolysis.¹³³ Chronic irradiation is thought to result in cytoplasmic accumulation of CCND1 protein, wherein it binds and thus sequesters Bax, leading to inhibition of mitochondrial-mediated cell death.¹³⁹ Consonantly, CCND1 overexpression is shown to be associated with poor prognosis in oral and head and neck cancers after radiotherapy or concurrent chemoradiotherapy.^140,141 The persistently high level of CCND1 during the S phase inhibits DNA replication by preventing replication fork progression, which will, in turn, trigger double-strand breaks.¹³³ The cell will then remove the aberrant replication fork and reconstruct the fork to resume DNA replication.¹³³ It is only CCND1 protein, but not the CCND1–CDK4 complex, that binds to the fork,¹³³ meaning that this function is irrelevant to CDK4.^142-144 Hence, the role of CCND1 varies among different cell types,¹⁴⁵ varies between acute and chronic irradiations,¹³³ and has CDK4-dependent^146,147 or -independent^133,148 mechanisms. The paradoxical roles of CCND1 in driving G₁ progression on the one hand, and in promoting DNA repair on the other, again shows its functional duality.

Association with Different Proteins Diversifies CCND, CDK4/6, or CCND-CDK4/6 Functions

At least 132 proteins can bind to CCND1 in breast cancer cells,¹⁴³ some of which bind to CCND1 in a way independent of CDK4, such as the DNA repair proteins RAD51, BRCA1, BRCA2, PCNA, and replication factor C.^143,144 BRCA2 brings CCND1 to damaged chromosomal sites, where CCND1 recruits RAD51 to perform homologous recombination (but not other types of DNA repair).^142-144 Another group of CCND1-binding proteins, which may be mechanistically related to its growth promotion,¹⁴⁹ belongs to transcription factors,¹⁵⁰ such as Sp1,^151,152 DMP1,¹⁴⁹ as well as steroid hormone and thyroid hormone receptors, as reviewed previously.⁵⁵ Interestingly, CCND1 binds to and activates estrogen receptor α,¹⁵³ but it binds to and inhibits androgen receptor.^154,155 Moreover, many CCND1-regulated genes encode molecular chaperones.^156,157

p16 and probably also other INK4 members form mainly binary INK4–CDK4/6 complexes. INK4–CDK4/6–CCND ternary complexes may also be formed at a lesser abundance^158,159 and probably mainly in senescent cells,¹⁶⁰ but inhibition of CDK4 kinase activity by p16 is not affected by whether CDK4 is alone or is bound to a CCND.¹⁶¹ Many proteins that contain ankyrin-repeat domain, such as IkBalpha, can bind to CDK4 as well.¹⁶² Gankyrin that contains 7 ankyrin repeats is known to compete with p16 in binding to CDK4, but it does not inhibit CDK4 activity; actually, this difference makes gankyrin an oncoprotein but makes p16 a tumor suppressor.^163,164 Survivin competes with p16 or p21 for binding to the CDK4¹⁶⁵ and brings CDK4 into the nucleus, which is a mechanism for its promotion of cell cycle entry and cell survival.^166,167 Cdc37 is a molecular chaperone important for the stability and activity of several protein kinases; like Hsp90, Cdc37 binds to the N terminus of CDK4 by competing with p16.¹⁶⁸ The portion of cdc37-bound CDK4 is largely devoid of CCND1, suggesting that this complex functions to prepare CDK4 for CCND1 interaction.^169,170 In addition, some viral proteins, such as the HTLV-1 Tax protein,¹⁷¹ can bind to CDK4 as well. Unlike CDK4 and other CDKs, CDK6 can bind to androgen receptor and activates its transcriptional activity, which does not require its kinase activity and is independent of CCND1.¹⁷² Actually, binding of CCND1 to CDK6 inhibits the transactivation of the androgen receptor.¹⁷²

Although p27 and p21 are similar in their CDK binding regions, only p27, but not p21, can bind to CDK4 alone, but in mid-G₁ p21 assembles CCND1–CDK4 complex by replacing p27 in binding to CDK4.¹⁶⁰ Binding of p27 to CCND1-CDK4 may or may not inhibit CDK4 kinase activity, depending on whether the Y88 and Y89 of p27 are phosphorylated, in part because p27 can also bind to CCND1, but not just CDK4.^173,174

It goes without saying that CCND–CDK4/6 holoenzymes bind to their substrates as well. In addition to RB1, p107, and p130, 68 proteins have been identified as CCND1–CDK4/6 substrates,¹⁷⁵ including FOXM1, Smad3, Cdt1, MARCKS, and PRMT5.^53,176 CCND1-CDK4 has also been reported to phosphorylate microtubule-associated protein Tau in neurons to promote neurite extension.¹⁷⁷ Filamin A can bind to CDK4 and may be a substrate of the CCND1–CDK4 complex as well, as regulation of its phosphorylation by CCND1 promotes cell migration and invasion.¹⁷⁸ Moreover, p220NPAT may be a CCND2–CDK4 substrate.¹⁷⁹ Thus, a gene has dual functions not only because it can produce multiple products, but also because one protein can have different partners and, if it is an enzyme like CDK4, different substrates.

Some Tumor-Suppressive Roles of CDK4 May Be Independent of CCND1 and Occur in the S and G₂/M phases

A question that has been baffling us for a long time is why CDK4 has mainly been shown to function at the G₁, while its main partner, CCND1, has already, as aforementioned, been elevated in the G₂-M of the previous cycle. Our cloning of a CDK4 mRNA variant that lacks exon 2, coined as ΔE2, provides some of the answers.¹⁸⁰ Initiation of protein translation of this variant is shifted from the canonical ATG start codon in exon 2 to an in-frame ATG in exon 3, resulting in an isoform lacking an N-terminal region of 74 amino acids that harbors not only the ATP binding sequence, but also the PISTVRE domain required for binding to the p16 and CCND. Co-immunoprecipitation assays confirm that this truncated protein has indeed lost the CCND1- and RB1-binding ability. Surprisingly, however, this ΔE2 protein can still drive the S–G₂/M progression as does the wt CDK4. More surprisingly, it is as potent as its wt in enhancing serum starvation-related apoptosis of RB1-inactivated 293T and Hela cells¹⁸¹ but inhibiting the death of RB1-proficient HEK293 cells. It is also as potent as its wt in arresting the Hela, but not the HEK293, cells at the G₁ when treated with a CDK4-inhibitory compound NPCD. These results suggest that CDK4 may use some CCND- and RB-irrelevant mechanisms not only in the well-established promotion of cell proliferation, but also in the unexpected inhibition of cell cycle progression and induction of cell death, somewhat similar to CDK6, which can activate androgen receptor transcriptional activity without involving its CCND1 binding and its kinase activity.¹⁷² A study on B-cell lymphoma has also suggested, though indirectly, the possible existence of cyclin-independent functions of CDK4.^182,183 Since Warenius et al. have reported that the 251–256th hexapeptide (PRGPRP) of human CDK4 has killing effects on cancer cells,¹⁸⁴ the growth arrest and apoptotic effects of the ΔE2 and the wt may be attributed to this C-terminal region. The dual-function nature, i.e., promotion of cell growth or death, has also been discerned for CDK8¹⁸⁵ and thus is not unique to CDK4.

The above-described novel properties of CDK4, i.e.: (1) its CCND- and RB-independent functions; (2) its promotion of S and G₂–M progression; and (3) its driving of G₁ arrest and apoptosis, occur in a cell line-specific manner and have so far been observed only in some situations, wherein cells were treated with CDK4 inhibitors or were serum starved and then replenished.¹⁸⁰ Since the ΔE2 isoform still differs from the wt CDK4 in some functions, it remains possible that some functions of the wt CDK4 at the G₂-M may still involve CCND1 that has already been high at the G₂ phase.^72,73 Regardless of whether a CCND is involved, there actually is some inkling of CDK4 activity at the S to G₂/M phases.^130,186-188 Ectopic expression of the wt p16, but not its mutant, which cannot inhibit CDK4/6, lengthens the S-phase progression of MCF7 and other cancer cell lines;^189,190 although the underlying mechanism is still unclear, the involvement of CDK4/6 remains highly possible in part because CDK4 is active throughout the cell cycle.¹⁸⁸ Transient transfection with either a wt p21 or a dominant-negative CDK4 construct can suppress doxorubicin-induced apoptosis at the G₂–M phase, indicating that the CDK4 mutant can act as a tumor suppressor, just like p21, during G₂-M.¹⁹¹ CDK2 is shown to be required for p53-independent checkpoint control of the G₂–M phase.¹⁹² Inactivation of RB1 by expressing polyomavirus large tumor antigen (PyLT) or SV40 increases CCNA2 expression and activates CDK2.^193,194 Since CDK4 indirectly regulates CDK2 activity via such means as sequestration of p27 binding to CDK2, CDK4 may indirectly regulate G₂/M checkpoint as well.

The CDK4 effects in the G₂-M may be more evident in fully differentiated cells. In postmitotic neurons, cell cycle activation is essential for DNA repair, as suppression of CDK4/6 activity attenuates DNA repair.¹⁹⁵ Likely, the CCND1–CDK4 complex is mainly localized in the cytoplasm of postmitotic neurons, because its internationalization into the nucleus causes apoptosis.¹⁹⁵ Disruption of RB1 accelerates the G₂–M progression in the presence of DNA damage by elevating the expression of a set of mitotic regulatory genes.¹⁹⁶ Conditional deletion of the Rb1 gene reactivates cell cycle-regulating genes in skeletal muscle, indicating that an RB1-independent mechanism preserves the postmitotic state of differentiated cells.¹⁹⁷ Adenovirus-mediated RB1 deletion in the liver of adult mice leads to DNA replication in the absence of productive mitotic condensation; the replication induced by RB1 loss is E2F-mediated and is associated with the induction of DNA damage and a non-transcriptional G₂/M checkpoint that targets the accumulation of CCNB1.¹⁹⁸ Inactivation of Rb1 by siRNA or other means can also achieve similar results.^108,199

Inhibition of CDK4 for Cancer Therapy May Also Have Weaknesses

Inhibition of CDK4 activity is considered an attractive strategy for cancer chemotherapy.^53,200-202 Many laboratory studies have indeed shown that chemical inhibitors of CDK4 kinase activity can cause cytostasis, senescence, and even death of cancer cells.^{175,200,203-212} Moreover, knockdown of CDK4 using shRNA also significantly increases radiation-induced apoptosis of malignant and nonmalignant breast cell lines without significantly altering cell cycle progression or post-irradiation DNA repair.²¹³ Several chemical inhibitors specific for CDK4/6 have been on clinical trials, including PD0332991 (Palbociclib), LY2835219, and LEE011,^200,214,215 of which the trials on PD0332991 have been promising for lymphomas and liposarcomas.^216,217 Studies with cancer cell lines and ex vivo breast^205,218-220 or prostate²²¹ cancer tissues all suggest that PD0332991 is efficient in RB-competent cases, but the RB-deficient counterparts are relatively refractory. Treatment of glioblastoma cells with PD0332991 in an xenograft mouse model also suggests that RB-proficient cells are more sensitive,²²² which makes sense, as RB-deficient cancer cells do not seem to require CCND1 for proliferation.^130,223 A single dose of 10 Gy irradiation therapy followed by 7 d of treatment with PD0332991 increased, by 19% in comparison with irradiation alone, the survival of the transgenic mice that bear brainstem glioma.²²⁴ PD0332991 has also been shown to have an additive effect on irradiation of prostate cancer cells in culture.²²¹

PD0332991 has been shown to reduce the overall irradiation toxicity without evidently compromising the therapeutic effect of the irradiation on autochthonous cancer in the mice,²²⁵ although it cannot be ruled out that some tumor cells are also protected. Co-administration of PD0332991 with carboplatin in mice bearing RB-competent breast cancer decreases the antitumor effect of the carboplatin, although the combination significantly attenuates the myelosuppression caused by carboplatin and thus is still beneficial.²²⁶ In contrast, RB-deficient breast tumors in mice are resistant to PD0332991, and concomitant administration of carboplatin does not have an additive effect.²²⁶ Treatment of RB-competent breast cancer cells in culture or in xenograft animals with PD0332991 in combination with genotoxic doxorubicin may have additive effects on cytostasis but actually antagonizes the doxorubicin’s cytotoxic effects.²¹⁹ An indolocarbazole-derived CDK4 inhibitor has been shown to sensitize MDA-MB231 but antagonize MCF7, both being breast cancer cell lines, to irradiation, whereas more evident growth inhibition and apoptosis by the inhibitor alone are discerned in MCF7 than in MDA–MB231 cells.²²⁷ These results suggest that the chemotherapeutic effect of the CDK4 inhibitor is disassociated from its addition or antagonism to irradiation, both being cell type-specific. A more disturbing observation is that when taxanes or doxorubicin induces DNA damage in triple-negative breast cancer cells, a subtype of breast cancer with poor prognosis, PD0332991 shifts the DNA repair from homologous recombination (HR) to non-homologous end joining (NHEJ).²²⁰ Since HR is error-proof whereas NHEJ is error-prone, this shift may, presumably, cause more mutations and thus may promote cancer progression toward more aggressiveness.

CCND1 can inhibit migration of several cell types including breast cancer cells.^{145,210,228,229} This tumor-suppressor function is in part attributed to its inhibition of epithelial–mesenchymal transition (EMT), which is a morphological feature for cancer progression toward metastasis.^229,230 A recent study shows that Panc-1 cells established from a primary pancreatic cancer are sensitive to PD0332991, whereas AsPC-1 cells established from ascites of a pancreatic cancer patient are resistant to it.²³¹ In AsPC-1 and another relatively resistant pancreatic cancer cell line (COLO-357), PD0332991 induces expression of those genes that promote EMT,²³¹ which not only suggests indirectly that the aforementioned inhibition of cell migration and EMT by CCND1 may involve CDK4 activity, but also implies that treatment with CDK4 inhibitors may facilitate cancer metastasis in at least breast and pancreatic cancers in the long run.

CCND1 has been shown to inhibit mitochondrial metabolism and biogenesis,^145,232 and both addition and loss of CCND1–CDK4 functions have been shown to increase mitochondrial superoxide production and decrease the cell’s lifespan, indicating that an imbalance in mitobiogenesis may lead to oxidative stress and aging.²³³ These effects on mitochondria and ensuing aging may facilitate not only the progression of the existing cancer cells, but also the development of a second malignancy.^234-236 Moreover, CDK4 inhibition may affect pancreatic β cells’ survival and insulin secretion,^234-236 and the long-term consequence of this effect to patients needs to be evaluated.

More Information Is Needed For Fully Assessing CDK4-Targeted Cancer Therapy

PD0332991 and other CDK4 inhibitory compounds do not function by lowering the CDK4 or CCND1 level, but by competing with ATP for binding to CDK4/6.^237-239 Actually, PD0332991 treatment increases the protein levels of CCND1^{211,212,219,231,240} as well as CDK4 and CDK6.²³¹ However, it remains esoteric whether the elevated CCND1 and/or CDK4/6 contribute to PD0332991-induced cytostasis, senescence, or death of cancer cells, and whether the elevation is because decreased RB phosphorylation triggers a feedback regulation, because PD0332991 stabilizes these proteins, or because of other reasons. Whether these increased proteins are in a free status or in a binary or ternary protein complex is also unknown, although the abundance of the CCND1–CDK4 complex is shown to be modestly increased in a study.²²¹ So far it is still elusive whether PD0332991-bound CDK4/6 has the same binding affinity to CCND or other partner proteins, despite the fact that the crystal structure of PD0332991–CDK6 complex has been studied.²³⁹ Persistently high level of CCND1 probably is more abnormal than that of CDK4 to the cell, since normally the CDK4 level and activity are relatively stable,¹⁸⁸ whereas the CCND1 level should be “oscillating” around the cell cycle. We need to know whether accumulated CCND1 and CDK4/6 in the presence of PD0332991 have any effect on the cells, so as to make a decision whether PD0332991 should be used with an approach that decreases the levels of these proteins. Since currently there has not yet been an approach available for clinical use that can specifically decrease CCND1 or CDK4/6 protein, some peers are considering designing peptides or compounds that block other functional, such as the substrate-binding, domains of CDK4/6 proteins.^241,242 Whether these alternatives, alone and in combination with PD0332991 or other ATP site blockers, are more effective is an intriguing question.

Treatment of the HepG2 liver cancer cells with a CDK4 inhibitor, flavopiridol, increases cellular ATP concentration up to 3-fold,¹⁶¹ whereas inhibition of CDK2-CCNA2 activity by p27 is known to be dependent on cellular ATP concentration.²⁴³ Therefore, it would be interesting to know whether CDK4 inhibitors attenuate the inhibitory effects of p27 and other endogenous CDK-inhibitory proteins by raising the ATP level. Why and how PD0332991 shifts drug-induced DNA repair from an error-proof to an error-prone mechanism²²⁰ should also be studied in more detail, as these details are prerequisite for making a correct decision on whether and how CDK4 inhibitors should be used in combination with irradiation and with genotoxic chemotherapeutic agents in cancer treatment.

Spontaneous tumors or cancers hardly regress spontaneously, but in experimental conditions, these lesions may be dependent on the inducer.^244,245 For instance, in animals, coal tar-induced tumors in the skin,²⁴⁶ estrogen-induced cancers in several organs,^247-251 as well as c-myc- and k-ras-induced lung cancer and lymphomas²⁵² may regress after withdrawal of the inducer(s). Similarly, although spontaneous senescence should be irreversible, induced senescence has frequently been shown to be reversible.^245,253-255 Blagosklonny’s group dichotomizes growth arrest to the earlier, reversible phase and the later, irreversible phase and points out that only the irreversible arrest is senescence.^{211,212,253,256,257} For those cells that are already at the end of their proliferative life span, i.e., are already at a permanent growth-arrested or senescent state, such as fully differentiated neurons and striated (both skeletal and cardiac) muscle, growth stimuli such as those acting via the MEK/MAPK pathways may still induce CCND1. For those cells that still retain a replication capacity but are growth-arrested, such as by p21 or p16 overexpression or by PD0332991, growth stimuli will induce CCND1 but eventually drive the cells to senescence if MTOR (mechanistic target of Rapamycin) is active to stimulate cellular mass growth and cellular super function.^{211,212,253,256,257} Likely in these situations, senescent cells display a super elevation of the CCND1 protein level as a biomarker for senescence.^{211,212,253,256,257} Entry into either a reversible growth arrest or a senescent state decreases the cell turnover, which is more efficient and affordable, relative to death from apoptosis, for some tissues to maintain the physiological cell number and thus physiological function.²⁵⁸ For those cells whose replication capacity is closer to the Hayflick limit, which is about 50 generations in vitro,^259,260 entry into a growth arrest status saves them replication capacity, while the high CCND1 level prevents DNA synthesis and thus prevents passing damaged DNA to the progeny. Hence, even if the cells already bear irreparable DNA damage, they are allowed to survive at an arrested or senescent state to continue their service. Therefore, growth arrest or senescence is an equally effective tumor-suppressive mechanism as is apoptosis for the organ or tissue to protect itself.²⁶¹ Nevertheless, it deserves further study in many more cell lines whether growth arrest induced by CDK4 inhibitors eventually turns to true, irreversible senescence, since senescent cells are, in general, relatively resistant to apoptosis^258,262 and thus may be refractory to other apoptosis inducers such as radiation.

The Functional Duality of Genes Arouses New Considerations on Cancer Therapies

Dichotomy of cancer-regulatory genes to “oncogenes” and “tumor-suppressor genes” has, on the one hand, greatly helped us in learning molecular details of tumor biology, but has, on the other hand, created a lot of confusion for some cancer researchers as well. This is largely because virtually all these genes are dual-functional in reality, with those along the CCND–CDK4/6–RB axis as tangible examples. Much of the confusion may be eliminated if we redefine each of the mRNAs, regulatory RNAs, protein isoforms, and protein modifications produced from the same genomic locus as a “gene”. For instance, by this definition, the oncogenic bcl-xL and suppressive bcl-xS are 2 different genes.¹¹, Unfortunately, this new gene definition still cannot eliminate all the perplexities, because dual-function phenomena occur at gene–gene and cell–cell interactions as well.

Cancer cells are characterized by their uncontrolled replication, which is often misunderstood as “faster growth”.^263,264 A cancer mass usually has a much larger percentage of cells that are in a proliferating status, compared with the normal part of the host tissue or organ and to most other normal tissues or organs of the patient.^263,264 However, this does not mean that individual cancer cells replicate faster than individual normal cells.²⁶⁴ Actually, the opposite is true in many occasions (Fig. 3). For example, many hematopoietic progenitor cells complete one round of the cell cycle in 12–24 h, while cells in many cancer types require 2–3 d for a cycle,²⁶⁴ with those dormant cancer cells and those cancer stem cells that proliferate only occasionally as the extremes. In other words, that some cancers grow very fast is often because they have a very large fraction of cells in the proliferating status, but not because individual cancer cells replicate faster. Cancer tissues usually have a larger fraction of dead cells than their host tissues or organs as well (Fig. 3),²⁶⁵ although the much larger fraction of proliferating cancer cells still makes the tumor mass larger and larger until, if without treatment, about 40 tumor doublings, when the patient capitulates.²⁶⁴ Despite that it is a misconception in many occasions, “cancer cells grow faster” has been a general rationale for anticancer drug development. Indeed, most classical chemotherapies target cell proliferation and have some side-effects in common, including low blood cell counts, skin itch, hair loss, and gastrointestinal (GI) tract symptoms. This is because the drugs also inhibit proliferation of those normal cells that are highly proliferating, including hematopoietic progenitor cells, skin epithelial cells, hair follicles, and GI epithelial cells. When most cells in a tumor lump do not actually replicate faster than the normal cells in these tissues, this type of side-effect is more severe.

Figure 3. Illustration of cell type distributions in normal and cancerous tissues and the locations of the 2 different types of adverse effects of chemotherapies. The data are rationalized from an average of most situations and thus may not be applied to a given case. A much larger percentage of cells in a cancer tissue is in cycling, i.e., in a proliferating status, relative to those normal tissues or organs that retain regeneration capacity. On the other hand, proliferating cells in many cancer types usually take a longer time to complete a cell cycle than many proliferating cells in normal tissues, such as those hematopoietic progenitor cells and those in the basal layer of skin epidermis and gastrointestinal mucosa. Common side-effects of traditional chemotherapeutic agents occur systemically, i.e., are the off-target effects on the proliferating cells in the normal tissues or organs. In contrast, the potential adversities of targeted therapies that are derived from the functional duality of targeted genes are attributed to the heterogeneity of the cancer cells, which, in turn, is related to the heterogeneous expression and function of the canonically defined target gene within the same cancer tissue.

In addition to the above-described adversities that occur systemically, i.e., outside the cancer, many targeted therapies, which target a particular gene such as CDK4 and are considered “magic bullets”,²⁶⁶ may also have another type of weakness, including the induction of resistance as well as invasion and metastasis, which occur within the cancer per se (Fig. 3). Cancer cells in any individual patient are heterogeneous and, even worse, continue to be more heterogeneous as therapy goes on, especially when the treatments are genotoxic. The dual-function nature of genes is one of the main reasons behind the heterogeneity and, thus, a main mechanism underlying these in-tumor adversities. Targeting one gene or one form of the gene products may be effective for some cancer cells, but in the meantime may promote survival or progression of other cells, eventually manifested as tumor recurrence or metastasis, besides the possible development of a second primary tumor. For example, the canonically oncogenic ERK/MAPK signaling also has a strong tumor-suppressive activity in culture cells, in animals, and probably in a subset of more aggressive cases of different human cancers as well, as discussed by Deschenes-Simard et al.²⁶⁷ Blocking this pathway may promote the growth and progression of this subset of cases. Hence, knowing the dual-function nature of genes and redefining “genes” may help us not only to clear up much of confusion, but also to steer our train of thought to a new track in finding targets for cancer therapy. For instance, targeting the bcl-xL or the splicing mechanism that converts bcl-xL to bcl-xS may be better than targeting the expression of the bcl-x gene.¹¹ A good combination of targeted therapies to simultaneously block 2 or more pathways may be synthetically lethal to some cancer cells,^268,269 but in the meantime they probably are synthetically promoting to some other cancer cells as well. Currently, most animal studies for testing drug efficacy use xenograft or orthotopic graft model that transplants culture cells or tumor tissue developed from a single cell line to an immunodeficient mouse. Our assumption rationalizes that using primary tumor tissue from an animal or a human patient as the tumor source, which contains heterogeneous cancer cell populations, will be a better animal model to test the efficacy of anticancer drugs.

A drawback closely related to selection of therapy targets is that we still know too little about cancer invasion and early metastasis and thus lack specific targets for these stages of cancer, while metastasis is the major reason why solid cancers are lethal. There is evidence suggesting that invasive and early metastatic cells, which locate in the invadopodia,²⁷⁰ are quiescent (G₀ phase) and thus are small in number.²⁷¹ While those cycling cells can be killed by chemo- or irradiation therapies, the quiescent cells are the ones that may recur later. Hence, identification of target “genes” (defined by our new concept) specific for those invasive and early metastatic cells in the invadopodia is imperative. Moreover, so far tumor shrinkage is still the major, if not the only, efficacy criterion for evaluation of whether a drug can enter into the pharmaceutical pipeline, which may be a manmade barrier that inhibits the development of metastatic-specific drugs, as such drugs may not cause tumor shrinkage in xenograft animal models, as pointed out by Rosel et al.²⁷² Therefore, there is also a need for evaluating invasion or metastasis as criteria for drugs to enter into the pharmaceutical pipeline.²⁷²

Summary

Genes defined canonically have dual functions, which are manifested at multiple levels: (1) Different mRNAs, regulatory RNAs, protein isoforms and protein modifications derived from the same genomic locus may function differently and even oppositely; (2) Genes may interact at different levels, such as by forming chimeric RNAs and by forming different protein complexes to exert different and even opposite functions in different situations; (3) High levels of tumor-suppressor genes in normal cells drive proliferation of cancer progenitor cells in the same organ or tissue by imposing compensatory proliferation pressure, which presents the functional duality of genes as a cell–cell interaction at the tissue level. This “mitoinhibition-resistant phenotype” principle can be further tested if one day technology allows the manipulation of genes specifically in cancer progenitor cells without affecting their normal surrounding cells in the same organ or tissue. All these manifestations can find tangible examples along the CCND–CDK4/6–RB axis. The dual-function nature of genes, which often creates confusion to hamper our understanding of tumor biology, may be a main reason behind the heterogeneity of cancer cells in individual patients. Redefining “gene” by considering each of the RNAs, proteins or protein modifications from the same genomic locus as an individual “gene” should help us in clearing up much of the confusion on tumor biology and in selecting target molecules for treatment of different subpopulations of cancer cells in individual patients.