Cross signaling, cell specificity, and physiology

J. E. Dumont, S. Dremier, I. Pirson, C. Maenhaut


The literature on intracellular signal transduction presents a confusing picture: every regulatory factor appears to be regulated by all signal transduction cascades and to regulate all cell processes. This contrasts with the known exquisite specificity of action of extracellular signals in different cell types in vivo. The confusion of the in vitro literature is shown to arise from several causes: the inevitable artifacts inherent in reductionism, the arguments used to establish causal effect relationships, the use of less than adequate models (cell lines, transfections, acellular systems, etc.), and the implicit assumption that networks of regulations are universal whereas they are in fact cell and stage specific. Cell specificity results from the existence in any cell type of a unique set of proteins and their isoforms at each level of signal transduction cascades, from the space structure of their components, from their combinatorial logic at each level, from the presence of modulators of signal transduction proteins and of modulators of modulators, from the time structure of extracellular signals and of their transduction, and from quantitative differences of expression of similar sets of factors.

  • signal transduction
  • effect of hormone modulations
  • isoforms
  • combinatorial logic

cell signal transduction encompassesall the biological and biochemical phenomena that lead from the perception of a signal by a cell to the response of the cell. The signal transduction machinery of a cell integrates all the signals it recognizes and translates them in a coordinated behavior. A signal for a cell is whatever is recognized as such by a receptor that itself initiates a response to this signal. A receptor is the structure that recognizes and reacts to the signal and interprets the specificity of the signal. These are circular definitions. Our physiology, which integrates the requirements of a living organism (for metabolism, growth, reproduction) and its responses to the outside world, uses thousands of signals. For each cell, these include hormones and neurotransmitters, signals from neighboring cells, soluble such as paracrine factors, or membrane bound such as ephrins, and signals from the inert substratum such as fibronectin. Because each signal may be recognized by different receptors (e.g., for norepinephrine or serotonin), the number of receptors is a multiple of the number of signals; hence the tremendous complexity and specificity of signals and their receptors. One category of receptors, the seven transmembrane receptors, are coded by ∼700 genes. i.e., ∼2% of the number of genes in the human genome. This contrasts with the rather limited repertoire of known signal transduction pathways that are modulated by such receptors, with only a few ubiquitous intracellular signal molecules (cAMP, cGMP, Ca2+, etc.), phosphorylation cascades, and other pathways [nuclear factor (NF)-κB, etc.].

The cross signalings (anthropomorphically called cross talks) between the various cascades further simplify the picture in appearance: everything seems to modulate everything (Fig.1). See, for example, a recent title: “Signaling networks—do all roads lead to the same genes?” (261). Furthermore, the opposite cross signalings between cascades, as reported, give a totally incoherent picture (141). It is striking that excellent reviews on known signal transduction proteins or cascades mention so many demonstrated effects obtained in different cell models that the authors have great difficulties in drawing a coherent picture or end up by concluding for each cascade, or even enzyme: “X is regulated by all signal transduction cascades and regulates almost all cellular processes, from gene expression to cell death” or “these results suggest that the pathway linking A to B involves the integration of numerous signal transduction steps by a highly complex network” (6, 18, 19, 25,26, 28, 42, 77, 92, 93, 121, 125, 179, 260, 271, 272, 275,298-300, 359, 360). The very restricted phenotypes of knockout mice models in which a supposedly essential protein is absent do not fit with such statements. The impression left was recently summarized: “elegant complexity coupled with hopeless confusion better defines our current state of knowledge” (309). In fact, such reviews give a comprehensive picture of all the interactions that may exist in mammalian cells, i.e., of the toolkit available for differentiation. On the other hand, attempts to give a unifying picture lead to unwarranted generalizations and/or selective consideration of the literature and scientific myths (e.g., “Ca2+ causes cell growth”). Thus a first paradox, which has been spelled out for a few cascades (226, 227, 230), opposes on the one hand the multiplicity and specificity of signals, their receptors, and their effects and on the other hand the nonspecificity or promiscuity of the few signaling cascades (78, 167, 205).

Fig. 1.

Insulin action: from 1977 to 2000. GH, growth hormone; IGF-I, insulin-like growth factor-I; IL, interleukin.

Another paradox arises from the use of simplified models for the study of normal physiology, namely, the behavior of the cell within the whole organism, preferably human. Because of the limitations of clinical investigation and ever-increasing restrictions of animal experimentation, one has to rely on models that, from experimental animals to reconstituted systems, may lose in physiological relevance what they gain in simplicity and in definition (Table1). In the 1960s hundreds of articles, published in the best journals, on the direct action of thyroid hormone on mitochondria or on various enzymes were undoubtedly true but physiologically irrelevant. There is no doubt that work on cell lines has tremendously enriched our knowledge of signal transduction, of the actors involved, and of their interactions. However, the literature on signal transduction shows that similar pathways may have different, sometimes opposite, effects in different cells. This suggests that, if we are interested in the human thyroid cell, it is this cell type that we must study. In fact, the choice of a model dictates our concepts, of which we become prisoners. If enough researchers use a model, they tend to forget about the caveats and to reject in their research or as referees the ugly facts that may question their way of life. They become a constituency of the model. Furthermore, it is easier to publish clean data and mechanisms on cell lines than more disperse and less clear-cut in vivo results. But then, as one mentor asked after a seminar on some peculiar properties of a much-used cell line model, “So what?” Of course, this should not detract from work on cell lines because this allows us to identify new partners and interactions and defines what is possible in signal transduction. On the other hand, work on physiological models defines what is relevant in the physiology of a given cell type at a given stage.

View this table:
Table 1.

Experimental systems and physiological characteristics they retain

In this review we analyze the physiological relevance of signal transduction data and discuss the mechanisms that, despite the apparently generalized “textbook” schemes, account for the exquisite specificity of in vivo cell signaling. Furthermore, we explore the possible biological consequences of a loosening of this specificity. For graphic representation of signal transduction pathways, we use a recently proposed system (267).


Attempts to synthesize any field of the signal transduction literature these days either simplify it in a personal, selective, and therefore distorted way or end up giving a very confusing picture. Part of the confusion may be clarified by considering the systems used to obtain the data and their possible artifacts (188, 309).

Loss of Biologically Important Information in Simplified Systems

Simplification of the systems used for the study of signal transduction is a double-edged sword. At each level from the in vivo study of humans to the precise molecular definition of proteins by X-ray crystallography, what we gain in precision, rigor, and definition we may lose in relevant biology (Table 1).

There are numerous examples of apparent discrepancies between findings in simplified systems, e.g., in vitro and the situation in vivo. For instance, expression of the “cell proliferation signal” epidermal growth factor (EGF) in transgenic mice causes growth retardation (41). Lack of phosphoinositol-3-kinase (PI3K)-γ, an essential element of proliferation cascades, leads to colorectal carcinomas (292). There are great differences, not commented on by the authors, between the specificities of G proteins and their βγ-subunits for their controlling receptors in membrane preparations and in whole cells (122). Similarly, the history of knockout mice is rich in genes whose protein product is essential for a cell process in vitro but not in vivo.

The crucial fact that teratocarcinoma cells aggregated with normal morulas give rise to normal cells in adult animals (265,148) or that bone marrow cells injected in heart regenerate myocardium (253) testifies to the importance of the proper tissue environment in the behavior of the cell. Similarly, the interrelations between cancer cells and their neighboring cells are fundamental to their biology (89). Normal ovarian stromal cells promote the growth of normal ovarian epithelial cells but inhibit the corresponding tumor cells in vivo (258). These observations led to the development of new transgenic models in which oncogenes can be activated by a spontaneous recombination event rather than systematically in all cells of a given type (161).

Autocrine and paracrine effects may differ in culture and in vivo because of the dilution of factors in vitro, the washout of factors in vivo, or the absence of a necessary factor in the medium. In COS7 cells, for instance, the stimulatory effect of insulin-like growth factor (IGF)-I on the mitogen-activated protein (MAP) kinase (MAPK) cascade is secondary to the autocrine release of EGF (287). The simple manipulation of cells in culture introduces new variables: change of medium or even minor mechanical stress causes ATP release and activation of the ubiquitous purinergic receptors (254, 255). Because many biological effects require the conjunction of several factors, simplified systems may lose properties, including the specificity of interactions. For instance, direct interactions of transcription factors with DNA in acellular systems are much more promiscuous than in vivo (43).

Yet, to define mechanisms, simplified systems and even pure molecular species are necessary. To fully understand the reaction of a pharmacophore with its target one must define the precise structures involved. Still, the biological relevance of the findings needs to be validated up to the level of the human organism. Animal models, transgenics or knockouts, general or local, permanent, permanently or transiently inducible, allow us to test precisely the role of a given gene in vivo. Defined human genetic diseases, when they exist, allow extension of the conclusions to humans.

Arguments In Favor of a Causal Relationship May Not Be Proofs

The best argument in favor of the hypothesis that a biological event is necessary for a signal transduction pathway is to show that its suppression inhibits the downstream events. Suppression can be achieved by pharmacological inhibitors, antibodies, dominant-negative or competing peptides or proteins, deletion of the protein by inhibition of its synthesis (e.g., antisense, RNA interference), or gene knockout. For inhibitors, the postulated suppression must take place in the system studied under the conditions used and it must be specific. Such controls are often missing. Many hormonal effects have been related to protein kinase A (PKA) because they are inhibited by the supposedly specific H89, which, in the same concentration range, inhibits MAP kinase-stimulated kinase (MSK1), a kinase downstream of MAPK (332).

In addition, the fact that an event is necessary does not necessarily imply that it is a required step in the causal relationship sequence (Fig. 2 A). Metabolites or an O2 supply are necessary for the survival of many cells and therefore for the operation of their signal transduction pathways, but they are not part of these pathways!

Fig. 2.

Causal sequences and their effects. A: suppression of an effect by a specific inhibitor (in this case the inhibition of B or D), even if specific, does not imply that the target of the inhibitor is in the causal sequence (e.g., D is not in the sequence). B: and/and control = coincidence detection. A, B, and C are necessary to cause D; either/or control: A, B, or C is sufficient to elicit D. C: biphasic control. A causes C. If the concentration effect curve of A vs. C is biphasic excess A will not cause C. (A), concentration of A; (C), concentration of C.

Demonstration that experimental induction of what is supposed to be the primary event causes the downstream steps of the pathway is an argument but no more. Constitutively active forms of signal transduction proteins are often used for such purpose, although their specificity may also have been altered, as shown for EGF receptors (EGFRs) (210). Failure to induce the consequences may just indicate that parallel events are also necessary (Fig. 2 B). Overexpression of the primary event will be ineffective or inhibitory if the effect is biphasic vs. concentration (182), i.e., in hormesis (36). This is the case for cAMP induction of proliferation in granulosa cells (282) (281) or for p53 and apoptosis (187) (Fig.2 C). Constant expression of the primary event will be ineffective or inhibitory in a sequential process in which each successive step requires the arrest of the previous one (e.g., in phagocytosis with extension of the membrane, engulfment, scission of the vesicle from the membrane, etc.; Refs. 12,228). Induction per se indicates that the initial event may cause the downstream consequences, but the fact that this event actually takes place in the pathway remains to be proven. Preferably, as laid out by Robison et al. (280) in their rules for cAMP causal relationships, activation/inhibition of each of the actors of a cascade should be shown in the intact cell, direct activation/inhibition of each actor by its upstream modulator should be demonstrated, and the kinetics and concentration effect relationships should be compatible with the proposed scheme.

Validity of Reported Relations May Be Restricted to Artificial Systems

Many interrelations within and between elements of the signal transduction pathways are protein-protein or protein-DNA interactions. The methods used to define such interactions have provided a tremendous yield of new information. However, their results should be properly assessed and their physiological relevance validated. Protein-protein interactions are difficult to study at the low concentrations that prevail in normal cells. For instance, coimmunoprecipitations depend on the affinity and specificity of the antibodies used, on the dissociation rate of the interaction, and on the concentrations of the targets to be demonstrated in Western blots. Of course, the investigator can diversify the antibodies and modulate their concentrations, the washings, etc. To overcome such difficulties the overexpression of one or several proteins in transfected systems has become very general. Factors of overexpressions up to 100-fold are common. At these concentrations, weak, nonphysiological interactions and effects may well take place and the specificity of action of isoforms may disappear (see, for example, pRb and p107; Ref.158, Fig. 3, Aand B). For example, proteins that associate with the GABA receptors and cluster them in transfected cells may not do so in their native neurons (171, 172). Moreover, interactions that are normally impaired by compartmentation may occur in such systems. If an interaction is constrained by stoichiometric binding in a protein scaffold, even a doubling of the concentration of one of the locked proteins may be sufficient to allow spillover outside (Fig.3 C). On the other hand, overexpression of a scaffold protein will segregate its binding proteins in separate complexes and thus have an inhibitory effect (Refs. 197, 252; Fig.3 D). Expression constructs can interfere with nuclear receptors or transcription factors. From such artificial interactions authors infer effects on the corresponding endogenous proteins. In this case, the cell is no more than a glorified test tube.

Fig. 3.

Effects of overexpression on protein-protein interactions.A: forcing a nonphysiological interaction (A and B). At normal concentration, A does not interact with B, which is on the membrane; when it is overexpressed, interaction takes place.B: forcing new interactions (A and C). A and B are on the membrane and interact. A does not interact with cytosolic C. If overexpressed, A spills over in the cytosol and interacts with C.C: interaction in the presence of a scaffold protein. A, B, and C are sequestrated in a scaffold protein so that B cannot interact with free D. A slight excess of B will saturate the scaffold binding site; some B will spill over and interact with C. D: absence of interactions in the presence of excess scaffold protein. A, B, and C are sequestrated but on different scaffold proteins.

Other confounding factors in transfection studies are the use of constitutively activated mutants whose persistent activity does not reproduce the temporally organized activation of the natural proteins or, conversely, the use of transient transfections that do not reproduce the sustained activity of oncogenes, or the use of dominant-negative mutants whose actions may be much more diverse than foreseen (309). Similarly, the whole field of gene regulation, which has relied on cotransfections of plasmids expressing transcription factors and promoters with reporter genes, is undergoing a reappraisal now that it is realized that genes in normal chromatin may behave differently, or at least in a more sophisticated way, than genes in the naked or poorly “chromatinized” plasmids. InDrosophila, for instance, there is little correlation between binding in vitro of transcription factors to DNA fragments and DNA binding in vivo (21).

In vitro acellular systems with proteins, purified or not, are also used and thus have little relation to physiology. They may allow us to estimate thermodynamic properties or to pinpoint possible interactions. However, the concentrations used may be much higher than in the cell and may generate many false positives. Moreover, the artificial system used may lack components that would confer specificity. Interactions of Hox transcription factors with naked DNA are much less specific than they are in vivo (150, 151). Gβγ complexes lose their target specificity in acellular systems (73).

Similar reservations hold for yeast double-hybrid systems. For instance, steroid receptors in double-hybrid systems respond to all ligands by activating the downstream promoters, whereas in vivo some ligands act as agonists and others as antagonists. To reproduce the in vivo situation, Yamamoto et al. (371) had to introduce also in yeast the needed coupling factors. Still, as a first step to define the repertoire of possible interactions, this method has allowed gigantic steps (150, 151).

All interactions proposed on the basis of transfection, double-hybrid, or acellular experiments should therefore be validated at the normal concentrations in the cells in which they are supposed to take place, which is easier said than done (247). In each case, of course, the investigator can overexpress to detect unfavorable interactions (for example, interactions that would require the nonexisting phosphorylation of the protein) or underexpress to indicate specificity. Conviction about the validity of interactions demonstrated in vitro arises from the convergence of independent arguments.

Possible Relations May Not Apply Because the Proteins Involved Are Not Expressed in the Same Cells in the Tissue or in the Same Compartment in the Cell

To interact, proteins must be expressed in the same cell. Evidence for this assumption may be invalid for several reasons. The detection by PCR of the corresponding mRNA in the cell may be due to minor “illegitimate transcription” as, with enough cycles of PCR reaction, every mRNA can be detected in every tissue. With such evidence, in the thyroid field, eye muscles would express thyroglobulin, thyroperoxidase, and thyrotropin (TSH) receptor and could be considered as pseudo-thyroid follicular cells! On the other hand, tissue distribution studies (for proteins or mRNA) do not indicate which cells contain the protein studied unless one relies on immunohistochemical or in situ hybridization evidence. Human thyroids contain many muscarinic receptors, but, in contrast to the dog thyroid, the receptors are not in their follicular cells.

For an effect to take place (e.g., DNA synthesis after growth factor action), all the elements of the corresponding signal transduction cascade should be present in the cell. In fact, illicit expression of protooncogenes in cells in which they are not normally expressed is a major cause of cancer (274). The finding that EGFR activation by G protein-coupled receptors (GPCRs) and the consequent cell division may require the processing of cell-bound pro-EGF by a metalloprotease, itself activated by the cascade downstream of the GPCR (269), explains why such a mechanism operates only in some cases (58). In addition, quite a few different mechanisms have been proposed in different cell types to explain GPCR activation of the growth factor receptor and MAPK cascades, with no attempt to disprove other hypotheses (120, 125, 170, 214, 215, 231, 294,354, 382). Most of these articles refer to “the mechanism of EGFR activation by GPCRs.”

Similarly, proteins may be restricted to different cell compartments or even to distinct macromolecular assemblies (scaffolds) that insulate them from others and thus prevent interaction, even though all are present in the same cell. This compartmentation generates functional modules, i.e., discrete entities whose function is separable from that of other molecules (133). By generating such a module with the elements of the MAPK cascade, the yeast scaffold protein Ste 5 confers to nonspecific enzymes the specificity of action of mating hormones (Fig.4; Refs. 61,95).

Fig. 4.

The STE5 scaffolding protein routes the mitogen-activated protein kinase (MAPK) module in yeast (90). The pheromone and its receptor release α- and βγ-subunits from the G protein, which will activate the first element (STE20) of the MAPK cascade, which by a linear path leads to mating behavior. STE11 corresponds to MAPK kinase kinase (MEKK), STE7 to MAPK kinase (MEK), and FUSS3 to MAPK in vertebrates.

Demonstrated Interactions May Only Occur in Some Cell Types, in Some Species, or in Model Cells, Which Explains Why a Cascade May Have Different and Even Opposite Effects in Different Cells

Differentiation in about 200 different cell types implies a specific program of protein expression for each of them. A clear indication of specificity is given by the numerous examples of opposite results of the same cascade in different cells. The same Ras oncogene product blocks proliferation in human fibroblasts and induces it in human thyrocytes and immortalized fibroblasts (57, 112, 113,250). It induces cyclin D in an intestinal cell line, while causing cyclin D1 phosphorylation and degradation in Rat1 fibroblasts (303). The same cAMP cascade inhibits cell proliferation in many cells (e.g., fibroblasts and other cells of mesodermal origin) but triggers it in some others (thyrocytes, somatotrophs, etc.); depending on the cell type, it activates, inhibits, or does not modulate MAPKs (285). In tadpoles the same hormone, triiodothyronine (T3), acting on the same T3receptor, induces cell death in the tail and cell proliferation in the rest of the body (308, 328). NF-κB inhibits apoptosis in most normal cells but induces it in some cancer cells (289). Even supposedly identical cells (e.g., endothelial or mesenchymal cells) are different in different tissues (2, 319). Blood vessels in endocrine tissues have a specific angiogenic mitogen (190).

Activation of the same pathway in the same cell type in different species may also lead to opposite results (209). The TSH receptor in human thyroid cells activates both the cAMP and phospholipase C cascades and accordingly activates thyroid secretion by the former and thyroid hormone synthesis by the latter, whereas in dog thyrocytes it only activates the cAMP pathway, which activates both functions (348). The same end result of TSH is obtained by different pathways in dog and human. In vertebrates as in viruses, evolution often conserves the function but may achieve it by different mechanisms.

Effects May Not Occur at All Times in a Given Cell

Receptors and signal transduction proteins are differentially expressed during embryogenesis, growth, and even under different physiological conditions. Effects may therefore differ at different stages. In fact, because signals during embryogenesis act through a few cascades, embryogenic development could not occur if these cascades were not interpreted differently by the target cells at different stages (62, 102). The whole early tissue organization inXenopus development depends on the well-defined, orderly time sequence of the action of four inducing factors on cells, which becomes different at each stage (314). The same Ras activation applied to Drosophila imaginal tissue at different stages leads to proliferation, apoptosis, or suppression of apoptosis (162). The evolution of cells in embryogenesis is now mimicked in vitro in embryonic stem (ES) cells (212). A maturating dendritic cell or T lymphocyte changes its panel of secreted cytokines and cytokine receptors, i.e., its signaling systems, within a few hours (27, 249).

The same stimulus may achieve the same result by different mechanisms at different times of the life of a cell, e.g., protein kinase C (PKC)-θ is required for T cell receptor-induced NF-κB activation in mature but not immature T lymphocytes (323).

Many effects depend on the cellular environment. Vasopressin, acting through its V1a receptor, activates Gq in proliferating Swiss 3T3 cells but Gq and G13 in cells in the G0/G1 phase of the cell cycle (1). In hepatocytes in culture, norepinephrine stimulates DNA synthesis through β-receptors at low density and through α-receptors at high density (166). Muscarinic receptors stimulate growth of quiescent NIH 3T3 cells but inhibit it when the cells are growing (245). The same increase in cytosolic Ca2+ may push or retract the growth cone of a nerve, depending on preexisting Ca2+ or cAMP level (378). Malignant mammary epithelial cells may revert to a normal phenotype in a specific intracellular matrix environment (23).

Cell Lines May Not Be Good Models of Their In Vivo Counterparts

Most articles on cell lines extrapolate their findings to the in vivo cell counterpart, e.g., extrapolating to the human thyroid cell what has been found in FRTL5 rat thyroid cell line. An ever greater part of the literature on cell signaling bears on cell lines as models in part because of the ease of working with them. In fact, cell lines are poor models of their in vivo counterparts. First, by definition, contrary to normal somatic cells, they are immortal, i.e., they reproduce indefinitely. The process by which they are obtained implies a selection over several generations of genetically altered cells having the desired properties, very different from those of the cells of origin. For instance, whereas tyrosine kinases of the Src family are necessary for the proliferating effect of platelet-derived growth factor (PDGF) in normal fibroblasts, they are not in NIH 3T3 cells or other cell lines (32). It is striking that similar rat thyroid cell lines selected by different criteria, the PCCl3 and the FRTL5 cell lines, exhibit quite different properties. For example, they require one (FRTL5) or two (PCCl3) oncogenes to become transformed. Finally, cell lines may evolve: the FRTL5 cells described at their origin required insulin and TSH to grow. Presently available samples require only one of the two or, in some laboratories, only insulin, TSH having just a complementary role. Some FRTL5 cells used in different laboratories are stimulated by EGF, some not; in some, serum enhances TSH mitogenic action, in some not. What is the interest for physiologists of the differences between two thyroid cell lines or between the same cell line having evolved in different laboratories? Another example of a molecular evolution of cell lines is the progressive accumulation and selection of cells with larger alleles due to unstable triplet repeats (117).

Similarly, cancer cell lines are often studied as representative of in vivo cancer cells even though they have developed new characteristics: p53 is mutated in thyroid and esophageal carcinoma cell lines but not in the corresponding primary tumors (327, 368). p16INK4 is inactivated in thyroid tumor cell lines but not in thyroid tumors (37). PTEN mutations are detected in melanoma cell lines but little in melanomas (379-381). Even in cancer cells the characteristics of disseminated cells can be very different from those of the tumor that releases them (177). How relevant to metastasis are human cancer cell lines injected in animals (318)?

Intuitively, one would guess that the variations between similar cells (e.g., thyrocytes of different species) would be much greater at the initial steps of the cascades than at their core mechanisms: there are hundreds of receptors modulating cAMP levels but only six isoenzymes of cAMP-dependent kinase, and these have the same substrates. However, this is not always true, as variations may also occur at the core of signal transduction pathways: in dog thyrocytes cAMP activates cyclin D/cyclin-dependent kinase (CDK)4 complexes without inducing cyclin Ds, whereas in FRTL5 cells it does so by inducing cyclin D1 (65).

Conclusions on Physiological Relevance of Reported Effects

It is therefore dangerous to extrapolate to other systems the data obtained in one species. One sees articles in which data on human, dog, or pig thyroid cells in primary cultures or rat thyroid cell lines are combined in reasoning about the nonexistent paradigmatic “the thyroid cell.” Although the existence of a given mechanism is often demonstrated only in one system, articles imply, implicitly or explicitly, that the mechanism described is general (e.g., “activation of the EGFR by seven transmembrane receptor,” “cAMP activates MAPK”). The specificity of cell signaling in different cells, even for the same or similar extracellular signals, and even through the same initial receptor, is demonstrated a contrario in reviews that attempt to synthesize our present knowledge. In a recent article on seven transmembrane receptors and cell proliferation, no two of the systems described work in the same way, and even when one receptor is considered, its mitogenic cascade differs from one model to another (125). The buzzwords of such reviews are “complex, pivotal, subtle…” (260, 261). This explains the wrong impression of confusion emerging from such reviews.

The map of all possible interactions and causal relations in signal transduction should therefore be considered as a map of possibilities, only few of which really take place at a given time in a given cell type. The exquisite cell- and stage specificity in signal transduction is fortunate for the pharmacologist (364) who aims at such a specificity for his drugs.


Cell Responses Depend on the Pattern of Their Protein and Isoform Expression

The specificity of response to one cascade in different cell types depends on its differentiation, i.e., on its protein composition and, therefore, on the genes whose promoters are accessible. For example, in kinase cascades the response to the same cAMP and cAMP-dependent kinase depends on the population of phosphorylable proteins in a specific cell type, i.e., on its differentiation. Similarly, the response to a similar transcription factor in different cells depends on the nature of the gene promoters that are accessible.

At each step of most cascades, several isoforms of proteins perform overlapping functions. They may be encoded by different genes or result from different mRNA splicing of the same gene or from postranslational processing. In mammalian cells there are at the present time 10 adenylate cyclases, more than 40 cyclic nucleotide phosphodiesterases (315), 70 A-kinase-anchoring proteins (AKAPs) (74), and 11 families of nonreceptor tyrosine kinases (279). Evolution multiplies the varieties of possible isoforms at each stage, from one kinase at each step of the MAPK STE pathway in yeast, Caenorhabditis elegans, andDrosophila to several kinases in mammalian cells (61). This explains why work on such simple model organisms is fundamental to demonstrate basic mechanisms and schemes, their actors, and their interactions and thus give the foundation of signal transduction. It also explains why direct extrapolation to mammalian cells is risky.

Because the isoforms have relatively similar properties, they are often, when discovered, lumped together as though interchangeable. In fact, more detailed investigations reveal different (sometimes opposite) and overlapping regulatory properties, e.g., the positive or negative effect of phosphorylation by MAPK on phosphodiesterase 4D isoforms (217) and the opposite and qualitatively different effects of p53 isoforms (239, 240). They also present cell type-specific expressions (30), intracellular localizations (206) as determined by specific docking domains (305), effectors (e.g., for receptors or kinases; Ref. 126), and controls in expression at the transcriptional, translational, and posttranslational level (70,123). These differences may give the isoforms entirely different physiological roles. They may also respond differently to direct and cross signalings. Isoforms of glucose transporters are usually tissue specific, with a conserved transmembrane catalyzing transport domain and different cytoplasmic tails allowing specific regulations (236-238, 333). The same Ca2+ signal will enhance or decrease cAMP accumulation depending on the type of adenylate cyclase present (147). Whereas the β- and γ-subunits of the G proteins transducing the action of seven transmembrane receptors were long thought to be interchangeable, it appears more and more that the response to a given receptor requires the presence of a defined set of α-, β-, and γ-subunits (140, 160, 278). E2F2 and E2F4 have opposite roles on cell differentiation (257). With only some of the possible isoforms present in a given cell, of all the possible controls only the few permitted by this selection will operate in this cell. This has been well demonstrated, e.g., for G protein α-subunits in the different cells of human fetal adrenal gland (30) or for AU-rich element (ARE) binding proteins that regulate the stability and translation of mRNA in embryos (142).

The nature of the expressed isoforms may itself depend on the physiological state of the cell: depolarization induces, through Ca2+ and Ca2+/calmodulin-dependent protein kinase, a different splicing of the pre-mRNA of the Slo channels and therefore the expression of different proteins with different allosteric properties (369).

The relatively low number of genes in the human genome seems to put a lid over the number of possible isoforms of any protein. In fact, any gene coding for an isoform of a signal transduction protein may also code for several isoforms by mRNA alternative splicing (24,119), e.g., each cyclic nucleotide phosphodiesterase gene for 3–10 alternatives (24). In these, the presence or absence of one motif of protein-protein interaction in an alternative form of a protein may channel a pathway in a given direction or not (24, 119) according to the protein recognition code (321). The presence of one or another intraprotein signaling module such as a protein phosphorylation motif may confer positive or negative regulation by PKA (Ref. 66; Fig.5) or by MAPK (340). A splice variant of phospholipase C behaves as a negative regulator of phospholipase Cδ (242). Truncation by alternative splicing of fosB into ΔfosB leads to a different transcriptional repertoire (290). Splice forms of an Eph receptor inverse the adhesion/repulsion response caused by this receptor (137). Variations in the upstream open reading frames of mRNAs also greatly change the life of the mRNA and its translocation efficiency (236-238).

Fig. 5.

Control of the dual enzymes phosphofructokinase/fructose 2–6 biphosphatase (PFK2-FBPase2). The enzymes have 2 opposite exclusive functions: the phosphorylation and dephosphorylation in 2 of fructose-6-phosphate. In the case of the liver, enzyme phosphorylation by cAMP-dependent protein kinase (PKA) in the NH2 terminus leads to activation of the phosphatase activity and inhibition of the kinase activity. In the case of the heart, phosphorylation by PKA of the COOH terminus of the enzyme leads to the opposite result. The isoforms are coded by different genes. Liver PFK2 has a PKA module in the NH2 terminal part. In the heart PFK2 has a PKA module in the COOH terminus.

Finally, the repertoire of possible effectors of each signal transduction protein, as presently known, will probably expand in the future. When an action of such a protein is discovered, it is often assumed to be the only one until we are shown otherwise. Thus we restrict the role of GPCRs to their effects on G proteins (129, 134) and the role of PTEN to its protein phosphatase activity (68) to discover later that other effects exist. The specificity of isoform expression thus goes a long way in explaining cell specificity of responses.

Operation of a Pathway May Depend on Spatial Structure of Its Constitutive Elements: Subcellular, Membrane Localizations, Multiprotein Complexes

Spatial structure in the cell refers to cell compartments and to supramolecular complexes. Nuclear or cytoplasm compartmentation prevents many possible interactions, sequestrating active molecules from each other. For example, sequestration of MDM2 in the nucleolus by p19ARF blocks its inhibition of p53. The regulation of the cell cycle and of transcription represents a ballet of nuclear to cytosol import-export dynamics (110, 232, 266). Similar dynamic controls of protein localization operate even in bacteria (154,155, 304). Thus the existence or not of a translocation mechanism or of its regulation may greatly differentiate the effects of a signal transduction pathway in different cell types. Loss of spatial structure of signal transduction pathways is a cause of several diseases (243).

The targeting or nontargeting of a protein at the membrane may change the whole pattern of its interactions. Such targeting may involve protein-lipid or protein-protein interactions (193). It may require specified mRNA localization and protein production (313). Nonprenylated Ras or, in some cells, nonmyristoylated cGMP-activated kinase does not activate its cascade (136, 219). Insulin and WNT both inhibit glycogen synthase kinase 3β, but this leads exclusively to increased glycogen synthesis for insulin and exclusively to increased availability of β-catenin for WNT (71). Integrin necessarily stimulates at defined contacts (298-300). Compartmentalization goes further with the segregation of some membrane proteins within or without lipid rafts and caveolae (107). GPCRs may have only access to their compatible G proteins in “raft” subdomains of the membrane (254). This could explain the discrepancy between TSH promiscuous effects on G proteins in isolated membranes and its more restricted effects in intact cells (5). The numerous proteins whose main function is to anchor signal transduction proteins (e.g., the AKAPs for cAMP-dependent protein kinase) to definite structures show the importance of such localizations (86).

Similarly, scaffold proteins also have the role of assembling supramolecular complexes, bringing together signal transduction proteins in one permanent or transitory functional unit, module, or “signalosome” (35, 133, 260, 261). Such complex functional multiprotein assemblies were first described for metabolic enzymes, accounting for metabolic channeling (256). Their properties are more than the sum of the properties of their individual constituents (106). For instance, by associating tightly phospholipase C and its G protein, the INAD scaffold protein allows the former to activate the GTPase activity of the latter and thus to shorten the signaling of the photoreceptor (51,95). For cAMP, inositol 1,4,5-trisphosphate (IP3), and even phosphatidylinositol-3,4,5-trisphosphate (PIP3), colocalization of the signal generation effector and remover allows highly localized effects in dendritic spines (169). The synapse or the neuromuscular junction is a permanent multimeric complex constituted sequentially (84,329-331) and dependent on specific targeting (44-46, 132, 171, 172). In yeast the Ste5 scaffold protein channels the activation of the MAPK pathway by mating factor to mating-specific genes (Refs. 61, 108; Fig.4). 14–3-3 Proteins have a similar role in vertebrates (241,370). Activation of the T cell or B cell receptor requires the constitution of a large integrated multimeric complex in membrane rafts (181, 326, 336, 358). The stability of such complexes varies from very transient to quasi-permanent (97).

Specificity of Response of Different Cell Types May Result from a Combinatorial Logic

From specific combination of elements involved in parallel.

In this type of regulation, a few factors may combine to specify many different instructions. A signal can be compared to a letter in a word: it has no meaning per se, only the combination of several letters has a meaning and the same letter used in a different word or combination may have a different or opposite meaning. Such regulations have been demonstrated, for example, for gene expression and for odor discrimination by the olfactory system (34, 109, 220). In the former, it is generally the combination of several complementary DNA regulatory elements and their specific transcription factors that confers specificity and strength to a promoter (204). Thus even broadly overlapping sets of regulated transcription factors may have very different end effects (Refs. 94,371; Figs. 6 and7). Assuming that in humans, as in other species, transcription factors may represent 5% of the 100,000 expressed mRNAs, the number of possible combinations of three (with repetitions) is 2 × 1010! The distribution of the 30 Ets transcription factors in different cell types can be considered as a fingerprint of each type (223). The expression of thyroid-specific genes depends on three transcription factors [thyroid transcription factor (TTF)1, TTF2, Pax8], each of which is expressed in the thyrocyte and in at least one other cell type, but all three are only coexpressed in the thyrocyte (60). Similarly different cascades with partially overlapping sets of induced genes may also exhibit the same combinatorial logic (Ref. 94; Fig.8).

Fig. 6.

Venn diagram depicting genes specifically dependent on components of the transcription machinery (small circles) in relation to the transcriptome. I, II, III, IV, components of the transcription machinery; N, sum of genes dependent on each component (I, II,III, IV) for transcription; n, sum of genes dependent on more than 1 component n1 (I,II), n2 (II, IV),n3 (II, III), n4(I, IV), n5 (I,II, IV), n6 (II,III, IV). All the genes in set I(N1) depend on transcription factor I, butn1 gene expression requires both I andII and n5 requires I, II, and IV.

Fig. 7.

Signals or receptors (A, B, C) acting on 3 cascades or genes (α, β, γ) positively, each cascade or gene resulting in a specific effect and each combination of cascades or genes resulting in the sum of individual effects and in new effects elicited by the combination. Even though there are only 3 primary actors and 3 intermediates and only positive effects are considered, the regulation has 7 outcomes. 0, None.

Fig. 8.

Overlapping and nonoverlapping inductions (+⧫) of genes by the cAMP cascade and the diacylglycerol (DAG) protein kinase C cascade. Both cascades activate (arrowheads) the cAMP response element-binding protein (CREB) transcription factors. Each cascade also activates distinct transcription factors X for cAMP and Y for DAG. Both cascades induce B through CREB, but cAMP also induces A, which requires both X and CREB, and DAG specifically induces C, which requires both Y and CREB.

This combinatorial logic and the various possibilities that it offers apply at other levels of signal transduction cascades. Even at the level of the few intracellular signals [cAMP, cGMP, Ca2+, diacylglycerol (DAG), ceramide, etc.] controlled in a binary mode (+ or −), 5 signals could allow 25 = 32 combinations. Examples abound in physiology and pathology. Different receptors acting on the Janus kinase (JAK) system act on different combinations of Jaks that will activate different signal transducer and activator of transcription (STAT) transcription factors (275). In macrophages, inflammation results from the activation of different Toll-like receptors, each triggering secretion of specific sets of chemokines with partially overlapping target receptors differentially expressed in target cells (Ref. 33; Fig.9). Heterodimerization of EGFR-like receptors generates many different receptors that discriminate between ligands and between effectors (251, 324). Control of cyclin-CDKs in different tissues results from various combinations of the CDK inhibitors, which explains why the different combinations of knockouts generate different types of tumors (101).

Fig. 9.

Combinatorial combinations of chemokines, their receptors (CCR1–6), and the target cells expressing the receptors (all positive controls). Each chemokine activates some receptors and therefore the target cells that express the receptors. MCP, monocyte chemoattractant protein; MIP, macrophage inflammatory protein.

The simplest case of combinatorial logic is the case of coincidence detection, whereby two factors acting in parallel are necessary for an effect (Fig. 2 B). Permissivity, in which one factor is necessary for the action of another, is analogous. The requirement of the conjunction of several stimuli to cause an effect (and/and control) is a general feature of signal transduction. The classic example is the requirement of activation of both the T cell receptor and CD28 for T cell stimulation (11, 181) or the necessity of both ATP and tumor necrosis factor (TNF)-α to induce maturation of human dendritic cells (295). Another example is the necessary complementarity of PI 3-kinase and Ras activation in the induction of metastasis by MET-hepatocyte growth factor (HGF) receptor (13). Such a requirement explains why general overexpression of only one protooncogene leads only to a few types of tumors. An example at the level of protein direct activation is the necessary coincidence of RacGTP and p67phox to activate O 2 generation by phagocyte p91phox(69) or at the level of protein phosphorylation, the necessary coincidence of phosphorylation by target of rapamycin (TOR) and 3-phosphoinositide-dependent protein kinase (PDK)-1 of p70 S6 kinase for full activation of the enzyme (149, 154, 155).

An alternative to the coincidence detection or and/and control is the either/or type of control by which any of the various parallel upstream cascades can elicit the cell response. This type of control obeys a different physiological logic: the cell response is so important that its occurrence is ensured by the redundancy of controls. The absence of phenotype of many gene knockout models testifies to the frequency of such controls.

From combination of unregulated parallel factors and of one regulated factor in each cell type: the triggering reaction or “switch.”

Many different signals operate in their target cell by inducing one or several rather ubiquitous transcription factors, i.e., early-immediate genes. C-Fos and Egr1, which are induced in many cells in response to all sorts of signals, are an obvious example (372). Cell-specific response is given by the other cell-specific transcription factors that are also necessary for induction of a specific gene in the cell. The nonspecific Egr1, in conjunction with SF1, leads in gonadotroph cells to the subsequent induction of LHβ gene (165, 338) and, in conjunction with WT1, induces Mullerian inhibiting substance in Sertoli cells (Ref. 310; Fig. 10). The interleukin (IL)-6 promoter in monocytes requires cAMP response element-binding protein (CREB), AP1, and cellular enhancer-binding protein (CEBP) but is activated by the newly released NF-κB (344). Just as in an electric circuit the response to an electric switch corresponds to the system downstream (light, air conditioning, heating, etc.), the transcription response to a signaling pathway and its general triggering transcription factor depends on the existing tissue-specific transcription factors and their regulatory elements. This concept would account for synexpression, i.e., the expression with a similar pattern of a set of genes in a cell type in response to a stimulus (124, 246), and in some cases for cell differentiation (124). It goes a long way in explaining the puzzling question of how a promiscuous signaling cascade can achieve unique effects in a given cell type. The phosphorylation of histones H1 by cAMP-dependent kinases in different cell types presumably causes a general loosening of the chromatin structure, which might facilitate later, more promoter-specific transcription effects, fits with the same concept. This concept of a single switch whose meaning is only determined by the specific existing elements of the cell also applies to early steps of signal transduction cascades. The effect of a general signal such as Ca2+ or cAMP in a cell depends on its complement of existing protein substrates of calmodulin or PKA (Fig. 11). Similarly, the pattern of gene expression induced by EGF in a single cell line depends on the composition of the cell matrix (373).

Fig. 10.

The triggering reaction or switch. The induction (+⧫) of the same early-immediate transcription factor egr1 induces LHβ in gonadotrophs in conjunction with SF1, and MIS in Sertoli cells in conjunction with WT1 (310, 338).

Fig. 11.

The same cascade has different effects depending on the protein substrates (P) of protein kinase A (PKA) which are expressed in different cells. The phosphorylation of P1, P2, and P3 in cell I will cause effect I; the phosphorylation of P3 and P4 in cell II will cause effect II; and the phosphorylation of P4 and P5 in cell III will causeeffect III.

From specific combination of sequential factors.

In the specificity of nuclear receptors action at least six different factors are involved, each of which can switch the sign (+ or −) of the response: the regulatory element in the promoter DNA, the nuclear receptor or the transcription factor binding to the regulatory element, the hormone binding to the receptor, the phosphorylation of the receptor, the other transcription factors present on the promoters, the coactivator or corepressor binding to the receptor, and their modulators and adaptors (157, 192, 221, 283, 371). Other independent or dependent factors regulating the outcome are the various feedbacks between transcription factors (276), the methylation of the promoter and enhancer, and the state of the chromatin (histone acetylation and phosphorylation, high-mobility group (HMG) protein binding, polycomb group protein binding, etc.), DNA methylation and chromatin structure being generally linked (40,47, 152, 277, 297, 339, 353). “Coordination of large sets of genes could be accomplished by affecting the function of specific components of the transcriptional machinery” itself (139). Moreover, some genes like that of the human fibroblast growth factor (FGF)-1 have four different promoters, differently regulated, directing the expression of four alternatively spliced transcript variants (48).

Thus each different combination of factors in a multimeric complex with two different possibilities for each factor (a binary switch) will lead to a specific result (Fig. 7) (one of the two possibilities can be the absence of a factor). For a complex of n factors with 2 possibilities for each factor, the number of possible combinations is 2n and with 3 possibilities 3n. To understand such a system the matrix of all possible combinations should be analyzed. As long as one factor is missing from the analysis, its results are doubtful.

The nature of the regulatory elements and their necessary transcription factors, coactivators, corepressors, etc. reflects the genetic background; the opening of the gene (methylation of the DNA, chromatin structure) and the presence of the receptor and transcription factors and especially of the modulators depend on the differentiation of the cell, and the presence of the hormone and the phosphorylation of the receptor transcription factors and modulators reflect the state of this cell, i.e., the physiological context. Similar regulations operate on enhancers and silencers (105, 152, 221, 284). The physical basis of the regulation by the nuclear receptor is its multifaceted character with different domains and functional surfaces, binding to different agents in the multiprotein transcription complex and signaling to each other and to the domain, which finally selects either a coactivator or a corepressor (143, 144). “Implicit in the concept of combinatorial regulation is the dynamic nature of regulatory complexes. The requirements for complex formation are that the component parts must be sufficiently flexible in their interactions to allow mixed assembly, yet must be sufficiently precise to insure decisiveness in their regulatory effects” (Ref. 371; see also Ref. 192). Within one nuclear receptor diversity may result from a flexible construct leading to diverse conformations. Decisiveness will result from precise assembly leading to specific conformations and supramolecular assemblies (31). Such fragile flexible structure may be protected by chaperones like heat shock protein (HSP)90 (288). At the level of mRNA stability and reading, a plethora of ARE binding proteins with either positive or negative effects also allows for numerous possibilities (363).

Examples of such logic also exist at other levels: given the occurrence of 2 A, 2 C, 4 B, at least 8 B′, 4 B", and 2 B“′ isoforms, a total of about 75 different dimeric and trimeric protein phosphatase (PP)2A holoenzymes can be generated (153). Also, the qualitatively different effectors and effects of c-Src depending on the nature of its activator (114, 115) reflect such sequential combinatorial logic. Multiple possible phosphorylation sites with distinct meanings, each with a binary choice (phosphorylated or not), may provide the same type of combinatorial logic at the level of one protein. For example, the same p53, depending on which of its five phosphorylation sites are phosphorylated, will activate different but partially overlapping sets of genes (146, 356). A similar phosphorylation code applies to ribosomal S6 kinase orN-formyl peptide receptor (80, 218). Similarly, histone modifications could also act in combination to form a histone code (320). The flexible multimodular protein model also applies to many signal transduction proteins (64, 174, 260,261).

The complexity of the regulations reviewed in this section involves the triggering or extinction of one gene, of one regulatory step. A further level of complexity involves the use of combinatorial regulations at several successive levels: for example, combinations of transcription factors activating a set of genes, the products of which are transcription factors that in defined combinations with newly generated or existing factors will govern other genes, etc. (184). Another example would be possible combinations of splicing factors regulating qualitatively and quantitatively which RNA species are exported and combinations of ARE binding proteins regulating the life and translation of messenger RNAs, etc. (70, 123).

Specificity of Response of a Cell to Different Cascades Having Apparently Similar Consequences May Result From Specific Combination of Biochemical Effects of Each Cascade: Common “Awakening” Reaction and Specific Effects

Some cell responses are common to many if not all signaling pathways in a given cell type. At the posttranslational level CREB and cAMP-responsive element modulator (CREM) phosphorylation and activation are caused by the cAMP cascade, intracellular Ca2+, and growth factors acting through MAPK, stress, and p38 and MAPK-activated protein (MAPKAP) kinase, etc. (63). Glycogen synthase kinase 3β is phosphorylated on the same serine and inactivated by protein kinase B (PKB) and PKA (202-204). At the transcription level the early-immediate gene c-Fos is induced by almost any cell stimulant from the growth factors inducing mitogenesis in fibroblasts to repetitive activation of nondividing neurons (145, 235). Such common responses to cascades having very different effects imply by definition that these responses are general and nonspecific. In fact, they suggest a sort of undifferentiated “awakening” reaction. Other more specific effects of the cascades or the specific complement of transcription factors (see From combination of unregulated parallel factors and of one regulated factor in each cell type: the triggering reaction or “switch”) will determine the direction or response the “awakened cell” will choose. For instance in the thyroid, as in other cells (25, 26), each cascade, besides its common activation of Rap1, also activates specific kinases with more or less specific substrates (PKA for cAMP, PKC for diacylglycerol, Ca2+/calmodulin-dependent protein kinases for Ca2+ and Ras MAPK and PKB for insulin and growth factors), which confers specificity to its action (Fig.12; Refs. 55,79, 81). Each cascade also induces a more or less specific panel of early-immediate genes besides c-Fos. At the level of chromatin, similar general rapid modifications suggested that “diverse pathways can then contribute to the modification of these proteins without the necessity of targeting these pathways to a particular chromosomal site” (365).

Fig. 12.

Overlapping and nonoverlapping biochemical effects of different cascades in the dog thyroid. An example of convergence and divergence of cascades (+arrowhead, stimulation; //, +arrowhead with 1 or more steps omitted). The 3 cascades activate Rap1, an awakening reaction, but each one also activates its specific protein phosphorylation cascade(s), which leads to specific effects. ACh, acetylcholine; PKC, protein kinase C; CmCaPK, calmodulin calcium-dependent protein kinase; PI3K, phosphatidylinositol-3-kinase; GF, growth factors (80, 348).

Specificity of Response to a Cascade May Depend on Timing of Stimulus

Timing may also explain specificity. A signal may be short or long, immediate or delayed, continuous or oscillatory. The multiple qualitative differences in effect that such modalities confer have been well studied in the case of the intracellular signal Ca2+(18, 19, 75, 82, 83, 99, 202-204, 233). For example, the frequency of calcium transients dictates NF-κB transcriptional activity and neuronal differentiation (143, 144, 317). Calcium influx and consequent pituitary hormone secretion depend on the step duration of spontaneous depolarization (346). Insulin stimulation of mitogenesis requires a longer occupancy of the receptor than its metabolic action (311). In dog thyroid cells IGF-I, which activates MAPK kinase for a short time and PI3K for longer, has a permissive effect on TSH mitogenic action, whereas EGF, which activates PI3K for a short time and MAPK for longer, has a mitogenic effect in the presence of IGF-I (55). Short-term stimulation of the MAPK pathway by EGF causes proliferation, whereas longer stimulation by nerve growth factor (NGF) causes differentiation in PC12 cells (337). Similarly, short-term activation of MAPK in Drosophila ommatidia causes growth and survival, whereas longer activation causes differentiation (128) and sometimes growth arrest (362). In regenerating liver, two waves of MAPK activation are necessary for the division of hepatocytes: an early wave for the awakening competence acquisition and a late wave for progression and cyclin D induction (325). In T cells long-term stimulation of Rap1 enhances the activation of Elk by the Ras cascade, whereas short-term stimulation inhibits it (59). The main mechanisms involved in the timing of effects in a signaling cascade are the positive and negative feedbacks and feedforward regulations (103). The presence of one such regulation may completely modify the timing of a cascade. For example, negative feedbacks and feedforward controls translate a constant stimulus in a sequence of wavelike causally related events (Fig.13). In a biphasic effect the relative strength of a negative feedback on an initial step may transform a positive to a globally negative effect and conversely. Constant infusion of the hypothalamic stimulatory hormone gonadotropin-releasing hormone (GnRH) leads, through desensitization of its cascade, to an overall inhibition of gonadotropic pituitary cells. Many cell processes imply an orderly causal sequence of events, some of which require the extinction of previous steps and the absence of subsequent events. This is evident for cell motion or phagocytosis (176). In a sequential process by which β-adrenergic receptor first activates Gs, then through arrestin Gi and Src (213-215), because Gs and Gi have opposite effects, differences in the kinetics of the sequence will lead to opposite results (231). The presence of a single positive feedback can confer bistability to a system (i.e., it is either on or off) and thus permanence to an effect (334). Similarly, the hysteresis of calmodulin-dependent protein kinase converts a short Ca2+ signal in a much longer protein phosphorylation (18, 19, 104). Thus differences in timing of a response to a signal result from the characteristics of the proteins involved, i.e., from the pattern of the proteins present.

Fig. 13.

Feedback controls can transform the simple kinetics of action of a stimulus on a cascade. In a linear cascade, the signal A should elicit D with similar kinetics and a small delay. In A, the negative feedback of D on B shuts off the generation of D despite the constancy of stimulus A [like protein tyrosine phosphatase induction by tyrosine kinase (379) or MAPK phosphatase by MAPK]. InB, the positive feedback of C on B transforms a pulse of A in a continuous stimulation of B and C. In C, the negative feedback of B on A and the negative feedforward controls of A on B to C effect and of B on C to D relation generate a wavelike response of B, C, and D to A. A: initial stimulus. +arrowhead, activation; −//, inhibition.

Cell Specificity of Response May Depend on Qualitative Difference of Protein Expression of Modulators

Modulators are proteins not involved in the signal transduction cascade themselves but which positively or negatively modulate the proteins of this cascade (Table2). When a given protein has opposite actions, differential modulations of them will lead to one result or the opposite, e.g., cMyc, which induces cell proliferation or apoptosis (50). Examples of such inhibitory modulating proteins abound: starting from natural antagonists of extracellular signals, soluble receptors that compete with extracellular signals for membrane-bound receptors, kinase inhibitors, G protein inhibitors [regulator of G protein signaling (RGS) proteins; Ref. 312], and Bin1 or Groucho inhibitors of MYC or LLT1/TCF transactivating action (85), among others. An excess of scaffold proteins can also cause inhibition (35). Other modulators enhance the activation of a cascade, e.g., DARPP32, which after phosphorylation by PKA inhibits PP1 and thus enhances and prolongs the phosphorylation and thus the activity of CREB (306). Tyrosine phosphorylation of calmodulin increases many of its activations (53). Targeting of a signal transduction protein, for example to the membrane, may also greatly modify its activity.

View this table:
Table 2.

Roles of modulators

Some modulators even change the receptivity of a protein from one signal to another. Receptor activity-modifying protein (RAMP) proteins switch the specificity of the cGRP receptor to one or the other hormone, RAMP1 to CGRP, RAMP2 to adrenomodulin (196,236-238, 302). For hormone nuclear receptors, besides the controls mentioned above, there are modulators of coactivators [e.g., pCIP for CREB binding protein (CBP)] and competitors. There are even modulators of modulators such as p34SEI-I, which antagonizes the inhibition by p16INK4a of cyclin-CDK complexes (322), and prothymosin-α, which sequestrates the repressor of estrogen activity (224). Finally, signal transduction cascades may induce, depending on the cell type, the synthesis and secretion of autocrine or paracrine factors or of proteases generating such factors, i.e., extracellular modulators from inactive precursors (183, 195). Just as the field of nuclear receptors is now discovering all the modulators of coactivators and corepressors that provide specificity of action in different cell types, the field of signal transduction cascades is now identifying its modulators. The role of multifaceted proteins should also be considered at this level.

Specificity of Response to a Cascade Depends on Quantitative Differences of Expression or Activity

Many cascades activate two or more different branches that may have parallel, synergic, or opposite effects. When the effects are opposite the expected behavior of the system may be very complex (341-343) (Fig. 14). Moreover, quantitative differences of expression may lead to opposite results: Fas receptor activates the caspase cascade, leading to apoptosis, and the MAPK cascade which inhibits it (138). Depending on the strength of the two effects, TNF or Fas receptor cause apoptosis in some cells but not in others (9, 10). The complex SMAD2/SMAD4, activated by transforming growth factor (TGF)-β receptor, induces expression of the target gene when it is bound to the coactivator p300/CBP and represses it when bound to the corepressor TGTF and histone deacetylase. Formation of one of these mutually exclusive complexes is determined by the relative levels of corepressors and coactivators in the cell (366). Somatostatin receptor sst2 activates proliferation of CHO cells through the PI3 kinase cascade and inhibits it through the p38 kinase pathway (301). Ca2+, through calcineurin, both activates and inhibits apoptosis in a myeloid cell line (211). CD45 phosphatase dephosphorylates the activating and inhibitory tyrosine phosphates of LCK (264). Similarly, M-Ras and p21 Ras positively and negatively regulate Elk1-dependent gene induction through MAPK and Ras pathway modulator (RPM), respectively (87). Moderate levels of cAMP promote, whereas high levels inhibit, the bone morphogenetic protein (BMP)2-induced sympathoadrenal cell development (22) and the proliferation of granulosa cells (281, 282). Thus, when an agent causes opposite effects, differences in the expression of the target proteins or of the concentration of the agent can switch the sign of the response. Quantitative differences of even a factor of 2 can have dramatic consequences as shown by numerous phenotypic expressions of haploid insufficiency or by the growth advantage of neurofibromatosis 1 (NF1) heterozygosity for astrocytes (8).

Fig. 14.

Opposite effects of a cascade.

Conclusion on Cell Specificity

Thus the mechanisms endowing cell types with an exquisite specificity in signaling in response to similar stimuli come down to the level and time structure of the stimuli, to the unique qualitative and quantitative pattern of expression of isoforms and modulators, and to their spatial arrangement in compartments and supramolecular complexes (293).


As discussed above, much work on signal transduction and “cross talk” is carried out on cell lines, mostly for reasons of convenience. Cross talk, originally designating interference in radiocommunication (247), may thus represent a loss of specificity in signaling. The question may then be raised whether this loss is not a characteristic of these cell lines and of the way they have been generated. This is suggested by many examples. Oncogenes would not have transformed NIH 3T3 cells and thus would not have been discovered if these cells had not been already half-transformed (130, 186). Whereas PDGF action on fibroblasts has a stringent requirement for Src-type kinases, this requirement is lost in NIH 3T3 cells or other cell lines (Ref. 32).

Our knowledge of thyroid cell lines such as the FRTL5 cell line supports such conclusions. Generated by continuous culture of rat thyroid primary cultures, the cells were selected by their property of multiplying only in the presence of serum, TSH, and insulin. In the first report the cells did not multiply in the presence of insulin or TSH alone. Now, as used after many generations, the cells multiply in the presence of either insulin or TSH. This is inconsistent with in vivo work showing that mice or human thyroid cells do not multiply in the absence of TSH or when its receptor is inactive and that they do not respond to TSH in the absence of IGF-I. Moreover, aging of the cell line leads to an increase in the unstimulated proliferation rate. At the biochemical level, whereas the pathways of insulin and TSH are clearly distinct in primary cultures (with the cAMP-PKA response specific to TSH and the Ras MAPK and PI3 kinase responses specific to insulin), the pattern blurs in the FRTL5 cells, in which TSH and cAMP have been reported also to activate the Ras-MAPK and the PI3K pathways. The specificity of the TSH-cAMP and IGF-I-PI3K pathways has completely disappeared (175). Similarly, Ras oncogene induces proliferation with no effect on differentiation in human thyroid cells in primary culture, but it enlarges its action to dedifferentiation in the rat thyroid cell lines (112, 113). Whereas the proliferation effect in human cells requires at least MAPK and PI3K activations and another factor, in the WRT cells either MAPK, PI3K, orRal GDS is sufficient (112, 113). It is obvious that such cells in which the activation of one pathway is sufficient for mitogenesis will be more prone to uncontrolled proliferation. Similar discrepancies have been reported for pituitary cell lines.

Thus the pattern of signaling is blurred in cell lines. This may result from genetic or, more likely, from epigenetic changes, from quantitative dysregulations leading to differences in expression, or from qualitative changes due to mutations. When specificity of an enzyme is constrained by its localization on a scaffold, even a slight increase of expression may lead to escape, new cross signalings, and scrambling of a network. The development of the embryo illustrates how a relatively shallow gradient, i.e., concentration difference, can lead to qualitatively different differentiation. These arguments should be related to recent findings in Drosophila. In these flies, mutations of HSP90 induce phenotypic variations in nearly all tissues, suggesting that HSP90 may be a part of a molecular buffering system that keeps cryptic signaling cassette variants silent (247). Disruption of such a system would lead to blurring in signaling.

There are indications that, besides protooncogene activation and antioncogene inhibition, a similar blurring is taking place in cancer cells. Illegitimate transcription leads to the synthesis and eventual secretion of heterotypic hormones in these cells. Heterotypic receptors also appear. Cancer in multiple endocrine neoplasia is accompanied by a loss of substrate specificity by tyrosine protein kinase Ret (in Men 2B) (291, 379-381). A clear example of the role of blurring signal transduction specificity in cancer is given by the loss of specificity of androgen hormone receptors and their corresponding abnormal responses in prostate cancers (361, 349, 377).

The normal hypermutation process of lymphocytes, when wrongly targeted, is at the source of many lymphomas (259). Such a blurring probably confers a selective growth advantage, as no inactivation, negative control, or checkpoint can stop a cell from proliferating when cross signaling ensures the bypass of any block. As the null phenotype of a cell is to grow and multiply, blurring of the controls should favor growth (229, 316). Moreover, if a loss of signaling specificity is a disadvantage for the growth of a cell, it will be selected against and disappear. If it is an advantage, the cell will outgrow the others in what becomes a Darwinian process (130). The same reasoning applies to random methylation of promoters, which accounts for some cancers (20, 194), presumably by loss of tumor suppressor gene activity (234), to the role of the conjunction of mutations in several cascades, and to the role of aneuploidy (225) in cancer. The condition is described as “gene addiction,” in which the full malignant phenotype depends on the continued interaction between pathways (357).

It would be interesting to explore whether aging per se induces such loosening of controls. Normal human mammary epithelial cells spontaneously escape senescence and acquire genomic changes akin to those of the earliest lesions of breast cancer (286). This would explain the multitude of benign tumors arising with age. Extensive cross signaling may therefore represent not only a consequence of evolution and selection, i.e., of the carcinogenetic process, but also a factor in the initiation or further proliferation of cancer cells or cell lines.

The blurring of specificity and extension of cross signaling in cell lines or cancer cells could be called signaling entropy. Its mechanisms have been little studied. However, one might speculate that they involve the loosening of all the controls described earlier in this article that ensure cell specificity of signaling, e.g., loss of space or time structure, broadening of the substrate repertoire of enzymes, and increased illegitimate transcription, that is, an increase of entropy at all levels of signal transduction pathways. This in turn could result from a loosening of the very strict specificity of transcription.


Among the infinite possibilities of cross signalings between signal transduction pathways in their superimposed layers of controls, only a restricted number will operate in a given cell at a given time and the pattern of these controls will be very different from one cell type to another, reflecting the diversity of function and constraints of the cells (133). One may wonder about the interest of these control networks. Their value may lie in their robustness: intuitively, we guess that the more complex regulatory web is also the more robust (14, 52, 207, 352). This has been proved experimentally in ecology (268) and by the Internet. However, such statements should be qualified. Communication networks display a high degree of robustness, the ability of their nodes to communicate being unaffected even by unrealistically high failure rates. “However, error tolerance comes at a price in that these networks are extremely vulnerable to removal of a few key nodes” (3). This and functional redundancies (178, 185,376) explain why null genotypes for so many genes have no or a very poor phenotype whereas a minority are lethal. In fact, in yeast, the proteins that have the largest potential repertoire of possible interactions, i.e., the so-called “nodes” of network, or more mundanely the common bottlenecks of regulation networks, are the only ones whose disruption is detrimental to the cell (156). Another role for cross signaling in signal transduction cascades is that, just as metabolic pathways, as they diverge and converge, they may be controlled at multiple steps for efficiency (52). Finally, complex networks may allow fine tuning in regulation, which when applied to many cells, may become very important at the level of the organism. Small advantages may be very important in long-term survival and thus in evolution. On the other hand, a converse explanation for some complex signalings may be that they have no role and are irrelevant, innocuous relics of evolutionary history. Simulating and possibly understanding such complicated networks will require new tools, sophisticated algorithms, perhaps similar to those used in the study of neural networks (99).


The significance of a signal on a given cell depends on the network in which it is inserted. The response of a cell to a signal depends on the timing and strength of the signal, on the cooperativity of the response, on the cell environment, on the cell population, on the subcellular localization of the signal transduction proteins, on the nature of the isoenzymes present at each level, on the positive and negative feedbacks and feedforward controls, and on the synergic and/and or either/or controls, etc., that is, on its program, i.e., the regulation network, namely, the nature, quantitative importance, and localization of its constituent proteins, the actors in the game. Proteomics, i.e., the definition of the protein population in a cell type in a given condition, could give this information (198,199) and thus, with the use of a database of direct or indirect protein-protein interactions and controls, theoretically allow us to draw the awfully complex regulatory network of a given cell at a given time in its history (91). There are several caveats to be considered. First, the existence of a regulatory pathway does not necessarily indicate an important physiological role. We and others have delineated the mechanisms of action of neurotransmitters on the dog and rat thyroids without ever being able to demonstrate a role for these controls. Second, given the fact that the addition of one factor or even one motif in a protein may change the whole result of a network, the task may be enormous. Third, anybody who has tried his hand at quantitative simulation knows that by choosing the right parameters in even very simple models almost any possible behavior can be predicted. Fourth, the requisites of Pollard (as quoted in Ref.39) for successful computational prediction are rather tough: molecular inventory, molecular structures, molecular partners, kinetic constants, genetic and pharmacological phenotypes. At each step enzymes may be location specific or not, inhibited and/or stimulated, controlling and/or controlled. In fact, the functioning of a network may only be predicted when it is completely known. As expressed in a recent review, “it ain't over til it's over” (309). Moreover, even simple cross signaling may elicit very complex behaviors (189, 341-343). Present simulations on general schemes based on data from different systems (163) may be useful didactic intellectual exercises but have little predictive value. The possibility of predicting the behaviors of a cell by knowing its protein composition will be tested by the very thorough work of the Alliance on a few cell types as pioneered by A. Gilman (111).

The research in signal transduction therefore follows, legitimately, a three-pronged approach, each prong of which supports the others but should not be confused with them. The study of simple models, from yeast to Caenorhabditis, or even Drosophila, defines basic mechanisms, actions, and their relations (273,355). After all, the basic steps of apoptosis and of the EGF pathway were demonstrated first in C. elegans andDrosophila, respectively (98). The study of various types of mammalian cell lines defines the many possible variations that evolution and the multiplication at each level of isoforms and new modulators and cross signalings have provided, i.e., the available toolkit for differentiation. It is precisely in these variations and cross signalings that different cell types will mainly differ. The study of a given cell type in its physiological context defines which of the almost infinite possible combinations of controls applies to this cell type.

Thus regulatory schemes can be presented and interpreted in two ways. The first way is as maps of all the possible interactions like the conventional metabolic pathway maps. Reviews on signal transduction pathways and protein belong to this category. They should provide information about all possible interactions, the “toolkits” of cell biology, but refrain from specific functional interpretation. Second, regulatory schemes can be presented and interpreted as the regulating scheme applied to one cell type at a given time in its history. In this case, the cell type, species, history, environment, and experimental condition should be defined. The exquisite specificity of cell types is illustrated by the exquisite cell specificity of the phenotypes of knockout models for supposedly ubiquitously important genes. Reasoning about the behavior of a given cell type should only be applied to the second type of scheme. In such reasoning one should never lose what Barbara McClintock called “the feeling for the organism,” i.e., in this case, how signal transduction in a cell fits in with physiology. One outstanding example of success of the study in depth of one cell type is the study of T lymphocyte signaling (181). Editors might recommend modesty in titles and conclusions of articles, restricting specifically the domain of application of conclusions to the systems really studied or at least specifying that the controls described are possible but not necessarily universal: “cAMP can activate the MAPK pathway” is tolerable; the “tabloid”-type title “cAMP activates the MAPK pathway” is not. Otherwise, as Bacon said, “words turn back and reflect their power upon the understanding” (as quoted in Ref. 164), i.e., vagueness of language leads to confused thinking.

In cases in which existing schemes do not explain the results, modulators of the proteins in a cascade and even modulators of modulators should be sought after. The research on nuclear hormone receptors has shown the way to the whole signal transduction field in this regard.

For physiological relevance the choice of the right model is paramount. For human physiologists, this is the human cell in vivo. Any departure from this material entails the risk of irrelevance. Who cares about the peculiar properties of the FRTL5 cells used in our laboratory? When we use model systems the validity of the scheme we outline should be assessed in “real cells,” physiological or pathological. To get our story on the role of TSH through cAMP in the control of thyroid function and growth accepted, we needed to generate transgenic mice in which constitutive activation of this cascade in the thyroid led to goiter and hyperthyroidism and to demonstrate the same mechanism and consequences in human autonomous thyroid adenomas (175, 285,347)! Similarly, the positive role of cyclin D1 on proliferation and its negative control by p27 have been definitely validated by double-knockout mice (335). This increasingly recognized need explains the expanding literature on gene knockouts and especially on cell-specific and inducible gene knockouts.

One example from our group illustrates the necessity of in vivo validation. An enzyme hydrolyzing the intracellular signal PIP3 had been cloned in our lab: SHIP2 (263). PIP3 has been involved in various cells in the stimulation of protein synthesis, proliferation, etc., in the prevention of apoptosis, and in the action of growth factors IGF-I and insulin. Our prediction was therefore that mice in which SHIP2 had been knocked out would present growth anomalies and tumors. In fact, the main effect of this knockout was not predicted: a greatly increased sensitivity to insulin with rapid death from hypoglycemia. The affected mice were born with normal weight (49). The phenotype is extraordinarily restricted, which suggests that the enzyme mostly deals with the PIP3generated by the insulin receptor! It is interesting that, whereas the 3′ phosphatase PTEN also inactivates PIP3 and in transfected cells inhibits insulin action (244), its half-knockout in mice does not, as the SHIP2 half-knockout, suffer from hypoglycemia. Thus for the physiologist the safest approaches to signal transduction may be at the two extremes of the experimental spectrum: on in vivo models (transgenics and gene knockout) and on the structure of the relevant proteins. With regard to the in vivo models, their use in the study of specific organs in laboratories specialized in these organs would provide a wealth of information for both these groups and the groups generating the models.

In fact, there is more coherence in the physiology of regulation systems than in the mechanisms that achieve this physiology. TSH activates both the synthesis and the secretion of thyroid hormones in all the species studied, but in some it activates both the phospholipase C and the cAMP pathways, which stimulate synthesis and secretion, respectively, whereas in others the hormone only stimulates the cAMP pathway, which there enhances synthesis and secretion (348). Similarly, somatostatin uses different cascades and mechanisms to inhibit cell proliferation in different systems (96). The diversity of mitogenic pathways used by GPCRs is even more striking, making any attempt to generalize illusory. Thus physiologists, as embryologists, rightly consider what regulates their system of interest and what the effects of these regulations are and define the elements of the signal transduction pathways black box later. For the physiologist, as well as for the pharmaceutical industry, those mechanisms operating in their cell of interest are most important. Cell specificity in signaling, even in very basic mechanisms, is the key to therapeutic targeting (375).


The authors thank R. Beauwens, G. Rousseau, and G. Vassart for critical reading of the manuscript.


  • The work of the group is supported by the “Service du Premier Ministre Affaires Scientifiques, Techniques et Culturelles SSTC” (PAI), the “Fonds National de la Recherche Scientifique,” “Fonds de la Recherche Scientifique Médicale,” “Fonds Cancérologique Fortis,” “Opération Télévie,” and “Fédération Belge contre le Cancer.”

  • Address for reprint requests and other correspondence: J. E. Dumont, Inst. of Interdisciplinary Research (I.R.I.B.H.N.), School of Medicine, Univ. of Brussels, 808 route de Lennik, B-1070 Brussels, Belgium (E-mail:jedumont{at}

  • 10.1152/ajpcell.00581.2001