The tandem CCCH zinc finger protein tristetraprolin and its relevance to cytokine mRNA turnover and arthritis
© BioMed Central Ltd 2004
Published: 8 October 2004
Skip to main content
© BioMed Central Ltd 2004
Published: 8 October 2004
Tristetraprolin (TTP) is the best-studied member of a small family of three proteins in humans that is characterized by a tandem CCCH zinc finger (TZF) domain with highly conserved sequences and spacing. Although initially discovered as a gene that could be induced rapidly and transiently by the stimulation of fibroblasts with growth factors and mitogens, it is now known that TTP can bind to AU-rich elements in mRNA, leading to the removal of the poly(A) tail from that mRNA and increased rates of mRNA turnover. This activity was discovered after TTP-deficient mice were created and found to have a systemic inflammatory syndrome with severe polyarticular arthritis and autoimmunity, as well as medullary and extramedullary myeloid hyperplasia. The syndrome seemed to be due predominantly to excess circulating tumor necrosis factor-α (TNF-α), resulting from the increased stability of the TNF-α mRNA and subsequent higher rates of secretion of the cytokine. The myeloid hyperplasia might be due in part to increased stability of granulocyte–macrophage colony-stimulating factor (GM-CSF). This review highlights briefly the characteristics of the TTP-deficiency syndrome in mice and its possible genetic modifiers, as well as recent data on the characteristics of the TTP-binding site in the TNF-α and GM-CSF mRNAs. Recent structural data on the characteristics of the complex between RNA and one of the TTP-related proteins are reviewed, and used to model the TTP-RNA binding complex. We review the current knowledge of TTP sequence variants in humans and discuss the possible contributions of the TTP-related proteins in mouse physiology and in human monocytes. The TTP pathway of TNF-α and GM-CSF mRNA degradation is a possible novel target for anti-TNF-α therapies for rheumatoid arthritis, and also for other conditions proven to respond to anti-TNF-α therapy.
The transcript encoding the tandem CCCH zinc finger (TZF) protein tristetraprolin (TTP) was discovered nearly 15 years ago in screens for genes that were rapidly turned on by exposure of cultured fibroblasts to insulin, serum, or tumor-promoting phorbol esters [1–4]. The transcript was rapidly but transiently induced by these and other stimuli. Because of the immediate-early response gene characteristics of this induction, the presence of two nearly identical putative zinc fingers in the predicted protein sequence, and early data indicating nuclear localization, TTP was thought likely to be a transcription factor.
During the succeeding several years, the protein was found to be translocated from the nucleus to the cytosol, and also phosphorylated, by the same agents that stimulated the accumulation of its mRNA [5, 6]. Its gene, labeled ZFP36 in humans (for zinc finger protein 36 ), was found to be highly regulated by known and novel transcription-factor-binding sites in both the classical promoter domain and in the single intron [8, 9]. Two other predicted proteins containing the characteristic TZFs were also identified during this time; their official gene names are ZFP36L1 and ZFP36L2 (for zinc finger protein 36-like 1 and 2) in humans, although the encoded proteins have numerous aliases (see  for a brief recent review).
Despite this progress, essentially nothing was known about the function of this small family of proteins with the unusual, hypothetical zinc fingers until the development of the TTP knockout (KO) mouse . These animals appeared normal at birth but soon developed a phenotype that included an erosive, polyarticular arthritis. A key observation was that this arthritis, as well as most other aspects of the phenotype, could be prevented by treating young mice with weekly injections of an antibody against tumor necrosis factor-α (TNF-α). This observation, and subsequent experiments on the role of TTP in TNF-α regulation, led to speculation that TTP might be involved in some aspects of human arthritis. In this brief review we will discuss the following: first, the characteristics of the TTP KO phenotype; second, recent experiments on genetic modifiers of this phenotype; third, recent data on the nature of the interaction between TTP and the TNF-α transcript; fourth, the current status of known TTP sequence variants in humans; and fifth, potential roles of the TTP-related proteins ZFP36L1 and ZFP36L2. This review is not meant to be comprehensive but instead will focus on our personal experiences with this family of proteins over the past 14 years.
As noted above, there seemed to be minimal, if any, excess prenatal mortality of TTP KO mice . However, the pups soon exhibited a slow-growth phenotype, which ranged from minimal to quite severe. This was associated with marked secondary loss of adipose tissue in all depots, and this cachexia was one of the key findings that led us to suspect a link with TNF-α, formerly known as 'cachectin'. Other aspects of the phenotype included patchy alopecia and dermatitis, conjunctivitis, and the 'kangaroo' hunched posture that is characteristic of many inflammatory syndromes in mice. However, one of the most striking aspects of the external phenotype was the apparent symmetrical arthritis, manifested by reddened, swollen paw joints. This had been seen previously in other TNF-α excess syndromes [12, 13] and was a second major clue that effective TNF-α excess might have a role in the disease pathogenesis.
This histological picture was similar to that seen in other models of TNF-α excess, although in most cases the marrow granulocytosis was either not found or not commented on [12–14]. One exception was the syndrome induced by periodic injections of TNF-α into mice, in which the authors reported a marked increase in marrow granulocytes .
Other pathological characteristics included granulocyte infiltration in the skin, accompanying the loss of subcutaneous fat; and foci of granulocytes in the liver, spleen, lymph nodes and other extramedullary sites. This was accompanied by marked splenomegaly and lymphadenopathy in most cases. In the blood, there was an approximately twofold increase in total white blood cell count; most of the increase could be accounted for by nearly a fourfold increase in Gr-1+ granulocytes, a threefold increase in F4/80+ macrophages, and increases in PK135+ natural killer cells. There were decreases in both B and T lymphocytes. Erythrocyte and platelet counts were essentially normal. There were also marked increases in myeloid progenitors in spleen and peripheral blood, but not in bone marrow, in the KO mice.
Serologically, autoimmunity was present in the form of high titers of antinuclear antibodies that stained in a homogeneous pattern, as well as antibodies against both double-stranded and single-stranded DNA. However, rheumatoid factor titers were normal (both IgG and IgM), as were anti-Sm antibody titers. There were some histological abnormalities of the kidneys, but no increase in IgG or IgM staining of the glomeruli was noted, and no proteinuria or azotemia was found in the KO mice.
Because many aspects of the mouse phenotype resembled previous models of chronic, systemic TNF-α excess, we performed an experiment in which hamster anti-mouse TNF-α monoclonal antibodies were administered to TTP KO mice as soon as they could be genotyped after birth . Remarkably, these injections prevented the development of essentially the entire TTP deficiency phenotype. Specifically, the cutaneous, joint, adipose tissue, and hematological abnormalities were prevented. This led to the conclusion that most if not all of the abnormalities noted in the KO mice were due to chronic 'effective elevation' of circulating TNF-α. This was strongly supported by later studies in which the TTP KO mice were interbred with mice lacking one or both types of TNF-α receptor (see below).
The next studies were aimed at elucidating the mechanism of this apparent TNF-α elevation. An important clue was that the phenotype could be transferred by whole bone marrow transplantation into Rag2 -/- recipient mice, but only after a rather long latent period, suggesting that the phenotype was not transferred with lymphocyte progenitors but instead by more slowly reconstituting cells such as those of the macrophage/monocyte lineage . We then demonstrated that KO macrophages derived from several sources, including fetal liver, adult bone marrow, and peritoneal cavity, all released considerably more TNF-α than did macrophages from their WT littermates . Critically, this was associated with an increase in the steady-state levels of TNF-α mRNA in the KO cells, implying either an increase in TNF-α gene transcription or a decrease in TNF-α mRNA turnover rates, or both.
Subsequent studies in bone marrow-derived macrophages demonstrated that there was a marked decrease in TNF-α mRNA turnover rate in KO cells stimulated with lipopolysaccharide (LPS) and then treated with actinomycin D to inhibit transcription, implicating TTP in the process of TNF-α mRNA turnover . To assess the possibility that TTP could be causing this effect by first binding directly to the transcript, we performed direct RNA binding studies with ultraviolet crosslinking of protein to RNA followed by immunoprecipitation with anti-TTP antibodies, and RNA gel-shift analyses. These studies revealed that TTP did indeed bind directly to the TNF-α transcript, in an AU-rich region long known to be an instability agent for this relatively unstable message, and that the TZF domain of TTP was the RNA-binding domain of the protein .
These studies showed that TTP was an AU-rich element (ARE)-binding protein that destabilized its target mRNA, in this case the mRNA encoding TNF-α. This led us to postulate a TTP-TNF-α feedback loop, in which both TNF-α and LPS stimulated the transcription of both the TNF-α and TTP genes, with the latter protein product feeding back to bind to the TNF-α mRNA and destabilizing it, thus potentially reversing the effects of the initial stimulus and the subsequent positive feedback effect that TNF-α has on its own transcription .
The mechanism of the effect of TTP remains to be determined, but an important clue came from our evaluation of a second mRNA containing a TNF-α-like ARE: that encoding granulocyte–macrophage colony-stimulating factor (GM-CSF) . TTP could bind to this element as well as to the TNF-α ARE. In addition, the transcript from KO bone marrow-derived stromal cells was essentially completely stable after treatment of the cells with actinomycin D, whereas the transcript was unstable in the corresponding WT cells. This indicated that the GM-CSF mRNA was likely to be another physiological target for TTP. Most importantly, however, the pattern of GM-CSF transcripts revealed by northern blotting was extremely informative. In total cellular RNA from the WT stromal cells, the GM-CSF mRNA existed as two species of approximately equal abundance, differing in size by about 200 bases. This size difference was shown to be due to the presence or absence of a poly(A) tail. Strikingly, almost all of the stable transcripts in the KO cells were in the fully polyadenylated form. This meant that the absence of TTP led to a failure of deadenylation, or removal of the poly(A) tail, from the GM-CSF transcript. This is thought to be the first step in vertebrate mRNA degradation [19–21], and it seemed likely that the inability of the GM-CSF mRNA to be deadenylated in the KO cells led directly to the increase in its stability. Conversely, it meant that TTP could be viewed as an mRNA-binding protein whose destabilizing abilities were likely to be due to its ability to promote the process of transcript deadenylation.
In subsequent studies, we developed both cell transfection assays and cell-free deadenylation assays to show that, indeed, TTP could promote the deadenylation and breakdown of mRNAs containing the characteristic ARE-binding sites [22, 23]. This ability of TTP to promote deadenylation was enhanced by increasing the concentrations of poly(A) exonuclease (PARN), either in intact cells or in cell-free assays, suggesting that the two proteins could act synergistically to promote the deadenylation of a relatively small number of transcripts containing characteristic ARE TTP-binding sites . The exact molecular nature of this interaction is the subject of continuing studies, but it should lead us to a molecular understanding of how TTP promotes the deadenylation and destabilization of a specific set of target proteins.
Now that the principal target of TTP relevant to the TTP-deficiency phenotype is known to be the TNF-α mRNA ARE, it is instructive to compare the TTP KO phenotype with others in the literature that have involved TNF-α specifically. In the first such model developed, Kollias and colleagues created transgenic mice in which human TNF-α was overexpressed in T lymphocytes . These animals developed many of the same phenotypes as the TTP KO mice, including wasting and cachexia, premature death, and universal inflammatory arthritis, all of which could be prevented by the administration of an antibody against TNF-α. Another directly relevant model was that created by specifically removing the TNF-α mRNA ARE (ΔARE) from the endogenous mouse TNF-α gene, resulting in a markedly more stable TNF-α mRNA . As expected, this led to the chronic oversecretion of TNF-α due to increased accumulation of the stabilized mRNA, in a manner analogous to that in the TTP KO mice.
In general, the phenotype of this mouse was considerably more severe than the TTP KO phenotype. For example, although these mice were born alive, the KO animals rapidly developed a marked slow-growth phenotype and died between 5 and 12 weeks of age. Pathologically, these mice were characterized by severe erosive arthritis and inflammatory bowel disease, but inflammation in other tissues was minimal. They also exhibited thymic atrophy and blurring of thymic cortical and medullary boundaries, also characteristic of the TTP KO phenotype. The arthritis developed as early as 12 days of age, and resembled the TTP KO arthritis in general character; however, the ΔARE mice exhibited elevated levels of mouse IgG and IgM rheumatoid factor, not seen in the TTP KO mice. The inflammatory bowel disease was severe and universal and affected both the colon and the terminal ileum in some cases. The hemizygous ΔARE mice had a less severe phenotype, but most aspects of the same syndrome were seen with a delayed onset relative to the homozygous mice .
In a striking difference from the TTP KO mice, both the ΔARE homozygous and hemizygous mice developed a severe form of Crohn's-like colitis, a syndrome that has not been seen in the TTP KO mice so far . We have recently shown that TTP protein is relatively highly expressed in normal mouse colon , and its absence from this tissue might be expected to have deleterious effects on this organ. Much remains to be determined about the specific cell types that express TTP in this tissue, which may not be directly relevant to the development of the colitis syndrome. Nonetheless, this remains an interesting difference between the ΔARE mice and the TTP KO mice that is yet to be explained.
We and others have evaluated the TTP-deficiency phenotype further by interbreeding the mice with other potentially informative genotypes. To confirm the data from the TNF-α antibody injection experiments, we interbred the TTP-deficient mice with mice deficient in either or both TNF-α receptors 1 and 2 (TNF-αR1 and TNF-αR2) . As predicted from the antibody experiments, TTP KO mice that were also deficient in both types of TNF-α receptor seemed essentially normal, with none of the cutaneous or joint manifestations of the TTP-deficiency syndrome. Most of the protective effect was due to the TNF-αR1 deficiency; the TTP/TNF-αR1 double KO mice were also essentially normal. Interestingly, the TTP/TNF-αR2 double KO mice were even more severely affected than the parental TTP KO strain, with very severe arthritis and growth retardation in early life leading to premature death in most cases. This supported the concept, developed in other studies, that TNF-α acting through TNF-αR2 can exert a protective effect on the TNF-α-induced inflammatory response in some situations. In any case, these studies confirmed the findings of the TNF-α antibody injection study; it seemed clear that most aspects of the TTP-deficiency phenotype were thus due to chronic, elevated levels of TNF-α.
However, observation of the normal-appearing TTP/TNF-αR1/TNF-αR2 triple KO mice for longer than about 1 year revealed that they were not quite normal. As these animals aged, there was the development of myeloid hyperplasia, particularly intramedullary, which eventually resembled that seen in the original TTP KO mice . This finding, coupled with the known effect of TTP deficiency to stabilize the GM-CSF mRNA and lead to enhanced secretion of this cytokine , suggested the possibility that the myeloid hyperplasia seen in the aging triple KO mice might be due to chronic increases in the concentration of either circulating or local GM-CSF. We are currently attempting to address this possibility by interbreeding the TTP KO mice, and the TTP/TNF-αR1/ TNF-αR2 triple KO mice, with mice deficient in GM-CSF . The first mice of these genotypes are approaching an age at which they can be analyzed, and should yield valuable information: first, the extent to which the myeloid hyperplasia of the original TTP KO mice was due to GM-CSF; second, the extent to which the late-onset myeloid hyperplasia of the triple KO mice is due to GM-CSF; and third, if the myeloid hyperplasia is prevented, these 'quadruple KO' mice might possibly yield valuable insights into any remaining physiologically relevant TTP target transcripts.
Besides interbreeding with TNF-αR1, TNF-αR2 and GM-CSF KO mice, several studies have been performed with other potentially informative genetic backgrounds. These studies are arduous because of the marked subfertility of the TTP KO mice, necessitating that the line be maintained by hemizygotes. Recently, for example, Phillips and colleagues  generated mice deficient in both TTP and another TNF-α synthesis regulator, TIA-1. TIA-1 is thought to inhibit TNF-α production primarily by interfering with the translation of existing transcripts. As expected, the TTP/TIA-1 double KO mice developed more severe arthritis than either genotype alone. An unexpected result of this study was that macrophages derived from the TTP/TIA-1 double KO mice secreted less TNF-α protein than cells from either single KO alone, leading the authors to speculate that the TNF-α secretory apparatus might somehow be interfered with in these macrophages. They also found that bone marrow cells from the TTP KO mice, particularly in combination with the TIA-1 KO, secreted considerable TNF-α in response to LPS, in contrast to the minimal secretion from marrow cells from either WT mice or TIA-1 KO mice. This enhanced secretion was also seen with isolated neutrophils, which accumulate markedly in the marrow and elsewhere in the TTP KO mice. The authors speculated that these neutrophils might represent the primary source of 'arthritogenic' TNF-α in this situation.
Other interbreeding experiments are either under way or have not yet been published. For example, J Rivera-Nieves has crossed the TTP-deficient mice with the TNF-α ΔARE mice alluded to above (personal communication). This is a potentially interesting cross because the TNF-α ΔARE mice develop a severe, Crohn's-like colitis, whereas the TTP-deficient mice have not been observed to do so, at least on the C57Bl6 background. Another interesting cross is the TTP-deficient mice interbred with a mouse in which TNF-α cannot be secreted because its biosynthetic cleavage is prevented, leading to the accumulation of cell-associated TNF-α but an absence of secreted TNF-α (G Kollias, personal communication). A third current study involves crossing the TTP KO mice with mice deficient in the p38 mitogen-activated protein (MAP) kinase target MK2, which seems to be involved in TNF-α mRNA translation and TTP post-translational modification (see below; M Gaestel, personal communication). Other studies with potentially interesting genetic modifiers are continuing, and the next several years should yield several interesting insights into regulatory molecules and pathways. In addition to these crosses into specific knockout mice, we are crossing the TTP deficiency genotype into other genetic backgrounds in the hope of identifying other modifiers.
Since the pioneering work of Shaw and Kamen , it has been clear that certain AREs in mRNAs can confer instability on those transcripts. These have been categorized more recently into classes based on the presence or absence of AU substructures, such as the AUUUA pentamer. The TNF-α and GM-CSF AREs, so far the only definite physiological targets for TTP, fall into the type 2 ARE of Chen and colleagues , or the type 2a of Willusz and colleagues , in which there are multiple copies of tandem and overlapping AUUUA pentamers.
Several recent studies have characterized a more specific TTP-binding site. In the first, Worthington and colleagues used the SELEX procedure to identify the nonamer 5'-UUAUUUAUU-3' as the optimal site from a random collection of oligonucleotides . Using a synthetic TZF domain peptide derived from the human TTP, known as TTP73, our group independently showed that this same nonamer sequence was optimal for TTP binding on gel-shift analysis, which could be accomplished with binding affinities in the low nanomolar range . In a heteronuclear single-quantum coherence nuclear magnetic resonance (NMR) analysis, we found that binding of the TTP peptide to target RNAs caused a major conformational shift in the resonances of the first zinc finger in the peptide, while simultaneously permitting the demonstration of second zinc finger resonances that had been apparently unstructured in the absence of RNA. Using progressive shortening of a longer, TNF-α ARE-based mRNA target, we found that the characteristic NMR resonance shift occurred in an identical manner down to the 5'-UUAUUUAUU-3' nonamer; however, loss of a single base in an octamer showed a deterioration of the NMR resonance pattern, and still shorter oligonucleotides were ineffective. These data suggested that the nonamer 5'-UUAUUUAUU-3' was the minimal complete binding site for the TTP peptide. This study also demonstrated that in longer AREs containing several repeats of this sequence, as are found in the TNF-α and GM-CSF AREs, several tandem molecules of the TTP73 peptide could bind to a single RNA strand, leaving open the possibility of multiple occupancy of the longer ARE by mature TTP protein molecules. The effect of these multiple tandem nonamer repeats on the effectiveness of TTP in promoting mRNA deadenylation is the subject of a continuing study.
Very recently, Hudson and colleagues determined the structure of the analogous TZF domain peptide from the TTP-related protein ZFP36L2 (TIS11D) in complex with this RNA nonamer . This novel structure confirmed the conformational change in the peptide after RNA binding, and revealed many interesting aspects of the interaction between the TZF domain and RNA. For example, each of the two zinc fingers formed very similar structures, each of which contacted an RNA 'half-site' 5'-UAUU-3', with the binding apparently mediated by electrostatic and hydrogen-bonding interactions. The binding is further influenced by 'stacking' of the conserved aromatic amino acid side chains with the RNA bases. The 'lead-in' motifs R(K)YKTEL of each finger also participate through hydrogen-bonding interactions with the 5'-most bases on each half site. Importantly for future informatics analyses, the 5'-most U of the 5'-UUAUUUAUU-3' nonamer was disordered in the structure, which might allow base substitutions at this position in physiologically relevant RNA-binding sites, as well as less conservation in the protein sequence at the protein face near this base. The structure of this RNA-binding domain was thought to be unlike any previously published structure.
One of the striking aspects of this model is the residues within the TTP TZF domain peptide that are identical to those in the ZFP36L2 peptide, which are shown in dark blue in Fig. 4. It is apparent that essentially all of the amino acid residues that are in direct contact with the RNA are identical to those in the ZFP36L2 peptide, whereas those residues that are progressively less well conserved (ranging in order of conservation from aquamarine [best conserved] through green, yellow, and orange [least conserved]) are far removed from the RNA contact surfaces. This model suggests that the RNA contact domain of the TTP peptide is likely to be identical in structure to that of ZFP36L2 (TIS11D), and that the differing residues on other faces of the peptide might be involved in potential specificity determinants, such as protein-protein interactions. The one potentially interfering interaction in this model is between the 5'-most U residue and the green amino acid in the upper left corner; however, from the data of Hudson and colleagues , this initial U is unstructured in the complex and thus can assume other conformations than that pictured, to prevent interfering interactions with the peptide.
Other aspects of this model fit some of our experimental observations. For example, it is possible to mutate the middle A (U5) to C while retaining some TTP binding and deadenylating abilities (WSL and PJB, unpublished data). This fits with the apparent position of U5 near a 'hole' in the middle of the structure shown in Fig. 4, which represents part of the inter-finger linker region. Conversely, mutations of either of the two A residues in this sequence (A3 and A7) were not well tolerated, and they are critical in the structure to stacking interactions with hydrophobic residues in the peptide. Mutations within these critical hydrophobic amino acids were also not well tolerated in terms of RNA binding and promoting mRNA deadenylation . It will be interesting in the coming years to extend these analyses to other residues in the TZF domain, particularly in the highly conserved inter-finger linker region, which contains two basic residues thought to be critical for the nuclear import of the peptide .
The situation is somewhat different with the GM-CSF transcript, because the common mammalian pattern is to have three densely overlapping nonamers (Fig. 5b). The pig is an outlier in this case, with three more widely spaced nonamers. However, GM-CSF also has a completely conserved 5'-UAUUUAU-3' heptamer, with miscellaneous bases in the 1 and 9 positions in the nonamer. On the basis of the structural features of the TZF-RNA complex, it seems possible that some variability in positions 1 and 9 can also be tolerated, and that this more 5' sequence in the GM-CSF mRNA might represent a physiologically relevant TTP-binding site.
From the point of view of identifying potential targets for these proteins by bioinformatics approaches, it seems that multiple copies of the canonical nonamer are optimal but that some base and length differences might be tolerable. For example, Brewer and colleagues  have recently shown that changing the 'inter-A' sequence from UUU to UU or UUUU changes the dissociation constant (K d) of the TTP73 synthetic peptide bound to synthetic RNA oligonucleotides from 3.2 nM to 18 or 6.4 nM, respectively, both close enough to the WT K d to make them possibly physiologically significant. However, just because an mRNA sequence meets these criteria does not necessarily make it a physiologically relevant target. There are many such potential TTP target AREs in the database , and TTP can be shown to bind to and deadenylate a variety of ARE-containing mRNAs in overexpression studies. In our view, the key criterion for demonstrating that a particular mRNA is a genuine target of TTP or one of its family members is a demonstration that the stability of the transcript is increased in cells deficient in the protein. At the time of this writing, only the TNF-α and GM-CSF transcripts have met this criterion and can therefore be considered true and validated physiological TTP targets.
TTP was initially discovered because of the rapid and marked inducibility of its mRNA in fibroblasts in response to insulin, phorbol esters, and serum. This rapid and profound induction is clearly one of the means by which cellular TTP levels are regulated. As with many genes of the 'immediate early response' type, TTP mRNA is quite unstable, and in many cell types it returns to basal levels only a few hours after stimulation, despite the continued presence of the original stimulus in the culture medium. Considerable work has been done on the TTP promoter and its mitogen- and cytokine-responsive enhancer elements, and it is clear that there are apparent transcription-factor-binding sites within the promoter and single intron that are involved in this rapid and transient transcriptional response, but the binding proteins remain to be identified .
In contrast to this rapid but transient induction of the TTP mRNA in response to various stimuli, the protein seems to be surprisingly stable. In a recent study from our laboratory in which a TTP antiserum was used to probe western blots from mouse RAW264.7 macrophages after stimulation with LPS , the protein was almost undetectable in normally growing, unstimulated cells. This was compatible with parallel studies in normal mouse tissues, in which extraordinary amounts of tissue protein were needed in western blots to be able to detect immunoreactive protein. However, within 30–60 min of stimulation with LPS, readily detectable protein began appearing in the cytosol, reaching a peak value after about 2–4 hours. Surprisingly, despite the relative lability of the mRNA in most cell types, the protein remained very stable in the cells for many hours thereafter; it was still much higher than baseline levels after 6 hours, and was still readily detectable after 24 hours. Therefore, de novo biosynthesis of protein is clearly a major mode of TTP regulation, with manyfold increases occurring within a few hours after stimulation with LPS. Nonetheless, the reversal of this process seems to be rather slow.
Another interesting aspect of the protein accumulation experiments is that, as the protein was being synthesized and accumulated in the cellular cytoplasm, its apparent molecular mass on SDS gels continued to increase in an incremental manner, compatible with a stoichiometric increase in its phosphorylation. This occurred more slowly than the generally accepted model of protein phosphorylation, in which a previously synthesized protein is acted upon within a few minutes by a newly activated protein kinase. Possible mechanisms for this slow but profound increase in phosphorylation include the following: first, the protein was initially in the wrong cellular compartment to serve as a kinase substrate, and only became accessible after a lag of 1–2 hours; second, the protein was protected from phosphorylation by binding proteins or even intramolecular folding events; third, the protein was phosphorylated by one or more protein kinases that were either slowly activated by LPS or whose biosynthesis was also stimulated by LPS; fourth, the phosphorylation of the protein was increased by the activity of constitutively active cellular protein kinases in the setting of phosphatases that were either inactivated or destroyed in response to stimulation with LPS; or fifth, some combination of these events.
Elucidation of these mechanisms is of increasing interest because the effect of protein phosphorylation on TTP behavior has begun to be elucidated in a flurry of recent studies. In our initial report on TTP phosphorylation , we demonstrated that both major bands of phosphorylated TTP expressed in intact fibroblasts, with or without stimulation with fetal calf serum, contained only phosphoserine. We identified a single major serine phosphorylation site in that study (S220 in the mouse, corresponding to S228 in the human protein) that was a substrate for MAP kinases both in intact cells and in cell-free systems. In both cases – that is, in intact cells and in cell-free assays – this phosphorylation event changed the migration behavior of the protein in SDS gels, compatible with a stoichiometric increase in phosphorylation. However, because nothing was known about TTP's physiological function at that point, we were unable to conclude anything about the effect of this phosphorylation event on protein behavior except to say that the absence of this phosphorylation site did not seem to affect TTP's ability to translocate from the nucleus to the cytosol in response to stimuli such as serum.
Several more recent studies have added to our knowledge of TTP phosphorylation and its effects on protein behavior. In one study we found that TTP could serve as a substrate for the p38 protein kinase and that global dephosphorylation of TTP in a cell-free system with alkaline phosphatase seemed to increase its binding affinity for an ARE-containing RNA substrate . Clark and colleagues  demonstrated that TTP could serve as a substrate for the MAP kinase-activated protein kinase 2 (MK2), although these authors did not detect direct phosphorylation of TTP by the p38 kinase. Johnson and colleagues  demonstrated that TTP could bind to the cellular protein 14-3-3, and that this binding could be influenced by the phosphorylation status of the protein, specifically at serine 178 in the mouse protein (corresponding to serine 186 of the human protein). They also demonstrated that the binding to 14-3-3 protein promoted the localization of TTP to the cellular cytoplasm. Very recently, Chrestensen and colleagues  identified the same serine 178 as a major phosphorylation site for MK2; they also identified serine 52 as another major site and showed that these two phosphorylations created functional binding sites for 14-3-3. Finally, Stoecklin and colleagues  showed that phosphorylation of TTP by MK2 on serines 52 and 178 led to 14-3-3 binding, which in turn led to the exclusion of TTP from arsenite-induced stress granules. Although the 14-3-3 interaction still permitted binding of TTP to RNA, it inhibited the TTP-dependent degradation of ARE-containing RNA substrates.
Taken together, these data suggest a model in which TTP activity against ARE-containing mRNAs can be modulated by the activation of p38 and the MK2-induced phosphorylation of TTP, leading in turn to 14-3-3 association and cytoplasmic sequestration as well as the inhibition of mRNA degradation. The relevance of direct p38 phosphorylation is unclear, although some of the minor sites identified by Christensen and colleagues  might have been sites for the p38 kinase. It will be interesting to determine how this model works in intact organisms; this should now be capable of investigation, given the availability of MK2-deficient mice . In fact, as pointed out by Mahtani and colleagues , mouse spleen cells from MK2-deficient mice seemed to express normal levels of TNF-α mRNA, which exhibited apparently normal stability after LPS stimulation .
The studies indicating association of TTP with 14-3-3 protein highlight another possible mode of regulation of TTP activity, namely association and dissociation from binding proteins. In addition to 14-3-3, previous studies have identified interactions of TTP with TIA-1  and the nucleoporin CAN/Nup214 ; in the latter study, the interaction again seemed to regulate the intracellular localization of TTP. Functional studies have suggested interactions with the nuclear export protein CRM1 , although direct binding interactions were not demonstrated in that study. Chen and colleagues  demonstrated an association of TTP with components of the exosome, suggesting a role for this structure in the degradation of deadenylated ARE-containing mRNAs. Finally, a recent two-hybrid analysis by our group (Blackshear PJ, unpublished data) identified a large number of previously unidentified potential interacting proteins, each of which needs to be painstakingly validated by other interaction methods. We expect to see the identification of many such interacting proteins in the immediate future, and many of these binding events will undoubtedly have effects on the cellular physiology of the protein.
It is possible that severe mutations in TTP coding sequences could prevent or decrease the expression of mature transcripts, interfere with splicing of the single intron, or lead to frame-shift or stop-codon mutations. These could have major effects on TTP's ability to bind to TNF-α transcripts and destabilize them. Linkage studies with less severe TTP polymorphisms could provide insight to the treatment and/or diagnosis of human disorders associated with excess TNF-α, such as rheumatoid arthritis or Crohn's disease. Although such linkage studies are yet to be completed, the first step in performing them is to identify polymorphisms. We have taken this first step and identified several single-nucleotide polymorphisms (SNPs) in the human TTP gene, ZFP36 .
Polymorphisms in ZFP36 encoding human tristetraprolin
Amino acid change
Variant allele frequency in EGP subjects (%) (n = 92)
Variant allele frequency in ESTs (%)
Polymorphisms in the promoter region and single intron of TTP are of particular interest because both of these regions are necessary for the proper regulation of expression [8, 9]. Four polymorphisms were identified in the promoter region. The polymorphism ZFP36*2 at base 359 in the promoter was the most common SNP identified; it was present at 47% in the EGP subjects. The other three promoter SNPs were ZFP36*1 (found in 1.8% of the EGP subjects), ZFP36*3 (3.1%), and ZFP36*4 (0.6%). Two intronic polymorphisms were also found in this group. ZFP36*5 and ZFP36*6 were present in the 92 EGP subjects at frequencies of 0.5% each.
Six polymorphisms were identified in the protein-coding domains by resequencing; all were in exon 2. Three of the SNPs were identified by resequencing the EGP subject DNA. These SNPs were present at frequencies of 0.6% for ZFP36*7, 6.2% for ZFP36*8, and 4.2% for ZFP36*9. Of these three SNPs, only ZFP36*7 resulted in an amino acid change from proline to serine. Because this represented a non-conservative change, the frequency of ZFP36*7 was determined in 422 North Carolina subjects of varying ethnicities; it was found in 2.4% of this population .
Three other potential protein-coding-domain polymorphisms were identified by EST searches. One of the SNPs, ZFP36*11, was found in 4.2% of the ESTs examined. This SNP resulted in a non-conservative amino acid change from glycine to cysteine. Although this SNP is within the TZF RNA-binding domain (in the 18 amino acid 'linker' between the two zinc fingers), it had no detectable effect on the RNA-binding ability of TTP in a cell-free gel-shift assay. The other two SNPs, ZFP36*12 (present in 4.1% of ESTs examined) and ZFP36*13 (11%), did not produce an amino acid change.
An ARE in the 3' untranslated region (UTR) of TTP confers instability on the mRNA, as determined by deletion studies (WSL and PJB, unpublished data). This region was also examined for polymorphisms. Two polymorphisms, ZFP36*10 and ZFP36*15, were identified within the 3' UTR. ZFP36*10 was identified by resequencing the EGP subject DNA. This polymorphism was the second most frequently occurring polymorphism (frequency 7.6% in EGP subjects). ZFP36*10 lies within a region of the transcript that is highly conserved among mammals and therefore might be significant in terms of altering TTP mRNA stability. The other 3' UTR polymorphism, ZFP36*15, was identified only by EST searches (present in 7.3% of the ESTs examined). Both of these SNPs are potentially interesting in that they lie with a region of the TTP mRNA 3' UTR that probably contributes to the stability of this message.
Since the completion of the initial resequencing study, we have embarked on a more extensive study in subjects with various potentially related diseases. These include subjects with TNF-associated periodic inflammatory syndrome (TRAPS); bronchial hyper-responsiveness to inhaled endotoxin; rheumatoid arthritis, both responsive and resistant to anti-TNF-α therapies; various subgroups of juvenile rheumatoid arthritis; psoriatic arthritis; and multiple sclerosis. If these subjects are added to the original 92 members of the EGP participants, then we will have resequenced the gene from a total of 507 subjects. Although analysis of these samples is not yet complete, we have identified an additional 20 potential SNPs, making the total number discovered so far about 35. Current efforts include assembling the known SNPs into haplotypes for association studies, determining SNP and haplotype frequency in a large number of normal subjects, and a biochemical examination of the SNPs and haplotypes for effects on the biosynthesis, activity and stability of the TTP protein itself.
It is of interest in this regard that long-term follow-up of the TTP hemizygous mice has shown that several mice older than 1 year have developed the full-blown TTP deficiency phenotype, apparently in a stochastic and unpredictable manner. This suggests that TTP hemizygosity in humans might be compatible with a completely normal life, but that in some cases the relative TTP insufficiency might lead to disease, perhaps in response to some environmental perturbation. We expect that a full-blown, autosomal recessive TTP deficiency syndrome would be severe and fatal in childhood in humans, but so far there are no known instances of such a condition in humans.
For Zfp36L1 in mouse, we have recently shown that its complete deficiency leads to universal intrauterine death of the KO embryos, usually at about embryonic day 9–11 . In most cases there was failure of chorioallantoic fusion, which undoubtedly leads to the death of the embryos. In the minority of cases in which fusion occurred, there seemed to be secondary failure of the placenta, leading to poor perfusion of the embryo, runting, widespread apoptosis, neural tube defects, and death. The presumed stabilized transcripts that lead to these abnormalities are not known, but they are the subject of continuing study in the laboratory. This embryonic lethal phenotype means, however, that elucidation of the physiological function of the Zfp36L1 protein in cells and tissues from adult mice will require conditional KO strategies.
For Zfp36L2 in the mouse, our attempts to create a conventional KO mouse led to a mouse in which a transcript is produced that contains a significant portion of the single intron as well as the complete second exon, apparently driven by the endogenous promoter (despite the presence of the neo gene in between) . The result is that the tissues and cells of this mouse produce a protein that lacks the amino-terminal 29 amino acids and is expressed at a variable fraction of endogenous protein levels in different cells and tissues. Despite this unsatisfactory situation, a mouse was produced that has a very specific phenotype: complete female infertility of the KO mice, apparently due to arrest of the embryo after the two-cell stage. This seems to be a maternal effect, because KO embryos from heterozygous mothers seem to develop entirely normally. This is an unusual phenotype and seems to implicate Zfp36L2, and particularly its amino terminus, in maternal aspects of the earliest stages of embryonic development. The development of more conventional and complete KO mice is currently under way.
As part of our evaluation of ZFP36 polymorphisms in humans, we have evaluated the transcript levels for all three family members in one human cell type, purified and cultured monocytes prepared from normal subjects and subjected to stimulation with LPS. It is difficult to compare expression levels of different proteins or transcripts by immunological and northern blotting procedures because of differences in antibody or probe affinity and for other reasons. However, with the use of real-time polymerase chain reaction (PCR) it is possible to make quantitative comparisons between different transcripts by using primer and probe sets that are carefully matched for PCR amplification efficiencies and fluorescence intensities. Using this technique, we have compared the expression profiles and levels of transcripts encoding TTP, ZFP36L1, and ZFP36L2 in normal human monocytes stimulated with LPS, to determine the approximate percentages of expression of each transcript in the control and stimulated states.
Purified human monocytes (from Dr Keith Hull and Dr Dan Kastner, National Institutes of Health) were treated for various times up to 24 hours with LPS (1 ng/μl) or phosphate-buffered saline (PBS) as a control. The cells were harvested, and the RNA was extracted, treated with DNAse, and reverse transcribed to cDNA; then 5 μl of the cDNA was subjected to Taqman analysis (Applied Biosystems 7900 instrument) using primers and fluorescently-labeled probes (Applied Biosystems Assays On Demand) specific for each of the genes of interest. The primer/probe sets were previously validated to detect only the gene of interest and to have similar PCR amplification efficiencies and fluorescence intensities, as determined by experiments showing that equivalent C t values were obtained with the same cDNA copy number for each gene and primer/probe set; this allowed comparisons to be made of the relative expression levels of each of the transcripts.
Initially, all data were expressed as a fraction of maximal expression, set at 1.0. Stimulation of human monocytes with LPS caused a rapid increase in the accumulation of TTP mRNA, reaching about 16-fold that in the control by 60 min, then rapidly declined again (Fig. 8a). This is similar to the profile seen in mouse macrophages in response to LPS .
Analysis of ZFP36L1 transcript expression patterns in the monocytes after stimulation with LPS showed a similar pattern to that of TTP mRNA, but the peak value was not reached until 90 min, and the return to baseline was slightly slower (Fig. 8b). In addition, ZFP36L1 transcripts did not decrease to basal levels, even after 24 hours. Although the temporal sequence of transcript accumulation was similar to that of TTP, the maximal increase at 90 min was only about 4.4-fold the control, unstimulated values, compared with the 16-fold increase with TTP.
Analysis of ZFP36L2 mRNA levels after LPS stimulation in the same samples exhibited an expression profile similar to that of TTP mRNA (Fig. 8c). Again, the overall fold increase in expression was much less than that of TTP, with transcript levels increasing to less than threefold that of the control at the maximal time point (Fig. 8c).
These data have important implications for potential treatments directed at TTP specifically as an approach to anti-TNF-α treatment. In resting, unstimulated monocytes, each of the family members may contribute approximately equally to the turnover of the ARE-containing target transcripts, whereas the TTP effect may become predominant after stimulation of the innate immune system. Interfering with TTP itself, for example by completely inhibiting its biosynthesis, might have less effect than expected because of partial compensation by other family members expressed in the same cell. From the TTP KO mouse experiments, there was no apparent compensatory increase in the expression of the other two family members in cells and tissues from the KO mice . Nonetheless, their presence at approximately equal molar concentrations suggests that they might well contribute to normal rates of mRNA deadenylation and stability in physiological circumstances. Many other factors could modify this conclusion, including major differences in the translation of the transcripts, differences in subcellular localization, differences in post-translational modification, and differences in associated binding proteins.
Nonetheless, these findings might help to explain why the phenotype of the ΔARE mice is so much more severe than that of the TTP KO mice . In the homozygous ΔARE mice there would be no target ARE in the TNF-α transcript for any of the three family members to bind to, whereas in the TTP KO mice the other two family members could bind to the TNF-α ARE and decrease its stability. As the other family members are knocked out, presumably by using conditional KO strategies, it may be possible to isolate macrophages deficient in one, two, and three family members to determine the potential additive effects of each member on TNF-α transcript stability. It will also be important to explore the expression of the various family members in diseased tissues of humans and mice, to determine whether certain cell types in, for example, the inflamed joints of rheumatoid arthritis might be overexpressing local TNF-α and other inflammatory cytokines because of a relative local insufficiency of TTP or its family members.
Although more than 6 years elapsed between the cloning of the cDNA for TTP and the discovery of its role in regulating TNF-α expression, the years since that connection was made have yielded many insights into the functions of this fascinating family of proteins. This work by many groups has culminated most recently in the striking structure of the TZF domain in complex with the nine-base ARE target . Can this interaction represent a novel target for anti-TNF-α therapies? It has been difficult to pursue studies of this type so far, because one would ideally be searching for small molecules that would mimic or potentiate the binding of TTP to its TNF-α ARE target, resulting in mRNA destabilization and decreased TNF-α secretion. Such molecular targets might also be difficult to use therapeutically, because agents that acted like TTP, or increased TTP's ability to destabilize TNF-α mRNA, might be expected to have similar effects on the GM-CSF mRNA in other cell types, perhaps with deleterious effects on GM-CSF functions. Conversely, inhibitors of this interaction might be expected to increase GM-CSF secretion, perhaps a beneficial response in neutropenic states, but at the same time have the potentially harmful side effect of inhibiting TNF-α mRNA degradation. Nonetheless, we view these approaches as potentially useful, particularly with the development of convenient, low-volume fluorescence assays for the binding of the TZF domain to RNA substrates [31, 38]; these could conveniently be adapted to high-throughput formats.
An alternative approach would be to try to identify small molecules that specifically penetrate macrophages and monocytes and stimulate TTP biosynthesis. It is clear that transcription of the TTP gene ZFP36 is regulated differently from that of the other two family members, and it might be possible to uncover compounds that stimulate its biosynthesis, specifically in macrophages, that would not affect either TNF-α biosynthesis or that of the other two family members. Such a molecule might be a useful prototype or lead compound for a novel approach to anti-TNF-α therapies, which have proven to be so useful in the treatment of rheumatoid arthritis and Crohn's disease in the past several years .
Environmental Genome Project
expressed sequence tag
granulocyte–macrophage colony-stimulating factor
nuclear magnetic resonance
polymerase chain reaction
tumor necrosis factor-α
tumor necrosis factor receptor
tandem CCCH zinc finger
zinc finger protein 36-like 1
zinc finger protein 36-like 2
We are grateful to Joe Krahn for help with the peptide modeling, to Peter Wright for communicating the TIS11d structure pdb file before publication, to Keith Hull and Dan Kastner for the human monocytes, and to many other colleagues and collaborators.