The JAK-STAT Pathway at Twenty

Abstract

We look back on the discoveries that the tyrosine kinases TYK2 and JAK1 and the transcription factors STAT1, STAT2, and IRF9 are required for the cellular response to type I interferons. This initial description of the JAK-STAT pathway led quickly to additional discoveries that type II interferons and many other cytokines signal through similar mechanisms. This well-understood pathway now serves as a paradigm showing how information from protein-protein contacts at the cell surface can be conveyed directly to genes in the nucleus. We also review recent work on the STAT proteins showing the importance of several different posttranslational modifications, including serine phosphorylation, acetylation, methylation, and sumoylation. These remarkably proficient proteins also provide noncanonical functions in transcriptional regulation and they also function in mitochondrial respiration and chromatin organization in ways that may not involve transcription at all.

Scientists who have followed the development of the signal transduction field over the past 30 years are likely to be aware that the two authors of this Perspective have had a close working relationship (ten joint papers published during the 1990s) as well as a warm friendship, especially as the understanding of the JAK (Janus kinase)-STAT (signal transducers and activators of transcription) pathway emerged. However, we each came to the study of the action of interferons (IFNs), the subject that uncovered the JAK-STAT pathway, in our separate ways. Furthermore, during the 1980s we were both studying gene induction by IFNs with sometimes parallel and other times quite divergent aims. We did frequently and freely discuss our plans and aims during these years, which were obviously sometimes almost identical. But by the late 1980s, our labs were emphasizing different immediate goals. Experiments in the Darnell lab featured the purification of site-specific, presumably regulatory DNA binding proteins that finally unearthed the STAT family. The Stark group, working closely in association with the group of Ian Kerr at the Imperial Cancer Research Fund in London, turned in the late 1980s to genetics of human cells in culture, with the ultimate goal of identifying all the major components of IFN-dependent signaling. In particular, Sandra Pellegrini, a postdoctoral fellow, developed a method to select mutant cell lines, followed by complementation (by cellular fusion) that allowed identification of eight different noncomplementing mutant cell lines, each of which was no longer capable of responding to IFN-α and IFN-β (IFN-I), IFN-γ, or both. Molecular complementation of these mutants uncovered the role of the JAKs in STAT activation and later also proved the steps in STAT activation and the imputed role of the STATs in IFN-I- and IFN-γ-dependent transcriptional activity.

In this Perspective, we each wish to speak in our own voices about how the early discoveries (1984–1994) came about (Table 1). Our separate historical sections are indicated. We then reunite to describe the completion of the catalog of all the JAKs and STATs, followed by an outline of the basic diagram of JAK-STAT functions, many accomplished by joint efforts with the mutant cell lines mentioned above. A final section features important recently recognized aspects of STAT function not covered in other reviews in this issue.

Table 1. Sequence of the Discoveries that Have Revealed the JAK-STAT Pathways.

Year	Milestone
1957	Isaacs and Lindenmann describe interferon
1975–1977	Oligonucleotide [2′-5′ oligoadenylates(s)] inhibitors of protein synthesis induced by IFN found
1979	Actinomycin-sensitive IFN-β-dependent new protein induction shown
1984	IFN-α-induced transcriptional stimulation of specific genes (ISGs) demonstrated; no new protein synthesis required
1986–1988	IFN-dependent promoters identified (ISREs, interferon-stimulated response elements)
1988–1989	IFN-α-induced ISRE binding protein complexes (ISGF3; E complex) in cytoplasm in 1–2 min; in nucleus in 5 min
1989	Genetic selection system for defective IFN-induced transcription described and first cell mutant selected
	IFN-γ-dependent promoters (GAS, gamma IFN-activated sequences, and GAF, gamma IFN-activated factor) identified
1989–1991	JAKs 1 and 2 and TYK2 identified
1990	ISGF-3 partially purified; identified subunits 113, 91, 84, 48
1991	Noncomplementing mutant cells unresponsive to both IFN-α and IFN-γ described
1992	cDNA clones sequenced later called STAT1 (a and b) and STAT2; RNA for IRF9 completing make up of ISGF3, establishing STAT family of proteins
	First IFN response mutant identified as Tyk2 by molecular complementation
	Upon IFN activation by IFN-α STAT1 and STAT2 are tyrosine phosphorylated; STAT1 also tyrosine phosphorylated after IFN-γ treatment
1993–1994	Major signaling events driven by IFN and IL-6 pinpointed by molecular complementation of mutant cells
1994	JAK3 described and sequenced
1994–1995	STAT3, 4, 5A, 5B, and 6 all described and sequenced
1995–1998	Functional and structural domains of STATs described
1996	Drosophila STAT (dSTAT92E) first described; later studied extensively genetically
	Mouse genetics identifies physiological functions for all STATs in various specific cells
1997	Negative regulation of pathways initially characterized
1998	First crystal structures of STATs
2000	Initial information that human mutations in JAKs and STATs and persistent activation of STATs cause disease
	First posttranslational modifications of STATs in addition to phosphorylations noted (methylation, acetylation, etc.)
2001	Comprehensive gene target sets identified

The Discovery of Interferon

The story of the JAK-STAT pathway begins with IFN. In 1957, Alick Isaacs and Jean Lindenmann reported on a phenomenon in the field of “virus interference,” generally used at the time to describe conditions that disrupted virus formation (Isaacs and Lindenmann, 1957; Isaacs et al., 1957). In a little-noticed paper published in French, Nagano and Kokima (1954) had reported earlier that rabbits injected in the skin with vaccinia virus developed a viral inhibitory factor after 5–7 hr that was insensitive to ultraviolet radiation and of relatively small size as determined by skin biopsies (not sedimented in the ultracentrifuge). However, it was not even clear that the infected cells secreted this material. Isaacs and Lindenmann found that influenza virus-infected chick embryo cells produced and released something in the surrounding fluid that would instigate resistance to infection of noninfected cells. They showed that this material was not, so far as they could determine, a “piece” of the originally infecting virus, but was a product released continually for many hours that in turn took a few hours to effect its protection on newly exposed cells. Their tests on its chemical properties and the relative size of the protective product suggested that it was likely to be a small protein, which they named “interferon.” As noted, the original experiments involved chicken embryo cells and influenza virus, but the interferon preparations were also capable of blocking the replication of several other RNA viruses, including Sendai virus and Newcastle disease virus and also vaccinia virus, a DNA virus. The Isaacs and Lindenmann discoveries revealed two remarkable broad conclusions. First, at the potentially practical level, an “innate” host defense system exists, triggered by virus infection, that causes infected cells to produce a substance that can protect cells not yet infected without waiting for antibody production. Second, at the basic level, a secreted, (probably) protein product can profoundly alter within a few hours the fundamental properties of the treated cells. This very important original work was the beginning of JAK-STAT pathway studies. It was 30 years later through studying the induction of specific mRNAs and proteins by IFN that the pathway was revealed.

Early Recognition of the Importance of IFN: The Darnell Contribution

For my part, I first learned about interferon from Royce Lockart, who worked for his PhD degree with Neal Groman at the University of Washington studying virus interference (Lockart and Groman, 1958). I had joined Harry Eagle's laboratory in 1956 and was directed by Eagle to use his advances in cell culture techniques to begin to study the biochemistry of virus infections in a homogenous population of cultured animal cells. When Lockart came to the Eagle lab in 1957, he told me about the interferon papers that were one of the wonders of the world of virology at the time but were beyond the ability of two neophytes like us to study.

Royce turned to work on IFN after his postdoctoral time with Eagle and was hired in the 1960s by E.I. du Pont in Wilmington, Delaware, to form and lead a group working on animal viruses. Several scientists in this new group, including a biochemist named Ernest (Pete) Knight were charged with determining whether interferon or a fragment of interferon might be developed as a commercially useful product. In 1961, my lab at MIT had begun work on cellular (HeLa cell) RNA and had begun the task of determining how eukaryotic mRNA was formed. Pete joined us in 1967 after we moved to the Albert Einstein College of Medicine and worked for a year to learn the primitive tricks of 1960s RNA biochemistry, knowing that IFN production might involve dealing with mRNA. Pete showed that the small 5S ribosomal RNA was not made from pre-rRNA contemporaneously with the 28S and 18S rRNAs (Knight and Darnell, 1967). He also discovered that a 5.8S (we called it 7S) RNA was synthesized as part of 45S pre-rRNA but remained hydrogen bonded to 28S RNA during processing (Pene et al., 1968). I kept in touch with Pete and Royce out of mutual interests and also, by that time, I was a du Pont consultant. These fortunate contacts eventually led me and my colleagues to work on IFN.

By 1980 our laboratory, now at Rockefeller University, had become engrossed in using cDNA libraries from polysomal mRNA to study transcriptional control of eukaryotic mRNA production. For the first time, individual members of cDNA libraries could be used to monitor the rate of synthesis of mRNA (mRNA precursors) from single eukaryotic genes.

We isolated a series of cDNA clones from rat liver mRNA and picked out a dozen or so that were abundant in liver and absent or very scarce in brain, kidney, or spleen (Derman et al., 1981). By using isolated nuclei to monitor chain elongation by [³²P]UTP incorporation, we showed that liver nuclei gave strong transcriptional signals for the liver-specific mRNAs, whereas nuclei from brain, spleen, or kidney did not (Derman et al., 1981; Powell et al., 1984). These were among the most convincing results at the time showing that cell-specific mRNA synthesis depended, at least as the first event, on cell-specific transcriptional control. This so-called run-on transcription assay with nuclei was first documented by Ronald Cox in 1976 (Cox, 1976) to represent short, already engaged RNA polymerase II chain elongation. Furthermore run-on transcription remains the most precise and straightforward means of determining differential rates of transcription of individual genes and would prove invaluable later in studying IFN action.

David Clayton, a Rockefeller University graduate student, showed that isolated cultured hepatocytes still had most of their liver-specific mRNA and produced liver-specific proteins ∼24 hr after culture but stopped transcribing mRNA precursors at the normal rates (Clayton and Darnell, 1983; Clayton et al., 1985). He went one step farther. Thin slices of liver (∼10 cell diameters) placed in culture continued to transcribe virtually normally after 24 hr. Hepatocytes are released as free cells for culture by perfusing the liver with EDTA-containing solution. If David perfused the liver with the disaggregating solutions but, instead of splitting the liver capsule to release individual cells, he reperfused the liver with serum containing culture medium and Mg²⁺ and Ca²⁺ salts, he could now slice the perfused liver without disaggregation and culture the slices, and liver-specific transcription was maintained. Maintaining cell contact seemed to be crucial for maintaining specific transcription.

We decided that to make further progress on transcriptional control, we must find a simple system in which a known extracellular agent, a protein presumably, could be used to initiate changes in specific transcription. By this time, Pete Knight and colleagues had made great progress in purifying IFN-β (one of a group of so-called leukocyte IFNs that also included many related IFN-α molecules). Moreover, Knight and a colleague, Bruce Korant, had shown that IFN-β caused human fibroblasts to produce new proteins (detected by 2D gel separations of radioactively labeled proteins) within several hours that were not present in untreated cells. These proteins were not produced in the presence of the RNA synthesis inhibitor actinomycin, although the majority of cell proteins continued to be made normally (Knight and Korant, 1979).

Pete, his colleagues, and I decided that we would try to identify specific mRNAs induced by IFN-β. With IFN-β purified by Knight and colleagues, Andy Larner in our lab made a cDNA library from HeLa cells with and without IFN treatment (Larner et al., 1984). Several cDNAs not present without IFN were indeed found in the library from the IFN-treated cells. We immediately tested by run-on analysis and found a large increase (>30-fold) in the transcriptional rate of these interferon-stimulated genes (ISGs). Two important points came from these experiments. The transcriptional activation came very quickly (within 15 min) and blocking protein synthesis with cycloheximide before IFN treatment did not affect the transcriptional increase. Thus, the proteins responsible for the IFN-dependent increase pre-existed in the cells and presumably were modified somehow to become active in stimulating transcription.

As related in the accompanying section of this review, George Stark and Ian Kerr and their colleagues (particularly Richard Friedman, a Stanford graduate student in the Stark lab), who were also pursuing the question of IFN induction of genes, did almost exactly identical experiments (Friedman et al., 1984). They in fact made larger libraries than we did and consequently found more IFN-dependent mRNAs than we did. They also showed by run-on analysis that increased transcription occurred for one of their seven IFN-dependent cDNAs.

By this time (∼1986), my lab had had success in identifying genomic DNA elements in genes expressed mainly in the liver that bound activating DNA-binding proteins, so-called transcriptional activators. Moreover, we were beginning to succeed in purifying these proteins and believed, therefore, that we could clone the genes for several of these liver-enriched transcription factors. Therefore, the task of solving IFN-dependent gene activation seemed straightforward—identify and purify the positive-acting transcription factor(s) driving IFN-responsive genes. David Levy had the first success on this path, finding a nucleotide stretch just 5′ to the start site of one of the cloned mRNAs in Andy Larner's collection that conferred IFN responsiveness in a reporter assay (Levy et al., 1986).

Knight and colleagues had purified and partially sequenced a ∼15 kD protein (ISG15, IFN-β-stimulated gene 15), made after IFN treatment of cells, and had obtained a cDNA clone (Blomstrom et al., 1986). With this clone, Nancy Reich in our lab proved that ISG15 was indeed transcriptionally controlled by IFN treatment (Reich et al., 1987). She then cloned the genomic ISG15 DNA and determined the location of an IFN-responsive DNA element. When Levy and Reich compared the immediate 5′ sequences of the two ISG genes, a highly conserved ∼25 base pair region was evident ∼100 nucleotides upstream of the start site for each of these mRNAs. Mutations within these “promoters,” termed interferon-stimulated response elements (ISREs), blocked the IFN-dependent reporter responses.

David Levy, Nancy Reich, and several others in our lab, with these IFN-dependent promoters, began in 1986 the arduous task of determining whether we might first identify and then purify a common protein or protein complex that was responsible for the IFN-α and IFN-β transcriptional increase. They used as a guide the run-on results (Levy et al., 1988). The protein(s) complex should not be found in uninduced cells, should be present within minutes after IFN treatment in the nucleus, and should not require new protein synthesis to be induced.

By this time, workers studying eukaryotic transcription had adopted a technique developed for bacterial extracts in the lab of Don Crothers at Yale by which DNA fragments containing protein-binding DNA regulatory sites could be retarded in electrophoretic migration through polyacrylamide gels by their cognate DNA binding regulatory proteins (Fried and Crothers, 1981). Such colloquially termed “gel-shift” experiments worked for Levy and colleagues with IFN-treated extracts. Several proteins were detected that bound the IFN-dependent promoters, the ISRE, but only one was IFN dependent, formed quickly, and remained during transcriptional increase (Levy et al., 1988, 1989). The factor was first detected in the cytoplasm after a 1 or 2 min IFN-β treatment and then accumulated in the nucleus. The complex appeared to be large because it migrated through a gel much more slowly than other complexes. Another protein complex that bound to the ISRE increased with time after IFN treatment, indicating perhaps that it contained IFN-dependent proteins but was not likely to be the IFN-dependent transcriptional activator. The large induced complex was christened ISGF3 for interferon-stimulated gene factor 3.

An important fact was uncovered that made purification practical. Most IFN-β-induced ISGs were not responsive to IFN-γ, which was under study by Thomas Decker and Danny Lew in our lab (Decker et al., 1989; Lew et al., 1989). But together with David Levy and Dan Kessler, a graduate student, they found that prior exposure of HeLa cells to IFN-γ enabled a much more vigorous subsequent IFN-α transcriptional response (Levy et al., 1990). Additionally, the prior IFN-γ treatment greatly increased the amount of ISGF3, the DNA promoter-binding complex produced in response to IFN-α and IFN-β. Moreover, the cytoplasm of the IFN-γ-treated HeLa cells contained a limiting amount of some component of ISGF3 because such cytoplasm would, when added to cytoplasmic extracts of IFN-α-treated cells, greatly increase the amount of ISGF3. Thus, some portion of ISGF3, but not the whole thing, seemed likely to be an IFN-γ-induced protein. The practical import of these results was that purification was going to be much easier after IFN-α treatment of HeLa cells that had been pretreated overnight with IFN-γ. And HeLa cells grown in suspension were the only cells that we could grow in large enough amounts to make purification an appealing prospect.

Settling on ISFG3 as the target, purification began. X.-Y. Fu, who came to us from Jim Manley's lab where nuclear proteins required for poly(A) addition to pre-mRNA were under study, began the ISGF3 purification (Fu et al., 1990). The best partially purified preparations contained ISGF3 concentrated by more than a hundred fold but were obviously still not pure. However, a tiny bit produced a robust ISFG3 signal. I suggested to Dan Kessler, a graduate student who was working with Fu and others, to try a short cut. Simply use enough of the partially purified extract to give a whopping ISGF3 binding signal detected with the ³²P-labeled promoter fragment, run the mixture on an acrylamide gel, identify the ISGF3 band by autoradiography without drying the gel, cut out the band, elute the proteins, and run a protein-stained acrylamide gel. Bingo—Dan identified clearly four proteins, nominally 113, 91, 84, and 48 kDa. The smallest protein proved to be the HeLa cell protein increased by prior IFN-γ treatment.

Fortunately at this point, Chris Schindler joined the group, regularized the protein purification, and obtained reproducible results with specific oligonucleotide (ISRE) affinity chromatography demonstrating that, after extensive additional purification, the 113, 91, 84, and 48 kDa species were the major proteins remaining and therefore were appropriate targets for sequencing (Schindler et al., 1992a). He made a preparation from 200 l of HeLa cells in which the 113, 91, 84, and 48 kDa proteins were clearly separated after gel electrophoresis. In early 1991, Reudi Aebersold took the isolated proteins and returned tryptic peptide sequences to us within a short while. Peptides specific for the 113 kDa and other peptides shared by 91 and 84 kDa showed that we probably had two isoforms of one protein and one additional large protein plus one considerably smaller 48 kDa protein. From sequences specific for 113, 91, and 84 kDa peptides, oligonucleotide probe libraries were prepared to search cDNA clones for a match. These standard techniques identified such clones and sequence analysis proved that the 113 and 91–84 kDa peptides were likely to be two members of a gene family. The cDNAs for 91 and 84 kDa showed a differential poly(A) choice, producing a different 3′ exon choice that explained their difference in size. By using the original peptide sequences furnished by Aebersold, Schindler prepared antibodies to both the 91 and 113 kDa proteins. (David Levy in his new lab at NYU finished off the complete sequencing of the 48 kDa cDNA [Veals et al., 1992].) With the 91 kDa or the 113 kDa antisera, Schindler showed that all of the proteins in the 113, 91, 84, and 48 complex coimmunoprecipitated. Thus, somehow we had blindly purified a multiprotein complex held together presumably by protein:protein interactions.

Before we had the complete sequences of the cDNAs, Schindler made another important discovery. Cells treated with a broad-spectrum serine kinase inhibitor still activated ISGF3. But staurosporin, which also inhibits tyrosine kinases, completely blocked IFN-α-dependent induction of ISGF3 (Schindler et al., 1992b). Based on this result, Schindler with others, particularly Ke Shuai, a new postdoc, and Vince Prezioso, a graduate student, then demonstrated with ³²P protein labeling and antibody precipitation that the vast majority of the ³²P label in all three proteins was in tyrosine, a big surprise. A single identical tryptic peptide labeled with ³²P was found in the 84 and 91 kDa proteins and a different phosphorylated peptide was present in the 113 kDa protein.

Thomas Decker and Danny Lew (Decker et al., 1991; Lew et al., 1989) had by this time identified an IFN-γ-specific promoter (GAS, gammaIFN-activated site) with a different sequence from the ISRE and a protein that bound to this promoter (GAF, gammaIFN-activated factor). Ke Shuai showed that this protein migrated differently than ISGF3 in a gel. Schindler, Shuai, and others showed that this protein was precipitated with the antiserum against the 91 kDa protein and was phosphorylated. The single tyrosine phosphorylation site was identified as Y701 in the 91 kDa protein, which could be activated by either IFN-α or IFN-γ (Schindler et al., 1992b; Shuai et al., 1992, 1993).

Together with Kerr and Stark, Shuai showed that one of their genetically selected defective cell lines (to be described shortly) could be caused to respond to IFN-γ by transfection of the 91 kDa cDNA clone (Shuai et al., 1992, 1993). At this point, we gave the name STAT1 (signal transducer and activator of transcription) to this protein and STAT2 to the larger 113 kDa family member that was found in ISGF3. The designation of the proteins as STATs came about during a conversation I had with my late wife, Jane, as we were driving from our home into New York and I was explaining our new results. She had heard (patiently) many times about transcription as the copying of genes into RNA and I was describing our newly isolated proteins as “dual function proteins.” These proteins received the activating signal probably at the cell surface that came from a protein outside the cell and carried the news to the nucleus to cause—activate—the gene transcription. Thus they were both signal transducers and then activators of transcription. She asked whether this was a quick act. I said yes. She said that STAT meant quick in a hospital and so—signal transducers and activators of transcription fell out of this conversation. STATs immediately had a pronounceable and accurate acronym name.

At this point Kerr, Stark, and I (Darnell et al., 1994) could describe a plausible tentative conclusion. IFN-α activated two different members of the same gene family, STAT1 and STAT2, that formed a complex with a 48 kDa protein which by that time had been sequenced in David Levy's new independent lab, together with workers in my group. A different receptor-ligand complex, IFN-γ and its cognate receptor, activated STAT1 but not STAT2, illustrating that differential transcription factor activation in response to two different ligands led to the formation of two different transcription complexes.

It is necessary to go back to the earlier experiments of George Stark and Ian Kerr to connect the JAKs to the STATs and complete the story.

Analysis of IFN-Dependent Signaling and Connecting the JAKs to the STATs: The Stark Contribution

I was trained in protein chemistry and enzymology (Stark, 2005) and gradually became more and more interested in eukaryotic cell biology and genetics. These two seemingly unrelated aspects of my research came together when we used PALA, a transition state analog inhibitor of aspartate transcarbamylase that we had synthesized, to block de novo pyrimidine nucleotide biosynthesis in mammalian cells (Kempe et al., 1976). The resistant mutant clones that emerged provided one of the first examples of gene amplification, overexpressing CAD, the three-enzyme protein responsible for the first three steps of the pyrimidine synthetic pathway (Wahl et al., 1979). This experience with selection of mutant mammalian cells came in handy when we were confronted later with the problem of dissecting IFN-dependent signaling. My interest in interferon research stems from my long friendship with Ian Kerr. We first met in 1964, when I was a fresh assistant professor in the Biochemistry Department at Stanford and Ian was a fresh postdoctoral fellow there with Bob Lehman. We had little scientific interaction until the mid-1970s, when Ian's lab made the marvelous discovery that 2′,5′-oligoadenylates (2-5A) were a vital second messenger in the interferon response, leading to degradation of RNA in infected cells and thus to inhibition of viral proliferation (Kerr et al., 1977). In 1977, I was ready for a sabbatical and imposed on our friendship to request that Ian put up with me for a year in his MRC lab at Mill Hill in London. I worked mainly on the 2-5A system, participating in the discovery with others that the 2-5A synthetases are capable of making, in addition to classical 2-5A, an interesting series of related oligoadenylates, the biological significance of which had long been mysterious (Cayley and Kerr, 1982; Reid et al., 1984).

Upon returning to Stanford in 1978, my lab continued for a while to pursue 2-5A biochemistry but gradually became more and more interested in how cells responded to IFNs. A breakthrough came through the efforts of Richard Friedman who, as a graduate student, made a lambda phage library from IFN-induced mRNAs and screened it with cDNA probes made from the RNAs of treated or untreated cells. More probe sequences present in the treated cDNA pools gave stronger signals from the induced clones, and we were thus able to isolate cDNAs corresponding to several strongly induced genes (Friedman et al., 1984). These were creatively named for the plate and clone numbers from the screen: 6-16, 1-8, 9-27, etc. Richard went on to look for homologous sequences in the promoter regions of several IFN-induced genes, recognizing the conserved ISRE (Friedman and Stark, 1985), which was later confirmed in functional experiments (Porter et al., 1988). This work was carried out as a distant collaboration with Ian's lab, and the attraction of working together more closely, in my favorite city of London, together with a significant contribution of midlife crisis, led me to move my lab to the Imperial Cancer Research Fund in London, where Ian had relocated in 1983. Our labs were conjoined and we collaborated extensively and closely during the nine very happy years I spent at ICRF, working out many details of IFN-dependent signaling.

With the cDNAs and promoter constructs corresponding to 6-16, 1-8, and 9-27, Trevor Dale, a graduate student working jointly with Ian and myself, carried out a series of experiments describing the IFN-induced E (for early) factor that bound to ISREs. In summarizing this work (Dale et al., 1989), we stated that “…these results suggest a model for signal transduction in which latent E factor, located in the cytoplasm, is activated or released from an inhibitor very rapidly upon binding of IFN-α to its receptor. Active E factor can then migrate to the nucleus, where it binds to the IFN-stimulated regulatory elements of IFN-regulated genes, activating their transcription.” “All” that remained to be done was to define the nature of E factor and its mechanism of activation. By happy coincidence, Jim Darnell's lab used their biochemical expertise to great advantage to isolate E factor (ISGF3) and show that it was comprised of STAT1, STAT2, and IRF9 (using modern terminology).

We decided to use a genetic approach (see PALA-resistant cells above). At first, we tried to use the growth-inhibitory effects of IFN, particularly strong for certain human cell lines, to isolate resistant mutants, but the clones that we selected were unstable and thus unsuitable for further work. We then took the bull by the horns and initiated a dedicated program to use chemical mutagenesis to create mutant cells in which some protein critical for IFN-dependent signaling was no longer expressed (Pellegrini et al., 1989). Many thought this approach was foolhardy. How could we mutate both alleles of a gene by extensive random mutagenesis, causing a very large number of changes in the DNA, and still have viable cells? We tested the idea by determining the frequency of mutating HPRT, an X-linked single-copy gene, with ICR 191, an efficient frame-shift mutagen. Fortunately, a powerful selection technique was available for cells that did not express HPRT. We found that we could grow a viable population of mutagenized cells from which HPRT-null variants could be recovered at a frequency of about 10⁻⁴, so we could expect to knock out both alleles of a typical gene with the square of this frequency (about 10⁻⁸) and, if 10 proteins were essential for IFN-dependent signaling, to recover a mutant with a frequency of about 10⁻⁷. Thus we forged ahead with a good sense of how many cells we would need to screen. The genetic selection utilized a construct in which the 6-16 promoter, responsive to IFN-α but not IFN-γ (Kelly et al., 1985), drove the expression of guanine phosphoribosyltransferase (GPT), which activates the prodrug 6-thioguanine to 6-thio-GMP, thereby killing cells in which GPT is expressed. If the IFN signaling pathway did not work, no GPT protein would be made in response to IFN and, voila!, mutant cells that could survive would be identified. Sandra Pellegrini, a postdoctoral fellow in the lab, developed this approach. Remarkably, she was able to isolate our first mutant with approximately the expected frequency (Pellegrini et al., 1989)! But it was not easy for Sandra to maintain enthusiasm while screening 10⁸ mutagenized cells over a period of many months. The first clone (again imaginatively named for the plate and clone number) was 11,1 (later designated U1A in a systematic nomenclature). The IFN-regulated GPT construct allowed for selection of revertant clones that were capable of growing in hypoxanthine-aminopterin-thymidine (HAT) medium. Sandra began investigating the 11,1 mutant at ICRF and then, working in her own new lab at the Pasteur Institute, managed the difficult job of complementing 11,1 with genomic DNA. Cosmid clones (large genomic DNA-bearing bacterial clones) were transfected into 11,1 cells to make them IFN responsive. After isolating the cosmid responsible, then piecing together from its sequence a translatable cDNA, we were able to announce in July 1992 that TYK2 was a required protein in the IFN pathway (Velazquez et al., 1992).

TYK2 was already known as a member of a new family of tyrosine kinases for which the substrate(s) had not yet been uncovered. Andrew Wilks and coworkers in Australia identified first one (Wilks, 1989) and then a second (Wilks et al., 1991) of these tyrosine kinases. The molecules had an unusual feature, namely, both a recognizable kinase domain and a second domain with great similarity to kinases, originally thought to be inactive because of amino acid changes at the putative active site. Because of the two domains, Wilks called the proteins Janus kinases 1 and 2, by analogy to the two-faced Roman god Janus. The name has been shortened colloquially to JAKs. (An apocryphal story is that JAK stands for “just another kinase.”) At almost the same time John Krolewski, in Riccardo Dalla-Favera's group, detected and then cloned another JAK family member, TYK2 (Firmbach-Kraft et al., 1990). These discoveries of the kinases were just in time.

Thus, the finding by Schindler and others in the Darnell lab that tyrosine phosphorylation was part of the activation of ISGF3 was determined to be a JAK-dependent event by Velazquez et al. (1992). With this success and the availability of a set of mutant cell lines, the floodgates were opened. Given a complete JAK cDNA (no mean feat at the time for a ∼6 kb mRNA), furnished separately by Andrew Wilks and James Ihle, the Kerr and Stark labs soon showed that IFN-dependent signaling required both JAK1 and TYK2 to successfully tyrosine phosphorylate the two substrates STAT1 and STAT2, in response to IFN-α, and JAK1 and JAK2 to phosphorylate STAT1 in response to IFN-γ (Müller et al., 1993). It was thought likely at the outset that noncovalent association of the JAKs with the IFN receptor (see for example Argetsinger et al., 1993) and cross phosphorylation of two associated JAKs resulted in tyrosine phosphorylation of first the kinases and then the receptor sites to which the SH₂ domains of the STATs were then bound, finally leading to tyrosine phosphorylation of the STATs.

With the experience of obtaining the first mutant, we realized that yet more extensive mutagenesis would increase the frequency of mutation. With the availability of expression constructs for the JAKs, STATs, IRF9, and receptor subunits, we were able to obtain and complement a total of eight different mutant cell lines, each lacking a single protein required for responses to IFN-α, IFN-γ, or both. Although most were isolated by using a drug-based selection, the mutant γ1A, lacking JAK2, was found by sorting cells that failed to activate expression of cell surface markers in response to IFN-γ (Watling et al., 1993). These cell lines made possible very productive collaborations, especially with the Darnell lab, in which many details of the signaling pathways were worked out. The use of cell lines that were mutant for the JAKs followed by mutagenesis of STAT proteins identified the receptor binding sites and the arginine in the pit of the STAT2 binding domain (reviewed by Stark et al., 1998, and Levy and Darnell, 2002). So by late 1993, the three of us, Stark, Kerr, and Darnell, could write a fairly comprehensive and still largely correct description of the IFN-α- and IFN-γ-dependent activation of the JAK-STAT pathway (Darnell et al., 1994). During 1994, strong hints arose that other cytokine-dependent signaling pathways utilized JAKs and STATs. The mutant cell lines have been made available to more than 500 additional laboratories (still counting), serving as invaluable reagents to define the roles of the missing proteins in the many different signaling pathways in which they are now known to participate.

Completing the JAK-STAT Menagerie

The catalog of the JAKs (only four in number) was completed more quickly than uncovering the remaining five STATs. During 1994 at least four or five laboratories identified JAK3 (the fourth and still final JAK to be discovered), which had limited cell distribution, in contrast to JAK1, JAK2, and TYK2 (Johnston et al., 1994; Rane and Reddy, 1994; Takahashi and Shirasawa, 1994; Witthuhn et al., 1994). Also early in the examination of lymphocytic cell lineages it was found that JAK3 mutations underlay long-recognized lymphocytic disease syndromes.

Rounding out the STAT collection began with a flurry of papers identifying phosphotyrosine-containing DNA binding factors that did not react with antisera against STAT1 or STAT2 (Larner et al., 1993; Ruff-Jamison et al., 1993; Sadowski et al., 1993; Silvennoinen et al., 1993). Furthermore, the growth factors EGF and PDGF, as well as other cytokines, including IL-6, IL-3, IL-5, IL-10, and G-CSF, also activated these phosphotyrosine-containing DNA binding proteins. Could all these factors be additional STATs?

The question was solved promptly. Zhong Zhong and Zilong Wen, two graduate students in the Darnell laboratory, using a lymphocyte cDNA library obtained from Michel Nussenzweig, cloned STAT3 and STAT4, establishing in the fall of 1993 by sequence analysis their membership in the STAT family. They showed that tyrosine phosphorylation of STAT3 occurred in response to IL-6 and EGF and, together with Ken Murphy's lab, that IL-12 activated STAT4 (Zhong et al., 1994a, 1994b; Jacobson et al., 1995). At virtually the same time other laboratories, notably Shizura Akira in Tadamistsu Kishimoto's group (Akira et al., 1994), also cloned STAT3, known to be activated by IL-6 and called earlier the acute phase response factor. This factor had been studied intensively by Wegenka et al. (1994), who recognized that the protein bound to a GAS element. Also, David Levy and colleagues, later in 1994, identified STAT3 as a relative of STAT1 (Campbell et al., 1995). Finally, Jim Ihle's group independently cloned STAT4, also in 1994 (Yamamoto et al., 1994).

Other workers in diverse systems around the world were attempting to identify the nature of DNA binding proteins. Bernd Groner followed the transcriptional response to prolactin and steroid stimulation of the β-casein gene in mammary gland tissue. The DNA binding factor of the prolactin response element was purified and the gene cloned and sequenced (Wakao et al., 1994). Jim Darnell was sent Bernd's paper with his colleagues for review only a month or two after Zhong and Wen had finished the sequence of STAT3 and STAT4. It was obvious that Groner and his colleagues had purified and cloned the gene for another STAT family member. He discarded his referee's anonymity and wrote Groner that he had found another family member and he might consider calling it STAT5. Groner demurred, sticking with MGF, mammary gland factor. Eventually the world at large, including Bernd, adopted STAT5. The cast was completed by Alice Mui and colleagues at the DNAX Research Institute in California (Mui et al., 1995), who isolated a distinctly different duplicated gene very similar to STAT5, encoded on the same chromosome. Groner and Lothar Henninghausen, also a bit later, independently cloned the STAT5B molecule (Liu et al., 1995). The two proteins, which became known as STAT5A and STAT5B, are expressed differently in different tissues. The seventh and final STAT (still only seven in mammals) was identified and sequenced at Tularik by J. Hou and U. Schindler and colleagues in the lab directed by Steven McKnight (Hou et al., 1994). STAT6, as this protein became named, was distributed mainly in bone marrow-derived cells (mainly lymphocytes).

Thus, in a cloudburst, essentially finished between 1991 and 1994, the cast of JAK-STAT family members and the outlines of the pathway were complete. Many continuing structural and functional protein studies and a huge number of physiological studies on the proteins of the pathway have been accomplished over the past 17 years (Stark et al., 1998; Levy and Darnell, 2002; Sehgal et al., 2003). The deep involvement of these proteins in many normal cellular functions—growth and development and homeostasis of virtually all tissues, the immunologic response, and acute primary response to infection and inflammation—are all critical functions that can go awry or be improperly regulated in disease. The pathway has become a major target of medical understanding and potential corrective therapy. Details of these subjects seem destined to continue to be uncovered, bringing deeper understanding of disease and, it is widely hoped, rational therapy for many ailments.

Building on History

A brief review of where this work stood by 1994 in the larger context of cell signaling seems in order. First, when we began chasing how IFN stimulated specific gene transcription (in 1984), very little was known about how mammalian cells changed course in response to extracellular polypeptides. It was established that transcription rates could be measured by run-on analysis with cell nuclei and that changes in transcription rates did occur in response to steroids. Of course, steroids entered cells and bound to cognate receptors with resulting increases in specific gene transcription (McKnight and Palmiter, 1979). But no connection between a specific cell surface receptor and a specific extracellular protein contact that resulted in targeted gene transcriptional changes had been established.

Two decades of work before the 1980s had identified the underlying biochemistry that caused fluctuations in internal “second messengers”—cAMP, diacylgylcerol, and phosphoinositides as well as Ca²⁺ ions. These changes were joined in the 1970s and 1980s by hundreds of studies on serine and tyrosine phosphorylation cascades, overactivation of which were attendant to oncogenic transformation. By the early 1980s, second messengers and phosphorylation cascades were invoked as the agents likely to be responsible for regulating transcription controlled in response to extracellular signaling proteins (Berridge, 1987; Bourne and DeFranco, 1990; Gill, 1990; Gilman, 1987; Hunter, 1990).

From the beginning of our interest, it seemed unlikely that the high degree of specificity of IFN:receptor interaction (indeed of any extracellular signaling protein and its specific receptor) followed by specific gene activation in the nucleus would depend on the relatively nonspecific fluctuations in second messengers (Levy and Darnell, 1990). Indeed, this has turned out to be the case. The tool that linked the JAKs to the STATs, namely cell lines mutant in different individual genes, proved that the identified members of the pathway functioned as proposed. The JAK-STAT pathway results were the first to close the loop—specific identified and cloned receptors to which specific ligands attached, activating specific members of an identified and cloned group of kinases, which in turn activated identified specific members of a latent cytoplasmic set of transcription factors that finally, after accumulation in the nucleus, directed increased transcription of specific genes.

The idea of sets of latent transcription factors responsive to sets of specific receptors for the purpose of directing transcription of specific sets of genes raised the question of how many such pathways there were. Although there are always surprises in store, it can be tentatively advanced that there are relatively few (reviewed by Brivanlou and Darnell, 2002).

NF-κB was the first latent cytoplasmic transcription factor to be recognized that required for its activation proteolytic destruction of a protein-bound cytoplasmic inhibitor. But the cell surface receptor systems that brought about the release weren't understood quite so early as the JAKs that respond to cytokines. Other latent cytoplasmic transcription factors followed, including the SMADs and several signaling pathways first identified in Drosophila. These include Notch, a cell surface protein, the cleaved internal domain of which furnishes the DNA binding component of composite transcription factors; the Wnt pathway, which preserves the β-catenins from destruction, which in turn serve as transcription factors; the hedgehog pathway, which preserves the Ci (cubitus interruptus) protein (called the GLI proteins in mammals) to serve as transcription factors; and finally the Tubby proteins, which are anchored at the cell membrane and require release by a lipase to enter the nucleus and act as transcription factors (Brivanlou and Darnell, 2002). The conclusion can be at least reasonably advanced that we now know the major pathways that are responsive to specific protein-protein ligand receptor interactions followed by induction of specific gene sets.

Posttranslational Modifications Other than Phosphotyrosine

As research on the STATs has progressed over the years, it has become clear that these proteins are subject to important posttranslational modifications in addition to tyrosine phosphorylation. Furthermore, some STATs function as transcription factors even without tyrosine phosphorylation, and some carry out important cellular functions independently of transcription. A brief review of these aspects is given here, to provide a flavor of many facets of JAK-STAT signaling that are still very active and important areas of ongoing research.

In 1995, phosphorylation of both STAT1 and STAT3 on S727 was shown in cell culture experiments to be necessary, in addition to tyrosine phosphorylation, for full activation of transcription in response to IFNs and gp130-linked cytokines such as IL-6, respectively (Wen et al., 1995). Experiments in mice in which these serine residues were mutated to alanines extended these cell culture results (Varinou et al., 2003; Shen et al., 2004). A different serine residue of STAT1, S708, is phosphorylated in response to type I IFNs by IKKε, a modification that is needed for ISGF3 to activate a subset of IFN-stimulated genes (Tenoever et al., 2007). More recent work from the Maniatis laboratory (Ng et al., 2011) reveals that S708 phosphorylation inhibits STAT1 homodimerization but not ISGF3 formation, thus regulating differentially the STAT1-dependent signals that are due to type I and type II IFNs. Along similar lines, Perwitasari et al. (2011) show that STAT1 tyrosine phosphorylation is required for STAT1 S708 phosphorylation and that this modification leads to expression of the IFIT2 gene, imparting innate immunity that restricts West Nile virus infection. A JAK- and receptor-independent mode of STAT activation that also affects immune function was revealed by the work of Chen et al. (2011). In many different cells, virus infection triggers STING (also named MITA or ERIS) to recruit STAT6 to the endoplasmic reticulum, leading to phosphorylation of S407 and Y641. Homodimers of tyrosine- and serine-phosphorylated STAT6 then activate specific target genes in the nucleus that mediate immune cell homing. Another example of JAK-independent STAT phosphorylation is provided by recent work of Gao et al. (2012) showing that STAT3 is phosphorylated in nuclei by dimeric pyruvate kinase M2, helping to promote cell proliferation in response to alterations in glucose metabolism.

STAT3 is dimethylated on K140 by SET9 in response to IL6 and demethylated by LSD1 (Yang et al., 2010). These enzymes, initially discovered based on their ability to modify histones, also modify wild-type STAT3 bound to the SOCS3 promoter but not the S727A mutant, showing that prior S727 phosphorylation is required for methylation, possibly by helping to recruit SET9 to tyrosine-phosphorylated STAT3 dimers bound to the promoter. Lysine methylation has not yet been demonstrated for other STATs, but has emerged as an important posttranslational modification for p53 and NF-κB (Stark et al., 2011). The methylation of STAT1 on R31 was reported by Mowen et al. (2001) to have important consequences for the function of tyrosine-phosphorylated STAT1 dimers, but this work has been controversial (Meissner et al., 2004; Komyod et al., 2005). However, two recent studies are supportive of R31 methylation. Iwasaki et al. (2010) show that arginine N-methyltransferase 2 links STAT3 methylation to the ability of leptin, an activator of STAT3 tyrosine phosphorylation, to regulate energy balance. In a fascinating study, Elizabeth Sampaio and colleagues (personal communication) have analyzed human autosomal-dominant gain-of-function STAT1 mutations that predispose to severe disseminated fungal infections. The mutated amino acid residues are far apart in the linear sequence of STAT1 but cluster near R31 in the three-dimensional structure. The mutations decrease the ability of R31 to be methylated, thus increasing the binding to STAT1 of PIAS1, which catalyzes SUMOylation of lysine residues (Shuai, 2006; Tahk et al., 2007). SUMOylation of STAT1 on K703 by PIAS1 inhibits its activation (Ungureanu et al., 2005) and also inhibits the formation of STAT1 nuclear paracrystalline arrays, which makes tyrosine-phosphorylated STAT1 more susceptible to inactivation by nuclear phosphatases (Droescher et al., 2011). SUMOylation is not known to occur in any other STAT, and therefore these are not prevented from forming paracrystalline arrays. Tyrosine-phosphorylated STATs in these arrays are protected from inactivating nuclear phosphatases, thus prolonging their activation.

Although Yuan et al. (2005) indicated that the reversible acetylation of STAT3 on K685 by the acetyltransferase p300 was essential for its cytokine-stimulated function, this work has been criticized (O'Shea et al., 2005). Several recent reports have shown that acetylation of K685 is important for functions of STAT3 that are independent of tyrosine phosphorylation, namely for the activation of STAT3 mediated by nuclear translocation of CD44 (Lee et al., 2009), for downregulating the ability of STAT3 to inhibit gluconeogenesis (Nie et al., 2009), and for the STAT3-dependent effects of IL-22 in keratinocytes (Sestito et al., 2011).

Functions of Unphosphorylated STATs

In the absence of tyrosine phosphorylation, STATs have at least three distinctly different functions, as transcription factors or modifiers of transcription factors, as effectors of mitochondrial function, and as effectors of chromatin structure. As reviewed recently (Cheon et al., 2011), the STAT1 gene responds to IFNs and the STAT3 gene responds to IL-6 (and related cytokines) through conventional GAS or ISRE elements in the respective promoters, leading to large increases in the concentrations of each protein after cytokine treatment. Although the initial responses to IFNs and IL-6 that are due to tyrosine phosphorylation are downregulated rapidly, through a variety of mechanisms but especially through the activities of the SOCS proteins, the responses to increased concentrations of U-STATs persist for many days. The STAT1 and STAT3 responses are distinct: the STAT1 gene does not respond to IL-6 and the STAT3 gene does not respond to IFNs, and completely different sets of genes are regulated to the two different U-STATs. The genes regulated by U-STAT3 are also distinct from those induced by tyrosine-phosphorylated STAT3, and several of them, such as MRAS and MET, have well-described roles in oncogenesis, thus potentially contributing to the major role of STAT3 in cancer (Yang et al., 2005). Some genes are regulated by U-STAT3 through a complex transcription factor formed in combination with NF-κB, in which NF-κB provides the DNA-binding and transactivation domains and STAT3 is responsible for nuclear translocation (Yang et al., 2007). U-STAT1 prolongs the expression of a subset of IFN-induced genes (Cheon and Stark, 2009), in combination with U-STAT2 and IRF9, the other two components of ISGF3, which are also induced in response to IFNs (Stark, data not shown). Thus, there are three forms of ISGF3: YP2-ISGF3 is formed when both STAT1 and STAT2 are tyrosine phosphorylated in response to type I IFNs; YP1-ISGF3 is formed when only STAT1 is tyrosine phosphorylated in response to IFN-γ (Morrow et al., 2011); and U-ISGF3 is formed when the concentrations of the three constituent proteins are increased in response to IFNs (Stark, data not shown). Many of the U-ISGF3-induced proteins have well-known antiviral or immune-regulatory functions (Cheon and Stark, 2009), and their continued expression, long after the initial response to IFN has been downregulated, serves to prolong resistance to virus infection for many days. Some of the U-ISGF3-induced proteins are also able to mediate resistance to DNA damage in the many cancers in which U-ISGF3 is overexpressed (Weichselbaum et al., 2008).

Two examples reveal additional unique transcription-related functions of U-STATs. Yue et al. (2010) show that angiotensin II-dependent signaling induces increased expression of U-STAT3 through an unknown mechanism, contributing to angiotensin II-induced cardiac hypertrophy, and Testoni et al. (2011) show that U-STAT2 is constitutively bound to many IFN-activated promoters before IFN treatment, contributing to their basal regulation.

Gough et al. (2011) have reviewed recent work showing important roles for STAT3 outside the nucleus, including in mitochondria, where STAT3 has been reported to support RAS-dependent oncogenic transformation (Gough et al., 2009), and to participate in cellular respiration (Wegrzyn et al., 2009). More recently, Szczepanek et al. (2011) have shown that mitochondrial STAT3 protects against stress-induced changes in the electron transport chain that result in the generation of reactive oxygen species. U-STAT5 now joins the party, as Lee et al. (2012) have shown that both U-STAT5A and U-STAT5B are associated with the Golgi apparatus and rough endoplasmic reticulum in vascular cells, with dramatic effects on the stability of these organelles after STAT5 knock-down.

There is only one STAT in Drosophila, and Yan et al. (2011) provide additional evidence for their earlier conclusion that U-dSTAT promotes heterochromatin stability through its interaction with heterochromatin protein 1 (HP1). Genetic manipulation to reduce the levels of either U-dSTAT or HP1 leads to reduced levels of heterochromatin and consequently to increased susceptibility to DNA damage and defects in chromosomal compaction and segregation during mitosis. The work of Christova et al. (2007) in human cells reveals that chromatin carrying the entire major histocompatibility complex (MHC) locus loops out in response to treatment with IFN-γ and that this phenomenon is dependent upon the tyrosine phosphorylation of STAT1. A plausible connection between the two observations is that depletion of the pool of U-STAT1, rather than formation of tyrosine-phosphorylated STAT1, might be responsible for the observed changes in the organization of the MHC locus. A recent example of negative regulation of gene expression in response to STAT activation is that STAT5 tetramers, formed in response to IL-7, bind to the Igκ intronic enhancer, recruiting the histone lysine methyl transferase EZH2, which in turn trime-thylates H3K27, creating a repressive mark that blocks Igκ expression (Mandal et al., 2011).

Each month brings new publications describing unexpected additional functions for the JAKs and STATs. These ubiquitous and amazingly adaptable ancient proteins have become linked to a wide variety of seemingly disparate functions over evolutionary time. What will the next 20 years bring?