Phosphorylation of STAT1 serine 708 by IKKε blocks formation of the STAT1 homodimer that constitutes GAF while leaving the STAT1 and STAT2 heterodimer of ISGF3 intact, biasing the shared STAT1 pool and the interferon-stimulated transcriptome toward the type I response.
Ng SL, Friedman BA, Schmid S, Gertz J, Myers RM, tenOever BR, Maniatis T. IκB kinase ε (IKKε) regulates the balance between type I and type II interferon responses. Proceedings of the National Academy of Sciences. 2011. Volume 108, issue 52, pages 21170-21175.
DOI 10.1073/pnas.1119137109. PMID 22171011. PMCID PMC3248534.
Phosphorylation of STAT1 serine 708 by IKKε blocks formation of the STAT1 homodimer that constitutes GAF while leaving the STAT1 and STAT2 heterodimer of ISGF3 intact, biasing the shared STAT1 pool and the interferon-stimulated transcriptome toward the type I response.
STAT1 is a shared component of two distinct interferon-driven transcription factor complexes. With STAT2 and IRF9 it forms ISGF3, which acts on interferon-stimulated response elements downstream of type I interferon. As a homodimer it forms GAF, which acts on gamma-activated sequences downstream of type II interferon. Because the two complexes draw on the same protein, the allocation of STAT1 between them determines which antiviral program a cell runs. Earlier work from these laboratories had shown that IKKε phosphorylates STAT1 at serine 708 and that this is required for an effective antiviral response, without explaining how. Here the authors ask how that phosphorylation acts. Using IKKε knockout fibroblasts and macrophages, gel shift assays, coimmunoprecipitation with conformation-reporting STAT1 mutants, recombinant kinase reconstitution, RNA sequencing and STAT1 ChIP sequencing, they find that loss of IKKε reduces ISGF3 assembly and DNA binding while increasing GAF assembly and binding, with a reciprocal shift in the gene expression program and increased permissiveness to influenza A virus. In a structural model of the GAF complex, serine 708 sits at the homodimer interface and makes hydrogen bonds there. Recombinant STAT1 phosphorylated by both a JAK kinase and IKKε failed to bind a gamma-activated sequence but still supported ISGF3. The work positions IKKε as a switch that partitions STAT1 between the two complexes.
Type I and type II interferons induce overlapping but distinct transcriptomes, and much of the overlap traces to the shared use of STAT1 and to promoters that carry both element classes. The IKK-related kinases IKKε and TBK1 were known to phosphorylate IRF3 and IRF7 during induction of interferon beta. Knockout mouse studies from several laboratories had, however, placed TBK1 rather than IKKε at the center of type I interferon induction, leaving the phenotype of IKKε loss unexplained by an induction defect. The immediately preceding work from these authors, published in 2007, showed that IKKε knockout mice make normal amounts of type I interferon after influenza A virus infection but respond abnormally to it, losing a subset of interferon-stimulated proteins and failing to clear virus, and identified STAT1 serine 708 as an IKKε substrate. That work also distinguished IKKε-dependent from IKKε-independent classes of interferon-stimulated response element. What remained unresolved was the molecular consequence of the serine 708 modification and why it should matter selectively for a subset of genes.
How does IKKε-mediated phosphorylation of STAT1 produce an effective type I interferon response, and specifically does that phosphorylation act by changing which STAT1-containing transcription factor complex assembles?
The design moves from phenotype to complex to residue and then back out to the genome. Loss-of-function fibroblasts and macrophages from Ikbke knockout and wild type mice provide the comparison throughout, with influenza A virus infection establishing that the signaling defect has a consequence for virus control. Complex-level questions are addressed by gel shift assays against three probe classes, an IKKε-dependent response element from the Adar1 promoter, an IKKε-independent element from the Irf7 promoter, and a gamma-activated sequence from the Irf1 promoter, plus size-exclusion chromatography to read the relative abundance of the two complexes directly. Gain of function is tested by adenoviral IKKε delivery into E1A-expressing HeLa cells, chosen because E1A blocks IRF3-driven interferon induction and so allows interferon input to be set by the experimenter. The residue-level question is approached through the published crystal structure of the phosphotyrosine STAT1 dimer on DNA, through coimmunoprecipitation using STAT1 mutants that cannot form the unstimulated antiparallel dimer and therefore report only on the activated parallel dimer, and through reconstitution with purified recombinant proteins phosphorylated in vitro by a JAK kinase and by IKKε. Finally, RNA sequencing and STAT1 alpha ChIP sequencing in bone marrow-derived macrophages test whether the model holds across the genome, with genes ordered along a beta to gamma mixture score that quantifies the relative response to the two interferons.
IKKε-deficient fibroblasts are more permissive to influenza A virus. Viral protein was detectable by 6 hours after infection in knockout cells against 12 hours in wild type, and interferon beta pretreatment did not significantly reduce viral load in the knockout, which the authors read as locating the defect in type I interferon signaling rather than in interferon production (Figure 1A).
The protein-level signature of IKKε loss is reciprocal. IFIT1 and IFIT2, both primarily type I driven, appeared later and at lower levels in knockout cells, while IRF1 and STAT1, both primarily GAF regulated, were not decreased and in the case of IRF1 appeared earlier (Figure 1B).
The same reciprocity holds at the transcript level. Ifit2, Mda5 and Viperin were reduced in knockout cells after interferon beta, while Irf1, Irf8 and Icam1 were elevated, and Stat1 message was higher in knockout cells after either interferon, more markedly after interferon gamma (Figure 2A,B).
Complex assembly and DNA binding shift in opposite directions. Binding of ISGF3 to the IKKε-dependent element was substantially reduced in knockout extracts while binding to the IKKε-independent element was indistinguishable between genotypes, including in competition titrations (Figure 3A and Figure S1). Binding of GAF to the gamma-activated sequence increased in knockout extracts after either interferon (Figure 3B). Size-exclusion chromatography showed ISGF3 exceeding GAF in wild type extracts with the ratio shifted in the knockout (Figure S2).
Adding IKKε produces the mirror-image result. Adenoviral IKKε in E1A-expressing HeLa cells did not reduce ISGF3 binding to the response element but did reduce GAF binding to the gamma-activated sequence, with an inverse relation between IKKε level and binding (Figure S3A-C). At the gene level, IKKε expression induced IFIT2 in response to type II interferon and sharply reduced IRF1 (Figure S4).
Serine 708 lies at the homodimer interface. In the published crystal structure of the STAT1 homodimer on DNA, serine 708 sits within the dimerization interface and participates in hydrogen bonding there (Figure 4A,B). The authors use this structure to motivate the hypothesis that phosphorylation at this position is incompatible with the homodimer but tolerated in ISGF3. The structural interpretation is theirs and is not tested by new structural work.
IKKε blocks activated STAT1 homodimerization but not the STAT1 and STAT2 interaction. Using STAT1 mutants unable to preassociate in the antiparallel conformation, interferon gamma induced dimerization was disrupted by wild type IKKε but not by GFP or by the catalytically dead K38A mutant (Figure 4C). The STAT1 and STAT2 interaction was not disrupted and in fact appeared stabilised before stimulation by IKKε expression (Figure 4D). A phosphorylation-dependent mobility shift in STAT2 led the authors to suggest that IKKε may also phosphorylate STAT2, which is an observation-based suggestion rather than a demonstrated modification.
Purified components reproduce the switch. Recombinant STAT1 phosphorylated by a JAK kinase formed GAF and bound the gamma-activated sequence, whereas STAT1 phosphorylated by both JAK and IKKε failed to bind, while IKKε phosphorylation did not disrupt ISGF3 formation (Figure S5). This is the most direct evidence in the paper that the effect is intrinsic to the modified protein rather than an indirect consequence of the knockout state.
Genome-wide expression follows the same axis. Among 538 interferon-stimulated genes in wild type macrophages, genes scoring toward the beta end of the mixture scale were expressed at lower levels in knockout cells after interferon beta, and genes scoring toward the gamma end were expressed at higher levels, with a general elevation of baseline expression for gamma-leaning genes in knockout cells (Figure S6A,B).
STAT1 occupancy changes correspondingly. Read density at STAT1 alpha peaks near Tlr9 and Ifit2 was higher in wild type macrophages, while peaks near Nos2, Gbp2 and Irf1 were not reduced in knockout cells and Nos2 was higher (Figure S7). The authors conclude that GAF binding is robust in macrophages and that the primary regulatory action of IKKε is enhancement of ISGF3 formation.
Sequence context distinguishes the two peak classes. The gamma-activated sequence motif was enriched in peaks near gamma-leaning genes relative to beta-leaning genes, while the response element consensus did not differ significantly between the two sets. Unbiased motif discovery recovered response element motifs, with the version from gamma-leaning genes carrying additional purine-rich contacts (Figure 5A,B). The authors relate this to their earlier proposal that extra DNA contacts can compensate for the absence of an IKKε-modified ISGF3.
The model the data support is one of competitive allocation of a shared subunit. STAT1 activated by JAK-mediated tyrosine phosphorylation can either homodimerise into GAF or partner with STAT2 and IRF9 into ISGF3. IKKε, itself induced during infection, phosphorylates STAT1 at serine 708 and thereby renders the homodimer interface unfavourable, so the available STAT1 is driven into ISGF3. The consequence at the genome is more binding and transcription at response elements, less at gamma-activated sequences, and a cell better able to restrict influenza A virus.
Several links in this chain are directly demonstrated. The reciprocal behavior of the two complexes in gain and loss of function, the requirement for IKKε catalytic activity in disrupting the activated homodimer, and the reconstitution with purified phosphorylated proteins all support the core claim. The structural explanation, that phosphorylation of serine 708 disrupts hydrogen bonding at the homodimer interface while being accommodated in the trimeric complex, is explicitly framed by the authors as a proposal based on an existing crystal structure of the homodimer. No structure of ISGF3 with or without the modification is presented, and the paper states that the structural differences in ISGF3 assembled with and without IKKε are unknown. The suggestion that a differential requirement for serine 708 explains why minimal response elements are IKKε-dependent while longer elements with additional purine contacts are not is likewise a proposal supported by correlation between motif content and genotype sensitivity, not by direct test.
The work converts a kinase and a phosphosite into a quantitative control point for the choice between two interferon programs. Rather than treating serine 708 phosphorylation as a general potentiator of STAT1 activity, it identifies a structural target, the homodimer interface, and a consequence, reallocation of a limiting shared subunit. That framing makes several things testable that were not before, including the prediction that the switch should be sensitive to the ratio of activated IKKε to STAT1, which the authors state explicitly, and the prediction that promoter element architecture determines which genes feel the loss of the kinase first. Methodologically, the combination of conformation-reporting STAT1 mutants with reconstitution from purified phosphorylated components provides a way to separate a complex assembly defect from a DNA affinity defect, and pairing RNA sequencing with STAT1 ChIP sequencing across a continuous beta to gamma score gives a genome-scale readout of a balance rather than of a single pathway.
The paper extends a line begun in the 2007 Science report from tenOever and colleagues, which identified serine 708 as an IKKε substrate and separated IKKε-dependent from IKKε-independent response elements without resolving the mechanism. It also connects to the 2010 work of Schmid, Mordstein, Kochs, García-Sastre and tenOever on transcription factor redundancy in induction of the antiviral state, which is cited here in support of the promoter-architecture argument. Taken together with those two papers the recurring question is how a limited set of shared transcription factors is apportioned so that a cell produces the right antiviral program rather than simply a large one. That framing is a category 3 synthesis across those three papers and is offered as such rather than as a claim made by any one of them.
The cellular work is in mouse embryonic fibroblasts and mouse bone marrow-derived macrophages, plus human cell lines for the overexpression and coimmunoprecipitation experiments, and the conclusions are not extended to whole animals in this paper. Loss of IKKε is constitutive rather than conditional or acute, so secondary adaptation in the knockout cells cannot be excluded, although the recombinant reconstitution and the gain-of-function experiments mitigate this. The overexpression experiments use adenoviral delivery at levels that are not physiological and depend on E1A to suppress endogenous interferon induction, which is itself a perturbation of the system. The structural claim rests on an existing homodimer structure rather than on structures of the phosphorylated protein or of ISGF3, and the paper says directly that the structural consequences within ISGF3 are unknown. The ChIP sequencing was performed with an antibody against STAT1 alpha and at a single 6 hour time point, so it does not resolve kinetics or distinguish complexes except by inference from the associated gene and motif classes. The suggestion that IKKε also phosphorylates STAT2 rests on a mobility shift and is not pursued. The quantitative relationship the authors propose between activated IKKε and STAT1 levels is a stated expectation rather than a measured parameter. Virus work is confined to one influenza A strain in fibroblasts.
A kinase induced during infection phosphorylates STAT1 at a dimer contact point, preventing one interferon complex from forming and channeling the protein into the antiviral alternative.
Two interferon pathways compete for the same protein, STAT1. Paired with STAT2 it drives the antiviral type I program, paired with itself it drives the type II program. The authors show that IKKε phosphorylates STAT1 at a residue sitting in the self-pairing interface, which blocks self-pairing but not partnering with STAT2. Cells lacking IKKε run the wrong program, bind the wrong promoters, and are more readily infected by influenza A virus.
STAT1 is shared between ISGF3, the type I interferon complex that acts at interferon-stimulated response elements, and GAF, the STAT1 homodimer that acts at gamma-activated sequences. The authors show that IKKε phosphorylation of STAT1 serine 708, a residue located in the homodimer interface of the published crystal structure, blocks assembly of the activated homodimer while leaving the STAT1 and STAT2 interaction intact. In IKKε-deficient fibroblasts and macrophages, ISGF3 assembly and response element binding fall while GAF assembly and gamma-activated sequence binding rise, transcript and protein signatures shift correspondingly, and influenza A virus replicates earlier. Adenoviral IKKε produces the inverse shift, and recombinant STAT1 phosphorylated by both a JAK kinase and IKKε fails to bind a gamma-activated sequence while still supporting ISGF3. Genome-wide, expression and STAT1 occupancy track a continuous score of relative type I and type II responsiveness. The structural explanation is proposed rather than solved here.
lab-led for Schmid 2010 and Schmid 2014, co-led with the Maniatis laboratory for Ng 2011, and training period for tenOever 2007, which was carried out with the Maniatis and García-Sastre laboratories and is not the independent program's work