Scientists have modified a medicine for high blood pressure into one that might tame misfolded protein diseases, they report in the April 10 Science. The new molecule, dubbed Sephin1, countered the effects of aggregating proteins in mouse models of amyotrophic lateral sclerosis and Charcot-Marie-Tooth disease. It might do so for other neurodegenerative disorders, speculated senior author Anne Bertolotti of the Medical Research Council Laboratory of Molecular Biology in Cambridge, England. Moreover, Sephin1 did this by selectively inhibiting dephosphorylation of a translation factor, something thought to be almost impossible because phosphatases have so many substrates.

Change for the Better.

Removing a chlorine from guanabenz generated Sephin1, which caused fewer side effects in mice. [Image courtesy of Science/AAAS.]

Protein phosphatase 1C (PP1c) dephosphorylates the eukaryotic translation initiation factor 2α subunit (eIF2α), among a plethora of other protein targets. As part of the unfolded protein response (UPR) in the endoplasmic reticulum, phosphorylated eIF2α temporarily shuts down translation, giving chaperones a chance to catch up on protein folding. The UPR often kicks in to counteract accumulation of misfolded proteins in neurodegenerative disease. Dephosphorylation of eIF2α turns off the UPR and allows normal translation to resume. PP1c removes the regulatory phosphate in conjunction with one of two accessory subunits that go by the complex monikers protein phosphatase 1, regulatory subunits 15A and 15B (PPP1R15A and PPP1R15B, or R15A and R15B for short). R15B is always around, but the cell only makes R15A during times of ER stress. By promoting eIF2α dephosphorylation, these subunits ensure the cell does not go too long without making proteins, which would be lethal.

In 2011, Bertolotti reported that the anti-hypertensive drug guanabenz binds to R15A, but not R15B, and inhibits eIF2α dephosphorylation. Guanabenz protected cells from misfolded proteins and tunicamycin, a drug that induces ER stress (Tsaytler et al., 2011). Other scientists have reported that guanabenz treatment slows down disease in mouse models of ALS (Wang et al., 2014; Jiang et al., 2014), multiple sclerosis (Way et al., 2015), and prion disease (Tribouillard-Tanvier et al., 2008).

Unfortunately, guanabenz has side effects such as drowsiness and lethargy, and even coma after overdose (Hall et al., 1985). This is due to activation of the α2-adrenergic receptor, causing a potent reduction in blood pressure. Guanabenz was designed and approved as an anti-hypertensive in 1982, but it fell out of use when better medications came along. Bertolotti and first author Indarjit Das sought to eliminate α2 binding, while preserving activity in the UPR.

From a panel of guanabenz derivatives, the authors zeroed in on one missing a chlorine (see image above). They christened it Sephin1, for “selective inhibitor of a holophosphatase.” Like guanabenz, Sephin1 only bound and inhibited R15A, not R15B. Sephin1 treatment rescued cells exposed to tunicamycin, but not cells missing R15A, confirming it worked via that subunit.

Maintaining this specificity for R15A was crucial, Bertolotti said. By subduing R15A, Sephin1 prolongs the phosphorylation of eIF2α in times of stress, but eventually R15B kicks in and dephosphorylates eIF2α, allowing translation to resume. In addition, Sephin1 did not interact with the α2-adrenergic receptor.

Those cell-culture experiments indicated that Sephin1 might avoid the undesirable side effects of guanabenz, which Das next tested by treating healthy 1-month-old mice with oral Sephin1, twice a day for a month. While guanabenz-treated mice toppled off a rotating rod due to lethargy, Sephin1-treated mice had no balance problems. The treated young mice gained weight as fast as untreated controls. Since eIF-2α plays a role in memory (see Apr 2015 news), Das also tested the mice for their ability to find a hidden platform in a water maze, and to associate a sound with a foot shock in fear-cue experiments. Treated mice had no problems with either. Bertolotti said the researchers have continued the Sephin1 treatment for as long as six months. “For those six months, we see no side effects whatsoever,” she said.

What about treating disorders of protein misfolding? The researchers tested Sephin1 in a mutant SOD1 mouse model of fast-progressing ALS. They selected this model, Bertolotti said, because there was already evidence from genetic studies that knocking down R15A protected a slower-progressing version of mSOD1 mice (Wang et al., 2014). Das treated the animals with daily Sephin1 or control vehicle solution for seven weeks, beginning at when they were four weeks old. The Sephin1 mice grew faster and balanced better on the rotating rod. “The SOD1 mice treated with Sephin1 behave almost like normal,” Bertolotti wrote in an email to Alzforum.

When examined a month later, the Sephin1-treated mice contained more motor neurons in the lumbar spinal cord than the untreated mSOD1 controls, again almost matching wild-type controls. The Sephin1 treatment greatly reduced the amount of insoluble, presumably misfolded, SOD1 in spinal cord extracts, though they had normal amounts of SOD1 protein overall. The Sephin1 mice also expressed fewer markers of ER stress. “It looks like we have improved [SOD1] folding ,” Bertolotti said.

Raymond Roos of the University of Chicago, who was not involved in the work, said he would have liked to see how Sephin1 affected survival. Given the effects on weight, motor neuron number, and SOD1 solubility, he said, “I would be surprised if there was not a significant survival effect.” Bertolotti told Alzforum that those studies, which require more animals, are now in the works. 

Sephin1 also helps mice that mimic a form of Charcot-Marie-Tooth neuropathy. Might it have broader effects against misfolding of neurodegeneration-related proteins, such as Aβ or tau? “Since they show this striking protective effect, it is worth trying in other neurodegenerative diseases,” commented Yang Hu of Temple University School of Medicine in Philadelphia, who did not participate in the study. However, he cautioned that Sephin1 is far from a shoo-in. Researchers have not confirmed that ER stress due to protein aggregates is a generalized cause of neurodegeneration, he pointed out. Bertolotti suspects that Sephin1 or a similar molecule would benefit more diseases than just ALS and CMT, but probably not all neurodegenerative conditions.

Hu and Roos were not entirely convinced that prolonging eIF2α phosphorylation was the only mechanism at work, either. Roos pointed out that R15A also dampens inflammation (Mesman et al., 2014). Bertolotti said R15A probably affects inflammation because of its influence on eIF2α. She is convinced that Sephin1 has no other mechanism of action because it mimics R15A knockout.

Jeroen Hoozemans and Wiep Scheper of VU University Medical Center in Amsterdam added that some studies suggest that eIF2α phosphorylation would be detrimental, not beneficial. Inhibition of PERK, the kinase that phosphorylates and activates eIF2α, protected mice from prion disease (see Oct 2013 news). “The control of eIF2α phosphorylation and dephosphorylation is very complex,” they wrote in an email to Alzforum. “For translation to human neurodegeneration, better understanding of how eIF2α is affected and controlled in different neurodegenerative diseases is required.”—Amber Dance


Make a Comment

Comments on News and Primary Papers

  1. The paper by Indrajit Das and colleagues supports accumulating evidence that regulating the phosphorylation or dephosphorylation of eukaryotic initiation factor 2α (eIF2α) is a potential therapeutic approach for treating neurodegenerative diseases. Regulation of phosphorylation of this translation initiation factor is central to the integrated stress response (ISR) and reduces protein synthesis under conditions of cellular stress. The authors designed Sephin1, an inhibitor of eIF2α dephosphorylation based on Guanabenz, a compound that inhibits the phosphatase subunit PPP1R15B. Like its predecessor, Sephin1 is selective for the stress-induced eIF2α phosphorylation, but it lacks side effects caused by interaction with the α2-adrenergic receptor. This paper raises the debate whether for neurodegenerative diseases can be treated by blocking the phosphorylation of eIF2α through the inhibition of eIF2α kinases such as PERK, or prolonging it by blocking phosphatases as in the current study. There is now substantial evidence that the approaches stimulate or reduce protein synthesis during stress, respectively (see Oct 2013 news).

    The control of eIF2α phosphorylation and dephosphorylation is very complex. It is important to increase understanding of how eIF2α is controlled in physiological conditions, but the real questions (and answers) lie in human pathology. Data is accumulating that the ISR is involved in many human neurodegenerative conditions. Although the key players of the ISR are involved in human pathology, it is unclear how. Is the ISR blocked, over-activated, or under aberrant control? Is this process similar or different among diseases? The study by Das et al. nicely addresses the potential of eIF2α as a therapeutic target for neurodegeneration and shows that therapeutic side effects can be reduced by more specific approaches. For translation to human therapy, better understanding of how eIF2α is affected and controlled in different neurodegenerative diseases is required.

    View all comments by Wiep Scheper
  2. In response to Hoozemans and Scheper's comments:

    Whether eIF2α phosphorylation is detrimental or beneficial is now very well understood.

    A transient phosphorylation of eIF2α is beneficial, but a persistent phosphorylation of eIF2α is detrimental. This is why mammals have evolved two eIF2α phosphatases: to avoid persistent phosphorylation of eIF2αand its deleterious effects. Because Sephin1 is a selective inhibitor of R15A, but not the related R15B, it is safe and doesn't cause deleterious side effects.

    This is in full agreement with genetic investigations of this pathway, carefully carried out in the past 15 years, mostly in the labs of David Ron and Randy Kaufman. Mice lacking R15A are viable and appear largely normal. Likewise, as we show, pharmacological inhibition of R15A is safe.

    To illustrate this mechanism in simple terms, it might be useful to compare the sophisticated adaptive eIF2α pathway to a diet and a hunger strike. A diet will be beneficial to someone who is overweight, but a prolonged hunger strike will inevitably be fatal.

    I hope this helps!

  3. We completely agree that the elegant design of the Sephin compound explains the lack of side effects. We thought it relevant to point out to readers that inhibition of eIF2α phosphorylation via genetic deletion or pharmacological inhibition of PERK protects when persistent eIF2α phosphorylation is observed, as in mouse models of prion disease and Aβ deposition (Ma et al., 2013;  Moreno et al., 2012; Moreno et al., 2013). True, such interventions have too many side effects for clinical application, but these proof-of concept studies raise hope for more subtle interventions to inhibit phospho-eIF2α-mediated translational attenuation, such as ISRIB ( Sekine et al., 2015; Sidrauski et al., 2015; and Apr 2015 news).

    In contrast, the opposite (prolonged eIF2α phosphorylation) is effected by Sephin and protective in ALS and CMT models as shown in the current paper. These paradoxical observations may, for example, relate to the stage of the disease process and the level of eIF2α phosphorylation at the start of treatment. 


    . Suppression of eIF2α kinases alleviates Alzheimer's disease-related plasticity and memory deficits. Nat Neurosci. 2013 Sep;16(9):1299-305. PubMed.

    . Sustained translational repression by eIF2α-P mediates prion neurodegeneration. Nature. 2012 May 24;485(7399):507-11. PubMed.

    . Oral treatment targeting the unfolded protein response prevents neurodegeneration and clinical disease in prion-infected mice. Sci Transl Med. 2013 Oct 9;5(206):206ra138. PubMed.

    . Mutations in a translation initiation factor identify the target of a memory-enhancing compound. Science. 2015 Apr 9; PubMed.

    . The small molecule ISRIB reverses the effects of eIF2α phosphorylation on translation and stress granule assembly. Elife. 2015 Feb 26;4 PubMed.

  4. This beautiful paper by Indrajit Das and colleagues led by Anne Bertolotti describes Sephin1, a compound modified from Guanabenz that selectively inhibits the stress-inducible phosphatase isoform PPP1R15A, but not the constitutive PPP1R15B (Das et al., 2015). In conjunction with their previous report (Tsaytler et al., 2011), they have provided significant evidence that targeting R15A phosphatase activity could be relevant to restore cell proteostasis.

    Phosphorylation of the α subunit of eukaryotic translation initiation factor 2 at serine 51 (eIF2α-P) attenuates global protein synthesis rates in a variety of stressful conditions, including ER stress and accumulation of misfolded proteins. It is widely acknowledged that a transient and physiological elevation in eIF2α-P restores proteostasis, while a more chronic eIF2α-P rise leads to cell death. Nonetheless, it is possible that extending protein synthesis inhibition might help cells recover proteostasis in the face of abnormal deposition of misfolded proteins.

    In this regard, the Bertolotti group has identified that maintaining higher eIF2α-P levels by inhibiting R15A phosphatase recovers functional deficits in two animal models of neurodegenerative disease (Charcot-Marie-Tooth 1B and ALS). Their results identify an elegant mechanism that specifically modulates eIF2α-P under stress, preserving normal eIF2α activity in conditions not affected by stress. This is indeed mirrored in normal behavior and physiology in non-diseased mice treated with Sephin1.

    Protein misfolding is an especially relevant issue for neurodegenerative diseases, given that many of them relate to abnormal protein deposition (e.g., Alzheimer’s, Parkinson, ALS, etc.). Thus, much effort has been recently made to understand how modulating proteostasis could serve as a target for neurological disease treatment. In fact, the past few years have been considerably rich in terms of insightful reports exploring the role of eIF2α-P in brain disorders. Now, the discovery of Sephin1 might raise the idea that this compound could be used for other neurodegenerative diseases.

    I wish to highlight that, at least in Alzheimer’s and prion disease models, prolonged eIF2α-P appears to lead to the observed behavioral impairment (Moreno et al., 2012; Lourenco et al., 2013; Ma et al., 2013; Moreno et al., 2013). As a proof of concept, strategies that reduce brain eIF2α-P levels in AD and prion mice improve behavioral phenotypes (Lourenco et al., 2013; Ma et al., 2013; Moreno et al., 2013; Halliday et al., 2015). We have further described that a mechanism involving pro-inflammatory signaling, stress kinase activation, and integrated stress response leads to increased eIF2α-P with deleterious consequences to memory in AD mice (Lourenco et al., 2013). Studies from the Eric Klann (Ma et al., 2013) and Ulrich Hengst (Baleriola et al., 2014) labs also support that excessive ISR/eIF2α-P/ATF4 signaling could be detrimental in AD. In a TDP-43 model of ALS, suppression of eIF2α-P also results in improved phenotype in Drosophila (Kim et al., 2013).

    I believe that the combination of the exciting results by Bertolotti and colleagues with the aforementioned reports strengthens the notion that neuronal proteostasis obeys an hormetic (U-shaped) pattern, in which either too low or too high rates of protein synthesis prevent cells from properly responding to insults. Although Das and colleagues still haven’t plunged deeper into mechanistic aspects, their observations are of clear importance for two reasons: They (1) identified a novel and apparently specific eIF2α phosphatase inhibitor and (2) contributed original data that support a role for proteostasis modulation in neurological disease, with potential translational impact in the future.

    Now it remains to be determined whether Sephin1 could be effective in animal models of diverse neurological conditions and, in order to fully establish the benefits triggered by Sephin1, it is essential that potential on- and off-target mechanisms be elucidated.


    . Axonally synthesized ATF4 transmits a neurodegenerative signal across brain regions. Cell. 2014 Aug 28;158(5):1159-72. PubMed.

    . Partial restoration of protein synthesis rates by the small molecule ISRIB prevents neurodegeneration without pancreatic toxicity. Cell Death Dis. 2015 Mar 5;6:e1672. PubMed.

    . Therapeutic modulation of eIF2α phosphorylation rescues TDP-43 toxicity in amyotrophic lateral sclerosis disease models. Nat Genet. 2014 Feb;46(2):152-60. Epub 2013 Dec 15 PubMed.

    . TNF-α mediates PKR-dependent memory impairment and brain IRS-1 inhibition induced by Alzheimer's β-amyloid oligomers in mice and monkeys. Cell Metab. 2013 Dec 3;18(6):831-43. PubMed.

    . Suppression of eIF2α kinases alleviates Alzheimer's disease-related plasticity and memory deficits. Nat Neurosci. 2013 Sep;16(9):1299-305. PubMed.

    . Sustained translational repression by eIF2α-P mediates prion neurodegeneration. Nature. 2012 May 24;485(7399):507-11. PubMed.

    . Oral treatment targeting the unfolded protein response prevents neurodegeneration and clinical disease in prion-infected mice. Sci Transl Med. 2013 Oct 9;5(206):206ra138. PubMed.

    . Selective inhibition of a regulatory subunit of protein phosphatase 1 restores proteostasis. Science. 2011 Apr 1;332(6025):91-4. Epub 2011 Mar 3 PubMed.

  5. Targeting stress-responsive signaling pathways that regulate cellular protein homeostasis (or proteostasis) is a promising strategy to potentially ameliorate pathologic protein misfolding associated with etiologically-diverse diseases including Alzheimer’s disease, Parkinson’s disease, and familial Amyotrophic Lateral Sclerosis (ALS) (Powers et al., 2009; Hetz et al., 2013). Despite this promise, few small molecules are available that target the activity of these stress-responsive signaling pathways. Furthermore, the small molecules that do are often non-selective and/or hampered by negative toxicity profiles, limiting the translation of this approach to intervene in human disease.

    In this manuscript, Das et al. identify a new small-molecule derivative of the α2 adrenergic receptor agonist guanabenz, called Sephin1, that acts as a modulator of stress-responsive signaling induced downstream of eIF2α phosphorylation. EIF2α phosphorylation is induced in response to diverse cellular insults, including ER stress (Ron and Walter, 2007). Phosphorylation of eIF2α leads to translation attenuation that reduces the load of newly synthesized, unfolded proteins as an initial response to cellular ER stress. This translation attenuation promotes cellular proteostasis by freeing chaperones and folding enzymes to protect the established proteome and prevent the proteotoxic accumulation of misfolded proteins. Phosphorylated eIF2α also promotes the activation of a network of transcription factors that induce expression of stress-responsive genes, including cellular proteostasis genes (e.g., redox enzymes), the pro-apoptotic transcription factor CHOP, and the eIF2α phosphatase regulatory subunit GADD34/PPP1R15A. GADD34 binds Protein Phosphatase 1 (PP1) to dephosphorylate eIF2α and restore translational integrity in a well-established negative feedback loop of eIF2α signaling (Ron and Walter, 2007). The transient translation attenuation and increased expression of cellular proteostasis factors promote cellular survival in response to stress. Alternatively, pro-apoptotic signaling mediated downstream of eIF2α phosphorylation, predominantly through CHOP, promotes apoptosis in response to severe or chronic stress. As such, eIF2α phosphorylation is a critical determinant in dictating cell fate in the context of many diverse neurodegenerative disorders (Hetz and Mollereau, 2014). 

    Sephin1 affects eIF2α phosphorylation stress-signaling by inhibiting the GADD34/PP1 phosphatase complex involved in the negative feedback loop. Importantly, Sephin1 does not inhibit the constitutive eIF2α phosphatase regulatory subunit CREP/PPP1R15B. This selectivity for GADD34/PP1 is important, because the addition of Sephin1 does not influence eIF2α phosphorylation in the absence of stress. The consequence of the selective inhibition afforded by Sephin1 is a delay in the translational recovery following the stress insult, extending the translational attenuation afforded by eIF2α phosphorylation. Interestingly, Sephin1 also reduces the translation of the pro-apoptotic transcription factor CHOP, which could attenuate downstream apoptotic signaling.

    Guanabenz also selectively inhibits GADD34/PP1 to promote cellular proteostasis in response to stress (Tsaytler et al., 2011), but the toxicity profile of this molecule (owing to its α2 adrenergic receptor activity) has largely precluded its use in protein-misfolding diseases. A critically important attribute of Sephin1 is that this molecule separates the GADD34 inhibitory activity of guanabenz from its α2 adrenergic receptor activity. Sephin1 shows no α2 adrenergic receptor activity in vitro, but still potently inhibits GADD34. Similarly, Sephin1 showed no detrimental phenotypes in mice that would reflect potent α2 adrenergic receptor activity. Furthermore, Sephin1 retained the desirable guanabenz bioavailabilty in the nervous system, suggesting that Sephin1 offers significant promise to ameliorate pathologic defects associated with eIF2α phosphorylation involved in protein-misfolding diseases.

    The potency of Sephin1 for GADD34 inhibition combined with its desirable bioavailability profile provides a unique opportunity to demonstrate the potential for this therapeutic strategy to intervene in protein misfolding diseases. Das et al. demonstrated this potential in two etiologically distinct diseases. In a mouse model of Charcot-Marie Tooth 1B, which expresses a destabilizing mutant of myelin protein zero, the addition of Sephin 1 rescued myelination, attenuated neuronal expression of the pro-apoptotic transcription factor CHOP, and restored motor coordination. Then, in a mouse model of ALS that expresses the aggregation-prone G93A mutant of SOD1, Sephin1 restored weight gain, rescued motor coordination deficiencies, decreased SOD1G93A aggregates, and attenuated neuronal expression of ER stress markers, including CHOP. Thus, Sephin1-dependent inhibition of GADD34 was able to attenuate pathologic phenotypes and promote neuronal function in these two distinct neurodegenerative disorders.

    The capacity for Sephin1 to attenuate degenerative phenotypes in distinct protein-misfolding diseases is extremely exciting, as it offers proof of principle that we can intervene in diverse protein misfolding diseases by targeting the activity of stress-responsive signaling pathways. The ability to separate the α2-adrenergic receptor activity of guanabenz from the GADD34/PP1 inhibitor activity through medicinal chemistry is a significant step toward translating this approach to the clinic to intervene in human neurodegenerative disorders including Charcot-Marie Tooth Type 1B and familial ALS. Furthermore, since stress-signaling downstream of eIF2α phosphorylation implicated in the pathophysiology of many other neurodegenerative disorders, including Alzheimer’s and Parkinson’s disease (Hetz and Mollereau, 2014), this study suggests that targeting GADD34/PP1 activity using Sephin1 could potentially provide a broadly-applicable strategy to intervene in many distinct neurodegenerative disorders.


    . Targeting the unfolded protein response in disease. Nat Rev Drug Discov. 2013 Sep;12(9):703-19. PubMed.

    . Disturbance of endoplasmic reticulum proteostasis in neurodegenerative diseases. Nat Rev Neurosci. 2014 Apr;15(4):233-49. Epub 2014 Mar 12 PubMed.

    . Biological and chemical approaches to diseases of proteostasis deficiency. Annu Rev Biochem. 2009;78:959-91. PubMed.

    . Signal integration in the endoplasmic reticulum unfolded protein response. Nat Rev Mol Cell Biol. 2007 Jul;8(7):519-29. PubMed.

    . Selective inhibition of a regulatory subunit of protein phosphatase 1 restores proteostasis. Science. 2011 Apr 1;332(6025):91-4. Epub 2011 Mar 3 PubMed.

Make a Comment

To make a comment you must login or register.


News Citations

  1. Memory Maker or Neuron Killer? For Transcription Factor, It May Depend on Mood
  2. PERKing Up Protein Synthesis May Prevent Neurodegeneration

Paper Citations

  1. . Selective inhibition of a regulatory subunit of protein phosphatase 1 restores proteostasis. Science. 2011 Apr 1;332(6025):91-4. Epub 2011 Mar 3 PubMed.
  2. . Guanabenz, which enhances the unfolded protein response, ameliorates mutant SOD1-induced amyotrophic lateral sclerosis. Neurobiol Dis. 2014 Nov;71:317-24. Epub 2014 Aug 15 PubMed.
  3. . Guanabenz delays the onset of disease symptoms, extends lifespan, improves motor performance and attenuates motor neuron loss in the SOD1 G93A mouse model of amyotrophic lateral sclerosis. Neuroscience. 2014 Sep 26;277:132-8. Epub 2014 Mar 31 PubMed.
  4. . Pharmaceutical integrated stress response enhancement protects oligodendrocytes and provides a potential multiple sclerosis therapeutic. Nat Commun. 2015 Mar 13;6:6532. PubMed.
  5. . Antihypertensive drug guanabenz is active in vivo against both yeast and mammalian prions. PLoS One. 2008;3(4):e1981. PubMed.
  6. . Guanabenz overdose. Ann Intern Med. 1985 Jun;102(6):787-8. PubMed.
  7. . An enhanced integrated stress response ameliorates mutant SOD1-induced ALS. Hum Mol Genet. 2014 May 15;23(10):2629-38. Epub 2013 Dec 23 PubMed.
  8. . Measles virus suppresses RIG-I-like receptor activation in dendritic cells via DC-SIGN-mediated inhibition of PP1 phosphatases. Cell Host Microbe. 2014 Jul 9;16(1):31-42. PubMed.

External Citations

  1. SOD1 

Further Reading


  1. . ALS-linked mutant SOD1 induces ER stress- and ASK1-dependent motor neuron death by targeting Derlin-1. Genes Dev. 2008 Jun 1;22(11):1451-64. PubMed.
  2. . Protein folding activity of ribosomal RNA is a selective target of two unrelated antiprion drugs. PLoS One. 2008 May 14;3(5):e2174. PubMed.
  3. . Sustained translational repression by eIF2α-P mediates prion neurodegeneration. Nature. 2012 May 24;485(7399):507-11. PubMed.

Primary Papers

  1. . Preventing proteostasis diseases by selective inhibition of a phosphatase regulatory subunit. Science. 2015 Apr 10;348(6231):239-42. PubMed.