A novel specific PERK activator reduces toxicity and extends survival in Huntington's disease models.

One of the pathways of the unfolded protein response, initiated by PKR-like endoplasmic reticulum kinase (PERK), is key to neuronal homeostasis in neurodegenerative diseases. PERK pathway activation is usually accomplished by inhibiting eIF2α-P dephosphorylation, after its phosphorylation by PERK. Less tried is an approach involving direct PERK activation without compromising long-term recovery of eIF2α function by dephosphorylation. Here we show major improvement in cellular (STHdhQ111/111) and mouse (R6/2) Huntington's disease (HD) models using a potent small molecule PERK activator that we developed, MK-28. MK-28 showed PERK selectivity in vitro on a 391-kinase panel and rescued cells (but not PERK−/− cells) from ER stress-induced apoptosis. Cells were also rescued by the commercial PERK activator CCT020312 but MK-28 was significantly more potent. Computational docking suggested MK-28 interaction with the PERK activation loop. MK-28 exhibited remarkable pharmacokinetic properties and high BBB penetration in mice. Transient subcutaneous delivery of MK-28 significantly improved motor and executive functions and delayed death onset in R6/2 mice, showing no toxicity. Therefore, PERK activation can treat a most aggressive HD model, suggesting a possible approach for HD therapy and worth exploring for other neurodegenerative disorders.

damaged, non-functional cells. We have previously reported that there is a strong ER stress induction and UPR activation in striatal neurons expressing mHtt 13 . Importantly, striatal neurons are the first to degenerate in HD patients. In mammals, the UPR includes three signaling pathways initiated by their sensors, the ER-resident transmembrane proteins PKR-like endoplasmic reticulum kinase (PERK), activating transcription factor-6 (ATF6), and inositol-requiring enzyme-1 (IRE1) 14 . Activation of PERK induces the phosphorylation of the eukaryotic translation initiation factor 2-alpha (eIF2α) and of transcription factor erythroid 2-related factor 2 (NRF2) 15 . NRF2 phosphorylation initiates a cell-protective anti-oxidant response, while eIF2α phosphorylation causes a transient global suppression of protein translation, reducing ER protein load and stress 16,17 . In addition, phosphorylated eIF2α enhances the translation of activating transcription factor-4 (ATF4), which upregulates beneficial UPR target genes required for autophagy, antioxidant response, and amino acid metabolism 18 . Direct PERK activation was never explored until recently, when neuroprotection by compound CCT020312 was reported in models of tauopathies 19 . These protein aggregation disorders share many common pathophysiological pathways with HD, particularly high ER stress status, where modulation of PERK could result in beneficial effects. PERK modulation has not been evaluated so far in animal models of HD. We had shown that the PERK pathway is strongly downregulated in striatal neurons compared to other cell types and brain regions in WT mice and is upregulated in HD 13 . We initially interpreted this activation of the PERK pathway as a possible cause of cellular dysfunction and toxicity. However, the results that we report here suggest that PERK activation in response to mHtt expression is an insufficient cellular attempt to restore homeostasis. Here we synthesized and tested novel structurally-related derivatives of a PERK modulator, finding several compounds that reduced mHtt cytotoxicity in a cellular HD model (a murine striatal cell line with knock-in of a full-length polyQ-expanded mHtt (STHdh Q111/111 ) 20 . The effective compounds are predicted to interact with the PERK activation loop. For the most effective one, MK-28, we determined that it is a potent and selective PERK activator that strongly reduces cytotoxicity and improves motor and executive functions, extending the lifespan of the most aggressive HD mouse model, the R6/2 with 160 CAG repeats.

MK-28 is non-toxic and neuroprotective in HD model striatal neurons. We had previously
reported protective effects of A4, a compound that reversed the toxicity of mHtt expression in a striatal neuron cell line 13 . A4 had been found in a screen for PERK inhibitors 21 . We now synthesized small molecule derivatives of the mother compound A4 and tested their ability to prevent cytotoxicity induced by pathogenic mHtt (Fig. 1). The derivatives were closely related analogues, in order to evaluate the effect of alternative substitution group patterns (Fig. 1A,B). We used the stable murine striatal cell line expressing full-length polyQ-expanded mHtt (STHdh Q111/111 ) for the rescue experiment. Although as an immortalized cell line these cells show low levels of apoptosis, we had observed that they are much more sensitive than their counterpart expressing WT Htt (STHdh Q7/7 cells) and than other cell types to the ER stressor and apoptosis inducer tunicamycin (Tun) 13 . We incubated STHdh Q111/111 cells with Tun and in parallel with increasing concentrations of the different PERK modulators. Apoptosis was measured after 48 h of Tun exposure and data analysis showed a strong cellular protection by MK-28 in a dose-dependent manner, reducing apoptosis by 5-fold at 100 μM (Fig. 1C, Supplementary Fig. S1). The protective effect of MK-28 was stronger than that of the mother compound A4 and of the other four derivatives tested (MPGL-1, MK26, MK-29, MK-30). MK-28 was particularly more effective than A4 at low concentrations, reducing apoptosis by 20% at 5 μM compared to no significant effect by A4. The activity appears to be highly dependent on the substitution pattern, with active compounds having an OH in position R1 and an OCH3 or preferably an OH in position R (Fig. 1A). When the effect of MK-28 was compared between STHdh Q111/111 and STHdh Q7/7 cells, a concentration was reached (50 μM) that showed identical levels of apoptosis (Fig. 1D). Therefore, MK-28 compensated the additional toxicity due to the expression of pathogenic huntingtin. MK-28 effect on the PERK pathway in striatal cells expressing mHtt. The effect of MK-28 was further tested in STHdh Q111/111 cells, compared to STHdh Q7/7 cells. The cells were treated for 1.5 hours with increasing concentrations of MK-28 and the levels of phosphorylation of the PERK substrate eIF2α (eIF2α-P) ( Fig. 2A) were determined. Reflecting their ER stress status 13 , STHdh Q111/111 cells showed three-fold higher levels of eIF2α-P than STHdh Q7/7 cells at baseline with no MK-28 exposure, as we had seen before 13 . MK-28 did not reduce eIF2α-P levels but instead caused an increase for both cell types, similar to that caused by an ER stressor, thapsigargin (Thap). The increase was significant, up to almost 2-fold, for STHdh Q7/7 cells (Fig. 2B). This was unexpected, because the mother compound had been found in a screen for PERK inhibitors. We then looked directly at PERK. This experiment was done with MEF cells, which showed a much higher PERK level than the striatal cells and could therefore be clearly visualized. MK-28 treatment caused a shift of part of the PERK population to a slower migration, consistent with PERK-P and similar to that obtained by cell treatment with Thap (Fig. 2C). The lower signal at 5 μM MK-28 is due to a shift of part of the population to higher MW, possibly dimers and oligomers of activated PERK ( Supplementary Fig. S6)). Downstream factors of the PERK pathway were also upregulated, including ATF4, CHOP and GADD34 (Fig. 2D-F). ATF4 protein levels were increased significantly, up to 2.5-fold, in STHdh Q7/7 cells at high MK-28 concentration, whereas there was a large variation in STHdh Q111/111 cells between experiments. CHOP and GADD34 mRNA levels were measured by quantitative real-time PCR (qPCR) and showed a significant increase in both cell types, up to 10-and 5-fold respectively (Fig. 2E,F).

MK-28 activates PERK in vitro.
To understand the mechanism of action of MK-28, the effect was analysed in vitro on the activity of isolated PERK (EIF2AK3) (performed at Reaction Biology Corp. (Malvern, PA, USA) using the HotSpot Kinase Assay, see Methods). MK-28 activated PERK, starting in the nanomolar range, with an EC50 of 490 nM (Fig. 3A, Supplementary Fig. S2). The results, especially in the high concentration range, may reflect allosteric effects, which should be further investigated in future studies. As expected, a commercial PERK inhibitor, GSK606414 22 showed a clear inhibitory effect in this assay. In addition, we also tested the effect of MK-28 on the other three eIF2α kinases. MK-28 had little or no effect on EIF2AK1 (HRI) or EIF2AK2 (PKR), but it activated EIF2AK4 (GCN2), although at almost one order of magnitude higher EC50 than with PERK (3.5 μM) ( Fig. 3B and Supplementary Fig. S2). The specificity of MK-28 was evaluated on a panel of 391 kinases, showing the highest hit for PERK activation (EIF2AK3, 81,4% increase over control), with only 3 other hits showing more than 30% increase over control (Fig. 3C, Supplementary Table S1).
Homology modeling and compound docking may provide an explanation for the activation of PERK, revealing poses for MK-28 and also for A4 that are predicted to interact with the kinase activation loop of PERK (Fig. 4A). Interactions (mainly hydrophobic) were predicted for MK-28 with several residues of the activation and novel derivatives. (C) Apoptosis rescue experiment using the murine striatal cell line expressing full-length polyQ-expanded Htt (STHdh Q111/111 ) incubated with the ER stress-inducer tunicamycin, shows that MK-28 and to a lesser extent A4 and MPGL1 are able to rescue the cells from apoptosis. Apoptosis was measured after 48 h of Tun exposure and PERK modulators were tested at different concentrations (0-100 μM). Results are expressed as average relative apoptosis rate ± SD of 3 independent experiments; apoptosis in the presence of tunicamycin and in the absence of other compounds was normalized to 100 to allow comparison between experiments. P values MK-28 vs. only Tun: 5 μM = 6 × 10 -5 , 50 μM = 1 × 10 -9 , 100 μM = 5 × 10 -13 ; A4 vs. only Tun: 50 μM = 8 × 10 -5 , 100 μM = 0.001. (D) Rescue of STHdh Q111/111 cells compared to STHdh Q7/7 cells by MK-28. The graph shows average relative apoptosis rate ± SD of 3 independent experiments. P values vs. only Tun: STHdh Q7/7 cells: 5 μM = 2 × 10 -5 , 10 μM = 0.0002, 50 μM = 0.004, STHdh 111/111 cells: 5 μM = 0.006, 10 μM = 0.008, 50 μM = 8 × 10 -6 . loop (Fig. 4B). Interestingly, the compounds that did not protect against cytotoxicity in STHdh Q111/111 cells ( Fig. 1) interacted with a region close to the catalytic pocket of PERK. This was true also for the PERK inhibitor (GSK2606414). These compounds showed no poses interacting with the region of the activation loop (Fig. 4A). These results would predict that also A4 might be a PERK activator. We tested this prediction using the in vitro kinase activity assay. Indeed, A4 activated PERK (EIF2AK3) in the nanomolar range, although to a small extent ( Supplementary Fig. S3).
We then compared the protective effect of MK-28 to that of a commercial PERK activator (CCT020312) on the survival of STHdh Q111/111 cells treated with Tun (Fig. 4C). Over three independent experiments, both compounds protected the STHdh Q111/111 cells similarly at high concentrations (>50 μM), but this was not the case at lower concentrations, where CCT020312 failed to be protective below 10 μM (Fig. 4C). In contrast, MK-28 reduced apoptosis by 40% at 10 μM and was also effective at lower concentrations. The IC50 calculated for inhibition of apoptosis in this experiment is 6.8 μM for MK-28 and 32.4 μM for CCT020312 ( Supplementary Fig. S4A). To evaluate whether the cytoprotective activity of MK-28 depends on PERK, we used PERK−/− MEF cells compared to WT MEFs. Both MK-28 and CCT020312 failed to protect the PERK−/− cells, indicating that the protective mechanism is indeed PERK-dependent (Fig. 4D). In addition, the protective effect was stronger in the WT MEFs than in the STHdh Q111/111 cells, with almost an order of magnitude difference in the efficacy of the compounds, IC50 = 1.7 μM for MK-28 and 11.1 μM for CCT020312 ( Supplementary Fig. S4B). We also tested whether any of the cytoprotective effect was due to activation of another PERK substrate, NRF2. The mRNA levels of a downstream target of NRF2-P, HMOX1, showed upregulation by MK-28, as expected, in STHdh Q7/7 and STHdh Q111/111 cells (Suppl. Fig. S5A-B). We then performed an experiment similar to that in Fig. 1C, but with the addition of increasing concentrations of ML385, an NRF2 inhibitor, to test whether it would block the beneficial effect of MK-28. ML385 caused some toxicity, as expected, in STHdh Q111/111 cells but did not reverse to any significant extent the cytoprotective effect of MK-28 (Suppl. Fig. S5C-D). Cell protection by MK-28 does not appear to be through NRF2.
In vivo evaluation of MK-28. We analysed pharmacokinetic aspects of MK-28 such as plasma stability, blood brain barrier (BBB) penetration and brain bioavailability in mice after a single IP injection of 10 mg/kg. The encouraging results showed a maximum concentration (Cmax) of 105 ng/ml and 30 min half-life in plasma, 40 min after the IP injection (Fig. 5A). In addition, very good brain bioavailability and BBB penetrance after 20 min of IP injection were also observed. The Cmax found in the brain was 57 ng/g and was obtained 40 min after the injection with a half-life of 80 minutes. Importantly, 57 ng/g is more than half of the Cmax found in plasma and the area under the curve (AUC) for the concentration in the brain was 22% of that in the plasma.
No toxic effects were observed by MK-28 administration (6 mg/kg, 6 times the dose used later for the therapeutic study) over 2 weeks of continuous delivery using subcutaneous Alzet pumps in B6 wild type mice. For therapeutic efficacy evaluation, the R6/2 HD mouse model with 160 CAG repeats was used, and subcutaneous Alzet pumps with 28-day compound delivery were placed in four-week-old mice (Fig. 5B). Given the pharmacokinetic  Supplementary Table S1 for complete results for all 391 kinases). www.nature.com/scientificreports www.nature.com/scientificreports/ properties of MK-28, we chose a dose of 1 mg/kg, which is well in the usual therapeutic range. Five groups were studied: 1) TG treated with MK-28 (1 mg/kg), 2) TG without treatment (pumps with vehicle), 3) Wild type treated with MK-28 (1 mg/kg), 4) Wild type without treatment (pumps with vehicle), 5) Wild type mice without pumps.
Weight evaluation during 7 weeks after pump implantation showed no significant differences between animals that received MK-28 or vehicle (p = 0.3292) (Fig. 4C). As anticipated, R6/2 mice exhibited a marked decrease in their weight in comparison to WT animals starting at week 5 after pump implantation (23.72 ± 0.59 g vs 23.99 ± 0.86 g, respectively) and until the end of the experiment by week 9 (16.34 ± 0.84 g vs 28.28 ± 1.23 g, respectively) (p < 0.001).
We assessed the motor behavior of the mice using rotarod analysis, at different time points within seven weeks, starting one week after the first MK-28 delivery. No significant differences were observed between the five groups during the first three weeks of treatment (p>0.05, Fig. 5D). Starting from the fourth week of treatment, at an age of eight weeks, significant differences in motor performance between transgenic (TG) and their control WT littermates was evident. Moreover, at this point MK-28 treated TG mice showed significant improvements in motor performance than those without treatment (p = 0.030, week 4). This effect was sustained for three weeks (p = 0.042 and p = 0.012, weeks 5 and 6, respectively) until both TG groups showed similar performances (p = 0.089) at week 7 after pump implantation (3 weeks after the end of the treatment). WT mice without any intervention performed similarly to those that received pumps filled with vehicle or MK-28 compound, showing no differences between groups and no compromised motor performance associated with MK-28 administration in WT animals.
Disease severity was evaluated using a composite phenotype test that includes the combined analysis of hind limb clasping, ledge walk, gait and kyphosis 23 . Our analysis revealed that TG and WT mice performed differently starting from week four after pump implantation (2.00 ± 1.08 vs 0.65 ± 0.79, respectively p = 0.0004, Fig. 5E), also in concordance with the rotarod results (Fig. 5D). From week 6 and until the end of the analysis (4 weeks after end of treatment), TG mice treated with MK-28 showed lower scores than those without treatment (3.14 ± 1.41 vs 4.33 ± 1.23, p = 0.016 for week 6 and 4.4 ± 0.55 vs 5.4 ± 0.89, p = 0.032 for week 8, respectively), indicating a positive effect of the treatment in gait, clasping and general motor function (Fig. 5E). Again, WT animals www.nature.com/scientificreports www.nature.com/scientificreports/ with or without pump or MK-28 performed similarly. As a reflection of both motor and cognitive abilities during the symptomatic phase, we measured mice vocalizations which are also affected in HD patients. The number of vocalizations were assessed in male mice when exposed to a new female in the same cage. The TG R6/2 mice produced on average 5 syllables in 5 min, while MK-28-treated mice exhibited a three-fold increase with 16 syllables in 5 min (p = 0.059, Fig. 5F). Although these results did not reach statistical significance, they show an interesting behavioral trend of improvement.
As a systemic parameter that is affected in HD patients and in some animal models during advanced stages, blood sugar levels were analysed. We found that during the asymptomatic phase no difference in blood sugar levels was observed between the TG and WT groups (144.80 ± 6.23 g/dl vs 159.14 ± 9.48 g/dl, respectively) (Fig. 6A). However, during the symptomatic phase TG mice blood sugar levels significantly increased compared to those of the asymptomatic phase (p = 0.028). In contrast, TG mice treated with MK-28 or WT control groups showed no significant increase compared to their asymptomatic levels. Analysis of the glucose increase ratio between disease stages confirmed a different behavior elicited by MK-28 treatment (p = 0.0045), where TG animals showed an increase, 144.30 ± 6.35% and TG + MK-28 maintained glucose levels similar to those of the presymptomatic phase and to their WT littermates, 109.60 ± 5.04% (Fig. 6B).
Lifespan evaluation showed that MK-28 treatment increased the median survival of TG mice by 8 days compared to controls (97d vs 89d respectively, Chi square p = 0.046; Gehan Breslow Wilcoxon p = 0.012) (Fig. 6C). In addition, death initiation was also delayed by 10 days. This observed effect on increasing survival and delaying death initiation occurred several weeks after the treatment ended.
Finally, results from the analysis of slices of mice brain striatum, at week 4 after pump implantation, show that eIF2α-P levels were slightly increased in the striatum of the TG animals when compared to the WT at baseline, but they were significantly increased upon treatment with MK-28, both for WT (~7-fold, p = 0.003) and for TG Analysis of the striatum in mice brain slices by immunofluorescence. eIF2α-P levels (red: total eIF2α, green: eIF2α-P) were slightly increased in the brain of the TG animals when compared to the WT at baseline, and they were significantly increased upon treatment with MK-28 in both TG and WT mice. Bar = 20 μm. The bars in the graph indicate medians and interquartile range. Significance **p (eIF2α-P WT vs. WT + MK-28) 0.003, p (eIF2α-P HD vs. HD + MK-28) 0.007, Two-tailed Ttest. www.nature.com/scientificreports www.nature.com/scientificreports/ (~6-fold, p = 0.007), consistent with MK-28 reaching the brain striatum and having an activating effect on eIF2α phosphorylation (Fig. 6D).

Discussion
ER stress is a common feature shared by diverse neurodegenerative diseases. Stress induced by accumulation of misfolded mHtt in the ER is observed in HD 7,9 . To cope with ER stress, cells activate the UPR to recover protein homeostasis or trigger apoptosis when adaptive measures are insufficient 18 . Despite our growing understanding about how the UPR alternates between those options, knowledge is still limited about their fine tuning and how to use those mechanisms for therapeutic interventions. Opposite strategies modulating ER stress through PERK pathway inhibition or activation have been tested positive in ameliorating the disease course of protein aggregation diseases 19,[24][25][26] .
We had previously found that striatal neurons have an unusually low level of eIF2α-P, and pathogenic mHtt expression causes a strong increase in eIF2α phosphorylation by PERK 13 . Therefore, we speculated at first that inhibition of PERK could be a way to restore homeostasis and achieve therapeutic results. A4, one of the top hits in an unbiased screen for PERK inhibitors 21 , had a protective effect in cells subjected to mHtt-induced ER stress 13 . We then synthesized 5 different but related small molecules derived from A4 (Fig. 1A,B). The fact that we found that our lead compound, MK-28, is a PERK activator was thus a real surprise. MK-28 was selected by being the most efficient in rescuing STHdh Q111/111 cells from ER stress-induced apoptosis (Fig. 1C). MK-28 activation of PERK using purified components (Fig. 3) suggests that also its activity in cells is directly on PERK and not by an upstream event such as causing ER stress. Reevaluation of the mother compound A4 showed that it is also a PERK activator although mild (Supplementary Fig. S3). Both MK-28 and A4 were predicted to interact with the activation loop of PERK (Fig. 4A,B). This is an interesting finding that should be verified experimentally in future studies. The activation of the eIF 2α kinase GCN2 (EIF2AK4) by MK-28. although at one order of magnitude higher than the one observed with PERK ( Fig. 3B and Supplementary Fig. S2), could also be beneficial. GCN2 is also involved in the integrated stress response and its activation under stress conditions exerts cell protective effects, for example in neurodegeneration caused by ribosome stalling [27][28][29] . The specificity of MK-28 was further demonstrated by a highly selective activation of PERK from a panel of 391 kinases ( Fig. 3C and Supplementary  Table S1) and especially important is the lack of protective effect in PERK−/− cells (Fig. 4D), showing that cell rescue from ER stress-induced apoptosis by MK-28 is exclusively through PERK.
EIF2α-P levels showed a consistent but small increase in both STHdh Q7/7 and STHdh Q111/111 cells upon MK-28 treatment, coherent with a PERK activation paradigm ( Fig. 2A,B). We observed a much stronger effect of MK-28 in the brains of treated R6/2 and WT mice. eIF2α-P levels were slightly increased in the striatum (the most sensitive region in HD) of R6/2 mice at baseline, similar to what we had seen before in another HD model, (N171-82Q mice) 13 . The significant further increase of eIF2α-P levels caused by MK-28 treatment (Fig. 6D) is consistent with an activating effect on eIF2α phosphorylation in vivo. The changes are in line with our results in STHdh Q7/7 and STHdh Q111/111 cells (Fig. 2), except that the increase in eIF2α-P levels was much more prominent in the mice. This is possibly due to the long-term effect of MK-28 in the mice, compared to the short-term treatment of the cells in culture. There might be a contribution of GCN2 activation to the eIF2α-P levels in the mice, although, as mentioned above, the results with PERK−/− cells suggest that the protective effects are exclusively through PERK. The detailed effects of MK-28 on the PERK pathway in vivo should be further explored in future studies. In the cultured cells it is also possible that the induction of GADD34 (Fig. 2) reduces eIF2α phosphorylation at the measured time of MK-28 treatment. Besides GADD34, we also observed upregulation of ATF4 and CHOP. The induction of CHOP by MK-28 appears to be still in the protective phase of the UPR, as it did not lead to apoptosis, but the contrary, cells were protected from ER stress. MK-28 exerts its cell-protective effect likely through the activation of the PERK pathway downstream of eIF2α-P, and not through NRF2 activation, as NRF2 inhibition did not affect cell rescue by MK-28 (Suppl. Fig. S5).
MK-28 exhibited remarkable pharmacokinetic properties in mice. It becomes readily available and is stable in plasma, penetrates the blood brain barrier and is stable inside the brain following a single IP injection (Fig. 5A). Brain levels of MK-28 were more than half of those detected in plasma, evidencing efficient bioavailability for brain tissue. We evaluated the effects of MK-28 in the HD R6/2 mouse model with 160 CAG repeats 3 , which exhibits a rapid progressive HD-like behavioral and neuropathological phenotype. This strain is recommended to be used as a gold standard for therapeutic trials in HD 30,31 , but given the massive and very rapid progression of the disease in this model with 160 CAG repeats, sometimes the therapeutic rescue window might be too short to elicit a significant change. R6/2 mice present dystonia with limb clasping by 6 weeks, body weight loss by 5-9 weeks, reduced hindlimb and forelimb performance by 9 and 11 weeks, rotarod changes by 5 weeks, seizures and diabetes 30 . Our results were consistent with these and remarkably, our treatment ameliorated all of them while also extending mice survival (Figs. 5, 6). MK-28 was delivered continuously in 4-week-old mice for 28 days by subcutaneous minipumps, using a concentration in the range of those used in clinical settings. MK-28 exhibited long lasting effects since it lasted for several weeks after the end of the treatment (Figs. 5, 6). MK-28 treated mice showed improvements in motor performance, effect most likely induced by delayed neuronal degeneration. We were able to quantify the temporal progression of the treatment by using a composite phenotype 23 , and showed that MK-28 effectively delayed the appearance of the symptoms (Fig. 5E). Furthermore, our 28-day treatment in young mice extended median survival and death initiation by 10% (Fig. 6C). Although the motor improvement and extension of survival exhibited by MK-28 delivery in our settings might not seem dramatic, they are consistent across different evaluations of neuromotor and systemic parameters. We believe that the combination of the rapid progressing model R6/2 160 CAG, a short therapeutic intervention window and the relatively short treatment, resulted in an apparently modest improvement. We envision that our therapeutic approach could perform even better in future studies in slow progressing models with a chronic supply of the compound, which could be closer to the human disease scenario. Speech is another function generally affected in HD, since choreiform (2020) 10:6875 | https://doi.org/10.1038/s41598-020-63899-4 www.nature.com/scientificreports www.nature.com/scientificreports/ movements and degeneration of the caudate nucleus, putamen and globus pallidus also disrupt systems involved in verbal communication 32,33 . While no signs are observed early, dysarthric speech evolve with the disease and sometimes a total communication impairment is observed 33 . In the seminal study by Mangiarini, the creators of the R6 strain, they describe chorea in face, jaw, and pharyngeal muscles affecting speech 3 . R6/1 mice emit short duration or nearly no vocalizations when compared to WT littermates 34 . We also found preliminary evidence, that R6/2 mice barely produced syllables if at all, while MK-28-treated mice exhibited a three-fold increase in the number of syllables produced within 5 min (Fig. 5F). The results are not statistically significant (p = 0.059), but show an interesting behavior of ameliorating the evaluated phenotype. We interpret this data as a trend and we treat it as merely preliminary, given the low number of animals tested. The number of animals tested was low because of the rapid death of R6/2 model mice and the late endpoint chosen for the analysis, resulting in being significantly underpowered from a statistical point of view. Future studies should include an increased number of animals to test and choose earlier endpoints, where the proportion of live animals is significantly higher. We speculate that vocalization improvements are a result of delaying neuronal malfunction and degeneration at different levels, likely cortical and striatal neurons, but we cannot pinpoint from our data whether the effects are on specific neuronal populations.
HD neuropathology is also accompanied by mHtt expression and aggregation in tissues other than the central nervous system 35 . HD subjects show diverse endocrine changes, abnormalities of carbohydrate metabolism and increased prevalence of diabetes [36][37][38] . R6/2 mice exhibit pancreatic inclusions associated with hyperglycemia and insulin-deficient diabetes 39,40 . We also found that R6/2 blood sugar levels were elevated during the symptomatic phase, but not before (Fig. 6A,B). MK-28 reduced these elevated levels to those found in WT and asymptomatic R6/2 mice. Although the number of animals that were present at both timepoints, asymptomatic (n = 3) and symptomatic stages (n = 6), was low to draw strong conclusions, the treated group behaved clearly differently from the untreated group and the difference reached statistical significance (p = 0.0045). This effect might occur through multiple mechanisms involving PERK activation.
Altogether our results suggest that MK-28-dependent PERK activation is an effective and selective strategy to reduce ER stress, reestablish ER homeostasis and increase the survival of cells and mice expressing mHtt (Fig. 7). Consistent with this, Bruch et al. recently observed that PERK activation using CCT020312 is neuroprotective in a model of tauopathies with high ER stress 19 . We compared CCT020312 to MK-28, to find that MK-28 was more efficient (with an IC50 ~3-fold higher) in rescuing mHtt-expressing striatal neurons from apoptosis (Fig. 4C). This mechanism is consistent with previous reports using another strategy for PERK pathway activation in other protein aggregation disorders, by inhibiting eIF2α-P dephosphorylation 26,[41][42][43] . This is accomplished by inhibition of the eIF2α-P phosphatase regulatory subunit GADD34 (PPP1R15A) or the constitutive subunit CREP (PPP1R15B). GADD34 is induced downstream of ATF4, resulting in recovery from the PERK-mediated block in protein synthesis 44 . This strategy was also beneficial in a cellular model of HD (PC6.3 cells expressing mHtt fragments) 10 and in an HD mouse model (N171-82Q) 43 . This approach may prolong the benefit of transient inhibition of protein synthesis by eIF2α-P, but has the caveat that restoration of protein synthesis in the long term by eIF2α-P dephosphorylation is also inhibited, and in turn, pro-apoptotic pathways may be activated. In fact, GADD34 inhibition had negative effects in at least one report, where a mutant SOD1 ALS mouse model was used 45 . In contrast, our approach of directly activating PERK stimulates the beneficial transient inhibition of protein synthesis, alleviating the ER load, and does not affect the activity of CREP or the activation of GADD34, consequently avoiding pro-apoptotic outcomes and restoring translation in the long run (Fig. 7).
The benefit of PERK activation is in line with the fact that loss of functional PERK (Wolcott-Rallison syndrome) or PERK missense mutations cause neurodegeneration 46,47 . Although PERK inhibition showed beneficial effects for some diseases 24,25,[48][49][50][51][52] , the side effects of chronic treatment are of major importance. A strong PERK inhibition as that caused by GSK2606414, can lead to a Wolcott-Rallison syndrome-like phenotype produced by the loss of function of PERK 53 . The benefit of transient inhibition of protein synthesis by eIF2α-P is not achieved, and this is detrimental and toxic to neurons as well as secretory tissues, such as the pancreas.
The compound that we developed, MK-28, was not toxic at high concentrations, although this was tested only in the short-term and should be analysed in the long term for the possibility of deleterious effects. MK-28 exhibited a notable pharmacokinetic profile and brain bioavailability in mice. Moreover, our compound showed a selective and efficient PERK activation effect that was sufficient to rescue cellular and animal models of HD. Our data support further evaluation of MK-28 in slow progressing HD models such as BACHD and knock-in models, since those more closely resemble the slow pathological progression observed in humans. Currently, there is a complete lack of selective therapeutic options for HD, thus we consider MK-28 as a potential safe therapeutic agent for this disease. Finally, since ER stress is also present and drives all of the most prevalent protein misfolding diseases, we speculate that MK-28 could also be effective in those related pathological scenarios.

Methods
Study design. The objective of this study was to evaluate preclinically the safety, kinetics and efficacy of a new compound that we synthesized, a derivative of a PERK modulator, in models of HD. We determined that the compound is a selective PERK activator. A therapeutic strategy based on the over-activation of PERK represents an unexplored and promising approach for treating HD, where protein aggregation drives the pathological progression of the disease and the development of ER stress is a main cytotoxic factor. We first investigated the effects of several small molecule derivatives of the PERK modulator A4 in a cellular model of HD and also under acute ER-stress. For all the in vitro studies, at least three independent experiments with internal replicates were performed. We then selected a compound (MK-28) with the best cytoprotective effects on a neuronal model of HD. This compound, which activated PERK with purified components in vitro, was further tested in vivo in a mouse HD model, the R6/2 strain harboring 160 CAG repeats. Power calculations determined that 16 mice per group was the minimum required to see at least a 20% increase (alpha 0.05, beta 0.2, power 80%) in motor www.nature.com/scientificreports www.nature.com/scientificreports/ performance and survival of R6/2 mice with MK-28 subcutaneous delivery compared to vehicle. Mice were coded and randomly assigned to different groups. Investigators were blinded to the different treatment groups during the experiment. Multiple endpoints were prospectively selected for the in vivo part, consistent with the analyses we here report. Data collection was ended before signs of animal suffering or inability to thrive appear. All in vivo experiments followed institutional guidelines and were approved by the Institutional Animal Care and Use Committee of Tel Aviv University.

Materials.
Tun, thapsigargin, ML385 and common reagents were from Sigma. GSK606414 was from APExBio and CCT020312 from Calbiochem. Mouse anti-actin and rabbit anti-calnexin antibodies were from Sigma. Mouse anti-eIF2α was from Cell Signaling. Rabbit anti-phospho-eIF2α (Ser51) was from MBL. Rabbit anti-PERK was from Abcam. Rabbit anti-ATF4 was a kind gift from David Ron (Univ. Cambridge). Goat anti-mouse IgG conjugated to 488 DyeLight, goat anti-rabbit IgG conjugated to 564 DyeLight, and goat anti-rabbit and anti-mouse IgG conjugated to HRP were from Jackson Labs. Normal goat serum was from Vector Laboratories (Burlingame, CA).

Compound generation. See detailed information in Supplementary methods.
Toxicity assay in cells in culture. STHdh Q111/111 cells (used before 13 ) or PERK+/+ or PERK−/− MEFs (kindly provided by David Ron, Univ. Cambridge) were incubated with Tun (10 μg/ml) for 48 h, which results in ER stress-induced cell death. Cells under the same conditions were incubated for the same times, together with Tun, with increasing concentrations of the different PERK modulators (0-120 μM) and the percent of apoptotic cells was measured by FACS cell cycle progression using propidium iodide (PI), considering the fraction beneath G0/G1. The relative apoptosis rate was calculated to be able to compare between compounds and experiments, where for each concentration curve the percent of apoptotic cells was considered 100 at concentration 0 of the Figure 7. Model of the effects of mutant Htt and the action of MK-28. Accumulation of misfolded mHtt inhibits ERAD, causing ER stress and inducing the activation of the UPR to either reestablish cellular homeostasis or trigger cell death. In the activation of the PERK pathway of the UPR, unfolded ER proteins bind to the chaperone BiP, allowing oligomerization and autophosphorylation of PERK, which phosphorylates eIF2α. EIF2α-P causes a transient global suppression of protein translation, reducing the ER protein load and ER stress. EIF2α-P also promotes the translation of ATF4, upregulating beneficial UPR target genes required for autophagy, antioxidant response, and amino acid metabolism. MK-28 activates PERK, extending the beneficial transient inhibition of protein synthesis and alleviating the ER load, promoting cell survival. This strategy does not affect the activation of GADD34, which promotes the restoration of translation in the long term by reduction of eIF2α-P. This, together with the reduction in ER stress prevents initiation of apoptosis.
Treatments and immunoblotting. STHdh Q7/7 and STHdh Q111/111 cells were treated with different concentrations of MK-28. The concentration range was chosen in accordance to previous studies with A4. Cells were lysed with PBS containing 1% Triton X-100, 0.5% Sodium deoxycholate with protease and phosphatase inhibitors cocktails. Cocktail 2 and 3 (Sigma) were added to the lysis buffer for detection of phosphorylated proteins and lysates were incubated on ice for 30 minutes, then centrifuged at 16,000xg for 30 minutes a 4 °C. The supernatants were boiled in loading buffer for 5 min and loaded on 12% reducing SDS-PAGE, except for PERK where 6 M urea was added before boiling and 8% SDS-PAGE was used. Immunoblotting, detection by ECL and quantitation was done as described previously 13 .
Real time PCR. Total cell RNA was extracted with TRIzol reagent (Invitrogen) according to the manufacturer's instructions. Reverse transcription (RT) was performed with QuantaBio qScript cDNA synthesis kit. Real-time qPCR was carried out using a CFX Connect Real-Time PCR Detection System (BioRad) and the amplifications were done using the SYBR Green PCR Master Mix (pcrbio). Reactions were run in triplicates in three independent experiments. Housekeeping gene GAPDH was used as an internal control to normalize the variability of expression levels.
In vitro hotspot kinase assay. In vitro kinase assays with purified components and full kinase panel analysis was done at Reaction Biology Corp. (Malvern, PA, USA) using the HotSpot Kinase Assay. Reactions were performed for 20 min at room temp. using 33 P-ATP (Specific activity 10 μCi/μl) with 1 μM total ATP, 20 μM substrate and kinases in the nanomolar range in 20 mM Hepes (pH 7.5), 10 mM MgCl2, 1 mM EGTA, 0.02% Brij35, 0.02 mg/ml BSA, 0.1 mM Na3VO4, 2 mM DTT, 1% DMSO and other required factors depending on the assayed kinase. GST-tagged, recombinant human, catalytic domain (amino acids 536-1116) of PERK, expressed in insect cells was used (Life technologies, Carlsbad, CA).

Pharmacokinetics and BBB penetration. See detailed information in Supplementary methods.
Huntington's disease mouse model. All experimental procedures were approved by and conducted in strict compliance with the Institutional Animal Care and Use Committee of Tel Aviv University. Mice were kept in a 12/12-hour light/dark cycle with food and water ad libitum. The HD mouse model used was the R6/2 TG mice model B6CBA-R6/2 (CAG 160) (Jackson Lab, catalog number #002810, USA). At three weeks mice were assessed for CAG repeat size using the following primer set: 5'-ATGAAGGCCTTCGAGTCCCTCAAGTCCTTC-3′ and 5′-GGCGGCTGAGGAAGCTGAGGA G-3′. Following Jackson laboratories guidelines, wild type (WT) and TG littermates were kept combined in the same minimally-enriched cages (with cardboard rolls and gauze pads), and after week 8 water and mushed food were placed at the ground level for all the cages. Male mice treatment groups were coded and remained blind to the observers throughout the experiment. Power calculations determined that 16 mice per group was the minimum required to see at least a 20% increase (alpha 0.05, beta 0.2, power 80%) in motor performance and survival of R6/2 mice with MK-28 subcutaneous delivery compared to vehicle. Multiple endpoints were prospectively selected, consistent with the analyses we here report. Data collection was ended before signs of animal suffering or inability to thrive appear.
Osmotic minipump preparation and implantation. Alzet pumps model 1004 (Alzet, USA), with a rate of 0.11 µl/hour and 28 days effective compound delivery was used. MK-28 compound was dissolved in 1:1 DMSO/PEG400 and prepared to delivery subcutaneously 6 mg/kg MK-28 for toxicity studies (n = 8) or 1 mg/kg for therapeutic evaluation in R6/2 mice (n = 72). For toxicity studies mice receiving six times the therapeutic dose of MK-28 or vehicle were weighed every two days for two consecutive weeks. For the therapeutic evaluation of MK-28, five groups of mice were studied: 1) R6/2 TG mice treated with MK-28, 2) TG without treatment (pumps with vehicle DMSO/PEG400 1:1), 3) WT treated with MK-28, 4) WT without treatment (pumps with vehicle DMSO/PEG400 1:1), 5) WT without pumps. Alzet pumps were implanted in 4 weeks old mice and MK-28 was administered at 1 mg/kg with continuous dosing for 28 days. All mice were anaesthetized with ketamine (80 mg/ kg IP) and xylazine (10 mg/kg IP) before subcutaneous pump implantation.
Rotarod test. Motor activity was assessed by the rotarod test (San Diego Instruments, CA, USA) (0-25 RPM). Relative slow acceleration was used to discriminate small differences between groups particularly at advanced stages of the disease. All mice were trained together before any experimental procedure was performed to set a common baseline. They were required to stay at least 2 minutes without falling in order to participate in the study. R6/2 mice and control littermates were tested every week after pump transplantation. Each test consisted of 3 consecutive measurements of 5 minutes with inter-trial recess of 5 minutes. The latency to fall was recorded and the best performance of the three trials was considered for the analysis.
Combined phenotype. A rapid and sensitive combined phenotype score was used to quantify disease severity in HD mouse model 23 . Measures include hind limb clasping, ledge test, gait and kyphosis. See detailed information in Supplementary methods. Brain immunohistochemistry. Under anesthesia, animals were perfused with 4% PFA (formaldehyde in phosphate-buffered saline (PBS)) and the brains were removed and soaked in 4% PFA overnight. Then, brains were incubated overnight in 30% sucrose solution and frozen in OCT compound Tissue-Tek by using dry ice. Sections (10 µm) of the striatum were obtained by cryostat. Brain sections were processed for immunochemistry with the kind help of Daniel Frenkel. They were washed with PBS for 10 minutes, incubated in 0.5% triton-X-100 for 10 minutes and then in blocking solution containing 1% bovine serum albumin (BSA), 10% normal goat serum and 0.25% triton for one hour at room temperature. The samples were rinsed with PBS and incubated overnight at 4 °C with the selected primary antibody diluted in blocking solution (eIF2α-mouse [1:400] and phosphorylated eIF2α-rabbit [1:400]). Samples were washed 3 times with PBS and incubated with the proper secondary antibody (564 DyeLight Anti-Rabbit [1:400] or 488 DyeLight Anti-mouse [1:400]) for 2 hours at room temperature. After washing with PBS, samples were placed in DAPI (4′,6-diamidino-2-phenylindole) diluted in PBS for 10 minutes, mounted and examined by confocal microscope Zeiss LSM-510, 3 slides were analysed from each subject. Image analysis. Signal quantification and cell counting were done using the ImageJ program. The value obtained for the phosphorylated eIF2α signal was divided by the total eIF2α signal in the same image. The result was then divided by the number of cells (identified by DAPI staining) for normalization. The final value for each subject was an average of images of 2-4 samples for each group. Two-tailed, paired sample student's t tests were performed. Statistical significance was determined at P < 0.05.
Computational homology modeling. The three-dimensional structure of human PERK without the activation loop (residues 958-984) is available at the Protein Data Bank (PDB code: 4G34) 22 . The binding site of the inhibitor GSK6924 is in the vicinity of the activation loop, thus this loop was homology modeled based on the non-phosphorylated tyrosine kinase (PDB code: 1OPJ). MODELLER 9.17 54 was used for the homology modeling of the activation loop using default parameters.
Computational docking of compounds. Docking studies were performed using Glide SP (standard precision) 55 as implemented in Maestro 11.3. Prior to docking the protein structure was prepared (i.e. missing hydrogen atoms were added, protonation states of titratable residues were determined, and restrained energy minimization of hydrogen atoms was performed to remove clashes) using the protein preparation 56 wizard in Maestro 11.3. The compounds (A4, MK26, MK28, MK29, and MK30) were 3D modeled using LigPrep, which produces chemically correct structures, and then were subjected to a conformational search protocol in Maestro 11.3. Representative conformations of the ligands were used as input for docking calculations. For each ligand, poses were selected based on Glide Emodel Score energies, and analysed for their interactions with binding site residues. The ranking of the interactions of the different analogues in the binding site was based on the Glide Score (gscore).

Statistical analysis.
The results are expressed as mean ± standard error unless otherwise stated. Student's t-test (two-tailed) was used to compare the means of two groups. Comparisons between several groups were performed using ANOVA with Scheffe posthoc analysis. Repeated tests (weekly rotarod) were also analysed using the Two-Way ANOVA test. Statistical significance was determined at P < 0.05 (*), P < 0.01 (**), P < 0.001 (***).
Ethics approval. All in vivo experiments followed institutional guidelines and were approved by the Institutional Animal Care and Use Committee of Tel Aviv University.

Data availability
All data generated or analysed during this study are included in this published article and its supplementary information files.