The small-molecule halofuginone activates the ISR kinase GCN2 and induces ISR-dependent mitochondrial elongation (Perea et al., 2023a; Keller et al., 2012) However, few other compounds are available that selectively activate other ISR kinases through mechanisms that allow for ISR-dependent mitochondrial remodeling (Perea et al., 2023a). For example, BtdCPU activates the ISR kinase HRI through a mechanism involving mitochondrial uncoupling, precluding its use for probing ISR-dependent protection of mitochondria (Perea et al., 2023a). To address this limitation and define the potential for pharmacologic activation of other ISR kinases to promote adaptive mitochondrial elongation, we established and implemented a screening platform to identify compounds that activated ISR signaling downstream of alternative ISR kinases (Figure 1A). In this screen, we used the ATF4-FLuc translational reporter of the ISR (Figure 1—figure supplement 1A; Yang et al., 2023a) We confirmed that ISR-activating stressors including the ER stressor thapsigargin (Tg) and the ATP synthase inhibitor oligomycin A (OA) robustly activated this reporter (Figure 1—figure supplement 1B). We then used this reporter to screen the ~3k nucleoside mimetic analog compound library (10 µM) and monitored ATF4-FLuc activity 8 hr after treatment. Our primary screen identified 34 hit compounds that activated the ATF4-FLuc reporter with a robust Z-score of greater than 3 fold. We then removed highly reactive compounds and pan-assay interference compounds (PAINS), which reduced the number of hits to 9 (Figure 1B). These compounds were re-purchased and then tested in dose response for ATF4-FLuc activation (Figure 1—figure supplement 1C). This identified compounds 0357 and 3610 as the compounds that most efficaciously activated the ATF4-FLuc reporter, albeit with low potency (EC₅₀ > 10 µM). We confirmed that co-treatment with the highly selective ISR inhibitor ISRIB (Sidrauski et al., 2013; Zyryanova et al., 2018) blocked ATF4-FLuc activation induced by these compounds, confirming this activation can be attributed to the ISR (Figure 1C and D). Further, we used qPCR to show that treatment with 0357 or 3610 increased expression of the ISR target genes ASNS and CHAC1 in HEK293 and MEF cells (Figure 1—figure supplement 1D and E; Kreß et al., 2023; Grandjean et al., 2019). Importantly, these compounds did not activate luciferase reporters of other stress-responsive signaling pathway such as the unfolded protein response (UPR; XBP1-RLuc) (Grandjean et al., 2020; Plate et al., 2016), the heat shock response (HSR; HSE-FLuc) (Calamini et al., 2011), or the oxidative stress response (OSR; ARE-FLuc) (Ibrahim et al., 2020; Figure 1E and F). Further, treatment with these compounds did not induce expression of the UPR target gene BiP, the HSR target gene HSPA1A, or the OSR target gene NQO1 in HEK293 cells (Figure 1—figure supplement 1F). Finally, treatment with 0357 or 3610 did not significantly reduce viability of HEK293 cells (Figure 1—figure supplement 1G). These results indicate compounds 0357 and 3610 preferentially activate the ISR compared to other stress-responsive signaling pathways.

Next, we sought to identify the specific ISR kinase responsible for ISR activation induced by these two nucleoside mimetics. We monitored the compound-dependent activation of an ATF4-mAPPLE fluorescent reporter stably expressed in HEK293 cells CRISPRi-depleted of each individual ISR kinase (Figure 1—figure supplement 1A; Guo et al., 2020; Perea et al., 2023a). We previously used this assay to confirm that halofuginone activates the ISR downstream of GCN2 and BtdCPU activates the ISR downstream of HRI (Perea et al., 2023a). Treatment with either 0357 or 3610 activates the ATF4-mAPPLE reporter in control cells (Figure 1G and H). CRISPRi-depletion of HRI, but no other ISR kinase, blocked ATF4-FLuc activation induced by these two compounds. This finding indicates that these compounds activate the ISR downstream of HRI. Collectively, these results identify 0357 and 3610 as nucleoside mimetic compounds that preferentially activate the ISR through a mechanism involving the ISR kinase HRI.

Halofuginone-dependent activation of GCN2 promotes adaptive mitochondrial elongation (Perea et al., 2023a). However, it is currently unclear if pharmacologic activation of other ISR kinases can similarly induce mitochondrial elongation. Here, we tested the ability of our HRI-activating compounds 0357 and 3610 to induce mitochondrial elongation downstream of the ISR. Previously, ISR-dependent mitochondrial elongation was quantified by manually classifying cells as containing fragmented, tubular, or elongated networks (Lebeau et al., 2018; Perea et al., 2023b). However, since 0357 and 3610 activate ISR signaling to lower levels than that observed for other compounds (e.g., halofuginone), we posited that these compounds may induce more modest mitochondrial elongation that may be difficult to quantify using this manual approach. To address this, we implemented an automated image analysis pipeline using Imaris software to quantify mitochondrial elongation in MEF cells stably expressing mitochondrial-targeted GFP (^mtGFP) treated with our compounds (Figure 2—figure supplement 1A); (Wang et al., 2012). We collected Z-stack confocal images of ^mtGFP-expressing MEF cells (MEF^mtGFP) and processed images using a deconvolution filter in FIJI to reduce the background and enhance the fluorescent signal. We used the ‘Surfaces’ module on Imaris to generate three-dimensional (3D) segmentation models of mitochondria visible in the deconvolved Z-stacks. Using the surfaces module, we quantified parameters defining mitochondrial shape, including bounding box length, sphericity, and ellipsoid principal axis length (Figure 2—figure supplement 1A). Treatment with conditions that induce mitochondrial elongation, such as the ER stressor thapsigargin (Tg) and the GCN2 activator halofuginone (HF), increased bounding box and ellipsoid principal axis length, while reducing sphericity (Figure 2A–D, Figure 2—figure supplement 1B–D) – all changes consistent with increases in mitochondrial length. In contrast, treatment with conditions that promote mitochondrial fragmentation, such as the mitochondrial uncouplers BtdCPU and CCCP (both compounds that activate HRI) (Fessler et al., 2020; Perea et al., 2023a), reduced bounding box length and ellipsoid principal axis length, while increasing sphericity (Figure 2A–D, Figure 2—figure supplement 1B–D) – all changes consistent with increased mitochondrial fragmentation.

We next applied this approach to define the impact of our pharmacologic HRI activators on mitochondrial morphology. Treatment with either 0357 or 3610 for 6 hr increased both the bounding box and ellipsoid principal axis length, while decreasing mitochondrial sphericity (Figure 2A–D). This result indicates that both these compounds induced mitochondrial elongation. Co-treatment with the selective ISR inhibitor ISRIB blocked these changes in mitochondria shape, indicating that these compounds induce mitochondrial elongation through an ISR-dependent mechanism (Figure 2E–H). ISRIB co-treatment also blocked mitochondrial elongation induced by the ER stressor thapsigargin (Tg) and the GCN2 activator halofuginone (HF), as predicted (Lebeau et al., 2018; Perea et al., 2023a). These results indicate that, like halofuginone, pharmacologic HRI activators also induce adaptive, ISR-dependent mitochondrial elongation. Further, these results suggest that the pharmacologic activation of different ISR kinases can induce protective elongation of mitochondria in the absence of cellular stress.

The ability for our pharmacologic GCN2 or HRI activators to induce adaptive mitochondrial elongation suggests that enhancing signaling through these kinases may suppress mitochondrial fragmentation induced by pathologic insults such as calcium dysregulation (Calvo-Rodriguez and Bacskai, 2020; Calvo-Rodriguez et al., 2020; Garbincius and Elrod, 2022; Matuz-Mares et al., 2022). Treatment with the calcium ionophore ionomycin induces rapid, DRP1-dependent mitochondrial fragmentation in cell culture models (Ji et al., 2015; Ji et al., 2017). We pretreated MEF^mtGFP cells for 6 hr with the GCN2 activator halofuginone or our two HRI-activating compounds (0357 and 3610) and subsequently challenged these cells with ionomycin. We then monitored mitochondrial morphology over a 15 min timecourse. As expected, ionomycin rapidly increased the accumulation of fragmented mitochondria in these cells, evidenced by reductions in both bounding box and ellipsoid principal axis length and increases of organelle sphericity (Figure 3—figure supplement 1A–D). Pretreatment with the ER stressor thapsigargin, which promotes stress-induced mitochondrial elongation downstream of the PERK ISR kinase, reduced the accumulation of fragmented mitochondria in ionomycin-treated cells (Figure 3A–D), as previously reported (Perea et al., 2023b). Intriguingly, treatment with halofuginone, 0357, or 3610 also reduced the accumulation of fragmented mitochondria in ionomycin-treated cells. Instead, mitochondria in cells pretreated with these ISR kinase activators and challenged with ionomycin demonstrated mitochondrial lengths and sphericity similar to that observed in vehicle-treated MEF^mtGFP cells (Figure 3A–D). These results indicate that pharmacologic activation of different ISR kinases can suppress the accumulation of fragmented mitochondria following ionomycin-induced calcium dysregulation.

Over 150 pathogenic variants in the pro-fusion GTPase MFN2 are causatively associated with the autosomal-dominant peripheral neuropathy CMT Type 2A (Cartoni and Martinou, 2009; Zaman and Shutt, 2022; Alberti et al., 2024). While these pathogenic variants can impact diverse aspects of mitochondrial biology (Zaman and Shutt, 2022), many, including D414V, lead to increases in mitochondrial fragmentation (Sharma et al., 2021b; Cartoni and Martinou, 2009; Zaman and Shutt, 2022; Alberti et al., 2024). This can be attributed to reduced activity of MFN2-dependent fusion associated with these variants and a subsequent relative increase of DRP1-dependent mitochondrial fission. We predicted that pharmacologic activation of ISR kinases could rescue mitochondrial network morphology in patient fibroblasts expressing the disease-associated MFN2 variant D414V (MFN2^D414V). To test this, we treated wild-type human fibroblasts and patient fibroblasts expressing MFN2^D414V with halofuginone or our two HRI-activating compounds 0357 and 3610 and monitored mitochondrial network morphology by staining with MitoTracker. As reported previously, MFN2^D414V-expressing fibroblasts showed shorter, more fragmented mitochondrial networks compared to control fibroblasts, reflected by reductions in both bounding box and ellipsoid principle axis lengths and increased sphericity (Figure 4A and E, Figure 4—figure supplement 1A–C; Sharma et al., 2021b). Treatment with halofuginone, 0357, or 3610 increased mitochondrial length in control fibroblasts (Figure 4A–D). These changes were inhibited by co-treatment with ISRIB, confirming these effects can be attributed to ISR activation. Intriguingly, all three compounds also increased mitochondrial length and reduced sphericity in MFN2^D414V-expressing patient fibroblasts to levels similar to those observed in control fibroblasts, with halofuginone showing the largest effect (Figure 4E–H). Again, this increase in mitochondrial elongation was reversed by co-treatment with ISRIB. These results show that pharmacologic activation of different ISR kinases can rescue basal mitochondrial morphology in patient fibroblasts expressing the disease-associated MFN2^D414V variant that causes dysregulation in various neurological functions, including ataxia, optic atrophy, and sensorineural hearing loss.

HEK293 cells (purchased from ATCC), HEK293 cells stably expressing XBP1-RLuc (Grandjean et al., 2020; Plate et al., 2016), HEK293 cells stably expressing HSE-FLuc, HEK293 cells stably expressing ATF4-FLuc (Yang et al., 2023a), HEK293 cells stably expressing ATF4-mAPPLE (a kind gift from Martin Kampmann’s lab) (Yang et al., 2023a) and CRISPRi-depleted of individual ISR kinases (HRI, PKR, PERK, GCN2; a kind gift from Martin Kampmann’s lab at UCSF) (Guo et al., 2020), and MEF^mtGFP (a kind gift from Peter Schultz) (Wang et al., 2012) were all cultured at 37°C and 5% CO₂ in DMEM (Corning-Cellgro) supplemented with 10% fetal bovine serum (FBS, Gibco), 2 mM l-glutamine (Gibco), 100 U/mL penicillin, and 100 mg/mL streptomycin (Gibco).

Primary fibroblast cells were isolated from partial thickness skin biopsy, as previously described (Sharma et al., 2021b; Martens et al., 2020), from a patient who provided written informed consent for research studies using human tissues (University of Calgary Conjoint Research Ethics Board REB17-0850). Cells were cultured in Medium Essential Media (11095080, Gibco), supplemented with 10% FBS (12483020, Gibco). Cells were maintained at 37°C and 5% CO₂. Clinical information regarding this participant was previously reported and included ataxia, optic atrophy, and sensorineural hearing loss (Sharma et al., 2021b). Exome sequencing in this participant identified a homozygous c.1241A>T variant in MFN2 (predicted to cause p.(Asp414Val)) and no other pathogenic variants.

The compounds used in this study were purchased from the following sources: thapsigargin (Tg; Cat# 50-464-294 Fisher Scientific), ISRIB (Cat# SML0843, Sigma), CCCP (Cat# C2759, Sigma), BtdCPU (Cat# 32-489-210MG, Fisher), halofuginone (Cat#50-576-3001, Sigma), and oligomycin A (S1478, Selleck). The nucleoside mimetic library was purchased from Chem Div. Hit compounds were repurchased from Chem Div.

HEK293 cells stably expressing the ATF4-FLuc, HSE-Fluc, or the ARE-Fluc reporter were seeded at a density of 15,000 cells per well in 384-well white plates with clear bottoms (Greiner). The following day, cells were treated with the indicated compound in triplicate at the indicated concentration for 8 hr. After treatment, an equal volume of Promega Bright-Glo substrate (Promega) was added to the wells and allowed to incubate at room temperature for 10 min. Luminescence was then measured using an Infinite F200 PRO plate reader (Tecan) with an integration time of 1000 ms. This assay was used to both screen the nucleoside mimetic library in triplicate and monitor the activity of hit compounds. HEK293 cells expressing the XBP1-RLuc reporter were tested using an analogous approach to that described above, monitoring RLuc activity using Renilla-Glo reagent (Promega), as previously described (Grandjean et al., 2020).

HEK293 cells stably expressing the ATF4-mApple reporter and CRISPRi-depleted of specific ISR kinases were seeded at a density of 300,000 cells per well in six-well TC-treated flat-bottom plates (Genesee Scientific). The cells were treated the next day for 16 hr with the compound at the indicated concentration. Cells were then washed twice with phosphate-buffered saline (PBS) and dissociated using TrypLE Express (Thermo Fisher). The cells were then resuspended in PBS and 5% FBS to neutralize the enzymatic reaction. Flow cytometry was performed on a Bio-Rad ZE5 Cell Analyzer monitoring mAPPLE fluorescence (568/592 nm) using the 561 nm green-yellow laser in combination with the 577/15 filter. Analysis was performed using FlowJo Software (BD Biosciences).

The relative mRNA expression of target genes was measured using quantitative RT-PCR. Cells were treated as indicated and then washed with PBS (Gibco). RNA was extracted using Quick-RNA MiniPrepKit (Zymo Research) according to the manufacturer’s protocol. RNA (500 ng) was then converted to cDNA using the High-Capacity Reverse Transcription Kit (Applied Biosystems). qPCR reactions were prepared using Power SYBR Green PCR Master Mix (Applied Biosystems), and primers (below) were obtained from Integrated DNA Technologies. Amplification reactions were run in an ABI 7900HT Fast Real Time PCR machine with an initial melting period of 95°C for 5 min and then 45 cycles of 10 s at 95°C, 30 s at 60°C.

MEF^mtGFP were seeded at a density of 15,000 cells/well in eight-chamber slides (Ibidi) coated with poly-d-lysine (Sigma) (Wang et al., 2012). The next day cells were treated with the indicated dose of compound for the indicated time. After treatment, cells were imaged on a Zeiss LSM 880 Confocal Laser Scanning Microscope equipped with a full incubation chamber for regulating temperature and CO₂ during live-cell imaging.

Patient fibroblasts were seeded at 70,000 cells per dish in 35 mm dishes with 20 mm glass bottoms (D35-20-1.5-N, Cellvis) for live-cell imaging. After 24 hr, the compound treatments were administered at the dosage and for the time points indicated in the figure legends. The mitochondrial network in patient fibroblasts was visualized using 100 nM MitoTracker Green (M7514, Life Technologies) (Kumar et al., 2001) for 45 min, following washing three times with culture media, according to the manufacturer’s instructions. The Z-stack images were acquired of patient fibroblasts using an Olympus Spinning Disc Confocal System (Olympus SD OSR) equipped with the Olympus UPlanApo 60XTIRF/1.50 Oil Objective using the CellSense Dimensions software. Acquired Z-stacks were analyzed using AI Machine Learning Segmentation (Imaris), as detailed below.

The Z-stack confocal images were processed in FIJI to reduce the background noise and enhance the fluorescent signal. The processed images are then introduced into the developed quantification pipeline in Imaris imaging software. In this approach, mitochondria are segmented in 3D using the ‘Surfaces’ module with a machine-learning algorithm that has been iteratively trained to detect foreground and background pixels in each Z-stack, filling blank holes within segments to generate 3D surfaces. The generated surfaces are filtered to include those above a minimum threshold of 250 voxels. While direct length measurements cannot be obtained through the Imaris surface module, indirect measurements of mitochondrial length are inferred from three separate calculations, including (1) object-oriented bounding box axis, (2) ellipsoid axis length, and (3) object sphericity (see Figure 2—figure supplement 1A). The object-oriented bounding box axis is calculated by measuring the length of the longest or principal bounding-box length of the smallest object-oriented rectangular box that encloses each 3D segmentation. The ellipsoid axis length is calculated by measuring the length of the longest or principal axis of each 3D segmentation. Sphericity is calculated by dividing the longest axis of each 3D segmentation by the length of the perpendicular axis.

Data are presented as mean ± SEM or as violin plots showing the mean and quartiles for the indicated number of measurements. Outliers were removed from datasets describing bounding box length and principal axis length, as appropriate, using the ROUT outlier test in PRISM 10 (GraphPad, San Diego, CA). Normality of datasets from our imaging studies was tested in PRISM 10 (GraphPad) using D’Agostino & Pearson, Anderson–Darling, Shapiro–Wilk, and Kolmogorov–Smirnov tests. Statistics were calculated in PRISM 10 (GraphPad) and analyzed by one‐way ANOVA with Tukey’s multiple correction test, Kruskal–Wallis or Mann–Whitney tests for data exhibiting a non-normal distribution, as indicated in the accompanying figure legends. Indications of nonsignificant interactions were generally omitted for clarity.

All materials detailed in this article can be provided by the authors upon reasonable request or purchased from the indicated supplier.