AGAP1-associated endolysosomal trafficking abnormalities link gene–environment interactions in neurodevelopmental disorders Open Access

Known genetic contributions to neurodevelopmental disorders are rapidly expanding and include both Mendelian and complex (non-Mendelian) phenomena. Mendelian inheritance patterns include classic autosomal dominant, recessive and sex chromosome-linked monogenic disorders, currently estimated to lead to ∼20% of autism spectrum disorders (Mahjani et al., 2021) and ∼25% of cerebral palsy cases (Moreno-De-Luca et al., 2021; van Eyk et al., 2021). Autosomal dominant inheritance patterns represent important contributors to neurodevelopmental disorders, often arising spontaneously as de novo variants (Brunet et al., 2021), and may exhibit complex, non-Mendelian characteristics such as incomplete penetrance and variable expressivity (Ahluwalia et al., 2009). Other complex inheritance elements include epigenetic contributions to disease, gene–gene (epistatic) interactions and gene–environment interactions. Despite the inherent challenges in studying complex inheritance, such patterns represent important contributors to neurodevelopmental disorders such as cerebral palsy (Fahey et al., 2017). Although substantial advances have been made in statistical and causal modeling of complex genetic factors (Madsen et al., 2011), few experimentally validated genes and mechanisms have been described for complex inheritance leading to neurodevelopmental disorders such as cerebral palsy.

Missense variants in the Arf1 regulator gene AGAP1 and heterozygous deletions in the 2q37.2 region that span this gene (Leroy et al., 2013) are associated with autism spectrum disorder (Cukier et al., 2014; Pacault et al., 2019) and cerebral palsy (Chopra et al., 2022; Jin et al., 2020; van Eyk et al., 2019). Despite a statistical enrichment in AGAP1 variants in a cerebral palsy cohort (van Eyk et al., 2019), there is also evidence for environmental factors contributing to disease in several cerebral palsy patients with AGAP1 variants. Therefore, the pathogenicity of these variants has not been definitively established and the disease–gene link for AGAP1 is still unknown. Additionally, although the frequency is very low, not all individuals harboring AGAP1 variants manifest neurodevelopmental disorders [Genome Aggregation Database (gnomAD), heterozygous loss of function (LoF)=0.008%; https://varsome.com/gene/hg38/agap1]. Such incomplete penetrance is particularly common in individuals with autism spectrum disorder (Geschwind, 2011), and many autism susceptibility genes have thus been referred to as ‘risk genes’ (Yuen et al., 2017). To better understand potential contributors to the apparent complex disease relationship of AGAP1, we sought to identify the underlying pathophysiology and characterize potential gene–environment interactions caused by AGAP1 loss of function.

AGAP1 is broadly expressed, with RNA levels highest in the brain and protein expression highest in the lung, endocrine tissues, male reproductive tissue, adipose tissue and bone marrow/lymphoid tissues (https://www.proteinatlas.org/ENSG00000157985-AGAP1/tissue). AGAP1 is expressed in the mouse and human brain throughout pre- and post-natal development, with subcellular localization to dendrites, axons and synapses (Arnold et al., 2016). This suggests that AGAP1 plays a role in brain development, although studies of its function are limited. AGAP1 is predicted to be intolerant to loss-of-function variants [observed/expected LoF=0.1169; LoF Z score=5.864; probability of loss-of-function intolerance (pLI)=1.0; https://varsome.com/gene/hg38/agap1] and potentially to missense variants as well (observed/expected missense ratio=0.8479; missense Z score=1.287; https://varsome.com/gene/hg38/agap1). AGAP1 contains a GTPase-like domain, Pleckstrin homology (PH) domain and GTPase-activating protein (GAP)-activity domain (http://www.ebi.ac.uk/interpro/protein/UniProt/Q9UPQ3/). The GAP activity of AGAP1 activates Arf1 hydrolysis to complete the GTPase cycle (Nie et al., 2003). The proper initiation and termination of Arf1 signaling is required for actin cytoskeleton polymerization (Davidson et al., 2015; Saila et al., 2020), which drives the formation of vesicles (Heuvingh et al., 2007). Overexpression of AGAP1 creates punctate endocytic structures and reduces stress fibers, demonstrating potential roles in regulating actin dynamics and trafficking from the endocytic compartment (Nie et al., 2003).

AGAP1 also regulates protein trafficking via the PH domain by recruiting subunits σ3 (AP3S3) and δ (AP3D1) of the AP-3 protein trafficking adaptor complex. AP-3 positively regulates the movement of proteins to the lysosome (Nie et al., 2003) and plasma membrane (Bendor et al., 2010). Immunostaining confirmed AGAP1 localization to early and recycling endosomes (Arnold et al., 2016) and to the Golgi (Cukierman et al., 1995). Both increases and decreases in AGAP1 levels decrease the rate of protein trafficking out of Rab11-positive endosomes (Arnold et al., 2016), and AGAP1-mediated endocytic recycling regulates the surface localization of M5 muscarinic receptor (CHRM5) and, consequently, neuronal function (Bendor et al., 2010). AP-3 is important for trafficking lysosome membrane proteins, such as LAMP1, from the Golgi to lysosomes (Chapuy et al., 2008). Therefore, AGAP1 is predicted to be a key regulator of protein trafficking from early endosomes to the Golgi, lysosomes or the plasma membrane.

To better understand how AGAP1 may contribute to nervous system function and expand on the potential AGAP1 disease link, we report clinical phenotypes of three individuals with microdeletion variants that had not been previously reported and compare phenotypes with prior reports. We have previously described impaired locomotion from a loss-of-function variant of CenG1a, the Drosophila ortholog to AGAP1 (Jin et al., 2020). CenG1a has well-conserved protein domains (Gündner et al., 2014) [Drosophila RNAi Screening Center Integrative Ortholog Prediction Tool (DIOPT), 13/16 tools support an orthologous gene-pair relationship; https://www.flyrnai.org/cgi-bin/DRSC_orthologs.pl] and Drosophila models have been widely used to dissect mechanisms of neurological disease (Ugur et al., 2016). Here, we investigated the role of CenG1a in neuronal morphology, endolysosomal distribution and composition, autophagy, and stress responses in this genetic model.

The dominant mutation assessment tool DOMINO (https://varsome.com/gene/hg38/agap1) predicts that heterozygous variants in AGAP1 will lead to haploinsufficiency [probability of haploinsufficiency or P(HI)=0.8548]. AGAP1 is also intolerant to genomic loss-of-function variants (pLI=0.9994). These observations provide complementary evidence for their role in AGAP1-associated neurodevelopmental disorders.

We identified three new cases in which patients were heterozygous for copy number variants that led to complete or partial AGAP1 gene deletion (exons 7-9 in one patient and exons 10-13 in another patient) (Table 1). We noted phenotype overlap with seven previously reported cases in which patients had AGAP1 variants or 2q37 deletions limited to the AGAP1 region (Table 2; Table S1). Key features include intellectual disability/developmental delay, language impairments, autism spectrum disorder, aggression, and impaired length, weight and cranial growth (Fig. 1). With lower frequency, we also observed epilepsy, dystonia and/or axial hypotonia, brain growth abnormalities, eye abnormalities, and skeletal defects. The shared clinical features of this expanded cohort provide additional support that AGAP1 variants can lead to a mixed neurodevelopmental disorder with systemic manifestations.

We noted movement disorders such as dystonia in patients with AGAP1 variants (Table 2) and previously showed locomotor impairments in CenG1a mutant flies (Jin et al., 2020). Patients with AGAP1 variants also exhibited microcephaly, impaired cortical growth and delayed myelination, and AGAP1 is known to be important for dendritic spine maturation (Arnold et al., 2016). We therefore screened CenG1a mutants for neuroanatomical phenotypes (Fig. 2A). We utilized a null allele (CenG1a^Δ9), given that AGAP1 putatively acts via loss of function. We found a reduction in the size of the axon terminal (Fig. 2B) without a corresponding decrease in the muscle area in CenG1a^Δ9 homozygotes. We confirmed that this was a recessive phenotype arising from loss of CenG1a as there was no change in heterozygous animals (Fig. S1) and the phenotype was also present in trans with a deficiency chromosome uncovering AGAP1 (CenG1a^Δ9/Df). CenG1aΔ9/Df hemizygotes also had an unexpected increase in muscle size (Fig. 2C). We did not detect a change in bouton number or density [bouton number divided by neuromuscular junction (NMJ) area], or the number of satellite boutons (Fig. 2D). Other than the decreased size, neurons appeared morphologically normal with no change in branch number (Fig. 2E). Pre- and post-synaptic marker colocalization was normal (not shown). A degenerative phenotype often has a post-synaptic signal with an absent pre-synaptic signal (Lincoln et al., 2015). In contrast, a synaptic maturation phenotype often has a pre-synaptic signal with an absent post-synaptic signal (Vasin et al., 2014). Given that CenG1a mutants only demonstrated reduced synapse size, our findings indicate that this likely represents an impaired neuronal growth or remodeling phenotype rather than representing degeneration or defective synaptogenesis.

AGAP1 is predicted to regulate protein trafficking out of the endosome (Arnold et al., 2016; Nie et al., 2003). We investigated whether AGAP1 loss of function disrupted trafficking within the endosomal compartment in the nervous system using the CenG1a knockout model. We examined the Rab7 marker of early-to-late endosomes (Shearer and Petersen, 2019) in the NMJ (Fig. 3A,B), using an area defined by anti-horseradish peroxidase (HRP) staining to distinguish between axon terminals and muscles. We noted increased Rab7 signal in CenG1a mutants (Fig. 3B′) compared to genetic controls (Fig. 3A′), particularly within the HRP-defined neuronal area. The area (Fig. 3C) and density of Rab7-positive puncta (Fig. 3C″) was increased in CenG1a mutants with no change in average size (Fig. 3C′) or intensity of signal (not shown). The increased endosome signal was localized to the motor neuron with no change in Rab7 area in the muscle, consistent with nervous system-specific expression of CenG1a. Thus, CenG1a appears to be important for regulating the number and size of endosomes in neurons.

We investigated whether increased endosome density was due to accumulated endosomes resulting from a failure of macroautophagy, reflecting disrupted endolysosomal degradation. We examined Arl8 (Fig. 3D,E), a marker of lysosomes (Rosa-Ferreira et al., 2018), and noted that its levels were also increased in CenG1a-mutant NMJs (Fig. 3E′) compared to its levels in genetic controls (Fig. 3D′). The lysosome-associated area (Fig. 3F) and density (Fig. 3F′) of Arl8 was increased in CenG1a mutants, although the average lysosome size was unchanged (Fig. 3F′). We subsequently found that there was increased colocalization between Rab7 and Arl8 puncta in the CenG1a mutants (Fig. 3G), reflecting a larger endolysosomal compartment. Taken together, these data demonstrate increases in average lysosome size and content, potentially from increased endosomal fusion (Jacomin et al., 2016), a defect in lysosomal degradation or disrupted endolysosomal reformation (Klionsky et al., 2008).

We next asked whether the increased endosomal–lysosomal colocalization was owing to impaired autophagic flux or increased autophagy in CenG1a^Δ9 mutants. We examined Atg8a, the Drosophila ortholog of LC3, which is lipidated in order to target membranes to the autophagosome. This modification is detected as a size shift in western blotting and calculated as lipidated/basal (LC3-II/LC3-I) levels. We found that CenG1a mutants had a higher ratio of LC3-II/LC3-I compared to the baseline in genetic controls (Fig. 4A,B). We induced starvation to assess the ability of CenG1a^Δ9 mutants to respond to an exogenous cytotoxic stressor. LC3-II/LC3-I ratios did not change in genetic controls after 24 h of starvation, indicating optimized balance between autophagy induction and protein degradation and clearance. In contrast, starvation decreased the basally elevated LC3-II/LC3-I ratios in CenG1a mutants.

We then tested whether autophagic flux was normal by treating larvae with chloroquine (Zirin et al., 2013), a weak base that blocks autophagy by altering the acidic environment of lysosomes and autophagosome–lysosome binding. We found that chloroquine feeding increased the ratios of LC3-II/LC3-I for both genetic controls and CenG1a^Δ9 mutants, as would be expected when blocking normal autophagic flux (Fig. 4D,E). This also suggests that lysosomal function is unaffected in CenG1a mutants.

We then assessed Ref(2)P, the Drosophila ortholog of p62 (SQSTM1), which recruits ubiquitinated cargos to autophagosomes; p62 accumulates when autophagy is impaired (Pircs et al., 2012). Consistent with elevated LC3-II/LC-I ratios, we found increases in Ref(2)P in CenG1a^Δ9 larvae compared to its levels in genetic controls (Fig. 4C,F). We did not detect a significant difference in Ref(2)P within genotypes between fed and starved conditions, nor with chloroquine treatment at 24 h. Taken together, our data demonstrate higher levels of autophagy induction in CenG1a mutants compared to the baseline in controls, but with preserved flux and increased clearance during starvation.

Disruptions of protein trafficking (Amodio et al., 2019) and clearance (Xiong et al., 2013) have been linked to activation of the integrated stress response. When the α subunit of the eukaryotic initiation factor-2 (eIF2α or eIF2A) is phosphorylated, global protein translation decreases, whereas the expression of stress response genes and autophagy components is selectively upregulated (B'chir et al., 2013; Humeau et al., 2020). We therefore investigated whether loss of CenG1a altered eIF2α phosphorylation (eIF2α-P) (Fig. 5A).

We found a significant increase in eIF2α-P in CenG1a^Δ9 larvae and young, 1-day-old adults (Fig. 5B). Control adults showed increased eIF2α-P between 1 and 14 days post eclosion, consistent with previous descriptions of normal aging (Lenox et al., 2015). In contrast, CenG1a^Δ9 animals showed elevated eIF2α-P early in development, but mutant animals did not show an age-dependent upregulation of eIF2α-P over time. These results indicate that eIF2α phosphorylation is chronically activated in mutant animals.

We next examined whether basal elevations in autophagy and eIF2α-P impaired responses to stress in a second-hit scenario. Amino acid starvation normally activates phosphorylation of eIF2α through the GCN2 kinase (Pakos-Zebrucka et al., 2016). When challenged with 24 h of starvation, control larvae exhibited increased eIF2α-P levels. In contrast, CenG1a^Δ9 larva with chronically elevated eIF2α-P levels did not show a further increase in its levels (Fig. 5C). This indicates a diminished capacity to respond to a second-hit cytotoxic stressor.

This suggested to us that CenG1a^Δ9 animals may have increased sensitivity to exogenous stressors. Accordingly, we exposed adult flies to tunicamycin, which blocks protein glycosylation. Tunicamycin impairs protein folding, induces the integrated stress response in the endoplasmic reticulum (ER), and triggers phosphorylation of eIF2α via PERK kinases (EIF2AK3) (Chow et al., 2013). Control animals exhibited increase eIF2α-P at 4 and 24 h of tunicamycin exposure. In contrast, eIF2α-P was equivocally upregulated in CenG1a-mutant adults compared to baseline levels in controls (P=0.06). In this context, CenG1a mutants exhibited a delay in their activation of stress response pathways, as there was a failure to increase eIF2α-P at 4 h, although phosphorylation was increased after 24 h (Fig. 5D). Thus, multiple stress response pathways are chronically active in CenG1a mutants and do not respond appropriately to acute stressors.

We confirmed that elevated eIF2α-P has functional consequences on the rates of protein synthesis with puromycin labeling. As expected, control larvae exhibited decreased global protein synthesis in the presence of tunicamycin-induced ER stress. In contrast, CenG1a^Δ9 mutants exhibited decreased levels of protein synthesis compared to baseline levels in controls and failed to further reduce the levels of protein synthesis in the presence of this stressor (Fig. 6A,B).

We next investigated whether impaired eIF2α-P creates functional impairments to stress responses in CenG1a mutants. Longevity studies revealed that CenG1a mutants died earlier than their genetic controls under normal rearing conditions (Fig. 6C). We sought to understand whether the inability of mutant animals to respond to a subsequent cytotoxic stressor would have important consequences at the level of the whole organism. We found that CenG1a mutants were more sensitive to ER stress-induced lethality, with tunicamycin-treated animals dying much earlier than genetic controls (Fig. 6D). Our findings illustrate a two-hit scenario wherein variants in CenG1a increases the susceptibility of animals to a subsequent insult.

We found evidence for an expanded endolysosomal compartment in CenG1a mutants. Accordingly, we also identified increased autophagy and eIF2α phosphorylation under basal conditions, suggesting chronic activation of the cytotoxic stress response. Autophagic flux was preserved in mutants, suggesting that lysosomal function was unaffected, but mutant animals had a decreased capacity to respond to ‘second-hit’ cytotoxic stressors. When challenged with such a stressor, mutants exhibited diminished survival.

Heterozygous deletions and predicted damaging missense variants in AGAP1 are associated with a mixed neurodevelopmental phenotype that includes autism spectrum disorder and cerebral palsy. AGAP1 is predicted to regulate AP3 trafficking proteins via the PH domain and patient phenotypes overlap with those caused by variants in the gene encoding an AP3-complex member, AP3D1. AP3D1 variants cause Hermansky–Pudlak syndrome 10; a patient with this syndrome shared several features with patients with AGAP1 variants, including global developmental delay, epilepsy, dystonia, hypotonia, poor feeding, diminished cortical volumes and poor myelination detected by neuroimaging, and microcephaly (Ammann et al., 2016). Consistent with the report on one patient with an AGAP1 variant (Jin et al., 2020), the patient with the AP3D1 variant had immunodeficiency and died of septic pneumonia at the age of 3.5 years. Disruptions to endosomal trafficking, such as to the retromer complex, are increasingly being linked to developmental disease, including hereditary spastic paraplegia and Ritscher–Schinzel Syndrome, with features spanning neurological, skeletal and immune systems (Saitoh, 2022; Yarwood et al., 2020).

We found decreased synaptic size in Drosophila with homozygous null variants in CenG1a, which was not apparent in a previous study decreasing CenG1a expression using RNAi knockdown or P-element gene disruption (Homma et al., 2014). We observed this phenotype in animals with a homozygous null allele or in trans with a deficiency, but not in heterozygotes. This suggests that the synaptic size phenotype is AGAP1 dependent, but with haploinsufficient inheritance of microcephaly in humans and recessive inheritance of NMJ size defects in the fly. This has been observed in related disorders (Gatto and Broadie, 2011; Tsai et al., 2012). It is therefore possible that the increased synaptic release found by Homma et al. (2014) may represent a mechanism to compensate for CenG1a partial loss of function to maintain synaptic function. We did not test whether AGAP1-mediated Arf1 activity or actin cytoskeleton regulation could underly the axon terminal size phenotype. Thus, the molecular mechanism by which AGAP1 leads to decreased axon terminal size needs further investigation. It is unlikely that decreased axon terminal size or increased synaptic release is due to increased autophagy; increased autophagy is associated with increased synaptogenesis in Drosophila and decreased synaptic release (Shen et al., 2015). Therefore, we conclude that the autophagy increase we observed compared to baseline levels is likely to be a compensatory mechanism.

In addition to an increased rate of basal autophagy, we observed an increase in basal eIF2α phosphorylation. Both indicate an activation of stress response pathways in the absence of external provocation. This argues that this basal activation is instead triggered in response to endogenous factors – the CenG1a genetic variant and the dysregulation of endolysosomal trafficking that ensues.

The activation of these elements of the integrated stress response likely serves as a compensatory mechanism; however, this mechanism is saturable. This sets up the potential sensitivity that leaves CenG1a mutants with a diminished capacity to respond to additional cytotoxic stressors, such as amino acid deprivation, tunicamycin or aging, and is observed here as increased mortality. Consistent with the idea of variants in AGAP1 increasing sensitivity to stress, we noted that one of the patients in this cohort had a history of prematurity, which may have also contributed to their disorder. We also noted that two patients inherited variants from a healthy parent. We further identified 4/7 cerebral palsy patients with AGAP1 variants from previously published studies who also had an adverse event in their medical history (Table S1). This further suggests that AGAP1 may contribute to disease in a gene–environment interaction by increasing sensitivity to diverse stressors in some cases. Our findings indicate that the gene–environment interaction is not limited to perturbations of protein trafficking, as we showed that starvation and ER stress both elicit phenotypes. The integrated stress response may thus represent a point of molecular convergence that connects several forms of genetic and environmental insults. Although this will require additional experimental validation, we suspect that this phenomenon may be broadly applicable to a variety of gene–environment interactions that contribute to human neurodevelopmental disorders and is not unique to AGAP1 variants.

Human subjects oversight for this study was provided by Phoenix Children's Hospital's Institutional Review Board #15-080. All participants (affected children and their parents) provided express written informed consent for participation in accordance with local rules and regulations as specified in the Declaration of Helsinki. Participants with variants in AGAP1 were identified through GeneMatcher (Sobreira et al., 2015) and de-identified phenotypic data were shared after local whole-exome sequencing and/or microarrays were performed. 2q37.2 microdeletion limited to AGAP1 was detected by microarray. Microdeletions were identified by array comparative genomic hybridization (Human Sureprint 2×105K, Agilent Technologies) and confirmed with semiquantitative PCR.

Drosophila melanogaster were reared on a standard cornmeal, yeast, sucrose food from the BIO5 media facility, University of Arizona. Stocks for experiments were reared at 25°C, 60-80% relative humidity with 12 h:12 h light/dark cycle. Cultures for controls and mutants were maintained with the same growth conditions, especially the density of animals within the vial. A mix of males and females were used for all experiments.

CenG1a^Δ9 is a null allele generated by ends-out gene targeting (Gündner et al., 2014) and was a kind gift from Michael Hoch (Department of Molecular Developmental Biology, University of Bonn, Life & Medical Sciences Institute). Df(2L)BSC252, w¹¹¹⁸ and Canton-S were acquired from Bloomington Drosophila Stock Center [National Institutes of Health (NIH), P0OD018537]. CenG1a^Δ9 was backcrossed with w¹¹¹⁸ for three generations and w¹¹¹⁸ was outcrossed with Canton-S and backcrossed with w¹¹¹⁸ for 12 generations while selecting for red eyes.

Imaging of NMJs of third instar wandering larvae was performed as previously described (Estes et al., 2011). Briefly, a larval fillet was dissected in HL-3 Ca²⁺-free saline (70 mM NaCl, 5 mM KCl, 22 mM MgCl₂, 10 mM NaHCO₃, 5 mM trehalose, 115 mM sucrose, 5 mM HEPES, pH 7.3) before fixation in Ca²⁺-free 4% paraformaldehyde. Washes were performed with PBS without or with 0.1% Triton X-100 (PBST). Blocking before and during primary and secondary antibody steps used PBST with 5% normal goat serum and 2% bovine serum albumin. The following primary antibodies [Developmental Studies Hybridoma Bank (DSHB) hybridoma monoclonal antibodies; details in Table S2] were incubated overnight at 4°C: mouse anti-Rab7 (1:10; deposited by S. Munro), rabbit anti-Arl8 (1:200; deposited by S. Munro) and mouse anti-DLG (1:400; 4F3; deposited by C. Goodman). Secondary antibodies were incubated for 1.5 h at room temperature and included: goat anti-mouse Cy3-conjugated IgG, goat anti-rabbit Cy3-conjugated IgG or donkey anti-rabbit Alexa Fluor 488-conjugated IgG (1:400; Thermo Fisher Scientific). Neuronal membranes were visualized with goat anti-HRP-Alexa Fluor 647 (1:100; Jackson ImmunoResearch) added with secondary antibody. Phalloidin-488 in PBS (1:300; Molecular Probes) was added as the final wash before mounting.

Imaging was performed with a 710 Zeiss confocal microscope using 1.0 µm z-stacks at 63× of the 1b 6/7 muscle in the A3 abdominal segment. Image parameters (laser gain and intensity, resolution, zoom) were kept constant between images of the same session; alternating between those for the control and mutants. Maximum-intensity projections were created for a given field of view using all images of the z-stack that included the axon terminal (defined by HRP staining) for all channels. Projections were converted to greyscale from RGB in Photoshop without adjustments to pixel intensity. Non-6/7 1b-HRP staining was identified based on the shape, size and HRP staining intensity and manually removed.

Images were analyzed using ImageJ software v1.50i (NIH). Images from individual channels were converted to black and white using the threshold feature with the threshold held uniform for images within an imaging session. Rab7 and Arl8 areas were measured using the ‘Analyze Particles’ tool with a mask created from the HRP image to define and measure the neuron-only area. NMJ area was measured by drawing around the area encompassed by both muscles and neuron in the field of view. Pearson correlation coefficient was determined using the coloc2 plugin from Fiji using the HRP channel to provide a mask for Rab7 and Arl8 channels. Orthogonal projections were converted to 8-bit black-and-white TIFFs without thresholding. Boutons were counted manually from 63× images stitched over the entire neuron area. Statistical analyses were performed and graphs created using R v4.1.0. For boxplots, boxes represent the 25th-75th percentiles, and whiskers represent the 10th and 90th percentiles.

Protein samples were prepared from ten to 20 males and females in protein extraction buffer plus 1% protease inhibitor (Thermo Fisher Scientific) and 1% phosphatase inhibitor (Sigma-Aldrich) as previously described (Emery, 2007). Western blotting was performed according to standard methods with detection on a 0.2 µm nitrocellulose membrane; antibodies are detailed in Table S2. For eIF2α, the phosphorylation-specific antibody was imaged first, then the membrane was washed and reprobed for total eIF2α. The following antibodies were used to detect autophagy: rabbit anti-Ref(2)P (1:500; Abcam, 178440) normalized using Ponceau staining (Santa Cruz Biotechnology, sc-301558), or mouse anti-β-actin (1:2000; Abcam, 8224) and rabbit anti-Atg8 (1:2000; Sigma-Aldrich, ABC974). The following antibodies were used to detect eIF2α-P: rabbit anti-phospho-S51 (1:1000; Cell Signaling Technology, 3597) and rabbit anti-eIF2S1 (1:500; Abcam 4837). Newly synthesized proteins were labeled with mouse anti-puromycin (1:1000; Kerafast, EQ0001) and their levels normalized using Ponceau staining. The following secondary antibodies were used: goat anti-rabbit or goat anti-mouse ECL IgGs (1:10,000; GE Healthcare, NA931 and NA934).

Chemoluminescence was captured using a FluorChem Imager (Biotechne) and quantified with Image studio Lite (LI-COR). Ref(2)P levels, LC3-II/LC3-I ratio and phosphorylated/total eIF2α ratio were normalized to those for wild-type or no treatment control for each biological replicate. Areas of the same size were used to measure total protein and puromycin signal for each lane to reduce variability. At least two technical replicates of each sample were used for quantification. Differences between genotypes were calculated using two-tailed paired t-test and graphs were generated using R v4.1.0. Detailed statistics are provided in Table S3.

For tunicamycin exposure, w¹¹¹⁸ and CenG1A^Δ9 male and female adults at 3 days post eclosion were anesthetized on ice and pooled into vials filled with 1.3% agar, 1% sucrose and 12 µM tunicamycin (Sigma-Aldrich), with a subset retained for immediate protein extraction. Flies were anesthetized on ice at 4 and 24 h for protein extraction.

Larvae were collected as second instar larvae and washed in water. For larval starvation, >20 male and female larvae were added to petri dishes made with 1.3% agar and 1% sucrose, with or without baker's yeast supplement, and returned to the 25°C incubator. Larvae were collected after 24 h; immobile or pupated animals were excluded. Chloroquine-treated food was prepared as described by Zirin et al. (2013) at a concentration of 3 mg/ml (Sigma-Aldrich). Larvae were collected as above and placed on food mixed with either chloroquine or water as a control for 24 h and then collected for protein extraction. Successful uptake of the drug was confirmed by examining larval gut for Bromophenol Blue that had been mixed into the food.

Puromycin labeling performed as described by Deliu et al. (2017). Briefly, third instar larva were floated out of food with 20% sucrose, washed in water, and inverted in batches of 15-20 mixed males and females. Larvae were incubated in Schneider's insect medium (Thermo Fisher Scientific) with either 12 μM tunicamycin or DMSO (Sigma-Aldrich) control for 3 h at 25°C. Then, 5 µg/ml of puromycin (Thermo Fisher Scientific) was added and incubated at 25°C for 40-60 min before protein extraction.

Drosophila were collected as late-stage pupae, separated by sex, and sequestered in standard food vials with ∼20 animals/vial. The numbers of live animals were monitored daily, with animals transferred to fresh food-containing vials once per week. For tunicamycin lethality, adults were sequestered in vials with 1.3% agar and 1% sucrose with 12 µM tunicamycin or DMSO as a control and monitored twice daily (Chow et al., 2013). Survival library in R v4.1.0 was used to generate Kaplan–Meier plots and log-rank test statistics.

We appreciate the participation by patients and their families for these studies. We acknowledge F. Nowlen and J. Liu for assistance with fly rearing and survival monitoring and A. Musmacker for western blotting. We thank Kurt Gusin, director of the Biomedical Imaging Core at the University of Arizona College of Medicine - Phoenix, for providing microscopy services and Jody Nyboer, director of the University of Arizona Bio5 media facility, for providing Drosophila food. Stocks obtained from the Bloomington Drosophila Stock Center (NIH, P40OD018537) were used in this study. The monoclonal antibodies, developed by S. Munro and C. Goodman were obtained from the Developmental Studies Hybridoma Bank, created by the National Institute of Child Health and Human Development of the NIH and maintained at the University of Iowa, Department of Biology, Iowa City, IA, USA. Biomedical Imaging Core microscopy facilities at the University of Arizona College of Medicine Phoenix were used in these studies. This study makes use of data generated by the DECIPHER community. A full list of centers that contributed to the generation of the data is available from https://deciphergenomics.org/about/stats. Funding for the DECIPHER project was provided by Wellcome (grant number WT223718/Z/21/Z).