Suppressive Genetic Interactions Between Haploinsufficient Mitochondrial Genes Encoded in the 22q11.2 Microdeletion Locus Define Brain and Cardiac Phenotypes

Abstract

Genomic copy number variations, such as the 22q11.2 microdeletion syndrome, cause pleiotropic disorders that affect diverse organ systems and disrupt neurodevelopment. Deletions of the 22q11.2 locus reduce the dosage of up to 46 protein coding genes, raising questions about the identity of haploinsufficient genes and their genetic interactions contributing to 22q11.2 phenotypes. Here, we dissect functional and molecular relationships between two genes encoded within the 22q11.2 locus: the mitochondrial ribosomal protein gene MRPL40 and the mitochondrial citrate transporter SLC25A1. We show that a MRPL40 null mutation disrupts mitochondrial translation, impairs respiration, and affects multiple components of the SLC25A1 interactome including factors required for lipid metabolism, mitochondrial ribosome subunits, and the mitochondrial RNA processing machinery. In silico coessentiality network analysis revealed correlated and anticorrelated fitness interactions linking MRPL40 and SLC25A1 to mitochondrial translation, intermediate carbon metabolism, and interferon signaling. We determined that Mrpl40-null mutations are embryonic lethal in mice, but Mrpl40^−/+ mice are viable and displayed embryonic cardiac development and adult behavioral phenotypes. Similarly, Slc25a1^+/− animals showed embryonic cardiac developmental defects but lacked the adult behavioral phenotypes observed in Mrpl40^−/+ mice. Surprisingly, transheterozygotic Slc25a1^+/−;Mrpl40^−/+ mice suppressed or mitigated cardiac development, behavioral, and brain transcriptome phenotypes observed in single heterozygotic animals. These results reveal that MRPL40 and SLC25A1 are haploinsufficient genes within the 22q11.2 locus that genetically and biochemically interact to define tissue development and physiology. Our findings provide a framework for understanding the complexity and type of gene dosage interactions within the 22q11.2 deletion syndrome locus.

Introduction

Human copy number variation (CNV) or microdeletion syndromes cause pleiotropic symptoms that affect multiple organ systems, including the brain, heart, immune and endocrine tissues, and craniofacial structures^1–4. This is exemplified by the 22q11.2 microdeletion syndrome, which can manifest as a constellation of multi-organ abnormalities or present primarily as a neurodevelopmental disorder, such as schizophrenia^5–7. The largest 22q11.2 microdeletion spans 3 Mb and reduces the dosage of 46 protein-coding genes, 27 pseudogenes, and 17 small non-coding regulatory RNAs^{8, 9}. Central questions arising from this genomic lesion include: which genes within the 22q11.2 CNV are dosage-sensitive contributors to disease phenotypes, and whether genes within the CNV genetically interact to specify these phenotypes? Current models consider the possibility of single dosage-sensitive “driver” genes within a CNV causing specific phenotypes; multiple dosage-sensitive genes contributing individually or through their genetic interactions; modifying loci residing outside the CNV influencing phenotypes; and imprinting mechanisms modulating gene effects^10–12. Genetic evidence in humans strongly supports the model that interactions among genes within and outside a CNV determine phenotypes¹⁰. Indeed, Drosophila studies modeling genetic interactions among genes contained within either the 3q29 or 16p11.2 CNVs show that intra-CNV interactions can either enhance or suppress phenotypes^{13, 14}. To our knowledge, intra-CNV genetic interactions have not been systematically tested using engineered mutants in mice.

Modifier genomic loci outside the 22q11.2 CNV are known to influence phenotypes associated with this syndrome. These include cardiovascular defects that could be modulated by genes outside the 22q11.2 locus, such as either SLC2A3 or epigenetic regulator genes^{9, 15}. Likewise, psychiatric phenotypes may be modified by genes outside the 22q11.2 locus, whose function is required for mitochondrial homeostasis, including PARK2 and SPG7¹⁶. Notably, the 22q11.2 microdeletion is unique among CNVs because it removes the largest known group of nuclear-encoded genes required for mitochondrial function. These eight genes include SLC25A1, TXNRD2, RTL10, MRPL40, PRODH, COMT, SNAP29, and TANGO2, all localized to mitochondria according to Mitocarta 3.0 or new evidence^17–20. In addition, genes not classically annotated as mitochondrial may nevertheless influence mitochondrial biology, including ZDHHC8, UFD1L, and DGCR8^17–19.

The abundance of mitochondrial genes within the 22q11.2 locus and the prominent mitochondrial brain proteome phenotypes in 22q11.2 mouse models^{21, 22} prompted us to test genetic interactions among 22q11.2 encoded mitochondrial genes. We asked if these mitochondrial genes could behave as haploinsufficient alleles and genetically interact to specify phenotypes in tissues with uniquely high mitochondrial bioenergetic demands, the brain and the heart^{23, 24}. We focused on two 22q11.2 mitochondrial genes Slc25a1 and Mrpl40. Slc25a1 encodes a citrate transporter located in the inner mitochondrial membrane that transports citrate from the mitochondrial matrix to the cytoplasm^{25, 26}. Mrpl40 specifies a polypeptide belonging to the large 39S subunit of the mitochondrial ribosome that binds a structural Val-tRNA^{27, 28}. Slc25a1 and Mrpl40 proteins participate in a shared protein–protein interaction network²⁹. We have shown that like Slc25a1, Mrpl40 is required for electron transport chain function and mitochondrial ribosome activity^{29, 30}. Because Slc25a1 and Mrpl40 converge on a mitochondrial protein interaction network, we predicted that their combined haploinsufficiency would exacerbate cardiac and brain phenotypes observed in the single-gene deficiencies. In contrast, and unexpectedly, we found that combined Slc25a1 and Mrpl40 haploinsufficiency suppressed cardiac and behavioral phenotypes observed in single haploinsufficiencies. These findings highlight the complexity of genetic interactions shaping CNV phenotypes and demonstrate the centrality of mitochondrial metabolism to heart development and behavior. We propose that, in addition to SLC25A1 and MRPL40, other dosage-sensitive nuclear-encoded mitochondrial genes contribute to the emergence of cardiovascular and psychiatric disease.

Results

We confirmed our previously reported association between SLC25A1 and its interactome with MRPL40²⁹, but this time using MRPL40 as a bait. We immunoprecipitated MRPL40 from detergent-soluble extracts of MRPL40-FLAG-expressing human neuroblastoma cells. MRPL40-FLAG coprecipitated endogenous SLC25A1 and proteins of the SLC25A1 interactome (Fig. 1A and Supplementary Fig. 1A). The specificity of these associations was tested by out-competition of immunoprecipitated antigens with an excess of FLAG peptide (Supplementary Fig. 1A, compare lanes 2–3). We then generated null alleles of MRPL40/Mprl40 in human neuroblastoma cells and in the mouse germline to assess the robustness of biochemical and genetic interactions between SLC25A1 and MRPL40²⁹. We resorted to engineer MRPL40 null cell lines because mouse mitochondrial ribosome subunit mutants of are lethal before gastrulation³¹. SLC25A1-null mutations modify protein levels of mitochondrial ribosome subunits, RNA processing machinery, and lipid metabolism pathways belonging to the SLC25A1 interactome²⁹. Thus, we reasoned that a MRPL40 knock out should alter the levels of these SLC25A1 interactome proteins (Fig. 1A). We generated six MRPL40 null neuroblastoma clonal lines by CRISPR gene editing. We targeted exon one, which encodes the first 17 residues of MRPL40 (Supplementary Fig. 1A). We used these cell lines and their controls to quantify protein levels of SLC25A1 and its interactome (Fig. 1A). Loss of MRPL40 was confirmed by MRPL40 immunoblotting (Fig. 1B). MRPL40-null cells also showed severely reduced levels of mitochondrial ribosome translated MT-CO1, 2, 3, and MT-ATP6 (Fig. 1C, G and Supplementary Fig. 1B), and reduced abundance of mitochondrial ribosome subunits belonging to the SLC25A1 interactome (Fig. 1D,G). The decreased levels of these proteins resulted in a marked reduction of mitochondrial respiration, even after challenging cells with carbonyl cyanide 4-(trifluoromethoxy)phenylhydrazone or FCCP, as determined by Seahorse oximetry (Fig. 1H). MRPL40-null cells had an increased rate of media acidification, suggestive of increased glycolysis under severely compromised respiration (Fig. 1H). SLC25A1 protein levels did not change across MRPL40-null clones (Fig. 1B, G). However, proteins belonging to the SLC25A1 interactome involved in mitochondrial lipid metabolism (CPT1A) were significantly increased to 300%, across multiple MRPL40-null clones (Fig. 1E, G). Similarly, we observed modifications of SLC25A1 interactome proteins belonging to the mitochondrial RNA machinery in MRPL40-null cells (Fig. 1F–G). We found decreased levels of FASTKD2, TFAM, and PNPT1 to 60, 46, and 35%, respectively. These RNA processing machinery perturbations resulted in a 2–3-fold increase in the levels of the mitochondrial rRNAs RNR1 and RNR2 (Fig. 1I), 9 of the 13 mRNAs encoded in the mitochondrial DNA, and unprocessed RNA junctions, including those encoded in the light DNA strand (Fig. 1J). These increased RNA species resulted in an increased mitochondrial content of dsRNAs, an innate immunity inductor, detected with J2 antibody (Fig. 1K). These mitochondrial transcriptome alterations correlate with a doubling of the number of mitochondrial genomes in MRPL40-null cells (Fig. 1L). We conclude that MRPL40-null mutations alter the levels of components of the SLC25A1 interactome concomitant with broad alterations in the mitochondrial transcriptome. These findings reveal a wide spectrum of gain- and loss-of-function biochemical phenotypes in MRPL40-null mutations that extend beyond the function of the mitochondrial ribosome.

Fig. 1. MRPL40 Null Mutants Disrupt Mitochondrial Ribosomes, Respiratory Chain, and Components of the SLC25A1 Interactome.

A. Proteins of the SLC25A1 interactome according to Gokhale et al²⁹. Edges were defined by the proximity ligation mitochondrial interactome according to Antonicka et al.⁸². B. MRPL40 and SLC25A1 immunoblots of wild type and MRPL40 null cell lines. C. Immunoblot of respiratory chain proteins encoded by the mitochondrial genome. D. Immunoblot of mitochondrial ribosome subunits. All subunits, except for MRPL44, are depicted in the SLC25A1 interactome. E. Immunoblot of SLC25A1 interactome proteins involved in acetyl-CoA and lipid metabolism. F. Immunoblot of RNA binding proteins present in the SLC25A1 interactome. HSP90 and ACTB were used as controls for B-F. G. Quantification of blots in panels B-F. Depicted is the ratio between mutant and wild type. Each dot represents an independent clone and/or independent experiment. p values were obtained with unpaired mean difference two-sided permutation t-test (italicized numbers represent p values). H. Metabolic activity in wild type and MRPL40 KO cells measured by the Seahorse Mito Stress Test (n=8 of each genotype). Oxygen consumption and extracellular acidification rates, OCR and ECAR, data are presented normalized to protein and analyzed by unpaired mean difference two-sided permutation t-test (italicized numbers represent p values). Basal, ATP-dependent, and maximal respiration were determined as described. Arrows indicate the sequential addition of oligomycin, FCCP and rotenone-antimycin. I. qRT-PCR quantification of mitochondrial rRNAs RNR1–2 and vimentin as control. Unpaired mean difference two-sided permutation t-test, n=3. J. Nanostring quantification of nuclear encoded and mitochondrial encoded RNAs. Junctions represent non-processed intermediaries derived from the polycistronic mitochondrial RNA. p values two-sided t test. Wild type n=6 and KO n=3. K. Immunofluorescent microscopy of with type and MRPL40 KO cells labeled for TOM20 and dsRNA. Scale bars correspond to 10 and 2.5 μm. Probability plot depicts the levels of dsRNA signal in 16 wild type and 20 MRPL40 KO cells. Kolmogorov-Smirnov test. L. Quantification of the ratio of mitochondrial and nuclear genomes. Unpaired mean difference two-sided permutation t-test. Wild type n=16 and KO n=20 cells.

The up- and down-regulation of components of the SLC25A1 interactome observed in MRPL40 KO cells suggests synthetic genetic interactions between the MRPL40 and SLC25A1 genes. Such interactions could either enhance and/or ameliorate phenotypes caused by deficiencies in either MRPL40 and/or SLC25A1 genes. We tested this prediction in silico by performing a coessentiality network analysis assessing genome-scale fitness correlations of SLC25A1 and MRPL40 across the publicly available Dependency Map (DepMap) screen datasets (Fig. 2)^{32, 33}. Correlated and anticorrelated genes reveal pathways and regulatory mechanisms connecting genes of interest. We performed this analysis using the FIREWORKS tool (Fitness Interaction Rank-Extrapolated netWORKs)³⁴. We identified two modules centered around MRPL40 and SLC25A1 that were connected by correlated and anticorrelated interactions with SOD2 and the zinc finger proteins ZNF577 and ZNF416 (Fig. 2A). These in silico-generated modules identified components of the SLC25A1 interactome that we determined experimentally²⁹, which were also altered in MRPL40 KO cells (red font, Fig. 2B). Within the MRPL40 module, the strongest positive fitness correlation was observed between MRPL40 and subunits of the mitochondrial ribosome and mitochondrial aminoacyl tRNA synthases (Fig. 2A, C). Similarly, the strongest positive fitness correlation within the SLC25A1 module included genes annotated to pyruvate and acetyl-CoA synthesis, which reside immediately upstream of citrate synthesis in the Krebs cycle. Importantly, the SLC25A1 and MRPL40 modules show anticorrelated genes involved in cytoplasmic protein synthesis (GO:0042273, RPL5;RPF2), nuclear encoded RNA processing (GO:0006396, LSM5;USP39), glycolipid biosynthesis (GO:0009247, PGAP2;ST8SIA1), and interferon signaling (GO:0060333, IRF1) (Fig. 2A, C). These in silico findings support the idea that these pathways could mediate SLC25A1 and MRPL40 genetic interactions in cells beyond the mitochondrial ribosome and citrate metabolism. Thus genetic interactions of similar complexity may occur in whole organisms where MRPL40 and SLC25A1 are simultaneously affected by a genetic defect.

Fig. 2. Coessentiality Network Analysis of SLC25A1 and MRPL40.

A. Coessentiality network analysis showing genome-scale fitness correlations of SLC25A1 and MRPL40 as inputs. Nodes correspond to first and second level interaction nodes. Network of correlated (red edges) and anticorrelated genes (blue edges) was built with the FIREWORKS tool. r= Pearson correlation. B. Subnetwork of correlated and anticorrelated genes linked to components of the SLC25A1 interactome (red font nodes). C. ENRICHR Gene ontologies inferred from first order nodes connected to either SLC25A1 or MRPL40. Red and blue fonts depict correlated and anticorrelated genes, respectively.

The magnitude or quality of mouse phenotypes observed in single mutants of either the Slc25a1 or Mrpl40 genes could be modified in double mutant mice, a prediction founded on their protein interactions, mitochondrial functional convergence, and in silico predicted genetic interactions (Figs. 1 and 2). To generate a Mrpl40-null allele in mice, we first quantitatively estimated the essentiality of the mammalian MRPL40/Mrpl40 gene by comparing the DepMap gene dependency score of human MRPL40 to TFAM and POLG, two proteins associated with the mitochondrial nucleoid^{32, 33, 35} (Fig. 3A). We chose TFAM and POLG as their null alleles are lethal at or before mouse embryonic day 10.5^{36, 37}. We used the pigmentation gene TYR/Tyr, as a reference since this gene is neither essential nor lethal in mammals³⁸. MRPL40 scored in between TFAM and POLG in essentiality, supporting lethality of homozygous Mrpl40 null mutations before birth in mice (Fig. 3A). This prediction agrees with the pre-gastrulation developmental stalling observed in embryos carrying null mutations of other mitochondrial ribosome subunit genes³¹. We generated a mouse Mrpl40 mutant allele by CRISPR gene editing. We engineered the Mrpl40 mutation in the heterozygotic Slc25a1^−/+ mouse strain we previously described to avoid genetic background variables when testing Slc25a1 and Mrpl40 transheterozygotic genetic interactions³⁰ (Fig. 3B). Mrpl40 editing resulted in a 7bp deletion in exon 3, which is shared by the two annotated Mrpl40 transcripts ENSMUST00000023391.16 (UNIPROT Q3UKS6) and ENSMUST00000119273.2 (UNIPROT D3Z7C0) (Fig. 3C). The founder Mrpl40 mutant mouse carries a premature stop codon after residue 53, thus truncating the remaining 153 residues, or 75%, of the Mrpl40 primary sequence encoded by transcript ENSMUST00000023391.16 (Fig. 3C). The mutant stop codon resides more than 50–55 nucleotides upstream of the final exon-exon junction making it unlikely to undergo nonsense-mediated mRNA decay³⁹ (see Fig. 6C and Supplementary Fig 3C). The deleted primary sequence contains a domain that binds a structural Val-tRNA present in the large subunits of the mitochondrial ribosome²⁷. Mrpl40^−/+ animals were backcrossed at least 6 times to remove any off-target CRISPR edits (Fig. 3B). We genotyped animals at embryonic date 14.5, recovering 50% of the animals either as wild type or Mrpl40^−/+ (Fig. 3D). We did not obtain Mrpl40^−/− embryos, indicative of lethality before embryonic date 14.5, as predicted by the gene dependency score (Fig. 3D). In contrast, we obtained all four genotypes at the expected Mendelian ratios in intercrosses of Slc25a1^−/+ with Mrpl40^−/+ animals (Fig. 3E). We did not observe any gross anatomical alterations in transheterozygotic Slc25a1^−/+;Mrpl40^−/+ embryos and the size of newborn Slc25a1^−/+;Mrpl40^−/+ animals was normal as compared with the wild type and their heterozygotic littermates.

Fig. 3. Generation of an Mrpl40 mutant mouse.

A. Estimation of mouse embryonic viability using DepMap gene dependency scores of human MRPL40, TFAM, POLG, and TYR. B. Experimental design to generate an Mrpl40 null allele. C. Mrpl40 genomic region, CRISPR mutagenesis strategy, founder mutation, and predicted protein truncation. D. Frequence of Mrpl40 genotypes at embryonic age 14.5 from timed Mrpl40^−/+ intercrosses. E. Diagram of Slc25a1^−/+;Mrpl40^−/+ timed intercrosses, frequency of genotypes, and crown-rump lengths of embryos at embryonic age 14.5. Chi square test of expected and observed genotypes.

Fig. 6. Cortical Transcriptome of Single and Double Slc25a1^−/+;Mrpl40^−/+ Heterozygotic Mice.

A. Uniform Manifold Approximation and Projection (UMAP) of cortical transcriptome before and after one-way ANOVA thresholding by p<0.01 of the indicated genotype transcriptomes. B. PCA of transcriptomes after one-way ANOVA thresholding by p<0.01. Lines represent Euclidean distance clustering. C. Normalized mRNA counts for Mrpl40 and Slc25a1 transcripts. Kruskal-Wallis test followed by Benjamini and Hochberg multiple corrections. D. Volcano plots and E. heat maps of paired comparisons between wild type animals with single or transheterozygotic animals. Insets in volcano plots show percentage of diverse types of RNAs differentially expressed. Heat Maps depict z-scored hierarchical clustering of transcripts after thresholding with a cut-off fold of change=1.5 and p<0.01 (Welsh t test). F and G depict violin plots of mRNAs either rescued (F) or worsened (G) in Slc25a1^−/+;Mrpl40^−/+ cortex. Kruskal-Wallis test followed by Benjamini and Hochberg multiple corrections. All panels wild type n=9 and all other genotypes n=7.

We took advantage of the fact that our Slc25a1^−/+ mouse model shows haploinsufficient cardiac development and metabolic phenotypes to test whether Slc25a1 and Mrpl40 genetically interact to modulate tissue phenotypes³⁰. As previously described, Slc25a1^−/+ mice show ventricular septal defects and ventricular noncompaction revealed by decreased compact myocardium and increased trabecular myocardium³⁰(Fig. 4). Some of these phenotypes are common with the 22q11.2 microdeletion, which causes conotruncal and ventricular septal defects⁶. We measured these Slc25a1^−/+ cardiac phenotypes in Mrpl40^−/+ and transheterozygotic Slc25a1^−/+;Mrpl40^−/+ hearts. We chose to study hearts at embryonic day 14.5 when septation and ventricular wall compaction should be complete^{40, 41}. We recapitulated Slc25a1^−/+ septal defects in Mrpl40^−/+ at a similar frequency as compared to Slc25a1^−/+ animals (20–31.3%, X²=0.47, Fig. 4A–B, arrows). However, the frequency of Slc25a1^−/+ septal defects was significantly reduced to 5.6% in Slc25a1^−/+;Mrpl40^−/+ mice (X²=0.049, Fig. 4A–B). This suppressive transheterozygotic interaction was also observed in the ventricular noncompaction phenotype (Fig. 4A and C, asterisks). We found a significant decrease in compact myocardium with a concomitant increase in the trabecular myocardium in Slc25a1^−/+ hearts but not in the Mrpl40^−/+ ventricular wall (Fig. 4A and C, asterisks). Importantly, the non-compaction phenotype observed in Slc25a1^−/+ hearts was ameliorated in transheterozygotic Slc25a1^−/+;Mrpl40^−/+ mice (Fig. 4A and C, asterisks). These results show partially overlapping cardiac development phenotypes in single heterozygotic mice, which can be reverted closer to wild type in transheterozygotic Slc25a1^−/+ and Mrpl40^−/+ tissue.

Fig. 4. Slc25a1^−/+;Mrpl40^−/+ Heterozygotic Mice Suppress Cardiac Defects Observed in Single Heterozygotes.

A. Representative images of E14.5 embryos hearts from the indicated genotypes stained with hematoxylin and eosin. Asterisk indicates noncompaction and arrow indicates ventricular septal defect. Scale bar is 0.5 mm. B Frequency of cardiac defects observed in E14.5 in embryos with the indicated genotypes. Chi square test of expected and observed genotypes. C Quantification of relative compact myocardium thickness. one-way ANOVA Fisher’s LSD.

Slc25a1^−/+ and Mrpl40^−/+ transheterozygotic genetic interactions observed in hearts could expand to other organs affected in 22q11.2 microdeletion syndrome, such as the brain⁶. Thus, we performed a battery of behavioral tests and collected microdissected adult brain cortex and hippocampus for bulk RNAseq from the same animals. We conducted Morris water maze (Fig. 5A), elevated plus maze, marble burying, nestlet shredding (Fig. 5B), light-dark cycle activity (Fig. 5C), and pre-pulse inhibition (Fig. 5D) to capture a wide span of behaviors in male and female mice (Supplementary Fig. 2). The Morris water maze, elevated plus maze, marble burying, and nestlet shredding were normal in all the genotypes tested irrespective of sex. Instead, we found that Mrpl40^−/+ males, but not females, have altered light-dark cycle activity with a bout of hyperactivity at dawn (mixed-effect model, genotype p=0.038, Fig. 5C). This phenotype was absent in Slc25a1^−/+ mutants as well as transheterozygotic Slc25a1^−/+;Mrpl40^−/+ males supporting a suppressive genetic interaction between Slc25a1^−/+ and Mrpl40^−/+. We further tested this finding by analyzing data with two-sided permutation statistics, which confirmed the phenotype in Mrpl40^−/+ mice whereas transheterozygotes were not different from wild type animals.

Fig. 5. Slc25a1^−/+;Mrpl40^−/+ Heterozygote Male Mice Suppress Mrpl40^−/+ Behavioral Phenotypes.

A. Swim distance, swim speed, and latency to the platform during acquisition trials of the Morris water maze in 8-week-old males of the indicated genotypes. Two Way Repeated Measures ANOVA n=5 per genotype. B. Anxiety-like behaviors tests (elevated plus maze, marble burying, and nestlet shredding) are normal in all genotypes tested. One Way ANOVA followed by Holm-Šídák’s multiple comparisons test. n=7–11 per genotype. C. Time course of 23-h locomotor activity of animals of the specified genotype. Grey column denotes dark period. Data represent mean ± SEM. Statistical comparisons were performed using a mixed-effects model (Restricted Maximum Likelihood) with factors genotype, time, and their interactions. Mixed effect model was followed by either Fisher’s LSD or two-sided permutation t-tests at the indicated time points. Violin plot shows two-sided permutation t-tests of one of those points. n=7–9 per genotype. D. Percent PPI, data represent mean ± SEM. Statistical comparisons were performed using a mixed-effects model (Restricted Maximum Likelihood) with factors genotype, prepulse intensity, and their interactions. Mixed effect model was followed by two-sided permutation t-tests. Violin plot shows two-sided permutation t-tests at 82 dB. n=8–11 per genotype. E. Percent PPI as in D. Two Way Repeated Measures ANOVA n=8. Factors are genotype, prepulse intensity, and their interactions. Violin plot shows two-sided permutation t-test at 82 dB. n=8.

A well-established phenotype in mouse models of 22q11.2 microdeletion syndrome is an impaired pre-pulse inhibition response, which has construct validity for abnormal sensorimotor gating responses in 22q11.2 microdeletion human subjects^{5, 6, 42}. We compared the effects of single and double heterozygosity in Slc25a1 and Mrpl40 (Fig. 5D) and compared responses in these animals to those in the Del(16Es2el-Ufd1l)217Bld (heretofore referred as Df1/+, Fig. 5E) mouse, an animal model of the 22q11.2 microdeletion syndrome that includes a copy number variation in the Slc25a1 gene^{43, 44}. We used these Df1/+ mice as a positive control and a way to assess phenotypic similarities between Df1/+ with either single or transheterozygotic animals. We found decreased pre-pulse inhibition in Mrpl40^−/+ mice as determined by the modification of the pulse response by the Mrpl40^−/+ genotype (mixed-effect model, pulse-genotype p=0.011, Fig. 5D). The effect of the Mrpl40^−/+ genotype was near significance (mixed-effect model, genotype p=0.051). Thus, we further scrutinized the pre-pulse inhibition phenotype in Mrpl40^−/+ animals using two-sided permutation statistics across pulse levels (Fig. 5D). This analysis showed strong and significant effects in Mrpl40^−/+ mice at 82 dB, which was reverted to wild type levels in transheterozygotic Slc25a1^−/+;Mrpl40^−/+ males. The extent of pre-pulse inhibition in Mrpl40^−/+ mice was comparable to the phenotype in Df1/+ mice (mixed-effect model, genotype p=0.036, Fig. 5E). Similar to the Mrpl40^−/+ mutants, the most robust Df1/+ phenotype was observed at 82 dB after two-sided permutation statistics.

Slc25a1^−/+ and Mrpl40^−/+ transheterozygotic genetic interactions at the behavioral level should correlate with modifications of gene expression. We measured gene expression by bulk RNAseq in microdissected adult brain cortex and hippocampus of wild type, single, and transheterozygotic mice to test this hypothesis prediction (Fig. 6 and Supplementary Fig. 3). Non-parametric Uniform Manifold Approximation and Projection (UMAP) of the whole bulk transcriptome revealed no separation across all genotypes suggesting discrete gene expression changes across genotypes. However, thresholding bulk transcriptomes by p<0.01 (one-way ANOVA) showed genotype clustering where the cortical Mrpl40^−/+ transcriptome and the hippocampal Slc25a1^−/+ transcriptome segregated from the other genotypes (Fig. 6A). Parametric data dimensionality reduction and Euclidean distance clustering (PCA) showed wild type, Slc25a1^−/+; and Slc25a1^−/+;Mrpl40^−/+ cortical transcriptomes co-clustered away from the Mrpl40^−/+ transcriptome in cortex (Fig. 6B). Similarly, wild type and Slc25a1^−/+;Mrpl40^−/+ hippocampal transcriptomes segregated from the Slc25a1^−/+ transcriptome (Supplementary Fig. 3B). These changes in the Mrpl40^−/+ transcriptome were evident even though we could not detect changes in the expression of Mrpl40 transcripts in both brain regions, likely the result of the discrete Mrpl40 7bp deletion sparing the mutant transcript from non-sense mediated decay (Fig. 6C). In contrast, Slc25a1^−/+ and Slc25a1^−/+;Mrpl40^−/+ transheterozygotic cortex and hippocampus showed a mean reduction in the Slc25a1 mRNA to ~60% of wild type (Fig. 6C). We selected differentially expressed mRNA by a fold of change >1.5 and a p value <0.01 (Welsh t-test) to identify less than 100 RNAs affected in either single or transheterozygotes as compared to wild type tissue (Fig. 6D–E). Of these RNAs, less than 40% were protein coding RNAs while the rest were non-coding transcripts (Fig. 6D). The paucity of differentially expressed coding RNAs prevented us from identifying pathways and ontologies enriched in these transcripts with either GSEA or ENRICHR tools. Importantly, we found that differentially expressed mRNAs in either Slc25a1^−/+ or Mrpl40^−/+ were attenuated or reverted to wild type levels in transheterozygotic cortex (Fig. 6E), as exemplified by the transcripts encoding Klra2, Lrrc69, Mhrt, and Oas2 in cortex (Fig. 6F) and Asb17, Ifit1bl2, and Ppp1r1c in hippocampus (Supplementary Fig. 3F). In contrast, other mRNA phenotypes became overt in transheterozygotic cortex, as illustrated by Car13, Ccr2, and Hsd17b13 mRNAs (Fig. 6G). These results show a complex transcriptional response where we observed two responses; first, we find differentially expressed cortical transcripts in single heterozygotic mutants that can be reverted to wild type levels in transheterozygotic Slc25a1^−/+;Mrpl40^−/+ cortex. A second response is an additive phenotype in transheterozygotic cortical tissue. We demonstrate that behavioral phenotypes and some differentially expressed transcripts observed in either Mrpl40^−/+ or Slc25a1^−/+ mutants can be suppressed in transheterozygotic Slc25a1^−/+;Mrpl40^−/+ mice. We conclude that Slc25a1^−/+ and Mrpl40^−/+ are haploinsufficiencient mitochondrial genes that genetically interact with a complex pattern of phenotypic outputs, some of which are suppressive. Our findings suggest that these and other genes within copy number variation loci interact to modulate the quality and magnitude of tissue phenotypes.

Discussion

A frequent strategy to assess the contribution of genes to copy number variation syndrome phenotypes is the identification of driver genes whose haploinsufficiency is sufficient to elicit phenotypes. One example of this is Tbx1 in the 22q11.2 microdeletion. Tbx1 haploinsufficiency causes conotruncal cardiovascular, cranial, and brain development, as well as behavioral phenotypes observed in 22q11.2 microdeletion syndrome models^{43, 45–47}. We previously expanded the repertoire of 22q11.2 genes required for cardiac development to include Slc25a1³⁰. Our findings add Mrpl40 to the list of haploinsufficiency genes affecting brain and cardiac phenotypes, supporting the concept that multiple loci within 22q11.2 determine tissue phenotypes. The focus on single driver genes, while an important approach, neglects the possibility that multiple genes within a microdeleted chromosomal locus could interact to specify tissue-specific phenotypes. Human genomic studies and direct hypothesis testing in Drosophila show that rather than driver genes specifying phenotypes, traits are determined by gene interactions that enhance or ameliorate a phenotype. This idea has been elegantly demonstrated in Drosophila where tissue phenotypes are modulated by interactions between pairs of genes encoded within a human CNV locus^{10, 11, 13, 14, 48}. Here we provide evidence of synthetic interactions between two of the eight mitochondrial genes located in the 22q11.2 microdeletion locus. We demonstrate that Mrpl40 and Slc25a1 genetically interact to specify cardiac development, behavior, and brain transcriptomes. Even though the two genes form part of a biochemical mitochondrial interactome and are both required for normal mitochondrial function, the combined haploinsufficiency of these genes is suppressive at the level of tissue phenotypes. These results reveal that the robustness of the principles governing genetic interactions among CNV genes discovered in Drosophila also apply to mammals.

The suppressive outcome in heart and brain of Slc25a1^−/+;Mrpl40^−/+ mice is surprising. Our biochemical studies show that proteins belonging to the SLC25A1 interactome are affected by knocking out MRPL40, thus revealing possible mitochondrial convergence points between SLC25A1 and MRPL40. Such convergence points include the function of the respiratory chain, the mitochondrial ribosome, the maturation of the mitochondrial transcriptome, possible activation of innate immune responses by excess mitochondrial nucleic acids, alterations in acetyl-CoA pools, and cellular processes that require this metabolite, such as epigenome modifications. These convergence points could result in synthetic interactions enhancing or ameliorating a phenotype. There is precedent for suppressive interactions even within components of the respiratory chain. For example, mutations in complex I subunits can be suppressed by intra complex I mutants⁴⁹. Presently, we do not know which one of these MRPL40- and SLC25A1-dependent mitochondrial mechanisms could account for suppressive effects. However, we speculate about three possibilities. Suppressive interactions may be related to pools of unassembled mitochondrial ribosome proteins, either within mitochondria or in the cytoplasm. We have shown that SLC25A1 KO cells have decreased integrity of the mitochondrial ribosome and reduction of the levels of mitochondrial ribosome proteins²⁹. If phenotypes observed in tissues are a consequence of unassembled mitochondrial ribosome proteins acting as dominant negative molecules, then reducing their levels would prevent such effect. Decreasing MRPL40 gene dosage could reduce the levels of unassembled mitochondrial ribosome proteins in an SLC25A1-deficient background, thus preventing a dominant effect. A second putative mechanism is heterogenous cell responses to Mrpl40 haploinsufficiencies. The levels of mitochondrial ribosome mRNAs are cell type specific, with interneurons expressing the highest levels of these transcripts^{29, 50}. Cell types with higher levels of MRPL40 expression could potentially have a greater suppressive effect due to greater decrease in mitochondrial ribosome gene expression. The regional specificity of the gene expression changes in the brain comparing cortex and hippocampus between Slc25a1^−/+ and Mrpl40^−/+ supports this idea. Finally, a third mechanism is impaired handling of mitochondrial or cytoplasmic calcium. A prior Mrpl40^−/+ mouse model shows increased mitochondrial calcium transients and short-term potentiation, the latter a phenotype that could be modulated by the expression of the mitochondrial ADP-ATP translocator Slc25a4. Slc25a1 and Slc25a4 biochemically interact and their knockouts have decreased calcium uptake into mitochondria, thus disruption of Slc25a1 could restore mitochondrial calcium homeostasis in Mrpl40^−/+ neurons^{22, 51}. We focused on genetic interactions between two of six mitochondrial genes encoded in the 22q11.2 microdeleted locus. However, defects in the 22q11.2 mitochondrial genes Txnrd2, Prodha, Snap29 and Comt cause defective neurodevelopment and/or behavioral alterations^52–58. Therefore, we postulate further synthetic interactions exist between additional 22q11.2 mitochondrial genes, which may be difficult to predict from studies in single gene heterozygosity models. Most neurodevelopmental syndromes are now understood to have a polygenic basis, which provides rationale for expanded interrogation of CNVs and multigenic models^{11, 12, 59}.

We found defective prepulse inhibition in our Mrpl40^−/+ animals, a phenotype that aligns with prior results in Danio rerio mrpl40^−/− displaying altered acoustic startle responses, and with mutations in the mitochondrial genome gene MT-TL1, which encodes a tRNA required for protein synthesis in mitochondria^{53, 60}. Heteroplasmic mice carrying this mitochondrial genome mutation also have reduced prepulse inhibition⁶⁰. Our results stand in contrast to a previous study in another Mrpl40^−/+ mouse model, which reported normal prepulse inhibition⁵¹. These mice share similar genetic backgrounds in these two Mrpl40^−/+ mice (MGI:3847513 and MGI:4431658), so we suspect the discrepancy in prepulse inhibition across these two models is the age at which tests were performed. Age is a variable with strong effects in prepulse inhibition, particularly in the C57BL/6N strain used in our studies^61–63. We tested our animals at 9–11 weeks of age, while Devaraju et al. tested them between 16–20 weeks. As further confirmation, we demonstrated that the magnitude of prepulse inhibition reduction in Mrpl40^−/+ mice is similar to that in Df1/+ mice, a 22q11.2 microdeletion animal model, which we tested with the same behavioral experimental paradigm and at the same age. In summary, our Mrpl40 mouse model adds to the list of genes within the 22q11.2 locus that impair prepulse inhibition. Our findings also suggest a more complex genetic regulation of prepulse inhibition responses among the genes affected in the 22q11.2 locus. Multiple CNVs of the mouse 22q11.2 synthenic region show defective prepulse inhibition, including Df1/+, Df(16)A+/−, LgDel/+, Df(h22q11)/+, Del(1.5 Mb)/+, and Del(3.0 Mb)/+^{42, 64–69}. The genetic defect in four of these mouse models spans Slc25a1 and Mrpl40, yet they still show robust reduction of prepulse inhibition (Df(16)A+/−, LgDel/+, Del(1.5 Mb)/+, Del(3.0 Mb)/+). If Slc25a1^−/+ can suppress a Mrpl40^−/+ prepulse phenotype, as we have shown here, then other genes within the 22q11.2 locus and their interactions must be causal of a prepulse inhibition defects. Among these genes, haploinsufficiency of either Tbx1^{44, 70, 71}, or the mitochondrial gene Prodh^{56, 58} cause this phenotype.

A limitation of our studies is the correlative nature of the transcriptome and behavioral phenotypes, as well as the fact that our transcriptomes do not include brain regions engaged in prepulse inhibition responses, including the amygdala and midbrain⁷². Slc25a1 and Mrpl40 are broadly expressed genes; thus, a reasonable assumption is that the genetic interaction we identify occurs in a cell-autonomous and anatomically localized manner. However, it is possible that the suppressive effect between these haploinsufficient genes could be distributed across the brain regions participating in prepulse inhibition. Slc25a1 and Mrpl40 cell-autonomous suppressive interactions are more likely in heart than brain as a result of the heart’s more homogenous cellular and anatomical landscape. Our emphasis has been in the most salient phenotypes caused by Slc25a1 and Mrpl40 haploinsufficiency and their suppressive interactions. However, even though brain transcriptomes show discrete genes whose expression is restored to wild type levels in combined heterozygotic tissue, the paucity of altered mRNAs prevents the identification of potential cell autonomous molecular mechanisms affected by the single and combined heterozygosity.

Together, our findings argue that the genetic architecture of the 22q11.2 microdeletion emerges from combinatorial interactions among multiple loci, including mitochondrial genes within the CNV. By uncovering a surprising suppressive interaction between Slc25a1 and Mrpl40, we show that mitochondrial gene dosage can buffer or reshape phenotypes in vivo, influencing cardiac development, brain transcriptomes, and sensorimotor gating. The presence of prepulse inhibition deficits in Mrpl40^−/+ mice, their alignment with mitochondrial ribosome and mitochondrial-genome mutants, and their persistence across multiple 22q11.2 CNV models underscore both the centrality of mitochondrial pathways to neural circuit function and the likelihood that additional 22q11.2 genes participate in genetically interacting modules that shape this behavioral outcome. Although the precise mitochondrial mechanisms driving suppression in brain and heart remain unresolved, our results highlight the complexity of mitochondrial genetic networks operating during development. Ultimately, this work advances a network-based model of CNV pathology in which interacting mitochondrial and non-mitochondrial genes collectively sculpt tissue phenotypes and modulate their penetrance.

Materials and Methods

Cell Culture and MRPL40 KO Cells Generation

Crispr edited SH-SY5Y cells were generated at Synthego with guide sequence UCACCCGCUAGUCGGGCGCA. Standard SH-SY5Y growth media was further supplemented with uridine (50 μg/mL), sodium pyruvate (110 μg/mL), and D-glucose (2 mg/mL). MRPL40 knock-out (KO) and control SH-SY5Y cell pools were cloned in-house. To verify isolated clones, DNA was extracted using QuickExtract DNA extraction solution (Lucigen QE0905T) according to manufactures protocol. The MRPL40 targeted genetic region was amplified with forward (GCAGCTGACACCCTAGGC) and reverse (TTCTCTCCCACTTCACAGGAAAAT) primers using AmplitaqGold 360 Master Mix (Life Technologies 4398876). PCR products were Sanger sequenced and Inference of CRISPR Edits (ICE) analysis (Synthego.com) was used to confirm genetic mutation. Western blot was used to confirm loss of MRPL40 protein.

All cells were maintained in DMEM supplemented with 10% FBS, and non-essential amino acids (NEAA) and uridine, D-glucose, and sodium pyruvate as described above. Control and MRPL40 KO clones were cultured in the supplemented media, in a humidified incubator at 37°C and 5% CO₂.

To generate stable cell lines, SH-SY5Y cells (ATCC, CRL-2266; RRID:CVCL_0019) were transfected with ORF expression clone containing C terminally tagged Myc-DDK MRPL40 (Origene, RC202166) or N terminally tagged FLAG-SLC25A1 (GeneCopoeia, EX-A1932-Lv1020GS) as described²⁹. The stable cell lines were maintained in DMEM media containing 10% FBS, 100 μg/ml penicillin and streptomycin, and neomycin 0.2 mg/ml (Hyclone, SV30068) or puromycin 2 μg/ml respectively (Invitrogen, A1113803) at 37°C in 10% CO₂

Antibodies and Immunoblots

For immunoblot, cells were washed twice with phosphate buffered saline (PBS) supplemented with 1 mM MgCl₂ and 100 μM CaCl₂, then lysed with Buffer A (10 mM HEPES pH 7.4, 150 mM NaCl, 1 mM EGTA, and 0.1 mM MgCl₂, with 0.5% Triton X-100 and Complete anti-protease) Lysates were incubated at 4°C for 30 minutes with periodic vortexing and clarified by centrifugation at 20,000 RCF for 15 minutes. The supernatant was collected for downstream use and the pellet was discarded. Clarified cell lysis was diluted to 1 μg/μL and Laemmli buffer added for SDS-PAGE. Equivalent amounts of sample were loaded onto 4–20% Criterion TGX Midi-gels for SDS-PAGE and transferred to PVDF at 1.5 mA/cm² for 65 minutes using a semi-dry method. Membranes were blocked for 30 minutes at room temperature with tris-buffered saline with 0.05% Triton-X 100 (TBST) containing 5% non-fat milk, washed well, and incubated overnight at 4°C with primary antibody diluted in antibody base (3% BSA and 0.2% sodium azide in PBS). The following day, membranes were washed in TBST and incubated with secondary HRP conjugated antibody or rhodamine conjugated anti-tubulin diluted in TBST with 5% non-fat dry milk, for 30 minutes at room temperature. Washed membranes were imaged either with ChemiDoc MP (BioRad) fluorescent detection or with Western Lighting Plus ECL substrate incubation (PerkinElmer, NEL105001EA) for chemiluminescent detection on GE Healthcare Hyperfilm. Quantification of western blot band intensity was determined with Fiji (ImageJ) or Image Lab (BioRad). The intensity of the KO band was divided by the average intensity of the WT bands on the membrane for that particular antibody.

Antibody	Supplier	Catalog #	Dilution
ACLY	Proteintech Group	67166-1-Ig	1:1000
ACTB	Sigma	A5451	1:1000
ATAD3A	Abnova	H00055210-D01	1:1000
CPT1A	Abcam	ab122558	1:1000
DHX30	Abcam	ab85687	1:2000
FASTKD2	Bethyl	A303-788	1:500
FLAG	Bethyl	A190-102A	1:1000
FLAG	Sigma	F3165	1:200
GRSF1	Abcam	ab205531	1:1000
J2 SCICONS	Exalpha	10010200	1:300
HSP90	BD Biosciences	610418	1:1000
LRPPRC	Proteintech Group	21175-1-AP	1:5000
MRPL11	Proteintech Group	15543-1-AP	1:500
MRPL12	Proteintech Group	14795-1-AP	1:500
MRPL14	Sigma	SAB4502786	1:500
MRPL40	Sigma	HPA006181	1:500
MRPL44	Proteintech Group	16394-1-AP	1:500
MRPS18B	Proteintech Group	16139-1-AP	1:1000
MRPS22	Proteintech Group	10984-1-AP	1:500
MT-ATP	Proteintech Group	55313-1-AP	1:500
MT-CO1	Abcam	ab203912	1:1000
MT-CO2	Abcam	ab110258	1:1000
MT-CO3	Proteintech Group	55082-1-AP	1:1500
PNTP1	Proteintech Group	14487-1-AP	1:2000
SLC25A1	Proteintech Group	15235-1-AP	1:500
TFAM	Abcam	ab131607	1:1000
TOM20	BD Biosciences,	612278	1:300
TUBB-Rhodamine	BioRad	12004166	1:5000
mouse-HRP	Thermo Fisher	A10668	1:5000
rabbit-HRP	Thermo Fisher	G21234	1:5000
rabbit-ICDYE 800CW	LiCor	92632211	1:5000

Immunoprecipitation Assays

Immunoprecipitation experiments were performed as described in Gokhale et. Al 2021. Briefly, SH-SY5Y cells expressing either FLAG-SLC25A1 or Myc-DDK MRPL40 were grown in 10 cm tissue culture dishes, placed on ice, and rinsed twice with cold PBS (Corning, 21–040-CV) containing 0.1 mm CaCl₂ and 1.0 mm MgCl₂. Cells were lysed in a buffer containing 150 mm NaCl, 10 mm HEPES, 1 mm EGTA, and 0.1 mm MgCl₂, pH 7.4 (Buffer A) with 0.5% Triton X-100 and Complete anti-protease (Roche, 11245200). The lysates were placed on ice for 30 min and centrifuged at 16,100 × g for 10 min. The clarified supernatant was retrieved, and protein concentration was determined using the Bradford Assay (Bio-Rad, 5000006). For immunoprecipitation assays, 500 μg of the soluble protein lysate was applied to 30 μl Dynal magnetic beads (Invitrogen, 110.31) coated with 1 μg of the mouse monoclonal FLAG antibody. This mixture was then incubated on an end-to-end rotator for 2 h at 4°C. As controls, the immunoprecipitation was outcompeted with the 3XFLAG peptide (340 μm; Sigma, F4799). After 2 hours, the magnetic beads were washed 6 times with Buffer A with 0.1% Triton X-100. Proteins were eluted from the beads with Laemmli buffer. Samples were then analyzed by immunoblot.

Seahorse

SH-SY5Y control and MRPL40 KO clones were plated at 32,000 cells per well in an Agilent Seahorse culture plate (Agilent, Seahorse FluxPak Kit 103792–100) with DMEM supplemented with 10% FBS, uridine, additional D-glucose, sodium pyruvate and NEAA and incubated overnight at 37°C, with 5% CO₂. The flux plate cartridge was hydrated with calibrant solution overnight at 37°C, without injected CO₂. The following day, culture media was removed, washed twice with Seahorse assay media (103574–100), adjusted to a final volume of 180 μL per well, then incubated for 1 hour before the start of the mitochondrial stress test, at 37°C without injected CO₂. Mitochondrial stress test drugs were loaded into injection ports at a 10x concentration to yield final concentrations of 2 μM oligomycin, 0.25 μM FCCP, 1 μM rotenone, and 1 μM antimycin A. The standard Agilent mitochondrial stress test protocol was used, including three baseline measurements prior to injection and three measurements following each injection of oligomycin, FCCP, and rotenone/antimycin A. Agilent WAVE software was used to measure and analyze oxygen consumption rates (OCR) and extracellular acidification rates (ECAR). Data were normalized by total protein content per as determined by BCA assay. Briefly, cells were washed with PBS supplemented with 1 mM MgCl₂ and 100 μM CaCl₂, then lysed in-well with 0.5% Triton X-100/Buffer A (as described above). BCA reagent was added to the lysis in each well and incubated at 37°C for 30 minutes before reading on the BioTech plate reader at 562 nm against BSA standards.

qRT-PCR

Total RNA was extracted from roughly 6.5 million control and MRPL40 KO SH-SY5Y cells with Trizol reagent, according to manufacturer’s protocol. First strand cDNA synthesis was done with the SuperScript III Kit (Thermo Fisher 12574030) with random hexamers and 5 μg of total RNA following manufactures protocol. RT-PCR of the resulting cDNA was performed in triplicate on the QuantStudio 6 Flex (Applied Biosystems) with LightCycler 489 SYBR Green I Master Mix (Roche 04707516001) and primers for MTRNR1, MTRNR2 and Vimentin as²⁹. Data analysis was completed using QuantStudio RT-PCR software version 1.2, with relative quantification based on standard curves generated from control SH-SY5Y cells.

Mitochondrial Genome Quantifications

Total DNA was extracted from approximately 2.5 million control and MRPL40 KO SH-SY5Y cells using the DNeasy Blood and Tissue DNA Extraction Kit (Qiagen 69504) according to manufactures guidelines. DNA concentration and purity were validated on the Nanodrop OneC. Following the protocol of Rooney et al. for determination of mitochondrial DNA copy number, 6 ng of total DNA was analyzed on the QuantStudio 6 Flex (Applied Biosystems) with LightCycler 489 SYBR Green I Master Mix (Roche 04707516001) and primers for mitochondrial specific (MT-tRNA-Leu) or nuclear specific (Nuc-Beta2-microglobulin) DNA. Ct values determined by QuantStudio RT-PCR software were used to calculate the relative mitochondrial DNA as shown by⁷³.

1. (ΔCt) = (Nuclear DNA C_t) – (Mitochondrial DNA C_t)

2. Relative mitochondrial DNA = 2 × 2^ΔCT

Nanostring Determinations.

Cells were solubilized in TRIzol (Invitrogen). RNA isolation and NanoString analysis was performed by the Emory Integrated Genomics Core. RNA quality was assessed by bioanalyzer. mRNA counts were normalized to the housekeeping gene CLCT for the MitoString using nSolver and then processed with Qlucore Omics Explorer Version 3.6 Software.

Confocal Microscopy

Cells were fixed in 4% PFA at 37°C for 10 minutes and permeabilized with 0.25% Triton X 100 for 5 minutes with 20U/mL SUPERase-In RNAse inhibitor. Cells were blocked with 10% fetal bovine serum in PBS and incubated with primary antibodies: J2 and TOM20 (BD Biosciences, 612278). Slides were imaged with a Nikon A14 HD25 confocal microscope and 60x Apo lens (NA=1.4, WD=140um). dsRNA localized to the mitochondria (TOM20) was quantified with ImageJ/Fiji software in the focal plane of maximal intensity. dsRNA was measured as pixel intensity per cell, restricted to TOM20 segmented by thresholding.

Animals

All animal procedures were approved by Emory University’s Institutional Animal Care and Use Committee. SLC25A1 KO mice (C57BL/6N-Slc25a1tm1a(EUCOMM)Wtsi/Mmucd) were purchased from the Mutant Mouse Resource and Research Centers (MMRRC 042258-UCD) and genotyped according to the MMRRC PCR protocol. The WT allele was detected by primers GAATTGGTCGTGGTCTCAGTAGCC and GGAGTGCCCAAGAGACTCTGAGC producing a 366 base pair product. The KO allele recognized primers GAGATGGCGCAACGCAATTAATG and TAGTGAGTTATGCTTTGAAGACTTCGC producing a 253 base pair product.

DF1 mice (B6.129S7-Del(16Es2el-Ufd1l)217Bld/Cnrm) were from Adriano Buzzati-Traverso at the Institute of Genetics and Biophysics, National Research Council through the European Mouse Mutant Archive (EMMA) managed by Intrafrontier (Intrafrontier EM02122). Genotyping was performed with primers TGGGCAATTGTTTAATCTTCC and TCTTTGTCAGCAGTTCCCTTT, where the mutant gene generated a 290 base pair product and the wild type gene provided a 150 base pair product.

We used the CRISPR/Cas9 system to generate the Mrpl40 knockout mouse line. One 20 base pair Mrpl40 guide RNA (gRNA) sequence, ACCTTCTTCTTCTTCCGCA, was designed at the syntenic loci in the mouse genome. The Emory Mouse Transgenic and Gene Targeting core injected 20 ng/μL of gRNA and 20 ng/μL of Cas9 protein into single-cell C57BL/6J zygotes. Embryos were cultured overnight and transferred to pseudopregnant females. The resulting pups were screened for Mrpl40 knockout via PCR. To confirm the desired mutation, Sanger sequencing was performed on purified PCR DNA from the potential mutant mice. Heterozygous mice progress to maturity and produce viable pups. Genotyped was done with primers CACTTGTTCCTACCACAGACATG and GACAGTGGACTAAGCTCGTGGAG. The resulting PCR product from the WT allele was digestible by AciI, producing bands of 237 and 136 base pairs. However, the MRPL40 mutant allele product was undigestible and produced a PCR product 373 base pairs in length.

Heart Histology and Analyses

Timed matings were conducted using Slc25a1^+/− male and female mice aged 2–3-months, with the detection of a copulation plug designated as embryonic day (E) 0.5. Embryos were harvested at E14.5, fixed in 10% formalin and embedded in paraffin. Paraffin-embedded embryos were sectioned at 6 μm, followed by deparaffinization, rehydration, and staining with hematoxylin and eosin (H&E). Slide were imaged using a Nanozoomer 2.0-HT whole-slide scanner (Hamamatsu). Myocardial trabecular and compact layer thicknesses was quantified on H&E-stained sections using QuPath software⁷⁴.

Behavioral Analyses

Morris Water Maze.

Morris Water Maze training took place in a round, water-filled tub (52 inch diameter) in an environment rich with extra maze cues and a small platform 1 cm below the surface (see below). White tempera paint (non-toxic) was added to the water to make it opaque so that the mice could not see the platform. Mice were placed in the water maze with their paws touching the wall from 4 different starting positions (N,S,E,W) in water that started at 25°C and typically declined to 22°C by the time a whole group of mice was tested. An invisible escape platform was in the same spatial location 1 cm below the water surface independent of a subjects start position on a particular trial. In this manner subjects were able to utilize extra maze cues to determine the platform’s location. Each subject was given 4 trials/day for 5 days with a 15-min inter-trial interval. The maximum trial length was 60 s and if subjects did not reach the platform in the allotted time, they were manually guided to it. Upon reaching the invisible escape platform, subjects were left on it for an additional 5 s to allow for survey of the spatial cues in the environment to guide future navigation to the platform. After each trial, subjects were dried and kept in a dry plastic holding cage filled with paper towels to allow them time to dry off. The holding cage was placed half-on, half-off a heating pad and mice were monitored closely. Following the 5 days of task acquisition, a probe trial was presented during which time the platform was removed and the amount of time and distance swam in the quadrant which previously contained the escape platform during task acquisition was measured over 60 s. All trials were videotaped and performance analyzed by means of MazeScan (Clever Sys, Inc.).

Elevated Plus Maze.

The elevated plus maze, which is constructed of Plexiglas, consisted of two open arms and two enclosed arms arranged in a plus orientation. The arms were elevated 30 inches above the floor, with each arm projecting 12 inches from the center. To begin each test, mice were placed in the center of the maze facing one of the open arms and allowed to freely explore the apparatus for five minutes, during which time their behavior were videotaped. Mice were returned to their home cage at the end of the 5 min test.

Prepulse Inhibition (PPI).

Prepulse inhibition was assessed in a sound attenuated chamber (San Diego Instruments). In the PPI test mice following a 5 minute habituation period, subjects were presented with 66 total trials in a pseudorandom order. Those trials consisted of startle stimulus alone trials (120 dB, 20 ms), prepulse alone trials (20 ms white noise at 70, 74, 78, 82, or 86 dB), and prepulse and startle stimulus combined trials (each of the 5 prepulse intensities followed by a 20 ms 120-dB startle stimulus). Each of the trial types was presented in a pseudorandom fashion such that each trial was presented 6 times and no two consecutive trials were identical. Mouse movement was measured by a piezoelectric accelerometer during 100 ms after startle stimulus onset for 100 ms. PPI (%) was calculated. At the end of the session, subjects were removed from the sound attenuated chamber and placed in their home cage.

Circadian Activity.

Mice were placed in plexiglass activity cages (with appropriate clean bedding) equipped with infrared photobeams (San Diego Instruments) for 23 hours. Ambulations (consecutive beam breaks) were counted by a computer. Food and water will be available ad libitum.

Marble Burying.

A standard mouse cage lined with corncob bedding was filled with 20 glass marbles (15 mm diameter), arranged in a 4 × 5 matrix and equidistant from one another. Mice were then placed in the clean cage with the marbles for 30 min. The number of marbles buried (>50% marble covered by bedding) was recorded as the primary dependent variable

Nestlet Shredding.

The mice were moved from their home cages to individual testing cages. Each testing cage contained 3.0 g of Nestlets (commercially available pressed cotton squares), which mice typically use to make nests. The cages did contain any other environmental enrichment items. Mice were left in the testing cages for 120 mins. At the end of the 120 mins, unused nest material was collected and weighed.

RNAseq

RNA Isolation.

Admera Health performed all RNA isolation, library preparation, and sequencing. Qiazol phase separation, followed by cleanup with RNeasy 96 was used to isolate RNA from Hippocampal and Cortical samples. Both RNA Tapestation assay and High Sensitivity RNA Tapestation assay (Agilent Technologies Inc., California, USA) and quantified by Infinite F Nano+ 200 Pro Tecan (Tecan, Switzerland) assessed the quality of the isolated RNA.

Coding RNA transcripts were isolated using NEBNext^® Poly(A) mRNA Magnetic Isolation beads as part of the NEBNext^® Ultra^™ II Directional RNA Library Prep Kit for Illumina^® (New England BioLabs Inc., Massachusetts, USA). cDNA synthesis and library preparation was performed using the same kit. Prior to first strand synthesis, samples are randomly primed (5´ d(N6) 3´[N=A,C,G,T]) and fragmented. The Protoscript II Reverse Transcriptase with a longer extension period, approximately 30 minutes at 42°C synthesized the first strand of cDNA. Final library quantification was performed by Qubit 2.0 (ThermoFisher, Massachusetts, USA) and quality assessment was made by TapeStation HSD1000 ScreenTape (Agilent Technologies Inc., California, USA). Final library size was about 450bp with an insert size of about 300bp. Illumina^® 8-nt dual-indices were used. Equimolar pooling of libraries was performed based on QC values and sequenced on an Illumina NovaSeq X Plus (Illumina, California, USA) with a read length configuration of 150 PE for 60M PE reads per sample (30M in each direction).

RNA sequencing analysis:

FastQC removed samples of poor quality⁷⁵). Trimmomatic (version v0.39) removed and trimmed adapter sequences and readings of poor quality. The web interface and public servers, usegalaxy.org and usegalaxy.eu, was used for mapping and analysis of RNAseq reads⁷⁶. The Galaxy server running Hisat2 (Galaxy Version 2.2.1+galaxy0), FeatureCounts (Galaxy Version 1.6.4), and Deseq2 (Galaxy Version 2.11.40.8+galaxy1) was used to map sequence reads^77–79. FeatureCounts files and raw files are available at GEO with accession XXX. Hisat2 was run with the following settings: paired-end, stranded, default settings (except for when a GTF file was used for transcript assembly). For GTF files, we used the Mus musculus (Mouse), Ensembl, GRCm39 build from iGenome (Illumina). The aligned SAM/BAM files were processed using Featurecounts with Default settings, except we used the Ensembl GRCm39 GTF file and output for DESeq2 and a gene length file.

Gene counts were normalized using DESeq2 (Love et al., 2014) followed by a regularized log transformation. Differential Expression was determined by DESeq2; the factors used were tissue type (hippocampus and cortex), Mrpl40 expression (+/+ and −/+), and Slc25a1 expression (+/+ and −/+). Pairwise comparisons were done across Mrpl40 and Slc25a overexpression status. Animal genotypes: Mrpl40 +/+ Slc25a1 +/+; Mrpl40 −/+ Slc25a1 +/+; Mrpl40 +/+ Slc25a1 −/+; Mrpl40 −/+ Slc25a1 −/+. All normalized tables, including regularized log transformation and variance stabilized tables, were generated using size estimation, where the standard median ratio was used; the fit type was parametric, and outliers were filtered using a Cook’s distance cutoff.

Data Access

RNAseq data are accessible through the GEO portal GSE315447

Statistics

ANOVA analyses were conducted with Prism Version 10.2.2. Data were tested for normality and log transformed when required for ANOVA analyses. Two-sided permutation t-test were done using the estimationstats engine⁸⁰ and Kolmogorov-Smirnov tests were conducted with the engine http://www.physics.csbsju.edu/stats/KS-test.n.plot_form.html.

Coessentiality Analysis was performed with the FIREWORKS engine³⁴ using SLC25A1 and MRPL40 genes as entries in the Pan-Cancer dataset as Context. The engine was run with the top 10 primary nodes and the top five secondary nodes positively and negatively correlated. Data were exported and analyzed in Cytoscape 3.10.2. Nodes enriched annotations were inspected with ENRICHR⁸¹.