Loss of heterozygosity exposes germline mutations in complex I and drives Warburg metabolism in oncocytic carcinoma of the thyroid

Abstract

Oncocytic (Hürthle cell) carcinoma of the thyroid (OCT) is characterized by widespread loss of heterozygosity (LOH), mitochondrial accumulation, and recurrent mitochondrial DNA mutations leading to impairment of complex I. Here, we establish and characterize a novel OCT cell line, UT946, which displays severe mitochondrial electron transport chain dysfunction and a Warburg metabolic phenotype. Using a series of cytoplasmic hybrids, we establish that the complex I defect in UT946 stems from a nuclear-encoded loss-of-function mutation in the complex I subunit NDUFS1. To our surprise, the mutation in NDUFS1 was inherited as a recessive germline allele that underwent LOH in the tumor to expose functional loss of complex I. A reanalysis of 91 OCT tumor genomes revealed that LOH-driven exposure of recessive germline mutations in complex I subunits was a recurrent mechanism underlying complex I inactivation in OCT. These findings unveil a previously unidentified germline-driven mechanism of complex I loss and metabolic reprogramming in cancer and provide further evidence of the selective pressure for complex I impairment in OCT.

Germline mutations in complex I induce aerobic glycolysis in oncocytic carcinoma of the thyroid through loss of heterozygosity.

INTRODUCTION

Central carbon metabolism, which includes glycolysis, the tricarboxylic acid (TCA) cycle, and the mitochondrial electron transport chain (ETC), is an interconnected series of biochemical reactions that allows cells to extract energy from nutrients such as glucose and produce the intermediate metabolites required for growth and proliferation (1). Defects in central carbon metabolism typically have profound effects on an organism’s growth and development. For example, mutations in components of the mitochondrial ETC can cause disorders including Leigh syndrome and mitochondrial encephalopathy, lactic acidosis, and stroke-like episodes, which present as systemic disorders affecting multiple organ systems including the central nervous and musculoskeletal systems (2).

Early studies by Warburg et al. (3) revealed that tumors produced excessive amounts of lactate even under aerobic conditions, a phenomenon later termed the “Warburg effect,” which he attributed to impaired mitochondrial respiration. Cellular respiration is performed by the mitochondrial ETC, a series of multiprotein complexes [reduced form of nicotinamide adenine dinucleotide (NADH) dehydrogenase, succinate dehydrogenase, cytochrome bc₁ oxidoreductase, cytochrome c oxidase, and adenosine 5′-triphosphate (ATP) synthase; also referred to as complexes I to V, respectively] that transfer electrons from reduced carriers such as NADH + H⁺ and FADH₂ to oxygen. Ultimately, ATP production is coupled to these electron transfer steps, allowing cells to efficiently generate energy from carbon-based fuels (1). While subsequent studies have confirmed that tumors and cultured cancer cells take up glucose and produce lactate at higher rates than healthy tissues, there is substantial evidence demonstrating that mitochondrial ETC activity is required for tumor growth and metastasis (4–11). Consistent with this, loss-of-function (LOF) mutations in mitochondrial ETC complex subunits, which are encoded by both the nuclear and mitochondrial genomes, are relatively uncommon in tumors (12–14).

However, specific mutations in TCA cycle and mitochondrial ETC components are protumorigenic, including gain-of-function mutations in isocitrate dehydrogenase 1 and 2 (IDH1 and IDH2) and LOF mutations in succinate dehydrogenase/mitochondrial ETC complex II (SDHA, SDHB, SDHC, and SDHD) or fumarate hydratase (FH) (15). The consequences of these mutations, which include the production of the oncogenic metabolite, (R)-2-hydroxyglutarate [(R)-2-HG] by mutant IDH, and the accumulation of the TCA cycle intermediates succinate and fumarate resulting from impaired SDH or FH activity, contribute to tumor development and progression through competitive inhibition of alpha-ketoglutarate (α-KG)–dependent enzymes such as dioxygenases and demethylases that ultimately regulate gene expression (15–18). These mutations are typically restricted to tumors arising from specific tissues, suggesting that the cell of origin may determine sensitivity to transformation by the metabolic consequences and gene expression programs induced by (R)-2-HG–dependent modulation of enzymes.

Our group and others previously reported that deleterious mitochondrial DNA (mtDNA) mutations affecting mitochondrial ETC complex I (NADH dehydrogenase) were enriched in oncocytic carcinoma of the thyroid (OCT) and maintained during tumor progression and metastasis (19–21). This contrasts with multiple other tumor types from diverse lineages that do not display enrichment of damaging mtDNA mutations despite the fact that somatic mtDNA mutations in tumors are relatively common (12, 14, 22). Oncocytomas, a class of tumors that includes OCT, are characterized by extensive mitochondrial accumulation and also display enrichment of deleterious mtDNA mutations (19–21). In addition to having an abundance of mitochondria, OCT tumors are glucose avid and display pronounced ¹⁸F-fluorodeoxyglucose (¹⁸F-FDG) uptake on positron emission tomography scans (23–25). These observations raise the possibility that genetic impairment of the mitochondrial ETC could underlie the mitochondrial accumulation and a metabolic reprogramming observed in OCT. Despite these common features, damaging somatic mtDNA mutations have been observed in approximately 60% of OCT tumors (19, 20), suggesting that 40% of OCT tumors either do not exhibit mitochondrial ETC impairment, or defects are driven by alternative genetic or epigenetic mechanisms. Here, we report the generation of a novel cell line derived from a patient with OCT that exhibits impaired complex I function. Our efforts to uncover the genetic basis of complex I impairment in this model led to the discovery of a previously unknown mechanism of complex I loss in OCT.

RESULTS

UT946 exhibits mitochondrial ETC dysfunction

We established a cell line derived from a primary OCT tumor with poorly differentiated/anaplastic features, UT946. Because previous studies have reported severe mitochondrial ETC dysfunction in patient-derived OCT models (26–28), we examined mitochondrial ETC function and cellular metabolism in a panel of thyroid cancer cell lines including UT946 (OCT), NCI-237^UTSW (OCT), TPC-1 (papillary thyroid carcinoma) and UT354 (poorly differentiated thyroid carcinoma). We performed Seahorse extracellular flux assays (29, 30) and observed that UT946, along with the OCT cell line NCI-237^UTSW, exhibited a severe respiration defect and increased extracellular acidification rate relative to non-OCT cell lines (Fig. 1, A and B). Cells with mitochondrial ETC defects require exogenous pyruvate and glucose for proliferation. Pyruvate sustains NAD⁺ regeneration and, ultimately, aspartate biosynthesis (31–34), whereas glucose supports cellular ATP production via glycolysis (35–37). We tested whether UT946 required exogenous pyruvate and glucose for viability and growth by culturing cells in pyruvate-free or glucose-free media. OCT cell lines, including UT946, required exogenous pyruvate for proliferation (Fig. 1C). In addition, we observed that UT946, and NCI-237^UTSW, exhibited a significant reduction in viability when grown in glucose-free media supplemented with galactose (Fig. 1D). By contrast, ETC-proficient thyroid cancer cell lines remained viable in media without glucose and sustained proliferation in pyruvate-free media (Fig. 1, C and D).

Fig. 1. UT946 exhibits ETC dysfunction and genetic features of OCT.

(A and B) Oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) of indicated OCT and non-OCT thyroid cancer cells after the addition of Oligo, oligomycin; CCCP, carbonyl cyanide 3-chlorophenylhydrazone; Rot., Rotenone; and Anti A, antimycin A [human plasma-like media (HPLM)] n = 6 to 8. (C) Fold-change in cell number of indicated cell lines in media ±100 μM pyruvate. n = 3 (HPLM). (D) Cell viability of indicated cell lines after 48 hours in media ±5 mM glucose or 5 mM galactose n = 3 (HPLM). (E) Schematic representation of ¹³C glucose labeled carbon incorporation into glycolysis and TCA cycle intermediates. Created in BioRender. McFadden, D. (2026) https://BioRender.com/8vk1adr. (F) Mass isotopomer abundance for the indicated metabolites in cells cultured with [U-¹³C₆] glucose for 6 hours (HPLM). Pyr., Pyruvate; Cit., Citrate; αKG, α-ketoglutarate; Fum., Fumarate; Mal., malate; Asp., Aspartate. (G) LOH plot showing variant allele fraction (VAF) of germline variants for UT946_{Cell Line}. Variants with VAF >0.5 are in red; variants with VAF <0.5 are in blue. (H) DNA content measured by propidium iodide staining in a diploid (HCT116) cell versus UT946_{Cell line}. Data are plotted as means ± SD of the indicated number of replicates. ****P ≤ 0.0001.

We assessed glucose metabolism in the panel of cell lines by culturing cells in media containing ¹³C-labeled glucose ([U-¹³C₆] glucose) and analyzing glucose carbon fate using gas chromatography–mass spectrometry (GC-MS). UT946 and NCI-237^UTSW displayed minimal incorporation of glucose-derived carbon into the citrate pool via pyruvate dehydrogenase and citrate synthase (Fig. 1, E and F) indicating overall reduced glucose-derived carbon movement through the TCA cycle. Together, we conclude that UT946 exhibited impaired mitochondrial ETC function and alterations in glucose metabolism consistent with a defect in complex I.

UT946 harbors key genetic features of OCT

We hypothesized that UT946 harbored an mtDNA-encoded complex I defect, like 60% of OCT tumors and all OCT cells lines reported to date (19, 20, 26, 27, 38). To establish a nuclear and mitochondrial genomic profile, we performed whole-exome sequencing (WES) of the primary tumor and matched cell line, referred to as UT946_Tumor and UT946_{Cell Line}, respectively. Normal thyroid tissue was not collected at the time of tumor resection. We analyzed exome sequencing datasets for regions of homozygosity, and we determined that UT946_Tumor displayed loss of heterozygosity (LOH) across Chr 1 to 4, 6, 8, 9, 11, 14 to 16, and 20 to 22, consistent with widespread LOH previously reported for larger cohorts of OCT tumors (fig. S1A) (19, 20, 26). UT946_{Cell Line} displayed more extensive LOH, with additional loss of one copy of Chr 7, 10, 13, 17, and X, while retaining heterozygosity for Chr 20 (Fig. 1G). To our knowledge, UT946 represents the only OCT reported to date that exhibits extensive LOH on Chr 7. While many OCT tumors develop near-haploid genomes as a result of widespread chromosome loss, OCT tumors frequently undergo whole-genome duplication (WGD) following chromosome loss, resulting in tumors with LOH and complex ploidies (19, 20). We analyzed the cellular DNA content of UT946_{Cell Line} using propidium iodide staining and flow cytometry and observed approximately 4N DNA content, suggestive of UT946_{Cell Line} having undergone WGD (Fig. 1H).

Because normal thyroid or blood was not collected from this patient, we identified putative somatic mutations from WES data by annotation of known germline single-nucleotide polymorphisms (SNPs) to exclude common germline mutations (table S1). Missense mutations of unknown functional significance from known SNPs in the DNA methyltransferase DNMT3A, [Chr2:25300227T>G; p.Glu30Ala; variant allelic fraction (VAF)^WES = 0.990], as well as in the DNA repair gene BRCA1 (Chr17:43104249T>C; p.Tyr105Cys; VAF^WES = 0.466) (fig. S1B) were found in UT946_Tumor. These mutations were maintained in UT946_{Cell Line}. Consistent with LOH on Chr 17 in UT946_{Cell Line}, the VAF of the BRCA1 mutation was ~1 (fig. S1C). This variant was heterozygous in normal thyroid (fig. S1D). Analysis of the mutational signature in the UT946 tumor and cell line did not reveal a BRCA1-associated signature (single-base substitution signature 3, SBS3; fig. S1, E and F). The lack of the BRCA1-associated signature suggested that this mutation was unlikely to represent an LOF variant.

Potentially deleterious missense or nonsense mutations in a series of tumor suppressors including CDKN2A (Chr9:21971156G>A; p.Ala68Val; VAF^WES = 0.565), PTEN (Chr10:87957915C>T; p.Arg233*; VAF^WES = 0.960), TP53 (Chr17:7674220C>T; p.Arg248Gln; VAF^WES = 0.950), and NF2 (Chr22:29661313C>T; p.Arg262*; VAF^WES = 0.992) were identified in UT946_{Cell Line}, as well as a noncanonical activating mutation in PIK3CA that created an arginine-to-glutamine substitution at amino acid residue 88 (Chr3:179199088G>A; p.Arg88Gln; VAF^WES = 0.451) (fig. S1C) (39–41). These mutations have previously been reported in advance forms of thyroid carcinomas, consistent with the pathological observations of poorly differentiated and anaplastic features (42, 43). Together with the observation of additional regions of LOH in UT946_{Cell Line}, the emergence of these mutations associated with advanced forms of thyroid cancer is consistent with the outgrowth of a tumor subclone during the generation of the cell line.

We next sought to identify mtDNA mutations present in the UT946 tumor and cell line. Off-target exome sequencing reads that mapped to mtDNA exhibited a mean coverage depth of 1212× and 123× in UT946_Tumor and UT946_{Cell Line}, respectively. Analysis of mtDNA-mapped reads identified two variants of interest in UT946_Tumor: a homoplasmic missense mutation in MT-ND5 causing a threonine-to-alanine substitution at amino acid residue 21 (ChrM:12397A>G; p.Thr21Ala; VAF^WES = 0.999) and a noncoding homoplasmic mutation in MT-TRNQ (ChrM:4336 T>C; VAF^WES = 1) (fig. S1G). These mutations were maintained at homoplasmy during establishment of UT946_{Cell Line}, and we identified three additional heteroplasmic mtDNA mutations including missense mutations in MT-ND1 (ChrM:4201G>A; p.Ala299Thr; VAF^WES = 0.232), MT-COX3 (ChrM:9804G>A; p.Ala200Thr; VAF^WES = 0.180), and MT-CYTB (ChrM:15884G>A; p.Ala380Thr; VAF^WES = 0.432) (fig. S1H). We concluded from our metabolic and genomic analysis that UT946 represented a bona fide OCT tumor:cell line pair.

Mitochondrial ETC dysfunction in UT946 is not mtDNA encoded

We hypothesized that the MT-ND5^T21A missense mutation represented an LOF mutation resulting in complex I impairment in UT946. To test whether this mutation was causal for the mitochondrial ETC dysfunction in UT946, we generated cytoplasmic hybrid (cybrid) cell lines that harbored different mitochondrial genomes in a common nuclear background (7, 44). Cybrids were prepared by fusion of mtDNA-depleted host cells (ρ⁰ cells) and enucleated mitochondrial donor cells (cytoplasts) followed by fluorescence-activated cell sorting (FACS) of putative cybrids and subsequent selection in uridine-free media (7). This strategy allowed us to isolate cybrid lines with different combinations of nuclear and mitochondrial genomes (Fig. 2A). To ensure that the results were not biased because of cell line subclones, we generated two or three independent cybrid cell lines for each mitochondrial:nuclear genome pair of interest. UT946, TPC-1, or 143B (an osteosarcoma cell line regularly used in cybrid experiments) enucleated cytoplasts were obtained by ultracentrifugation against a Percoll gradient. ρ⁰ cells were obtained by serial passaging of UT946 and TPC-1 cells in 2′,3′-dideoxycytidine (ddC), a nucleoside analog that inhibits mtDNA replication (45). To ensure mtDNA depletion in ρ⁰ cells, we measured mtDNA levels by quantitative polymerase chain reaction (qPCR) using primer pairs directed against mtDNA targets (MT-ND1 and MT-CO3) and a nuclear genome target (ACTB). ddC treatment caused progressive depletion of mtDNA up to 500-fold relative to untreated cells. Removal of ddC led to a slow but consistent recovery of the mtDNA over time (fig. S2, A and B). After cybrid generation and expansion, we confirmed mtDNA repopulation in cybrid lines by qPCR. UT946-based cybrids displayed nearly 100% mtDNA repopulation, whereas TPC-1–based cybrids were not fully repopulated (fig. S2, C and D). TPC-1–based cybrids displayed mitochondrial ETC function similar to parental TPC-1 cells (discussed below), suggesting that the level of mtDNA repopulation was sufficient to restore mitochondrial function. To validate the nuclear and mitochondrial genomic integrity of the cybrids, we used short-tandem repeat (STR) profiling and Oxford nanopore long-read sequencing of PCR-amplified mtDNA fragments, respectively (table S2 and fig. S2, E and F). We discarded cybrids containing incorrect or mixed nuclei and/or mtDNAs, and we successfully established several cybrid lines to investigate ETC dysfunction in UT946_{Cell Line} (Fig. 2A).

Fig. 2. Mitochondrial ETC dysfunction in UT946 is not mtDNA-encoded.

(A) Summary table of generated cybrids and the subsequent technical validation. Nuc., nucleus. Mito., mitochondria. STR, short tandem repeat. Seq., sequencing. SH – Resp, Seahorse respiration. (B) Schematic representation of TPC cybrids. Created with BioRender.com. Mcfadden, D. (2026) https://BioRender.com/ztqllao. (C) Normalized OCR in indicated cybrids and cell lines n = 8 to 12. (D) Fold change in cell number of indicated cell lines relative to reconstituted Dulbecco’s modified Eagle’s medium (DMEM) containing ± 100 μM pyruvate (Pyr), ± uridine (Uri, 50 μg/ml), ± 5 mM glucose/galactose (Gal) n = 3. (E) Schematic representation of UT946 cybrids. Created with BioRender.com. Mcfadden, D. (2026). https://BioRender.com/qsgtyvl. (F) Normalized OCR in indicated cybrids and cell lines n = 8 to 12. (G) Fold change in cell number of indicated cell lines relative to reconstituted DMEM containing ± 100 μM pyruvate (Pyr), ± uridine (Uri, 50 μg/ml), ± 5 mM glucose/galactose (Gal) n = 3. Data are plotted as means ± SD of the indicated number of replicates.

To determine whether UT946 mtDNA encoded a functional ETC, we introduced UT946 mitochondria into TPC-1 nuclear donors (TPC^Nuc946^mito). If the MT-ND5^T21A, or any other variant, encoded in the UT946 mtDNA acted as an LOF event, TPC^Nuc946^mito cybrids should exhibit a respiration defect. In contrast, we observed that TPC-1–based cybrids harboring UT946 mtDNA displayed oxygen consumption and extracellular acidification rates similar to the parental TPC-1 cell line or TPC-1 cybrids expressing functional 143B mtDNA (TPC^Nuc143B^mito) (Fig. 2, B and C, and fig. S2G). Consistent with intact mitochondrial ETC function, we also observed that TPC^Nuc946^mito cybrid lines exhibited no growth defect, relative to parental TPC-1 cells or a TPC^Nuc143B^mito cybrid line, when cultured in pyruvate- or uridine-free media, or in galactose-containing media (Fig. 2D). We concluded that, contrary to our initial hypothesis, the UT946 mitochondrial genome encoded a fully functional complex I.

Mitochondrial complex I dysfunction in UT946 is nuclear encoded

We next tested whether the UT946 nuclear genome encoded a defect in mitochondrial respiratory ETC. We introduced functional mitochondria from 143B and TPC-1 cells into UT946 nuclear donors, generating UT946-based cybrids (946^nuc143B^mito and 946^nucTPC^mito). These UT946-based cybrids exhibited oxygen consumption defects and extracellular acidification rates, which closely matched the parental UT946 cell line or UT946 ρ⁰ cells repopulated with UT946 mitochondria (946^nuc946^mito) (Fig. 2, E and F, and fig. S2H). All UT946-based cybrids displayed proliferation defects when cultured in pyruvate-free or galactose-containing media, but not uridine-free media, suggesting a mitochondrial ETC defect linked to complex I (Fig. 2G). Together, these findings indicated that the ETC defect in UT946 cells originated from a nuclear-encoded defect in complex I rather than a damaging mtDNA mutation.

To identify potential genetic causes of complex I dysfunction in UT946_{Cell Line}, we reexamined the WES data and focused on nuclear-encoded complex I subunits. We identified a missense mutation in the complex I subunit, NDUFS1 (Chr2:206146957A>G; c.683T>C; p.Val228Ala) that generated a valine-to-alanine substitution at position 228. In both the cell line and tumor sequencing datasets, this variant existed at an allele fraction of 1.0, suggesting that the Chr2 LOH we observed (Fig. 1G) led to the loss of the wild-type (WT) allele. Although the chemical change between valine and alanine was modest, previous studies had identified NDUFS1^V228A as a disease-associated mutation in compound heterozygous patients with complex I deficiency (46–48). In addition, computational predictions CADD (28.1) (49), REVEL (0.944) (50), and PolyPhen-2 (0.999) (51) indicated a high likelihood of pathogenicity for this variant.

To examine the effect of NDUFS1^V228A mutation on complex I function, we disrupted NDUFS1 using CRISPR-Cas9 in respiration-competent TPC-1 cells. Cells with NDUFS1 deletion displayed reduced oxygen consumption relative to WT TPC-1 cells or TPC-1 cells in which we targeted the neutral AAVS1 locus (Fig. 3, A to C), confirming the essentiality of NDUFS1 gene for complex I function. We next expressed an exogenous copy of either WT NDUFS1 or NDUFS1^V228A in NDUFS1 knockout (KO) cells (Fig. 3A). Reexpression of WT NDUFS1, but not NDUFS1^V228A, was sufficient to restore respiration and decrease acidification rates, establishing NDUFS1^V228A as an LOF mutation (Fig. 3C and fig. S3A). However, whether this mutation was solely responsible for the complex I defect observed in UT946 cells was not certain.

Fig. 3. Mitochondrial complex I dysfunction in UT946 is nuclear-encoded.

(A) Schematic representation of NDUFS1^V228A validation strategy. NDUFS1 gene was knocked out (KO) in TPC-1 respiration competent cells followed by the reexpression of the WT or V228A form. Created with BioRender.com. Mcfadden, D. (2026) https://BioRender.com/3qg6hpb. (B) TPC-1–edited cells were collected at indicated passages (p3, p5, and p7) after transduction with CRISPR-Cas9 system. KO cells were infected as indicated viral dilution (1:1, 1:6, 1:9) with either the WT or V228A form of NDUFS1. Protein lysates were immunoblotted for indicated proteins. (C) Normalized OCR in indicated TPC-1–edited cell lines 1:1, 1:9 refers to viral dilution upon NDUFS1 reexpression, n = 8 to 12. (D) Schematic representation of the repair strategy. HDR, Homologous-directed repair. Created with BioRender.com https://BioRender.com/s888kcd. (E) UT946^Repair cells were collected after selection. Protein lysates were immunoblotted for indicated proteins. (F) Sanger sequencing traces of the genomic region surrounding the V228A mutation in the indicated cell lines. PAM, protospacer adjacent motif. (G) Normalized OCR in indicated UT946-edited cell lines, n = 8 to 12. Data are plotted as means ± SD of the indicated number of replicates.

We therefore used CRISPR-Cas9 to revert the NDUFS1^T683C (p.Val228Ala) mutation. We introduced Cas9 ribonucleoprotein (RNP) complex targeting NDUFS1 with a single-stranded oligodeoxynucleotide (ssODN) template encoding the WT NDUFS1 sequence at nucleotide 683 (Fig. 3D) into UT946 cells. Cells harboring WT NDUFS1 (UT946^Repair) were enriched via a metabolism-based selection strategy wherein cells were cultured in glucose-free (galactose-replete) media. If reversion of the NDUFS1^V228A mutation to the WT form led to restoration of complex I function, then these cells should survive in glucose-free media whereas cells with mutant NDUFS1 should die. We observed that substantial, but not complete cell death after UT946^Repair cells were shifted to glucose-free conditions. Surviving cells were expanded and pooled. We observed increased NDUFS1 protein levels in UT946^Repair cells compared to UT946 cells harboring the NDUFS1^V228A mutation (UT946^V228A; Fig. 3E). This finding raised the possibility that a complex I defect in UT946^V228A cells resulted either from destabilization of the NDUFS1 protein or indirectly through destabilization of the entire complex as has been previously reported following mutation of individual subunits (52). We assessed NDUFS1 mutation status in UT946^Repair cells by Sanger sequencing of PCR amplicons surrounding NDUFS1 nucleotide 683, and we observed reversion of the V228A mutation to the WT allele (Fig. 3F). We subjected UT946^V228A and UT946^Repair cells to Seahorse extracellular flux assays to determine whether repair of the mutation was sufficient to restore complex I and ETC function. We observed restored respiratory activity in UT946^Repair cells (Fig. 3G and fig. S3B). These results established that NDUFS1^V228A was singularly responsible for the complex I defect observed in UT946^V228A cells, and demonstrated that at least in some cases, nuclear mutations in complex I subunits can drive metabolic alterations in OCT.

NDUFS1^V228A induces metabolic adaptation to complex I dysfunction

These experiments led to the generation of an isogenic cell line pair in which a naturally arising complex I defect was reverted to the reference allele, resulting in restoration of mammalian complex I function (UT946^V228A and UT946^Repair cells). We used this system to interrogate several metabolic dependencies that arise in the context of impaired respiration. First, to isolate the effect of V228A mutation on cell proliferation independently of metabolic stress, we evaluated cell growth in nutrient-replete conditions. We quantified cell doublings over several passages in UT946^V228A and UT946^Repair and TPC-1 cells stably reexpressing either NDUFS1^WT or NDUFS1^V228A. The presence of NDUFS1^V228A mutation did not affect proliferation compared to NDUFS1^WT counterpart (fig. S3, C to F). Therefore, in nutrient-replete culture conditions in which glucose and pyruvate were abundant, NDUFS1^V228A-driven complex I impairment did not limit cell growth.

Several studies have demonstrated that NAD⁺ regeneration limits proliferation in cells lacking complex I function (31, 32, 53, 54). As a result, cells rely on alternative pathways for biogenesis and energy metabolism, including glycolysis, glutamine anaplerosis, and de novo nucleotide biosynthesis (31, 37, 53, 54). Repair of the NDUFS1^V228A mutation in UT946 cells was sufficient to revert auxotrophy for glucose and pyruvate (fig. S3G). Similarly, TPC-1 KO cells expressing NDUFS1^V228A exhibited a dependency on glucose and pyruvate (fig. S3H).

In some cancer cells, ETC defects suppress de novo purine synthesis causing dependency on purine uptake and salvage to maximize growth (54). However, addition of purine nucleosides (adenosine, inosine, and guanosine) to media lacking pyruvate was not sufficient to rescue proliferation of UT946^V228A or TPC-1 NDUFS1^V228A-expressing cells. The same was true under glucose withdrawal or in media containing limiting glutamine (0.1 mM; fig. S3, G and H). Tumor cells with mitochondrial dysfunction have also been reported to use glutamine-dependent reductive carboxylation to replenish TCA intermediates (53). Complete glutamine withdrawal arrested proliferation but did not induce cell death in UT946^V228A and UT946^Repair cells (fig. S3I). Glutamine limitation also did not exacerbate the growth defect caused by pyruvate withdrawal in UT946^V228A or TPC-1 cells expressing NDUFS1^V228A. In contrast, pyruvate supplementation rescued the proliferation defect associated with low glutamine (fig. S3, G and H). UT946^V228A and UT946^Repair cells also exhibited similar sensitivity to the glutaminase inhibitor CB839 with both lines exhibiting intrinsic resistance [median inhibitory concentration (IC₅₀) > 20 μM; fig. S3J]. Together, these results suggested that complex I impairment in UT946^V228A and TPC-1 KO cells expressing NDUFS1^V228A induced a broad metabolic defect that was not rescued by glutamine or purine nucleoside supplementation.

OCT tumors were reported to exhibit elevated polyunsaturated fatty acid levels and increased dependency on glutathione peroxidase-4 (GPX4), a phenotype linked to complex I deficiency (26). We noted that GPX4 protein was reduced in UT946^Repair cells relative to UT946^V228A cells (fig. S3K). In addition, UT946^V228A cells treated with the GPX4 inhibitor RSL3 had increased lipid peroxidation relative to UT946^Repair cells (fig. S3, L and M). Consistent with these results, we noted a small difference in sensitivity to the GPX4 inhibitor ML210 (55), with the UT946^V228A cells exhibiting greater sensitivity (IC₅₀: UT946^V228A = 1.61 μM versus UT946^Repair = 4.85 μM; fig. S3N). These findings suggested that ETC dysfunction increased lipid peroxidation following GPX4 inhibition and sensitized, albeit slightly, to GPX4 inhibition in our culture conditions.

We also used an orthogonal model of ETC restoration by expression of the yeast complex I, NDI1. We stably expressed FLAG-tagged NDI1 in UT946 cells (UT946^NDI1; fig. S3O). NDI1 expression restored oxygen consumption in UT946 cells to levels comparable to those of UT946^Repair cells, while UT946^mCherry controls remained indistinguishable from parental UT946^V228A cells (fig. S3P). Consistent with NDI1 bypassing complex I, the increase in maximal respiration upon carbonyl cyanide 3-chlorophenylhydrazone (CCCP) uncoupling in UT946^NDI1 cells was rotenone resistant. UT946^NDI1 cells did not exceed the respiration of UT946^Repair cells, which maintained higher basal and maximal respiration. NDI1 expression did not affect proliferation under standard nutrient-replete conditions as UT946^NDI1 and UT946^mCherry cells grew equivalently (fig. S3, Q and R). In pyruvate-free conditions, UT946^NDI1 cells exhibited rescue of cell growth compared to UT946^mCherry cells and similar to UT946^Repair cells (fig. S3, G and S). Last, glucose withdrawal, even in the presence of galactose, imposed a much more severe constraint on viability. Notably, while no UT946^mCherry control cells survived 5 days of glucose deprivation, a subpopulation of UT946^NDI1 cells remained viable under these conditions, demonstrating a partial but meaningful rescue even under this stringent metabolic challenge (fig. S3S). The incomplete rescue observed relative to UT946^Repair cells (fig. S3, G and S) might be due to selection for UT946 cells following repair of NDUFS1^V228A using glucose withdrawal, leading to a cell population specifically adapted to survival in glucose-deplete conditions. Together, these findings supported a causal role for complex I dysfunction in the metabolic adaptations of UT946^V228A cells and demonstrated that restoration of NADH oxidation through NDI1 was sufficient to rescue both respiratory function and survival under nutrient stress.

LOH of recessive germline mutations unmasks complex I deficiency in OCT

Complex I is a multisubunit enzyme composed of 45 proteins, of which 38 are encoded by the nuclear genome and 7 by the mitochondrial genome. Although sequencing studies of OCT tumors have consistently identified complex I mutations in approximately 60 to 70% of cases, complex I mutations have only been reported in mitochondrial-encoded complex I genes (19, 20). We hypothesized that the UT946 tumor represented a rare example of a somatic mutation in a nuclear-encoded complex I subunit. However, because matched normal thyroid tissue was initially unavailable, we could not establish somatic origin of the NDUFS1^T683C (p.Val228Ala) mutation. We retrieved paraffin-embedded tumor and adjacent normal thyroid tissue from the clinical tissue blocks. We performed targeted PCR amplification and Sanger sequencing of NDUFS1 from normal thyroid and tumor to determine whether the NDUFS1^V228A mutation arose specifically in the tumor. We observed NDUFS1^T683C (p.Val228Ala)in normal thyroid as a heterozygous mutation, which rejected our hypothesis that NDUFS1^T683 was acquired during tumorigenesis (Fig. 4A). Although we could not discriminate between a germline variant and a developmental field mutation, we favored the hypothesis that NDUFS1^T683C was an inherited variant. NDUFS1 is encoded on chromosome 2, which underwent whole-chromosome loss in the UT946 primary tumor and cell line. This LOH event resulted in the loss of the WT NDUFS1 allele, exposing NDUFS1^V228A and complex I impairment.

Fig. 4. LOH of recessive germline mutations unmasks complex I deficiency.

(A) Sanger sequencing traces of the genomic region surrounding the V228A mutation in UT946 normal thyroid and tumor. DNA obtained from paraffin blocks. (B) CoMut plot showing somatic mitochondrial encoded mutations in complex I genes (mtDNA) and recessive germline variants that underwent LOH in the tumor targeting complex I genes encoded in the nucleus (nDNA). Silent (validated) mutations are missense mutations validated by SH to encode a functional complex I. (C) Normalized OCR in TPC-1 NDUFV1–edited cell lines, n = 8 to 12. (D) Protein lysates from TPC-1 NDUFV1–edited cells were immunoblotted for indicated proteins. (E) Normalized OCR in TPC-1 NDUFAF6–edited cell lines, n = 8 to 12. (F) Protein lysates from TPC-1 NDUFAF6–edited cells were immunoblotted for indicated proteins. (G) Normalized OCR in TPC-1 NDUFS8–edited cell lines, n = 8 to 12. (H) Protein lysates from TPC-1 NDUFS8–edited cells were immunoblotted for indicated proteins. Data are plotted as means ± SD of the indicated number of replicates.

The observation of a recessive LOF allele (NDUFS1^T683C) in tumor-adjacent normal thyroid prompted us to explore whether LOH-driven exposure of recessive germline variants represented an alternative mechanism of complex I impairment in OCT. Tumor genome data are typically analyzed to identify true somatic mutations; consequently, variants present in matched normal tissue or buffy coat samples are excluded. We reanalyzed WES data from previously published studies of OCT, comprising a total of 91 patients with OCT (Fig. 4B and table S3) (19, 20). We specifically annotated germline variants that underwent LOH in our analysis. We observed a total of 21 of 91 (23%) cases that harbored germline coding variants in complex I subunits that underwent loss of the WT allele in the tumor samples (table S3 and fig. S4A). Of these, 8 of 91 (9%) represented clear LOF mutations, including four frameshift mutations and one nonsense mutation in subunits required for complex I function (56–59): NDUFV2 (p.His21Argfs*6), NDUFA6 (p.Met104Cysfs*35), NDUFAF7 (p.Gln384Alafs*7), NDUFAF3 (p.Gly164Serfs*29), and NDUFB6 (p.Tyr84*) (fig. S4A). In addition, we identified a nonstart splicing-altering mutation in NDUFAF6 (chr8:95046993 G>A) and two start codon mutations in NUBPL and NDUFAF4 (60–63). In at least three of these cases, buffy coat was used at the normal tissue control, which established a germline origin for these events (fig. S4A).

We assessed whether mutations in nuclear-encoded complex I genes were mutually exclusive with mtDNA-encoded mutations, which would support a similar selective advantage of these events. We reanalyzed exome-sequencing reads that mapped to mtDNA to identify somatic mtDNA events. To identify mutations that were likely to affect complex I function, we selected missense and LOF mutations in protein-coding sequences with VAF >0.7 in tumors and VAF <0.5 in matched normal (Fig. 4B and table S3). Although our initial pipeline failed to call 10 variants reported in the original studies, after individual review, we confirmed and included each of these events in our analysis (Fig. 4B). Of the 21 individuals with nuclear-encoded complex I mutations, 19 (90%) had no detectable mtDNA mutations or had mutations in genes unlikely to have functional consequences for complex I function. Only two (HCTC-102 and SRS3086539) carried LOF mutations in both nuclear- and mitochondrial-encoded complex I genes (Fig. 4B and table S3).

We observed 13 of 91 (14%) cases that harbored a missense germline complex I mutation that underwent LOH in tumor samples. Notably, two independent patients harbored the same NDUFS1^V228A mutation we identified and studied in UT946. In one case, NDUFS1^T683C (p.Val228Ala) was detected as a heterozygous mutation in the buffy coat, confirming germline origin (fig. S4A). The NDUFS1^V228A mutation was therefore detected in 3.26% (3 of 91) of patients with OCT, representing an ~250-fold enrichment, compared with a reported frequency of 0.013% in the general population (https://gnomad.broadinstitute.org/variant/2-206146957-A-G?dataset=gnomad_r4).

We also identified 10 additional missense mutations of unknown significance affecting complex I core, supernumerary, and assembly factors (fig. S4A). NDUFV1 (p.Leu138Pro), NDUFS8 (p.Trp61Arg), NDUFV3 (p.Pro4Arg, p.Arg26Gln), NDUFA6 (p.His108Tyr), NDUFAF6 (p.Ala178Pro), NUBPL (p.Asn198Thr), TMEM126B (p.Asp133Asn), and TMEM186 (p.Leu133Arg, p.Ala159Val). Each of these subunits, except NDUFV3 and TMEM186, was reported to be required for complex I function (52, 59–61, 64–66). The necessity of these subunits for complex I function raised the possibility that some of these variants induced a functional complex I defect in OCT tumors. To determine whether these mutations encoded a complex I defect, we used CRISPR-Cas9 to disrupt each of the genes in respiration-competent TPC-1 cells and subsequently reexpress the corresponding WT or mutant subunit. Multiple in silico tools supported the deleterious nature of NDUFV1^L138P and TMEM126B^D133N variants, including high scores from CADD (31.0 and 29.2), REVEL (0.936 and 0.861), and PolyPhen-2 (0.991 and 1), respectively. Reexpression of NDUFV1^L138P failed to restore respiration in NDUFV1 KO cells, consistent with L138P representing an LOF variant (Fig. 4, C and D, and fig. S4B). In contrast, TPC1-TMEM126B^KO cells expressing TMEM126B^D133N exhibited similar respiratory activity as cells expressing the WT protein, suggesting that D133N was a silent mutation (fig. S4, C to E). NDUFAF6^A178P was predicted to be a deleterious mutation; however, computational evidence was less compelling compared to variants such as NDUFV1^L138P, CADD (26.2), REVEL (0.73), and PolyPhen-2 (0.979). Nevertheless, reexpression of NDUFAF6^A178P in NDUFAF6 KO cells failed to rescue respiration (Fig. 4, E and F, and fig. S4F) establishing this variant as a bona fide LOF mutation. NDUFA6^H108T and NUBPL^N198T variants were not predicted to be pathogenic. Consistently, reexpression of the mutant alleles in the corresponding KO cells restored respiration (fig. S4, G to N). Last, the NDUFS8^W61R variant was not present in gnomAD or clinVar databases. We observed that upon reexpression of the mutant form, respiration in NDUFS8 KO cells was partially restored, suggesting that, within the limit of our experimental strategy to precisely control expression levels, NDUFS8^W61R might act as a hypomorphic variant (Fig. 4, G and H, and fig. S4O). In all cases, deletion of the target complex I subunit resulted in destabilization of other complex I subunits, including NDUFS1, as has been previously reported. Consequently, NDUFS1 protein levels served as a surrogate marker for NDUFAF6 KO in the absence of a high-quality NDUFAF6-specific antibody (fig. S4H). As anticipated, not every missense mutation led to functional impairment of complex I. Nonetheless, 13 of 91 (14%) OCT tumors carried definitive LOF mutations in nuclear-encoded complex I genes, with one additional tumor harboring a hypomorphic mutation. Together, these findings suggest that ETC dysfunction in OCT generally arises through one of two distinct mechanisms: (i) somatically acquired mutations in mtDNA-encoded complex I subunits or (ii) LOH-driven exposure of germline LOF mutations in nuclear-encoded complex I subunits.

DISCUSSION

Here, we report the generation of an OCT tumor cell line that exhibits the key features of these uncommon cancers: impaired mitochondrial respiration and widespread loss of chromosomes (19–21). The generation of UT946 represents an advance for OCT research, considering that only three patient-derived models have been reported (26, 27, 38). Our efforts to determine the genetic basis of complex I impairment in UT946 fortuitously led to the discovery of a recessive LOF germline allele in NDUFS1 that underwent LOH in the tumor to expose the mutant phenotype. We put forth that this study provides an example of the power of studying rare disease outliers with extreme phenotypes (and genotypes). By probing the genetic basis of complex I impairment in this tumor-cell line pair derived from a single patient, we uncovered an unexpected genetic mechanism of somatic complex I impairment leading to the Warburg effect. This mechanism, however, is not uncommon in OCT and represents a recurrent genetic mechanism that explains the basis of complex I impairment and Warburg metabolism in 14% (13 of 91) of our OCT cohorts. Together with the approximately 60% of cases that harbor putative LOF mutations in the mtDNA, we now understand the genetic basis of complex I loss in approximately 75% of cases. Whether the remaining 25% harbor mutations in proteins not yet known to control complex I function, or in other pathways that lead to increased mitochondrial biogenesis that is pathognomonic of OCT will require additional studies. Testing for germline events that undergo LOH might be warranted in clinical genotyping assays should therapies be developed that target complex I, as has been reported in laboratory models (26, 27). As demonstrated here, the UT946 isogenic pair in which repair of NDUFS1^V228A fully restored complex I function might serve as an ideal experimental system to interrogate metabolic dependencies of complex I–deficient OCT cells.

LOH that exposes germline mutations represents a classical mechanism of tumor suppressor inactivation. Mutations in complex II (SDH complex) underlie familial paraganglioma syndromes (67). However, SDH also acts as a TCA cycle enzyme, and tumor development following SDH loss is reported to stem from an accumulation of the SDH substrate succinate to levels that impair α-KG–dependent enzymes (15). This mechanism is similar to mutations in other TCA enzymes including FH and IDH, in which the enzyme substrates accumulate and impair α-KG–dependent dioxygenases and other enzymes. Complex I loss, which does not lead to accumulation of TCA substrates, induces distinct metabolic alterations compared to mutations within TCA cycle enzymes (68). These differences and the absence of recurrent SDH, FH, or IDH mutations in OCT suggest that complex I loss imparts different selective advantages compared to defects in TCA cycle enzymes.

Our finding of recurrent LOF germline mutations in complex I that undergo LOH to expose the mutant allele provides further evidence of the selective advantage of complex I loss in OCT. This observation parallels our previous report of allelic enrichment of complex I mutations encoded in the mtDNA in OCT, in comparison to most cancers in which complex I loss is selected against (12, 26). We also note that the NDUFS1^V228A mutation was identified in three independent patients in our cohort (~3%). Considering gnomAD reports an allele frequency of 1.3 × 10⁻⁴ for this variant (https://gnomad.broadinstitute.org/variant/2-206146957-A-G?dataset=gnomad_r4), we observe more than 250-fold enrichment for this allele in our small cohort of patients. Positive selection for complex I mutations has also been demonstrated in experimental models (69, 70). However, experimental evidence demonstrating that complex I loss provides a growth advantage is limited, suggesting that complex I loss could be under selection in very early stages of tumorigenesis or that complex I loss becomes fixed in the tumor at a very early stage of tumorigenesis. OCT does not classically present in a familial pattern, and in the absence of a family in which complex I germline mutations are strongly associated with tumor incidence, we specifically avoid claiming that these mutations represent bona fide tumor suppressor events akin to SDH mutations. We favor that inheritance of a recessive complex I LOF allele might predispose to the development of oncocytic features in a thyroid tumor, and in this sense represents a tumor modifying allele as opposed to a tumor suppressor mutation. Nonetheless, additional studies of patients with recessive complex I alleles will be required to determine whether these mutations predispose to thyroid nodules, benign, or malignant, and whether oncocytic features are more common in nodules that arise in these patients.

Why have not these germline mutations in complex I been previously described? Cancer genome sequencing studies generally seek to identify tumor-specific somatic mutations. Therefore, germline events are excluded in analysis pipelines to identify somatic mutations. Despite the fact that a majority of the OCT genome undergoes LOH, all prior studies, including our own, followed a standard analysis approach and failed to report germline events (19, 20). In addition, identification of recurrent mutations in genes that encode subunits of large multiprotein complexes is more complex and requires detailed knowledge of subunit and complex assembly, stability, and function. This study exemplifies the power that the germline retains over tumor genotype and phenotype and reminds us to review germline variation in tumor genome studies, particularly in tumors, like OCT, in which a large fraction of the genome undergoes LOH (19, 20). We hypothesize that continued analysis of tumor genomes to systematically identify LOH of germline alleles will uncover additional genes and pathways that contribute to tumorigenesis and modify tumor phenotypes.

MATERIALS AND METHODS

Patient-derived materials and human cell lines

UT946 (female) cell line was derived from a primary OCT tumor with evidence of anaplastic/poorly differentiated transformation. Snap-frozen and fresh tissue samples were collected under IRB Protocols STU 072017-103 (principal investigator, D.G.M.) and STU 102010-051 (principal investigator, C.L.). Written informed consent for tissue collection and research use was obtained from the patient before sample collection. Fresh tissue was minced and placed in 5-ml Dulbecco’s Modified Eagle Medium (DMEM, Sigma-Aldrich, D6429) containing 200 μl of Enzyme H, 100 μl of Enzyme R, and 25 μl of Enzyme A that were prepared according to the manufacturer’s instruction (Miltenyi, 130-095-929) and incubated for approximately 30 min at 37°C with agitation every 5 to 10 min. The resulting cell suspension was filtered through a 40-μm nylon mesh filter, pelleted, and resuspended in DMEM supplemented with 10% fetal bovine serum (FBS, Sigma-Aldrich, F0926), 2 mM l-glutamine (Sigma-Aldrich, G7513), 1% penicillin-streptomycin (Sigma-Aldrich, P4333), and uridine (50 μg/ml, Sigma-Aldrich, U3003). NCI-237 and UT354 cell lines were generated as previously described in Frank et al. (27). TPC-1 (female) was obtained from S. Parangi (Massachusetts General Hospital, Boston, MA, USA). Human embryonic kidney (HEK) 293T/17 were purchased from American Type Culture Collection (CRL-11268). HCT-116 (male) was obtained from D. Nijhawan (University of Texas Southwestern Medical Center, TX, USA). 143B osteosarcoma cells were purchased from American Type Culture Collection (CRL-8303).

Cell lines were maintained at 37°C with 5% CO₂ and cultured in DMEM (Sigma-Aldrich, D6429) supplemented with 10% FBS, 2 mM l-glutamine, and 1% penicillin-streptomycin unless stated otherwise. For metabolic assays (Fig. 1), human plasma–like media (HPLM) were used. Cell lines were adapted to culture in HPLM supplemented with 2% FBS, 1% penicillin-streptomycin, and 1× insulin-transferrin-selenium for at least five passages. HPLM pool stocks were prepared and stored as previously described in (27). Prepared HPLM was used within 3 to 4 days, and supplements were added daily before use. All cell lines were subjected to STR profiling performed by the Eugene McDermott Center for Human Growth and Development Sequencing Core. Cell lines were periodically tested for mycoplasma contamination using a PCR-based assay. Before conducting cell culture experiments, cell lines were passed at least once after thawing, and experiments were performed within 10 to 25 passages of being thawed.

Respiration analysis

Mitochondrial respiration measurements were performed using an Agilent Seahorse XFe96 Analyzer according to the manufacturer’s instructions for Mito Stress kit. Briefly, before seeding the cells, 96-well plates were coated with ESGRO Complete Gelatin Solution (Sigma-Aldrich, # SF008). Then, the cells were plated at 20,000 cells per well in 80 μl of culture media and incubated overnight. The following day, the cells were rinsed twice using Seahorse assay medium [DMEM (Sigma-Aldrich, D5030) with 10 mM glucose, 2 mM l-glutamine, 1 mM sodium pyruvate, and 1% penicillin-streptomycin], and 150 μl assay medium was added after the second wash. Oxygen consumption rate and extracellular acidification rate (ECAR) were measured after equilibration in a CO₂-free incubator for 45 min following the sequential addition of oligomycin (1.5 μM), CCCP (3 μM), rotenone (0.5 μM), and antimycin A (0.5 μM). For normalization (10 μg/ml) Hoechst 33342 was added directly into the wells and incubated for 15 min. Live cells were counted using the Celígo 4 channel image cytometer (Nexcelom Bioscience).

Cell proliferation assays

For assays in Fig. 1, cells were plated in six-well plates (Corning) in 2 ml of HPLM media at the following densities: NCI-237 (30,000 cells per well), UT946 (30,000 cells per well), TPC-1 (30,000 cells per well), and UT354 (100,000 cells per well), and allowed to adhere overnight. The following day, the cells were washed once with 2 ml of phosphate-buffered saline (PBS, Sigma-Aldrich, D8537), and 4 to 8 ml of HPLM growth media was added to the respective wells. A separate “Pool 9 stock” solution that lacked pyruvate was prepared and used to prepare HPLM without pyruvate before pH adjustment and sterile filtration.

For assays in Fig. 2, TPC-1–derived as well as UT946-derived cybrids, were seeded in duplicate 12-well plates (Corning) at a density of 25,000 cells per well in 1 ml of media. The following day (designated as day 0), one plate was washed once with PBS (Sigma-Aldrich, D8537), trypsinized, and counted to establish baseline cell numbers. Then, the remaining plate was washed with PBS and incubated for either 3 (TPC-1) or 5 (UT946) days in 3 ml of the corresponding media. Media lacking specific nutrients (pyruvate, uridine, or glucose) were prepared using glucose-free, pyruvate-free DMEM (Thermo Fisher Scientific, #11966) supplemented with 10% dialyzed FBS, 2 mM l-glutamine (Sigma-Aldrich, #G7513), and 1% penicillin-streptomycin as a base. Reconstituted media were supplemented with 1 mM sodium pyruvate (Sigma-Aldrich, #P2256-10MG), uridine (50 μg/ml; Sigma-Aldrich, #U3003-50G), and 5 mM glucose (Sigma-Aldrich, #G7021). To generate nutrient-depleted conditions, the corresponding component (pyruvate or uridine) was simply omitted. For glucose-deprivation experiments, glucose was replaced with 5 mM galactose (Sigma-Aldrich, #G5388-100G). Conditions were assayed in triplicate.

For assays in fig. S3, the same procedure as for Fig. 2 was followed. Media lacking specific nutrients (pyruvate, glucose, and glutamine) were prepared using glucose-free, pyruvate-free glutamine-free DMEM (Thermo Fisher Scientific, #A14430) supplemented with 10% dialyzed FBS and 1% penicillin-streptomycin as a base. Reconstituted media were supplemented with 1 mM sodium pyruvate (Sigma-Aldrich, #P2256-10MG), 6 mM glutamine (Sigma-Aldrich, #G7513), 5 mM glucose (Sigma-Aldrich, #G7021), and 50 μM of a mixture of purine nucleosides [adenosine (Sigma-Aldrich, #A9251), inosine (Sigma-Aldrich, #I4125), and guanosine (Sigma-Aldrich, #G6752)]. To generate nutrient-depleted conditions, the corresponding component (pyruvate, glutamine, or purine nucleosides) was simply omitted. For glucose-deprivation experiments, glucose was replaced with 5 mM galactose (Sigma-Aldrich, #G5388-100G). For glutamine-deprivation experiments, 0.1 mM or no glutamine was used. Conditions were assayed in triplicate.

Viability assays

Cells were plated in six-well plates (Corning) in 2 ml of HPLM media at 100,000 cells per well and allowed to adhere overnight. The following day, the cells were washed once with 2 ml of PBS, and 8 ml of HPLM growth media was added to the respective wells. Growth media lacking glucose were prepared by supplementing the media with 5 mM galactose or 5 mM glucose from sterile 500 mM stocks. The cells were cultured for 48 hours before all cells (floating and adherent) were collected by trypsinization and pelleting at 500g for 5 min. The cells were resuspended in 500 μl of PBS + propidium iodide (0.5 μg/ml; Life Technologies, P1304MP) and stained in the dark at room temperature for 10 min before data acquisition and data analysis on a Guava easyCyteTM flow cytometer (EMD Millipore). Conditions were assayed in triplicate and experiments were performed at least twice.

Cell doubling experiments

To assess growth rate, UT946-edited cells were seeded at 1.5 million cells per 150-cm² dish for 2-day passages and at 1.8 million cells per dish for 3-day passages, and cell numbers were determined over five to six consecutive passages. TPC-1–edited cells were seeded at 0.8 million cells per 150-cm² dish for 2-day passages and at 1.0 million cells per dish for 3-day passages. In all cases, the cells were cultured in maintenance medium consisting of DMEM (Sigma-Aldrich, D6429) supplemented with 10% FBS, 2 mM l-glutamine, and 1% penicillin-streptomycin.

¹³C glucose labeling

Cells were plated in six-well plates at the following densities: NCI-237 (300,000 cells per well), UT946 (300,000 cells per well), TPC-1 (400,000 cells per well), and UT354 (800,000 cells per well), in 3 ml of the media and allowed to adhere overnight. The following day, HPLM was prepared and supplemented with 5 mM [U-¹³C₆] glucose before sterile filtration. Following incubation at 37°C and 5% CO₂, the cells were collected as described in (27). Briefly, the cells were extracted in methanol:water (80:20) and vacuum-dried before derivatization. Derivatized samples were transferred to Agilent GC-MS autosampler vials, and data acquisition was performed using an Agilent G2579A MSD coupled to an Agilent 6890 gas chromatogram. Data were analyzed using Agilent MSD Chemstation software and a MatLab script for area-under-the-curve analysis, total ion count determination, and natural isotopomer abundance correction. Conditions for all GC-MS experiments were assayed in triplicate.

Compound dose-response curve

UT946-edited cells were seeded into Corning 3903 96-well plates at a density of 2000 cells per well in 100 μl of medium. After 24 hours, the medium was aspirated and replaced with 200 μl of fresh medium. CB839 (Selleckchem, #S7655) or ML210 (MedChemExpress, #HY-100003) was added as a threefold serial dilution using a D300e Digital Dispenser (Tecan). Following 72 hours of compound exposure, cell viability was measured with the ATP-based luminescence assay CellTiter-Glo (Promega, G7570). Before reagent addition, plates were equilibrated to room temperature for 30 min. Next, 70 μl of CellTiter-Glo reagent, prepared according to the manufacturer’s instructions and diluted 1:1 with PBS plus 1% Triton X-100, was added to each well. The plates were then shaken orbitally at 120 rpm for 10 min at room temperature, and luminescence was recorded using Cytation 5 plate reader.

Generation of SBS mutation profiles

WES variants from UT946_Tumor and UT946_{Cell Line} were used for mutational signature analysis. Because no matched normal sample was available, rare single-nucleotide variants were selected using dbSNP151 allele frequency <0.001 and gnomAD allele frequency <0.001 to enrich for likely somatic mutations. SBS96 mutational profiles were generated using trinucleotide sequence context from the hg38 reference genome, compared with COSMIC reference signatures, and analyzed with SigProfilerAssignment using COSMIC v3.5 in exome mode with artifact signatures excluded. Analysis codes are available at https://github.com/jiwoongbio/Annomen/.

WES analysis–LOH analysis

For the reanalysis of published OCT data (19, 71), raw reads were trimmed with Trim Galore (https://bioinformatics.babraham.ac.uk/projects/trim_galore/) to remove adapter sequences and low-quality bases and then aligned to the human reference genome (hg38) using Burrows-Wheeler aligner (v0.7.17) (72). PCR duplicates were removed with Picard (2.21.3) (https://broadinstitute.github.io/picard), and base-quality scores were recalibrated with Genome Analysis Toolkit (GATK, 4.1.4.0) (73, 74). Variant calling followed GATK best-practice recommendations and the resulting Variant Call Formats (VCFs) were filtered (QD <2, FS >60, MQ <40, DP <10, GQ <20) and annotated with RefSeq, dbSNP, COSMIC, ClinVar, gnomAD, and CADD using the custom Annomen pipeline (https://github.com/jiwoongbio/Annomen). Genome-wide LOH and runs of homozygosity were inferred from variant allele frequency profiles with AutoMap (75) and validated with PLINK (--homozyg) (76). Broad LOH regions were visualized through per-chromosome variant allele frequency plots generated by a custom R script. Previously published OCT datasets were obtained from Sequence Read Archive [#SRP136351 for Ganly et al. (19)] and from the database of Genotypes and Phenotypes [dbGaP:phs001580.v1.p1 for Gopal et al. (20)].

Ploidy

For assessing ploidy in UT946 and HCT116, 500,000 cells were stained using Guava cell cycle reagent (#4500-0220) following the manufacturer’s instructions. Briefly, cells were centrifuged and washed once with PBS. Then, the cells were added to 70% ethanol to a concentration of 5 × 10⁵ cells/ml for fixation and incubated for 3 hours at −20°C. An aliquot of fixed cells was then centrifuged and washed with 1× PBS. Last, cells were resuspended in 200 μl of Guava cell cycle Reagent and incubated for 30 min before data acquisition and data analysis on the Guava easyCyteTM flow cytometer (EMD Millipore).

Lipid peroxidation measurement

For assessment of lipid peroxidation, 1 million UT946-edited cells were incubated with BODIPY 581/591 C11 (Thermo Fisher Scientific, D3861) in the presence or absence of 10 μM RSL3 (Cayman Chemical #19288) in 100 μL of serum-free medium for 30 min at room temperature followed by 15 min at 37°C. The cells were then pelleted, washed twice with PBS, resuspended in 300 μl of serum-free medium, filtered, and analyzed on a Guava easyCyte flow cytometer (EMD Millipore). Lipid oxidation was quantified by measuring the shift in BODIPY 581/591 C11 fluorescence from red to green, detected in the 488 of 510 and 58 of /591 channels, respectively. No stained cells were used as negative controls.

Cytoplasmic hybrid (cybrid) generation

Cybrids were generated using a modified version of the protocol described in Shelton et al. (7). The day before cybrid generation, ultracentrifugation tubes were prepared with a Percoll (Sigma-Aldrich, P1644-25ML):media gradient. The media consisted of DMEM (Sigma-Aldrich, D6429) supplemented with 10% FBS, 2 mM l-glutamine, 1% penicillin-streptomycin, and uridine (50 μg/ml). The gradient was layered as follows: 10 ml of 50% Percoll, followed by 4 ml each of 35, 30, 26, and 22% Percoll. Tubes were equilibrated overnight at 37°C and 5% CO₂. The next day, one 150-mm plate containing 5 to 20 million donor cells was trypsinized, collected, and counted. Three million cells were set aside as positive and negative controls. The remaining cells were centrifuged at 200g for 5 min, resuspended in 4 ml of 12.5% Percoll:media solution containing cytochalasin B (10 μg/ml; Cayman, 11328), 100 nM MitoTracker Green (Cell Signaling Technology, #9074P), and Hoechst 33342 (10 μg/ml. The suspension was gently layered on top of the prepared gradient. Tubes were balanced and centrifuged at 25,000 rpm for 60 min at 37°C with minimal braking. During centrifugation, one million of the reserved cells were stained with either Hoechst or MitoTracker (single-color controls), and one million remained unstained as a negative control. After centrifugation, the cytoplast-enriched band appearing between gradient layers (~1 ml) was carefully collected and washed in 49 ml of DMEM (Sigma-Aldrich, D6429). Cytoplasts and controls were centrifuged at 650g for 10 min, resuspended in FACS buffer (Hanks’ balanced salt solution, Gibco 14175079) containing 2% FBS, 1% penicillin-streptomycin, and 0.5 mM EDTA (Invitrogen, 15575020), and adjusted to ~5 million cells/ml. Cytoplasts were sorted using a BD FACSMelody Cell Sorter by selecting MitoTracker-positive [fluorescein isothiocyanate (FITC)] and Hoechst-negative (BV421) populations. In parallel, a 100-mm plate of recipient ρ⁰ cells was harvested and counted. Approximately one million ρ⁰ cells were set aside as negative controls. The remaining cells were stained in media containing 50 nM SYTO62 (Invitrogen, #811344) for 1 hour at 37°C. The cells were then washed, resuspended in 1 to 2 ml of DMEM, and counted. Following cytoplast sorting, equal numbers of cytoplasts and SYTO62-labeled ρ⁰ cells were mixed in 2 ml of serum-free DMEM and incubated at room temperature for 10 min to facilitate membrane contact. A 200 μl of aliquot of the mixture was reserved as a nonfused control. Both the fusion mixture and nonfused control were centrifuged at 650g for 10 min. The nonfused pellet was resuspended in 300 μl of FACS buffer. For the fusion reaction, the supernatant was carefully aspirated from the pellet to avoid disruption. A total of 100 μl of PEG solution was added directly to the pellet, and the cells were gently homogenized. After 1 min of incubation, prewarmed serum-free DMEM was gradually added: 100 μl in the first minute, 200 μl in the second, and up to 500 μl total over several min. The cell suspension was then incubated at 37°C for 15 min in a water bath, followed by centrifugation at 650g for 10 min. The resulting fused cell pellet was resuspended in 500 μl of FACS buffer and sorted on the BD FACSMelody Cell Sorter. Successfully fused cybrids were identified as double-positive for MitoTracker Green (FITC) and SYTO62 (PE-Cy5). Sorted cells were centrifuged again at 650g for 10 min, resuspended in DMEM (Sigma-Aldrich, D6429) supplemented with 10% FBS, 2 mM l-glutamine, 1% penicillin-streptomycin, and uridine (50 μg/ml), and plated into a 96-well plate. Media were refreshed the following morning to allow cybrid attachment. Once the cells were established and expanded over several passages, cybrids were transitioned to uridine-free media to eliminate any residual ρ⁰ cell contamination. Uridine-free media were prepared using DMEM (Sigma-Aldrich, D6429) supplemented with 10% dialyzed FBS, 2 mM l-glutamine, and 1% penicillin-streptomycin. Cybrids were cultured in this selective media for four to six passages, after which mitochondrial and nuclear DNA content was assessed. All cybrids were subjected to STR profiling performed by the Eugene McDermott Center for Human Growth and Development Sequencing Core to validate the nuclear entity of the new cybrid line.

ρ⁰ cell generation

Cells were cultured in complete medium consisting of DMEM (Sigma-Aldrich, D6429) supplemented with 10% FBS, 2 mM l-glutamine, 1% penicillin-streptomycin, and uridine (50 μg/ml). To deplete mtDNA, cells were treated with 10 μM ddC (MedChem, HY-17392) and passaged every 2 days. mtDNA depletion was monitored by qPCR and considered successful when reduced by more than 500-fold relative to WT cells. Since mtDNA levels slowly recover after ddC removal, ρ⁰ cells were used shortly after drug withdrawal, and mtDNA abundance was routinely quantified during cybrid generation.

DNA extraction

Genomic DNA was isolated from tissue culture samples and snap-frozen tissue samples by resuspension in approximately 500 ml of genomic DNA extraction buffer [10 mM tris-HCl (pH 8.0), 25 mM EDTA (pH 8.0), 100 mM NaCl, and 1% SDS] supplemented with proteinase K (0.5 mg/ml; Zymo, D3001-2-20) and incubated overnight at 55°C. DNA was subsequently purified by phenol-chloroform extraction followed by ethanol precipitation as previously described (77). The resulting DNA was resuspended in nuclease-free water or TE buffer [1 mM EDTA (pH 8.0) and 10 mM tris (pH 8.0)] and stored at −20°C for downstream applications. For genomic DNA isolation from Formalin-Fixed, Paraffin-Embedded (FFPE) samples, consecutive 10-mm sections were mounted on clean slides. Tumor and normal thyroid margins were demarcated by a trained pathologist (J.A.B.). The respective tumor and normal thyroid sections were gently scraped from slides using a clean scalpel and transferred to clean tubes. Samples were processed using the QIAamp DNA FFPE Tissue Kit (Qiagen 56404) according to the manufacturer’s instructions.

RNA extraction and cDNA synthesis

Total RNA was extracted using TRIzol reagent (Life Technologies, 15596-018) following the manufacturer’s protocol. One microgram of RNA was reverse-transcribed into cDNA using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, #4368813) according to the manufacturer’s instructions. Synthesized cDNA was stored at −20°C for future analyses.

Quantitative PCR

qPCR was performed in triplicate using the GoTaq qPCR Master Mix (Promega, A6001) in a total reaction volume of 10 μl, containing 3 ng of DNA or cDNA. For mtDNA quantification, the nuclear-encoded gene ACTB was used as a reference for nuclear DNA. Mitochondrial genes ND1 and COX3 served as proxies for mtDNA content. NDUFA6 was used for gene expression analysis. See table S4 for primers.

mtDNA sequencing validation

To validate mtDNA in cybrids, long-range PCR was performed using total DNA as template and KAPA HiFi HotStart DNA Polymerase (Roche, 07958927001). Amplified mtDNA fragments were submitted to Plasmidsaurus for linear PCR sequencing using Oxford Nanopore Technology, with custom analysis and variant annotation (table S4).

PCR amplification and sanger sequencing

The NDUFS1 V228A mutation was validated by PCR amplification of exon 8 using total DNA as a template. PCR products were subjected to Sanger sequencing with the forward primer used as the sequencing primer. To confirm the expression of the correct coding sequences (WT versus mutant) for the seven missense mutations expressed from PGK-driven plasmids in TPC KO cells, total DNA was amplified by PCR using the corresponding primer pairs (table S4). All PCR reactions were performed with KAPA HiFi HotStart DNA Polymerase (Roche, 07958927001). Mutations were verified by Sanger sequencing using at least the forward primer; when necessary, additional primers were used to confirm the presence or absence of the mutation of interest.

Lentivirus production

Lentiviruses were produced by cotransfecting HEK293T/17 cells with the lentiviral expression plasmid of interest along with the packaging plasmids psPAX2 and pMD2.G at a 5:3:2 mass ratio (vector:psPAX2:pMD2.G), using TransIT-LT1 transfection reagent (Mirus, MIR 2300). Viral supernatants were collected at 48 and 72 hours posttransfection, pooled, and filtered through 0.45-μm Polyethersulfone (PES) filters. Supernatants were either used immediately or aliquoted and stored at −80°C for future use.

Generation of CRISPR-Cas9–edited cells

To KO the desired genes (NDUFS1, NDUFV1, NDUFAF6, NDUFS8, NDUFA6, NUBPL, and TMEM126B), three single guide RNAs (sgRNAs) per gene were designed using CrisPick (Broad Institute) and/or based on the Brunello library and cloned into the LentiCRISPR-v2 backbone (Addgene, #52961) (table S4). Briefly, LentiCRISPR-v2 was linearized with BsmBI-v2 (New England Biolabs, #R0739L) according to the manufacturer’s instructions, and the linearized plasmid was gel-purified using the Macherey-Nagel Gel Extraction Kit (#740609.50S). In parallel, sgRNA oligonucleotides were annealed and phosphorylated with T4 Polynucleotide Kinase (New England Biolabs, #M0201S) and ligated into the linearized vector using T4 DNA Ligase (New England Biolabs, #M0202S). Recombinant plasmids were transformed into Stbl3 (Zymo Research T3001) competent cells and grown under ampicillin selection (100 μg/ml). Plasmid DNA was isolated using the Promega Miniprep Kit (#A1222). Lentiviruses encoding sgRNAs were produced as described above. TPC-1 cells were seeded at 4 × 10⁵ cells per well in six-well plates, and lentiviral supernatants were added at a 1:1 ratio, unless indicated otherwise, with fresh medium at the time of plating. After overnight incubation, the medium was replaced, and 48 hours posttransduction, the cells were transferred to selection medium containing puromycin (2 μg/ml; Sigma-Aldrich, P8833). Selection was maintained for three to five days to generate pooled CRISPR-Cas9–edited populations. Cell survival following puromycin selection was typically 80 to 100%, as estimated visually. Protein depletion was verified by Western blotting, and the most efficient sgRNA for each gene (listed in table S4) was selected for downstream experiments. When population-level results were inconclusive, single-cell clones were isolated from the total population. Briefly, 500, 1000, or 2000 TPC-1 cells were seeded in independent 150-cm² plates in triplicate. After approximately 2 weeks, individual colonies were manually picked, expanded, and validated by western blotting. Verified clones were subsequently used for re-expression of either the WT or mutant form of the gene of interest.

Cloning strategy for stable expression of complex I genes and NDI1

To stably express WT or mutant forms of human complex I genes (NDUFS1, NDUFV1, NDUFAF6, NDUFS8, NDUFA6, NUBPL, and TMEM126B) in CRISPR-Cas9–edited cells, point mutations were introduced by PCR amplification. These included silent mutations in the protospacer adjacent motif sequences to prevent reediting, as well as the desired missense mutations. Reference human cDNA templates were obtained from the McDermott Center for Human Growth and Development (Ultimate ORF Lite human cDNA collection, Life Technologies) for NDUFS1, NDUFV1, NDUFA6, NDUFS8, and NUBPL and used as PCR templates. For NDUFAF6 and TMEM126B, cDNA templates were obtained from GeneScript (clone ID: OHu19552D and OHu29970D, respectively). Amplified cDNA fragments were inserted into a PGK-IRES-Blast lentiviral vector previously digested with BamHI and XBaI (New England Biolabs, #R3136S and #R0145S) using Gibson Assembly (New England Biolabs, #E2621S), generating the final expression constructs. The assembled plasmids were transformed into chemically competent Escherichia coli Stbl3 cells (Zymo Research T3001) by heat shock for 40 s, followed by recovery in 300 μl of LB medium for 10 min at 37°C without antibiotic selection. A total of 100 μl of the transformation mixture was plated onto LB agar plates supplemented with ampicillin (100 μg/ml), and individual colonies were picked for plasmid expansion. Construct identity was confirmed by plasmid purification followed by long-read sequencing. For reexpression of WT or mutant constructs in TPC-1 KO cells, lentiviral transduction was performed as described for CRISPR-Cas9–mediated gene editing. Forty-eight hours posttransduction, the cells were plated directly into medium containing blasticidin (5 μg/ml; InvivoGen, ant-bl-1) and maintained under selection for 7 to 10 days to generate pooled reexpressing cell lines. Cell survival following blasticidin selection was typically 80 to 100%, as estimated visually.

To stably express NDI1 and mCherry in UT946 cells, NDI1 and mCherry fragments were PCR amplified from previously generated lab plasmids. Fragments were introduced into previously linearized pLVX-IRES-Puro plasmid with XbaI and BamHI using Gibson Assembly. Lentiviral production and transduction was performed as previously described for CRISPR-Cas9–mediated gene editing. Forty-eight hours posttransduction, the cells were plated directly into medium containing puromycin (2 μg/ml) and maintained under selection for 3 to 7 days to generate pooled expressing cell lines.

Repair of NDUFS1^V228A using IDT Alt-R CRISPR-Cas9 system

Repair of the NDUFS1^T683C (p.Val228Ala) point mutation directly in UT946 cells was achieved via homology-directed repair (HDR) using the Alt-R CRISPR-Cas9 System (IDT) and ssODNs. HDR donor templates flanking the V228A mutation in exon 8 of NDUFS1 were designed using the Alt-R HDR Design Tool (IDT) (78). The protocol involved codelivery of the HDR donor ssODN and a CRISPR-Cas9 RNP complex via electroporation using the 4D-Nucleofector System (Lonza). For nucleofection, 1.5 million UT946 cells were washed with 1× PBS and resuspended in 71.4 μl of Buffer SE (Lonza). The cell suspension was combined with the RNP complex and HDR donor template, transferred to a well of a Nucleocuvette Plate (Lonza), and electroporated using program EO-100. After nucleofection, the cells were rested for 10 min at room temperature and then transferred to 100-mm dishes containing prewarmed complete media. Cells were incubated at 37°C and 5% CO₂, and media were replaced the following morning. After several passages and complete recovery, selection for successfully repaired cells was performed by culturing cells for 24 hours in glucose-free, galactose-containing medium [DMEM D5030 supplemented with 10% dialyzed FBS, 1 mM sodium pyruvate, 5 mM galactose, 2 mM l-glutamine, 1% penicillin-streptomycin, and uridine (50 μg/ml)]. Remaining viable clones were expanded in regular media and validated by Sanger sequencing.

Immunoblotting

Cells were rinsed with ice-cold PBS and lysed in buffer containing 10 mM Hepes (pH 7.4), 150 mM NaCl, 2 mM MgCl₂, and 1% SDS, supplemented with 0.08 units of Universal Nuclease (Thermo Fisher Scientific, #88702). Lysates were incubated for 10 min at room temperature, followed by boiling at 96°C for 10 min. Protein content was measured with bicinchoninic acid protein assay (Thermo Fisher Scientific, 23222), and protein was denatured by the addition of sample buffer, boiled for 5 min, resolved by SDS–polyacrylamide gel electrophoresis, and analyzed by immunoblotting. Western blot analyses were performed according to standard procedures. Antibodies were visualized by using Odyssey Infrared Imaging System (Application software version 3.0.30) LI-COR Biosciences. Antibodies from Cell Signaling Technology were used for detection of β-actin (#4970, RRID: AB_2223172) dilution 1:5000, P-AMPK (#2535, RRID: AB_331250) dilution 1:1000, and GPX4 (#52455S, RRID: AB_2924984) β3-tubulin (#5568S, RRID: AB_10694505). Antibodies from Proteintech were used to detect NDUFV1 (#11238-1-AP, RRID: AB_2149040) dilution 1:1000, vinculin (#66305-1-Ig, RRID:AB_2810300) dilution 1:5000, dilution 1:1000, and NDUFA6 (#15445-1-AP, RRID: AB_2267065) dilution 1:500. Antibodies from Abcam were used to detect NDUFS1 (#ab169540, RRID: AB_2687932) dilution 1:1000, NDUFS3 (#ab110246, RRID: AB_10861972) dilution 1:1000 and NUPBL (#ab171741) dilution 1:1000. Antibodies from Santa Cruz Biotechnology were used to detect NDUFS8 (#sc-515527) dilution 1:500 and antibodies from Millipore Sigma were used to detect NDUFAF6 (#HPA050545, RRID: AB_2681166) and Anti-flag (#F3165, RRID: AB_259529).

Supplementary Materials

The PDF file includes:

Figs. S1 to S4

Legends for tables S1 to S3

Table S4

Legend for data S1

Other Supplementary Material for this manuscript includes the following:

Tables S1 to S3

Data S1