Genome-wide synthetic lethal screen unveils novel CAIX-NFS1/xCT axis as a targetable vulnerability in hypoxic solid tumors

To identify synthetic lethal interactions with CA9, we knocked the gene out in SUM159PT TNBC cells using CRISPR-Cas9 (fig. S1, A to C). We then screened the genome-wide GeCKO lentiCRISPRv2 library A consisting of 65,383 single-guide RNAs (sgRNAs) targeting 19,050 genes in a dropout screen to identify synthetic lethal interactions (Fig. 1A). Control cells expressing a nontargeting sgRNA (NTC) and CA9 (carbonic anhydrase IX) knockout (KO) (CA9^KO) cells were maintained in 1% O₂ following transduction with the lentiviral library and passaged for 20 generations to allow adequate gene dropout time. Changes in sgRNA abundance corresponding to each gene between the initial (T₀) and the final time point (T₁₁; representing 20 generations) were determined by next-generation sequencing. We identified 762 statistically significant hits that we classified by biological function using gene ontology (Fig. 1B and table S1). Among the significantly enriched processes affected by loss of CA9 were RNA metabolism, nonsense-mediated decay, ribosomal RNA processing, selenocysteine synthesis, selenoamino acid metabolism, tricarboxylic acid (TCA) cycle, and electron transport pathways (Fig. 1B and fig. S1D). Further refining the CA9 synthetic lethal interactions into a network of functional pathways exposed vulnerable gene networks associated with cytoskeleton, cell cycle and mitosis, ribosome biogenesis, RNA processing, mitochondrial organization, DNA damage repair (DDR) and nucleic acid (NA) metabolism, and redox homeostasis (Fig. 1C). The redox homeostasis network is particularly interesting, as CAIX function has not been previously linked to redox homeostasis. Among the genes we identified involved in redox homeostasis were thioredoxin (TXN), glutathione S-transferases (GSTM2 and GSTM5), glutathione reductase (GSR), and the molybdenum cofactor (MOCS3), all genes playing a critical role in antioxidant defense (26). In addition, a set of genes involved in iron-sulfur cluster biogenesis including the cysteine desulfurase (NFS1), iron-sulfur cluster scaffold protein (ISCU), iron-sulfur cluster adapter protein (ISCA2), adenosine 5′-triphosphate–binding cassette subfamily B member 7 (ABCB7), and glutaredoxin 5 (GLRX5) were identified (27). Ranking hits according to their fitness score, one of the top synthetic lethal genes in the screen was NFS1 (Fig. 1D). NFS1 catalyzes the first step in the generation of iron-sulfur clusters, cofactors for proteins involved in a number of cellular functions, including the TCA cycle and electron transport chain, by removing the thiol group from cysteine generating alanine and a persulfide group on the scaffold protein ISCU (28, 29). Moreover, NFS1 was recently implicated in protection from the iron-dependent, oxidative form of cell death, ferroptosis (30). Given the evidence demonstrating that CAIX affects cellular metabolism and the abundance of enzymes involved in cell metabolism requiring iron-sulfur clusters as cofactors, we decided to pursue this synthetic lethal interaction (29, 31–35).

With little being known about the importance of NFS1 expression across various cancers, we interrogated datasets from The Cancer Genome Atlas (TCGA) for NFS1 expression. We identified a number of solid cancers where NFS1 expression is elevated relative to normal tissue expression, including breast cancer (fig. S1E). Since the screen was carried out in a TNBC cell line, we took a closer look at the TCGA invasive breast cancer cohort after stratifying the patient population according to molecular subtype and identified CA9 and NFS1 expression that were correlated in the basal subtype (Fig. 1E). Furthermore, patients with elevated CA9 and NFS1 mRNA expression in their tumors in this cohort survive for shorter periods (Fig. 1F). To confirm the transcription data from the TCGA datasets extended to expression of the two proteins, we stained a small breast tumor tissue microarray consisting of triple-negative cases for CAIX and NFS1 (Fig. 1G). We and others have previously shown that CAIX is highly expressed and is a marker of worse patient prognosis in breast cancer (11, 14). CAIX was expressed in ~65% of the TNBC cases in this array, whereas NFS1 was expressed in nearly all cases (fig. S1F). These data suggest that both CAIX and NFS1 are expressed in a subset of patients with breast cancer where cotargeting may be beneficial.

To validate NFS1 as a hit from the screen, we took the top two performing sgRNAs targeting NFS1 from the library pool and generated NTC and CA9^KO cells expressing each sgRNA. We then transduced each line with Cas9 to create double (CA9^KO) or single (NTC) NFS1 KOs and assessed viability (fig. S2, A and B). In agreement with the pooled library screen, double KO cell lines had two- (sgNFS1_1) and threefold (sgNFS1_2) more cell death than the single KOs over a 7-day growth period (fig. S2, A and B). As an orthogonal approach, we used small interfering RNA (siRNA)–mediated suppression of NFS1 in SUM159PT CA9^KO cells (Fig. 2, A and B, and fig. S2C). Depletion of NFS1 in CA9^KO cells with two independent siRNAs led to 3.5- and fourfold increases in cell death relative to NTC cells (Fig. 2, A and B), thus providing additional, independent confirmation of this synthetic lethal interaction. With NFS1 critically involved in iron-sulfur cluster biogenesis, we assessed whether mitochondrial iron levels were affected upon NFS1 depletion in CA9^KO cells using rhodamine B-[1,10-phenanthrolin-5-yl)aminocarbonyl]benzyl ester (RPA), a dye quenched in the presence of Fe²⁺ (36). To evaluate iron levels before the induction of cell death following NFS1 depletion, we assessed these cells after a 72-hour expression of the siRNA versus the 96-hour time point where cell death was observed. We observed a decrease in RPA fluorescence in NFS1-depleted CA9^KO cells indicative of increased iron levels (Fig. 2, C and D). Increased cellular iron levels can react with H₂O₂ via the Fenton reaction to produce ROS, leading to lipid peroxidation, a hallmark of ferroptosis (37). To address whether the observed increase in mitochondrial iron levels affected lipid peroxidation, we stained cells with BODIPY C11. In agreement with the increased iron levels observed in the NFS1-depleted CA9^KO cells, we observed a concomitant increase in lipid peroxidation (Fig. 2, E to G).

To assess whether pharmacological inhibition of CAIX could recapitulate the phenotype associated with genetic loss of CA9, we suppressed NFS1 expression with siRNA in SUM159PT cells and administered concentrations of SLC-0111, a clinically evaluated small molecule inhibitor of CAIX/XII (17), that effectively inhibits CAIX activity in cells (Fig. 3, A and B, and fig. S2D) (32, 34). In agreement with previous findings, a reduction in NFS1 levels alone resulted in very little cell death in these hypoxic cultured cells (Fig. 3, A and B, and fig. S2E) (30). However, combining NFS1 depletion with SLC-0111 triggered a dose-dependent increase in cell death (Fig. 3, A and B). Treating this combination with the radical trapping antioxidant and ferroptosis inhibitor, ferrostatin-1 (Fer-1) (20), completely blocked the increased cell death (Fig. 3, A and B). We confirmed these findings in additional models of TNBC [MDA-MB-231 LM2-4 (LM2-4) and 4T1] as well as in models of bladder (UC-16) and colon cancer (HT-29) (Fig. 3C and fig. S2, F to H). Moreover, the combination of NFS1 suppression and SLC-0111 resulted in increased lipid peroxidation (Fig. 3, D and E, and fig. S2, I and J). To demonstrate that this effect was not unique to SLC-0111, we used an additional ureidosulfonamide inhibitor of CAIX/XII, compound 13, which also effectively inhibits CAIX/XII activity in cells (fig. S2K) (38). Combining compound 13 treatment with NFS1 depletion in SUM159PT (Fig. 3F) and LM2-4 (Fig. 3G) resulted in increased cell death in both cell lines in a dose-dependent manner, thus providing evidence that, in addition to genetic loss of CA9, inhibition of CAIX/XII activity results in increased cellular dependence on NFS1 for survival and that combined loss of both results in ferroptosis.

To evaluate this relationship in vivo, we generated SUM159PT cell lines expressing doxycycline (Dox)–inducible short hairpin–mediated RNA (shRNA) to NFS1 (fig. S2L) and verified that inducible depletion of NFS1 in combination with CAIX inhibition resulted in enhanced cell death (fig. S2M). Cells were implanted orthotopically, and tumors were allowed to form before initiation of treatment. Tumors expressing control shRNA (shNS) or NFS1 shRNA (shNFS1) in the absence of Dox grew at similar rates (Fig. 3H and fig. S2N). CAIX/XII inhibition (shNS + Dox + SLC-0111) or NFS1 depletion (shNFS1 + Dox + vehicle) alone had little impact on tumor growth in this model. However, combining both treatments (shNFS1 + Dox + SLC-0111) caused significant suppression of tumor growth (Fig. 3, H and I). Together, these data suggest that targeting both CAIX and NFS1 is an effective therapeutic strategy to overcome resistance to CAIX/XII inhibition in solid tumors (Fig. 3J).

We next sought to determine the mechanistic details behind the enhanced ferroptosis resulting from cotargeting CAIX/XII and NFS1. Since the combination increases lipid peroxidation and can be blocked in the presence of Fer-1 (Fig. 3, A to C), we next assessed whether cellular ROS levels were affected in the absence of CAIX expression or activity. We identified increased ROS in CA9^KO and SLC-0111–treated and CAIX-depleted cells (Fig. 4, A to C, and fig. S3, A to C). Restoring CAIX expression in CAIX-depleted cells returned ROS to basal, hypoxic levels (Fig. 4C). We then evaluated whether NFS1 depletion would further affect ROS production (fig. S3, D and E). Depletion of NFS1 did not significantly affect cellular production of ROS, whereas inhibition of CAIX/XII did irrespective of NFS1 levels. Moreover, as expected, mitochondrial iron increased following NFS1 depletion, inducing an iron starvation response increasing transferrin receptor levels, but blocking CAIX/XII activity had no further impact (Fig. 4D and fig. S3, F and G). To address whether the increased ROS was responsible for the increased cell death observed upon cotargeting CAIX and NFS1, we combined NFS1 knockdown with SLC-0111 in the presence or absence of the antioxidant Trolox (Fig. 4E). Similar to what we observed with Fer-1, treatment with Trolox suppressed the increased cell death accompanying this combination. These findings are in agreement with those previously published, revealing that NFS1 does not affect cellular ROS levels, yet NFS1-depleted cells were more sensitive to treatments inducing accumulation of ROS (30). Furthermore, treatment with the iron chelator deferoxamine (Fig. 4E) also significantly reduced the efficacy of this combination, suggesting that the increased ROS accompanying CAIX/XII inhibition creates a cellular vulnerability to the increased iron following NFS1 suppression, thereby triggering ferroptosis (Fig. 3, A to C). To test this directly, we treated SUM159PT cells with physiological concentrations of iron in the form of ferric citrate (Fig. 4F). Notably, ferric citrate alone did not affect cell viability. However, combining ferric citrate with CAIX/XII inhibition led to robust cell death at a concentration of SLC-0111 capable of increasing ROS levels. Furthermore, in addition to SLC-0111, combining compound 13 or compound 11 with ferric citrate led to significant increases in cell death that could be reversed by treatment with Fer-1 or Trolox in both SUM159PT and LM2-4 cells (Fig. 4, G and H).

We previously demonstrated reduced metastatic capacity of breast cancer cells with suppressed CAIX expression (14). Oxidative stress has been identified as a hurdle to systemic metastatic spread (39). Furthermore, ferroptosis was recently found to limit hematogenous metastatic spread of cells with high cellular ROS levels (40). Given the vulnerability to physiological iron concentrations following CAIX/XII inhibition, we addressed whether the increased ROS associated with CAIX depletion impeded the hematogenous metastatic capacity of these cells (Fig. 4, I to K). In agreement with our previous findings, depletion of CAIX affected the metastatic capability of 4T1 cells. The pretreatment of shCAIX cells with Trolox significantly reduced cellular ROS levels (Fig. 4C) and enhanced their metastatic capacity (Fig. 4, I to K). Collectively, the data suggest that CAIX protects cells from ferroptosis by limiting ROS production and cotargeting CAIX with chemical agents that perturb iron or antioxidant homeostasis and may be an effective therapeutic strategy (Fig. 4L).

We next sought to assess whether combining CAIX/XII inhibition with limiting cyst(e)ine availability, the substrate of NFS1, through inhibition of the cystine glutamate antiporter xCT, would lead to a similar enhancement of ferroptosis. To target xCT, cells were treated with the small molecule inhibitor erastin (20). Notably, at doses higher than 0.5 μM, erastin is lethal on its own in SUM159PT cells (fig. S4A). To identify whether CAIX/XII inhibition could enhance erastin-induced cell death, we treated cells with SLC-0111 and a concentration of erastin that did not induce cell death following a 72-hour treatment (Fig. 5A and fig. S4A). Notably, SLC-0111 also causes very little cell death on its own in this model. However, the combination treatment resulted in a significant increase in cell death that could be reversed in the presence of Fer-1, Trolox, and 2-mercaptoethanol (2-ME) but not inhibitors of necroptosis (Nec1s) or apoptosis (ZVAD-FMK) (Fig. 5, A to C), suggesting that ferroptosis is the primary mode of cell death induced by the combination and restoring cysteine uptake or buffering ROS alleviates this. Moreover, the SLC-0111–erastin combination resulted in marked accumulation of lipid peroxidation (Fig. 5D). To assess the range at which this combination was effective, we treated SUM159PT with multiple doses of SLC-0111 and erastin (Fig. 5E). CAIX/XII inhibition potentiated erastin-induced ferroptosis in a dose-dependent manner. We extended this finding to additional models of TNBC, pancreatic, and bladder cancer (Fig. 5, F and G, and fig. S4, B and C). In addition, this finding was not unique to the SLC-0111 compound, as cells treated with the same titration of erastin in the presence of compound 13 or compound 11 evoked similar dose-dependent increases in cell death (Fig. 5, H and I). Alternatively, to assess whether this effect was unique to erastin, we carried out similar combination experiments with SLC-0111 and an additional inhibitor of xCT, sulfasalazine (Fig. 5J) (41). As sulfasalazine is a much less potent inhibitor of xCT than erastin, much higher doses were needed before sulfasalazine caused any cell death on its own (Fig. 5J). Nevertheless, the addition of SLC-0111 was able to enhance sulfasalazine-induced cell death in a dose-dependent manner across multiple TNBC models (Fig. 5J and fig. S4, D and E). To further confirm the on-target nature of the CAIX/XII inhibitors, we depleted CAIX, CAXII, or both simultaneously and treated with erastin or sulfasalazine (fig. S4, F to H). In support of the findings from the pharmacological inhibition studies, suppressing CAIX or CAXII expression increased the ferroptosis resulting from xCT inhibition.

To evaluate the clinical relevance of cotargeting CAIX and xCT, we assessed the expression of CA9 and SLC7A11, the gene encoding xCT, across human cancers in the TCGA. Positive correlations were observed between the two genes in cohorts of patients with breast and pancreatic ductal adenocarcinoma (PDAC) (Fig. 5K). To cotarget CAIX/XII and xCT in a clinically relevant model, we treated CAIX expressing organoids derived from liver metastases of patients with metastatic PDAC with erastin and SLC-0111 (Fig. 5, L and M, and fig. S4I). Similar to our findings in cell line models, this combination led to substantive increases in cell death in organoids from multiple patients outperforming treatment with either compound alone (Fig. 5, L and M). Therefore, cotargeting CAIX/XII and xCT enhances ferroptosis and is an effective strategy to pharmacologically exploit the synthetic lethal interaction between CA9 and NFS1 (Fig. 5N).

To determine the mechanistic details surrounding the cell death occurring as a result of combined CAIX/XII and xCT inhibition, we first assessed the impact of CAIX activity on pH_i (Fig. 6A). Since CAIX is involved in regulation of pH_i through the production of bicarbonate (7), we assessed pH_i in hypoxic SUM159PT cells following treatment with SLC-0111, compound 11, and compound 13 (Fig. 6A). As expected, we identified that all three inhibitors led to the acidification of pH_i. Moreover, the bicarbonate produced by CAIX/XII can be imported by bicarbonate transporters, and CAIX has been found to associate with the sodium bicarbonate cotransporter NBCn1 (12). Therefore, we blocked sodium bicarbonate cotransporters with S0859 (42), reducing pH_i (Fig. 6B), and evaluated whether the bicarbonate generated was important in protecting cells from ferroptosis (Fig. 6, C and D, and fig. S4J). Treating hypoxic SUM159PT (Fig. 6C) and LM2-4 (Fig. 6D) cells with S0859 in combination with erastin induced a dose-dependent increase in cell death to levels similar to those achieved by cotargeting CAIX/XII and xCT, suggesting that CAIX/XII-driven pH_i regulation is critical for protection against ferroptosis. To further implicate pH in influencing ferroptosis downstream of cotargeting CAIX/XII or sodium bicarbonate cotransporters with xCT, we added NH₄Cl to the combinations to raise pH_i (Fig. 6, B, E, and F). Combining NH₄Cl with SLC-0111 and erastin (Fig. 6E) or with S0859 and erastin (Fig. 6F) reduced ferroptosis in a dose-dependent manner, providing additional support that acidic pH_i potentiates ferroptosis.

We next sought to identify the impact on cell signaling following treatment with SLC-0111 and erastin (Fig. 6G). CAIX has been shown to affect cellular metabolism (31–35), and adenosine 5′-monophosphate–activated protein kinase (AMPK) signaling is central to maintaining metabolic homeostasis. Activated in response to energy stress, ROS, and hypoxia (43), AMPK signaling has been implicated in protecting cells from ferroptosis (44). The combination of SLC-0111 and erastin at sublethal concentrations led to inactivation of AMPK following reduced phosphorylation of Thr¹⁷² (Fig. 6G). This inactivation was echoed by a reduction in phosphorylation of the AMPK target acetyl–coenzyme A (CoA) carboxylase 1 (ACC1) on Ser⁷⁹, suggesting that the combination was activating ACC1 (Fig. 6G). We then assessed whether the combination of S0859 and erastin also affected AMPK activity (Fig. 6H). Similar to SLC-0111 treatment, combination of S0859 and erastin decreased phosphorylation of AMPK on Thr¹⁷² (Fig. 6H). This reduction in AMPK phosphorylation was mirrored by a reduction in phosphorylation of ACC1 on Ser⁷⁹. To evaluate whether these changes in AMPK/ACC1 activation were relevant to the increased ferroptosis resulting from the SLC-0111–erastin and S0859-erastin combinations, we simultaneously activated AMPK with 5-aminoimidazole-4-carboxamide ribonucleotide (AICAR) (Fig. 6, I and J). Activation of AMPK reduced the enhanced cytotoxicity of the SLC-0111–erastin and S0859-erastin combinations in a dose-dependent manner. In parallel, we inhibited ACC1 with 5-(tetradecyloxy)-2-furoic acid (TOFA) (Fig. 6, I and J). Inhibition of ACC1 decreased cell death of both erastin combinations in a dose-dependent manner, suggesting that the signal transmitted from AMPK to ACC1 is biologically relevant to the ferroptosis induced by both the SLC-0111–erastin and S0859-erastin combinations. Furthermore, we inhibited acyl-CoA synthetase long-chain family member 4 (ACSL4), which is downstream of ACC1, with rosiglitazone (Fig. 6, I and J). ACSL4 expression is predictive of susceptibility to ferroptosis (45–47), and its activity is responsible for the conversion of polyunsaturated fatty acids into polyunsaturated fatty acyl-CoAs for incorporation into phospholipids (48, 49). We found that rosiglitazone treatment also reduced the cytotoxicity of the combinations (Fig. 6, I and J). To provide further confirmation of the importance of ACC1 in potentiating ferroptosis downstream of CAIX/XII or sodium bicarbonate cotransporter and xCT inhibition, we suppressed ACC1 expression with siRNA and treated cells with the SLC-0111–erastin and S0859-erastin combinations (Fig. 6, K and L, and fig. S4K). Reducing ACC1 expression partially suppressed the level of cytotoxicity induced by the SLC-0111–erastin (Fig. 6K) and S0859-erastin (Fig. 6L) combinations dependent on the extent of knockdown by the siRNAs. Thus, these cytotoxicity studies suggest that the observed reduction in AMPK activity following cotargeting CAIX/XII and xCT is responsible for the enhanced ferroptosis from this combination by activating ACC1. Together, these data suggest that pH_i modulation is important in regulating ferroptosis. Furthermore, our data also implicate the importance of the AMPK/ACC1/ACSL4 axis downstream of cotargeting hypoxia-induced CAIX and xCT (Fig. 6M).

The SUM159PT cell line has been previously described (63) and was a gift from S. Gorski (BC Cancer Research Institute). MDA-MB-231 LM2-4^luc+ cells were a gift from R. Kerbel (Sunnybrook Research Institute, Toronto) (64). 4T1^luc+ cells have been previously described (65). UC-16 cells were a gift from P. Black (Vancouver Prostate Centre). HT-29 cells were a gift from A. Minchinton (BC Cancer Research Institute). Human embryonic kidney (HEK) 293T cells were a gift from S. Aparicio (BC Cancer Research Institute). All cells lines were validated by short tandem repeat analysis (Genetica) and mycoplasma testing using the LookOut Mycoplasma test. All cell lines were maintained in a humidified incubator at 37°C in an atmosphere containing 5% CO₂. Cell lines used in hypoxia (1% O₂) were routinely monitored for hypoxia-induced up-regulation of CAIX.

Tumor tissue was collected from consented patients. The study was approved by the University of British Columbia research ethics board (REB) (H16-00291 and H18-03701). Patient-derived organoids were generated as previously described (66). Briefly, fresh liver metastatic biopsies from patients with PDAC were isolated and placed in RPMI 1640 medium supplemented with fetal bovine serum (FBS). The biopsy material was dissected and minced with scissors in the 10% FBS-supplemented RPMI 1640 medium. The resulting tissue chunks were washed, spun, and digested with collagenase/dispase (1 mg/ml). The collagenase/dispase was deactivated with 1% bovine serum albumin (BSA)–supplemented Dulbecco’s modified Eagle’s medium (DMEM); the tissue was washed and spun again and then digested with Accutase (500 U/ml). The digestion was again deactivated with 1% BSA-supplemented DMEM, and the resulting cell suspension was filtered through a 20-μm filter. The resulting cell suspension was spun and resuspended in complete organoid medium before being plated on 12-well plates coated with Matrigel. The organoids were grown at 37°C with 5% CO₂ and passaged every 2 weeks.

SUM159PT cells are cultured in DMEM containing 10% FBS, 1× nonessential amino acids, hydrocortisone (1 μg/ml), and insulin (5 μg/ml). The MDA-MB-231 LM2-4 and PK-8 cells were cultured in RPMI 1640 containing 10% FBS. 4T1 cells were cultured in DMEM containing 10% FBS and 1× nonessential amino acids. HT-29 cells were cultured in DMEM containing 10% FBS. UC-16 cells were cultured in minimum essential medium containing 10% FBS, 1× nonessential amino acids, and 1 mM sodium pyruvate.

Erastin (Selleckchem, S7242), Fer-1 (Cayman Chemical, 17729), Nec-1s (BioVision, 2263), Z-VAD-FMK (Selleckchem, S7023), sulfasalazine (Selleckchem, S1576), deferoxamine mesylate (Sigma-Aldrich, #D9533), Trolox (Sigma-Aldrich, #238813), ferric citrate (Sigma-Aldrich, #F3388), AICAR (Selleckchem, S1802), TOFA (Selleckchem, S6690), rosiglitazone (Selleckchem, S2556), S0859 (Cayman Chemicals, 18497), SLC-0111, compound 11, and compound 13 have been previously described (32, 38); RPA (Squarix Biotechnology, ME043), BODIPY 581/591 undecanoic acid (Thermo Fisher Scientific, #D3861), CM-H₂DCFDA (Thermo Fisher Scientific, #C6827), CellROX Deep Red (Thermo Fisher Scientific, #C10422), SYTOX Green nucleic acid stain (Thermo Fisher Scientific, #S7020), HCS NuclearMask red stain (Thermo Fisher Scientific, #H10326), nigericin (Sigma-Aldrich, #N7143), and NH₄Cl (Sigma-Aldrich, #A9434). Antibodies to AMPK (cs2532; 1:1000), AMPK^T172 (cs2535; 1:500), ACC (cs3662; 1:500), and ACC^S79 (cs3661; 1:500) were purchased from Cell Signaling Technology. Antibodies to CAIX were purchased from R&D Systems (AF2188; 1:1000) and BioScience Slovakia (M75; 1:1000). Antibodies to CA12 (ab195233; 1:1000) and ferrochelatase (ab137042; 1:1000) were purchased from Abcam. Antibodies to NFS1 (sc365308; 1:500) and CD71/TFRC (Transferrin receptor 1) (sc-32272; 1:500) were purchased from Santa Cruz Biotechnology. Antibody to vinculin (MAB3574; 1:1000) was purchased from Millipore.

The procedure for generation of the CA9^KO cell line is similar to what has been previously described for the generation of the MDA-MB-231 CA9^KO using CRISPR-Cas9, and the same procedures were followed with the SUM159PT cell line (12). Briefly, cells were transfected with CG01 plasmid (GeneCopoeia) expressing Cas9 and sgRNA to CA9 using Lipofectamine 2000 as per the manufacturer’s suggestions and then expanded for 5 days. Cells were sorted for red fluorescent protein expression by flow cytometry using an AriaFusion and then cloned out from single cells in 96-well plates. Western blotting for CAIX expression following 72 hours of incubation in 1% O₂ provided first pass filtering for positive KO clones. Genomic DNA was then extracted from these lines using QuickExtract solution (Lucigen, QE09050) and the CA9 locus around the sgRNA-targeted region amplified by polymerase chain reaction (PCR). Amplicons were sequenced by Sanger sequencing to screen for null mutations.

HEK293T cells were seeded at 5 × 10⁴ cells/cm² on 15-cm plates and transfected the following day with TransIT-LT1 (Mirus Bio, 2300) with 20 μg of human GeCKOv2 lentiCRISPRv2 KO pooled plasmid library A (a gift from F. Zhang; Addgene #1000000048), 10 μg of pVSVg, and 15 μg of psPAX2. After an overnight incubation, the medium was changed, and 30 ml of DMEM containing 10% FBS and 1% BSA (Sigma-Aldrich) was added to each 15-cm plate. After 60 hours, the medium was collected and centrifuged at 3000 rpm at 4°C for 5 min. The supernatant was then filtered through a 0.45-μm filter and concentrated 300× by centrifugation at 25,000 rpm at 4°C for 90 min.

A total of 7 × 10⁸ SUM15PT NTC or CA9^KO cells were transduced at a multiplicity of infection of 0.3, representing 300× coverage of the library, in SUM159PT medium containing polybrene (800 μg/ml). The following day, the medium was removed, and the plates washed 2× with phosphate-buffered saline and fresh SUM (SUM159T) medium added containing puromycin (7 μg/ml). After 48 hours, the medium was removed, and the cells washed and trypsinized. Two aliquots of 2.5 × 10⁷ cells were collected for T₀ samples. A total of 2 × 10⁷ cells were seeded onto T1000 HyCell multilayer flasks (Corning) in triplicate for culture in 1% O₂ and passaged for 20 generations. Samples of 2 × 10⁷ cells were collected every 72 hours. Genomic DNA was extracted using the QIAamp DNA Blood Kit (QIAGEN, #51194) as described by the manufacturer. Illumina adapters and barcodes were added to samples by PCR as previously described (67). Samples were sequenced on an Illumina HiSeq2500 by the Michael Smith Genome Sciences Centre, BC Cancer, Vancouver, Canada. Indexed reads were demultiplexed before analysis.

Pooled sgRNA screening data were analyzed as previously described (68). Before analysis, the read counts from the demultiplexed sequencing files were generated using the MAGeCK software package (69). To calculate the differential cumulative change (DCC) between NTC and CA9^KO, we normalized the read counts using cyclic loess normalization and sgRNAs with low read counts at the initial time point T₀ were removed. The screen was performed as an end point assay with two time points (T₀ and T₁₁). Accordingly, the following formula was usedwhere $x_{t11,\text{rep}}^{\text{KO}}$ is the normalized read counts at T₁₁ and $x_{t0,\text{rep}}^{\text{KO}}$ is the normalized read counts at T₀ in replicates rep ∈ (1..2) for KO-B4 cells. Likewise, $x_{t11,\text{rep}}^{\text{NTC}}$ and $x_{t0,\text{rep}}^{\text{NTC}}$ are for NTC cells. To measure the gene level DCC, we used the two best sgRNAs with the maximum dropout for that gene. We used the Student’s t test in conjunction with the permutation test P value to classify sgRNA and its corresponding genes that were substantially different between the CA9^KO and NTC cells. The permutation test was performed by estimating the frequency of randomized, shuffled DCC with maximum dropout compared to the observed gene level DCC value. Pathway enrichment analysis was performed on hits using the iDEP package (70). The gene network image was generated using Cytoscape software (71) for genes enriched in multiple pathways.

Comprehensive data analysis on gene expression from the TCGA database was performed by exploring the RNA sequencing expectation maximization–normalized expression data downloaded from https://portal.gdc.cancer.gov/ site. The normalized data were log-transformed (log₂) to bring them to same scale. Spearman rank correlation was used to compute the correlation between CA9 and NFS1 or SLC7A11 in data of patient with breast and pancreatic cancer. Further subsetting the TCGA BRCA (breast cancer) dataset according to molecular subtype was performed using the PAM50 classifier (72). Statistical significance between normal and tumor samples for 24 different cancer types for NFS1 expression analysis was calculated using Mann-Whitney U test.

Dox-induced shRNA to NFS1 (V2THS_67142; ATGTCATTGACATCAAGTG) using the TRIPZ system (Dharmacon) was introduced by lentivirus. siRNA to NFS1 (GS9054), CA9 (GS768), CA12 (GS771), and ACACA (GS31) were purchased from QIAGEN. FlexiTube GeneSolution sets of four siRNAs were screened for each gene to identify two efficacious siRNAs to use in our studies. siRNAs were reverse-transfected using siLentFect (Bio-Rad, 1703360) or Lipofectamine RNAiMAX (Thermo Fisher Scientific, 13778075) as per the manufacturer’s instructions. The following day, the cells were split into experiments and cultured for an additional 72 hours. Protein levels were routinely monitored by Western blot.

pH_i measurements were carried out as previously described (32). Briefly, cells were seeded into 96-well plates (3000 cells per well) in quadruplicates and then treated with SLC-0111, compound 13, compound 11, S0859, or NH₄Cl and incubated in hypoxia for 72 hours. The pH_i measurements were carried out using the Fluorometric Intracellular pH Assay Kit (Sigma-Aldrich, catalog no. MAK150) according to the manufacturer’s instructions. Briefly, growth medium was removed, and cells were loaded with 50 μl of BCFL-AM dye loading solution for 30 min at 37°C in hypoxia, followed by treatment with fresh inhibitor for 15 min. Ratiometric measurements were then carried out at λ_ex = 505, λ_em = 535 and λ_ex = 430, and λ_em = 535 using a SpectraMax i3x microplate reader (Molecular Devices). A calibration standard curve was prepared using pH calibration buffers (pH 6.0 to pH 8.0; increments of 0.5 pH units) containing 10 μM nigericin. Dye loading, drug treatment, and calibration were performed using Hanks’ balanced salt solution (HBSS) with 20 mM Hepes, without sodium bicarbonate, provided with the kit. A sigmoidal 4PL nonlinear regression model was used to fit the calibration curve and interpolate the experimental pH_i values.

Cells were lysed in radioimmunoprecipitation assay buffer and quantified by bicinchoninic acid assay. Five micrograms of protein was separated on a 4 to 12% bis-tris gradient gel in 1× Mops buffer or a 3 to 8% tris-acetate gel in 1× tris-acetate SDS buffer. Gels were transferred to polyvinylidene difluoride for 1 hour at 100 V in ice-cold tris-glycine transfer buffer containing 20% methanol. Membranes were fixed in methanol for 20 s and dried before incubation with primary antibodies overnight in 1× TBST (Tris-buffered saline containing 0.05% Tween-20) (pH 7.6) containing 2% BSA. Horseradish peroxidase (HRP)–conjugated, species-specific, secondary antibodies (mouse, Cell Signaling Technology, #cs7076; rabbit, Jackson ImmunoResearch, #111-035-003; and goat, Jackson ImmunoResearch, #805-035-180) were added the following morning for 1 hour at room temperature and detected using enhanced chemiluminescence on a ChemiDoc (Bio-Rad).

Cells were seeded at 1 × 10⁴ cells/cm² in six-well plates, treated, and incubated in 1% O₂ for 72 hours in respective growth medium. Following the 72-hour incubation, the cells were washed with 1× HBSS and stained with BODIPY 581/591 C11 (2 μM), CM-H₂DCFDA (0.5 μM), or CellROX Deep Red (5 μM) for 15 min at 37°C or with RPA (0.5 μM) for 15 min at 37°C, followed by a 30-min incubation in HBSS at the same temperature. Cells were then washed and trypsinized. Collected cells were washed with HBSS + 2% FBS, resuspended in the same buffer, and analyzed on a BD LSRFortessa and collected with BD FACSDiva. Fluorescence following excitation with the 488-nm laser (BODIPY 581/591 C11 and CM-H₂DCFDA), 568-nm laser (RPA), or 633-nm laser (CellROX Deep Red) was monitored in the blue 530/30 detector, yellow/green 610/20 detector, or red 670/14 detector, respectively. Data were analyzed in FlowJo v10 (FlowJo LLC). For BODIPY 581/591 C11 analyses, we report the percentage of the 10,000 events analyzed with increased green fluorescence (BODIPY C11_Ox) signal above control samples. For CM-H₂DCFDA, CellROX Deep Red, and RPA analyses, we report mean fluorescence intensity values for each respective fluorochrome.

IN Cell CAIX catalytic activity assays were carried out as described previously (12) with the following modifications: Assays were performed using SUM159PT cells following a 72-hour incubation in 1% O₂ and were carried out on ice. For each sample, 5 × 10⁵ cells were suspended in 100 μl of CO₂-free isotonic buffer [20 mM Hepes, 130 mM NaCl, and 5 mM KCl (pH 8.0)] in a 2-ml flat bottom vial equipped with an 8-mm magnetic stir bar. Inhibitors (SLC-0111, compound 11, and compound 13) were added to the cell suspension to a final concentration of 50 μM or an equivalent volume of dimethyl sulfoxide (DMSO) vehicle added, and the sample was incubated at room temperature for 30 min. An additional 700 μl of ice-cold buffer was then added to the cell-inhibitor mixture, and a narrow pH electrode (Accumet, catalog no. 13-620-850) was immersed in the sample and equilibrated for 3 min. A total of 200 μl of CO₂-saturated water was added to initiate the assay, and pH readings were recorded at 5-s intervals for up to 150 s. The change in pH was plotted as function of time. To determine the rate of spontaneous CO₂ hydration, measurements were performed on cell-free (“buffer”) samples. The increase in hydration rate above the spontaneous rate is a measure of carbonic anhydrase activity. Three replicates were assayed for each condition evaluated in the assay.

All studies were performed in accordance with University of British Columbia and BC Cancer Research Institute Animal Care Committee guidelines under approved animal ethics protocol no. A18-0132.

This study was supported by the BC Cancer Tumor Tissue Repository, Victoria, BC, a member of the Canadian Tissue Repository Network in accordance with REB approval under human ethics protocol no. H18-03701. A 48-specimen array was constructed with 0.6-mm cores, spotted in duplicate, containing 28 triple-negative cases and 20 high-grade ductal cases. Arrays were stained manually with antibodies to CAIX (M-75; 1 μg/ml; BioScience Slovakia, AB1001) or NFS1 (20 μg/ml; Santa Cruz Biotechnology, sc365308). Following deparaffinization in xylene and gradual rehydration in graded ethanol baths, antigen retrieval was performed by microwave on high for 10 min in citrate buffer. Antibodies were added overnight at 4°C, and ImmPRESS HRP-conjugated, universal, secondary antibodies were added the following morning according to the manufacturer’s instructions (Vector Labs, #MP-7500), followed by DAB (3,3′-Diaminobenzidine) detection (Vector Labs, #SK-4100). Arrays were scored by two independent observers (S. Chafe and O. Nemirovsky). CAIX scoring was performed using a presence or absence approach as previously described (14), and NFS1 scoring was binned into high (2/3) and low (0/1) samples.

Data were plotted using GraphPad Prism version 8.0. All statistical analyses were performed using GraphPad using a threshold of P < 0.05 for statistical significance.