Complex I protein NDUFB9 is a metabolic vulnerability in triple negative breast cancer brain metastases

Mingxi Lin; Zhexu Wen; Cheng Zeng; Yizi Jin; Teng Zhou; Yuxin Yan; Shenglin Huang; Xin Hu; Xiaoxiang Guan; Xichun Hu; Jian Zhang

doi:10.1038/s41467-026-72927-2

. 2026 May 9;17:6262. doi: 10.1038/s41467-026-72927-2

Complex I protein NDUFB9 is a metabolic vulnerability in triple negative breast cancer brain metastases

Mingxi Lin ^1,^2,^#, Zhexu Wen ^3,^#, Cheng Zeng ^1,², Yizi Jin ^1,², Teng Zhou ^1,², Yuxin Yan ^1,², Shenglin Huang ^2,^4,⁵, Xin Hu ^2,⁶, Xiaoxiang Guan ^7,^8,^✉, Xichun Hu ^1,^2,^✉, Jian Zhang ^1,^2,^9,^✉

PMCID: PMC13376752 PMID: 42103753

Abstract

Triple-negative breast cancer (TNBC) brain metastases (BrMs) remain a therapeutic challenge. We depict the discrepancies between primary tumors and BrMs, and examine patient-matched cerebrospinal fluid and plasma to provide detailed profiles of BrMs’ metabolic microenvironment. High-throughput in vivo loss of function CRISPR screens identify NDUFB9 (NADH: Ubiquinone Oxidoreductase Subunit B9) as a brain-specific metabolic vulnerability. NDUFB9-knockout selectively inhibits the BrMs outgrowth without affecting extracranial metastases. Mechanistically, TNBC cells exhibit an imbalance between aspartate upstream supply and downstream biosynthetic demand. NDUFB9-knockout disrupts mitochondrial complex I and reduces intracellular aspartate, but this alone is insufficient to inhibit TNBC proliferation. Instead, the lower asparagine concentration in the brain microenvironment induces compensatory upregulation of asparagine synthetase, which further diverts aspartate toward asparagine biosynthesis. This dual-hit mechanism exhausts the aspartate pool and restricts nucleotide biosynthesis, thereby selectively suppressing BrM outgrowth. Our findings uncover a therapeutic strategy for TNBC BrMs.

Subject terms: Breast cancer, Metastasis, Cancer metabolism

Triple-negative breast cancer brain metastases present a significant clinical challenge. Here, the authors show that targeting the complex I protein NDUFB9 selectively inhibits tumor growth in the brain.

Introduction

Breast cancer brain metastases (BrMs) remain a therapeutic challenge due to their increasing incidence and poor prognosis. Patients with HER2-positive or triple-negative breast cancer (TNBC) are at significantly higher risk of developing BrMs compared to those with the luminal subtype¹. Local treatments, such as stereotactic radiotherapy (SRT) and neurosurgery, are the mainstay of BrMs treatment². However, these approaches are limited in addressing imaging-invisible lesions, leading to high rates of intracranial progression. Additionally, over 70% of patients have multiple BrMs lesions, making them ineligible for SRT or neurosurgery³. In such cases, whole-brain radiotherapy (WBRT) remains the standard treatment option, though the median overall survival was only three to six months^3,4. In recent years, advancements in targeted therapies have improved outcomes for HER2-positive breast cancer BrMs. However, treatments for TNBC BrMs remain ineffective. The median PFS for chemotherapy of physician’s choice (TPC) was just 1.6 months⁵, and even with the anti-Trop2 antibody-drug conjugates (ADCs) sacituzumab govitecan (SG), median PFS was only 2.8 months⁵. Notably, approximately 40% of TNBC exhibit HER2-low expression⁶. For patients with HER2-low BrMs, trastuzumab deruxtecan (T-DXd) offers a treatment option, but median PFS remains limited at 4.1–5.4 months⁷. These outcomes underscore the urgent need for effective systemic therapies for TNBC BrMs.

The brain possesses a distinct microenvironment, characterized by unique anatomical structures, specialized cell types (such as neurons, microglia, astrocytes), and metabolic constraints. Previous studies have mainly focused on reciprocal interactions between BrMs and brain-resident cells. For example, astrocyte-derived exosomal microRNAs have been shown to induce PTEN loss, increase CCL2 secretion and promote BrMs growth, suggesting that targeting CCL2 could inhibit BrMs⁸. Similarly, polyunsaturated fatty acids secreted by astrocytes activated PPARγ signaling in tumor cells, facilitating BrMs progression, while PPARγ antagonists significantly reduced BrMs⁹. Additionally, the M2 polarization of microglia via the IL6/JAK2/STAT3 signaling pathway promoted the colonization of BrMs, indicating that blocking this pathway might offer therapeutic benefits¹⁰. Beyond cellular interactions, the brain’s metabolic microenvironment is uniquely challenging for tumors. The blood-brain barrier (BBB) and blood-tumor barrier (BTB) restrict the influx of certain metabolites, leading to a significant deficiency of neurotransmitters like glycine and glutamate in cerebrospinal fluid (CSF) compared to plasma^11,12. Moreover, the brain contains much lower levels of triacylglycerols (TAGs), the common storage form of long-chain fatty acids, than extracranial organs¹³. To survive and grow within this microenvironment, tumors might need to undergo specific metabolic adaptations that differ significantly from those required in extracranial organs. Targeting these adaptations might solve the treatment dilemma in TNBC BrMs. Despite these insights, the metabolic vulnerabilities of TNBC BrMs remain poorly understood.

In this work, we analyze discrepancy between TNBC primary tumors and paired BrMs, conduct a comprehensive analysis of the metabolic microenvironment of TNBC BrMs, and identify NDUFB9 as a brain-specific metabolic vulnerability through high-throughput in vivo CRISPR-Cas9 screenings.

Results

High-throughput in vivo loss-of-function CRISPR screens identify potential treatment targets of TNBC BrMs

To identify discrepancy between TNBC primary tumors and BrMs, we included five patients with TNBC brain oligometastases (no extracranial metastases). While challenging to find, brain oligometastases could better help to study the mechanism of brain colonization. Comprehensive clinical data, including treatment regimens and survival outcomes, were detailed in Supplementary Data 1. Archival formalin-fixed paraffin-embedded (FFPE) tissues of primary tumors and paired BrMs were collected and analyzed using RNA-Seq and mass spectrometry (MS) based label-free proteomics.

To systematically identify potential therapeutic targets of TNBC BrMs, we developed a high-throughput CRISPR-Cas9 loss-of-function library, referred to as the triple-negative breast cancer brain metastases CRISPR-Cas9 knockout library (TNBCBM-CK library). This CRISPR-Cas9 knockout library targeted 499 genes that were upregulated (fold change >1.5) in TNBC BrMs compared to primary tumors, as identified by both RNA-seq and proteomics (Fig.1A). In total, the library comprised 3420 sgRNAs, including 2988 sgRNAs targeting these 499 genes and 432 non-targeting sgRNAs as controls (Supplementary Data 2). We transduced the lentiviral TNBCBM-CK library into Cas9-expressing MDA-MB-231 cells. The cells were then subjected to intracranial injection (BrMs group), mammary fat pad injection (orthotopic tumors group), or in vitro culture (in vitro group) to impose distinct selection pressures (Fig.1A). The BrMs group and orthotopic tumors group each included five mice, with sufficient cells injected per mouse to maintain ~100-fold coverage of all sgRNAs in the library. After eighteen days, the tumors were harvested for sgRNA library sequencing and analyzed by Model-bases Analysis of Genome-wide CRISPR-Cas9 Knockout Robust Rank Aggregation (MAGeCK RRA) tool (Supplementary Data 3).

Fig. 1 — A Overview of the experimental workflow for the in vivo CRISPR screens using the TNBCBM-CK library. Created in BioRender. Zhang, J. (2026) https://BioRender.com/akzn9ib. B, C Genes in the library are ranked by negative selection scores from MAGeCK RRA analyses. The top ten ranked genes, associated with tumor growth in the brain (B) and mammary fat pad (C), are annotated. D Nine-quadrant plot comparing MAGeCK RRA scores for BrMs (x-axis) and orthotopic tumors (y-axis). Genes that specifically promote BrMs (lower right quadrant; red), genes specifically promoting orthotopic tumors (upper left quadrant; blue), and nonspecific genes affecting general tumor growth (upper right quadrant; black) are annotated.) The relative abundance changes of sgRNAs targeting NDUFB9, COX6A1, TOMM40, MRPL28, and SLC25A10 in the BrMs (E) or orthotopic tumors (F), compared to in vitro controls. The axis represents log₂fold change. The histogram displays the distribution of all sgRNAs. Red bars represent targeted sgRNAs and gray bars represent all other sgRNAs. MOI, multiplicity of infection. Source data are provided as a Source Data file.

We analyzed the sgRNAs that were relatively reduced in BrMs compared to in vitro cells, identifying key genes that promote tumor growth within the brain microenvironment (Fig.1B; Supplementary Fig. 1A). The top-ranked genes included NDUFB9, RAN, PTMA, PPP4C, and HMGCS1. Similarly, we examined sgRNAs relatively reduced in orthotopic tumors compared to in vitro cells, identifying genes that promote tumor growth within the mammary fat pad microenvironment. These genes included MTHFD2, TRMT112, CDC20, and ATP2B2 (Fig.1C). These findings highlight the distinct genetic dependencies of TNBC in different microenvironments.

To identify genes that specifically affecting TNBC BrMs, we analyzed the sgRNAs that were relatively reduced in BrMs compared to orthotopic tumors (Supplementary Fig. 1B). However, this approach has a limitation: some genes may rank highly not because they strongly impair BrM growth, but rather because their inhibitory effect on orthotopic tumor growth is even weaker, thereby creating an apparent difference between orthotopic tumors and BrMs. Such genes are not the targets we aimed to identify. Therefore, we conducted combined analyses (BrMs vs. in-vitro and orthotopic tumors vs. in-vitro) to identify gene targets that both effectively and specifically affecting TNBC BrMs (Fig.1D, lower right quadrant). Notably, the majority of these genes were mitochondria-related genes, including NDUFB9, SLC25A10, TOMM40, MRPL28, and COX6A1. To validate the robustness of the screening results, we analyzed the abundance distribution of sgRNAs corresponding to each identified gene (Fig.1E-F). Multiple sgRNAs targeting the same gene consistently showed depletion in BrMs while remaining relatively unchanged in orthotopic tumors. This behavioral consistency suggests that the findings are unlikely to result from PCR bias or random clonal deletion. These results highlight the potential importance of mitochondria-related genes in TNBC BrMs, with NDUFB9 emerging as the top-ranked candidate (Fig.1B) for further validation.

NDUFB9-knockout significantly inhibits TNBC BrMs

To evaluate the role of NDUFB9 in TNBC BrMs, we knocked out NDUFB9 in cell lines. Knockout efficiency was confirmed by Western blotting (Supplementary Fig. 1C). We intracardially injected control and NDUFB9-knockout MDA-MB-231 cells into mice. Three weeks post-injection, BrMs burden was evaluated. Mice injected with NDUFB9-knockout cells showed reduced BrMs burden by HE histology (Supplementary Fig. 1F) and bioluminescence imaging (BLI)—~100-fold lower than in the control group (Fig. 2A–C). Consistent with the findings in MDA-MB-231 cells, NDUFB9-knockout suppressed BrMs growth in HCC1806 (Supplementary Fig. 1I-L) and murine TNBC cell line 4T1 (Fig.2D-G). Add-back of NDUFB9 in knockout cells rescued their ability of BrMs (Supplementary Fig. 1I-L). These data suggest that NDUFB9-knockout inhibits TNBC BrMs.

Fig. 2 — A–C Schematic of the experimental design (A). BrMs tumor burden quantification by BLI at day 21 after intracardiac injection of MDA-MB-231 cells transduced with sgCtr, sgNDUFB9#1, or sgNDUFB9#2. n (# of mice) =7(sgCtr); 8(sgNDUFB9#1 and sgNDUFB9#2) (B). Representative BLI showing BrMs tumor burden in different groups (C). D–G Schematic diagram of the experimental design (D). Representative BLI showing whole-body (upper) and brain-specific (lower) tumor burden at day 18 after intracardiac injection of 4T1 cells transduced with sgCtr, sgNDUFB9#1, or sgNDUFB9#2 (E). Quantification by BLI to measure BrMs (F) and extracranial (G) tumor burden in different groups. n (# of mice) = 6 (sgCtr); 5 (sgNDUFB9#1 and sgNDUFB9#2). (**H-K**) Schematic of the experimental design to assess the post-colonization outgrowth of tumor cells in the brain (H). Representative BLI showing intracranial tumor burden in different groups (I). Kaplan-Meier survival analysis of mice bearing intracranial MDA-MB-231 cells transduced with sgCtr, sgNDUFB9#1, or sgNDUFB9#2 (J). Quantification of intracranial tumor burden by BLI in different groups. n (# of mice) = 8 for each group (K). P values were determined by the Kruskal-Wallis test with Benjamini-Hochberg multiple test correction (B, F), one-way ANOVA with Benjamini-Hochberg multiple test correction (G, K), or log-rank test (J). Data are presented as median with interquartile range (B, F, G) or means ± SEM (K). BLI, Bioluminescent imaging. Source data are provided as a Source Data file.

NDUFB9-knockout impairs TNBC post-colonization outgrowth in the brain

The brain metastatic cascade involves two key stages: (i) intravasation of cancer cells into the circulation, survival in the bloodstream, and extravasation across the BBB (“seeding stage”); and (ii) survival and proliferation of cancer cells in the brain microenvironment (“post-colonization outgrowth stage”)¹⁴. To assess whether NDUFB9-knockout affects the “post-colonization outgrowth stage”, we bypassed the “seeding stage” by injecting MDA-MB-231 cells directly into the brain parenchyma. Mice injected with NDUFB9-knockout cells exhibited reduced BrMs burden and extended survival compared to the control group (Fig.2H-K; supplementary Fig. 1H). These findings suggest that NDUFB9 is needed for the post-colonization outgrowth of TNBC cells in the brain microenvironment. Moreover, we generated cells expressing a doxycycline-inducible shRNA targeting NDUFB9 or a non-targeting control (Supplementary Fig. 1M). Doxycycline was administered either during days 1–6 (the “seeding stage”) or days 7–12 (the “post-colonization outgrowth stage”). The time point of seven day allows circulating tumor cells to cross the BBB and enter the brain parenchyma without post-colonization outgrowth occurs^15,16. Knockdown of NDUFB9 at days 7–12 reduced BrMs burden compared to control shRNA, and to NDUFB9 knockdown during days 1–6 (Supplementary Fig. 1M-Q), indicating that NDUFB9-knockout impairs TNBC post-colonization outgrowth in the brain.

NDUFB9-knockout specifically inhibits TNBC BrMs without affecting extracranial tumors

We next examined whether NDUFB9-knockout affected extracranial tumors. In MDA-MB-231 cells, NDUFB9-knockout did not inhibit the growth of orthotopic tumors (Supplementary Fig. 2A-D), lung metastases (Supplementary Fig. 2E-H, M), bone metastases (Supplementary Fig. 2I-L, N), and liver metastases (Supplementary Fig. 2O-Q). Similarly, NDUFB9-knockout did not suppress extracranial disease progression in 4T1 (Fig. 2D-G) and HCC1806 (Supplementary Fig. 1I-L) cells. These findings demonstrate that NDUFB9-knockout attenuates BrMs while leaving extracranial tumor growth unaffected.

NDUFB9-knockout disrupts respiratory chain complex I and reduces intracellular aspartate levels

NDUFB9 encodes one of the 30 supernumerary subunits of respiratory chain complex I, which do not directly participate in catalysis as core subunits. Therefore, it remains unclear whether NDUFB9 is involved in mitochondrial respiration and, if so, how it contributes to this process. We conducted metabolomic analyses (Supplementary Data 4), showing that tricarboxylic acid (TCA) cycle was the most significantly downregulated metabolic pathway in NDUFB9-knockout cells compared to controls (Fig. 3A). Metabolites associated with the TCA cycle, including succinate, malate, and α-KG were significantly decreased (Fig. 3B). Aspartate levels, which depend on a functional electron transport chain (ETC) for biosynthesis, were reduced (Fig. 3B).

We also evaluated the normalized mitochondrial membrane potential (calculated as the ratio of mitochondrial membrane potential to mitochondrial mass). NDUFB9-knockout cells exhibited impaired mitochondrial polarization compared to controls (Fig.3C-D; Supplementary Fig. 3A-B). We next measured the oxygen consumption rate (OCR), an indicator of aerobic respiration, and observed that NDUFB9-knockout significantly reduced basal OCR, ATP-coupled OCR, and maximum OCR (Fig.3E-F; Supplementary Fig. 3C-D). Additionally, NDUFB9-knockout induced a metabolic switch, characterized by an increase in glycolysis ATP production rate and a decrease in mitochondrial ATP production rate, while maintaining the total ATP production rate (Fig.3G). Morphological analysis of mitochondria in NDUFB9-knockout cells revealed that the mitochondria appeared more circular, and the number of cristae (deep invaginations critical for mitochondrial respiration) was significantly decreased, indicating the defects in the mitochondria^17,18 (Fig.3H-O). These findings confirm that NDUFB9-knockout results in ETC dysfunction and aspartate depletion without affecting overall ATP production.

Addition of TCA cycle intermediates failed to restore proliferation (Supplementary Fig. 3F), whereas aspartate supplementation restored the growth of NDUFB9-knockout cells (Supplementary Fig. 3F). Blue-native polyacrylamide gel electrophoresis (BN-PAGE) analyses confirmed that NDUFB9-knockout reduced the abundance of complex I (Supplementary Fig. 3G). Complex I performs two essential functions: it catalyzes NAD⁺ regeneration and drives proton translocation across the inner mitochondrial membrane, thereby supporting ATP synthesis. Although NDUFB9-knockout reduced mitochondrial ATP production, total ATP production rate remained unchanged (Fig.3G). NDUFB9-knockout significantly reduced NAD⁺/NADH ratio (Supplementary Fig. 3I) and NAD⁺ is needed for TCA cycle and aspartate de novo biosynthesis. Therefore, we hypothesized that NDUFB9-knockout impair TCA cycle and aspartate biosynthesis by impairing NAD⁺ regeneration. NADH oxidase from Lactobacillus brevis (LbNOX) oxidizes NADH without reducing coenzyme Q, thereby decoupling NAD⁺ regeneration from proton-pumping function. We expressed LbNOX targeted specifically to either the mitochondria (mito-LbNOX) or cytosol (cyto-LbNOX)¹⁹ (Supplementary Fig. 3J). Mito-LbNOX increased NAD⁺/NADH ratio and significantly restored the intracellular aspartate level (Supplementary Fig. 3K-L). These results indicate that the decrease in aspartate observed after NDUFB9-knockout is, at least in part, due to impaired NAD⁺ regeneration within mitochondria. In the above section, we confirmed the function of NDUFB9. However, it remains to be determined why NDUFB9-knockout selectively impairs TNBC BrMs, and whether NDUFB9 possesses a unique role compared with other subunits of mitochondrial complex I.

Comprehensively metabolic microenvironment atlas of TNBC brain metastases

To understand why NDUFB9-knockout specifically attenuates TNBC BrMs, we hypothesized that NDUFB9-knockout cells might fail to adapt to the unique astrocyte- or microglia-rich brain microenvironment. However, co-culture experiments with astrocytes or microglia did not affect the proliferation of NDUFB9-knockout cells (Supplementary Fig. 4A-C). We then hypothesized that NDUFB9-knockout cells might fail to adapt to the metabolic microenvironment of the brain. To comprehensively depict the BrMs metabolic microenvironment, we collected the matched CSF and plasma from three patients with TNBC BrMs (Fig.4A). CSF and plasma served as proxies for the interstitial fluid (ISF) of BrMs and primary tumors, respectively^20,21. We identified and quantified 100 common metabolites (Supplementary Data 5). Metabolites such as serine (P = 0.023), proline (P = 0.020), asparagine (P = 0.042), aspartate (P = 0.006), 3-Methyl-2-oxovaleric acid (P = 0.009), and L-carnitine (P < 0.001) were significantly lower in the CSF, while inositol (P = 0.032) was more abundant.

Fig. 4 — A Schematic of sample collection and experimental procedures (upper). Heatmap showing metabolite levels (Z-score) in paired CSF and plasma collected from three patients with TNBC BrMs (lower). Metabolite classes are grouped as indicated on the right of the heatmap. B, C Cell proliferation of MDA-MB-231 (B) and HCC1806 (C) cells expressing sgCtr, sgNDUFB9#1, or sgNDUFB9#2. Cells were cultured in plasma-like medium, CSF-like medium, or CSF-like medium individually supplemented with each indicated metabolite to match its plasma concentration. n (# of independent experiments) = 4 for each group. D Asparagine concentrations in paired CSF and plasma from three patients with TNBC BrMs. E, F Cell proliferation of sgCtr, sgNDUFB9#1, and sgNDUFB9#2 MDA-MB-231 (E) and HCC1806 (F) cells cultured in CSF-like medium (6 μM ASN) or CSF-like medium supplemented with asparagine to match its plasma concentration (50 μM ASN). n (# of independent experiments) =4 for each group. P values were determined by one-way ANOVA with Benjamini-Hochberg multiple test correction (B, C, E, F), or two-sided paired t-test (D). Data are presented as means ± SEM. ASN, asparagine; Val, valine; Phe, phenylalanine; Lys, lysine; Leu, leucine; Trp, tryptophan; Gly, glycine; Pro, proline; Ser, serine; Asp, aspartate; Asn, asparagine; Glu, glutamate; Pyr, pyruvate; Glc, glucose; Gln, glutamine; CSF, cerebrospinal fluid. Source data are provided as a Source Data file.

NDUFB9-knockout selectively impairs TNBC BrMs due to intracranial–extracranial differences in asparagine rather than aspartate availability

We then developed two culture media—“CSF-like” and “plasma-like”—that recapitulate the concentrations of glucose, amino acids, and other small-molecule metabolites found in human CSF and plasma (Supplementary Data 6). In line with our in vivo results, NDUFB9-knockout significantly inhibited cell proliferation in CSF-like media but not in plasma-like media (Fig.4B-C). To pinpoint the metabolite responsible, we individually supplemented CSF-like media to match each metabolite’s plasma concentration. Because extracellular aspartate has poor cell permeability, modest changes in extracellular concentration did not substantially alter intracellular aspartate levels (Supplementary Fig. 3E). Supplementing CSF-like medium with aspartate to match the 18 μM concentration in plasma-like medium did not rescue cell proliferation (Fig. 4B–C), possibly due to the limited cell permeability of aspartate.

We found that only when supplementing CSF–like medium with asparagine, NDUFB9-knockout did not inhibit cell proliferation (Fig.4B-C). These findings indicate that NDUFB9-knockout impairs TNBC outgrowth specifically in the asparagine-limited brain microenvironment. In patients with TNBC BrMs, CSF asparagine levels were approximately 6 μmol/L, which was 8-fold lower than matched plasma levels (50 μmol/L; Fig.4D). Similarly, in mice with TNBC BrMs, CSF asparagine levels were significantly lower (7 μmol/L) compared to plasma levels (53 μmol/L; Supplementary Fig. 4D). We then generated asparagine-limited media (CSF-like medium with 6 μM ASN) and asparagine-abundant media (CSF-like media supplemented with 50 μM asparagine to match its plasma concentration). Similarly, NDUFB9-knockout impairs TNBC proliferation specifically in the asparagine-limited (6 μM ASN) media (Fig.4E-F). Asparagine is cell-permeable²². However, human cells lack the enzyme necessary to convert asparagine into aspartate²², ruling out the possibility that extracellular asparagine directly rescues the proliferation of NDUFB9-knockout cells by increasing aspartate pool. Therefore, why NDUFB9-knockout suppresses proliferation under asparagine-limited media remains unclear.

The upregulation of asparagine synthetase induced by asparagine-limited brain environment, coupled with aspartate reduction induced by NDUFB9-knockout, synergistically restrains nucleotide biosynthesis

Previous studies have shown that asparagine sustains mTOR complex I (mTORC1) activity, which promotes cell proliferation by driving anabolic processes needed for cell division²³. We hypothesized that low extracellular asparagine might fail to maintain the mTORC1 activity required for NDUFB9-knockout cells. However, phosphorylation of S6K on Thr389, a marker of mTORC1 activity, was similar in NDUFB9-knockout cells cultured in asparagine-limited (6 μM ASN) or asparagine-abundant (50 μM ASN) media (Supplementary Fig. 5A). This indicates that reduced mTORC1 activity is not the cause of the proliferation inhibition observed in NDUFB9-knockout cells under asparagine-limited conditions.

Activating transcription factor 4 (ATF4) plays a vital role in maintaining amino acid homeostasis under nutrient-depleted conditions by activating the transcription of genes involved in amino acid synthesis and transport, such as asparagine synthetase (ASNS) and serine hydroxymethyltransferase 2 (SHMT2)^24,25. However, chronic ATF4 activation can lead to apoptosis²⁶. In NDUFB9-knockout cells, ATF4 and its target genes (ASNS and SHMT2) were upregulated in asparagine-limited (6 μM ASN) media but not in asparagine-abundant (50 μM ASN) media (Fig.5A). We hypothesized that activation of ATF4 and/or its downstream target genes might contribute to the proliferation inhibition observed in NDUFB9-knockout cells under asparagine-limited conditions. To test this, we generated ATF4-silenced cells. ATF4 knockdown restored cell proliferation under asparagine-limited (6 μM ASN) conditions in the NDUFB9-knockout cells (Supplementary Fig. 5B-C). However, ATF4 is the transcriptional activator of ASNS and SHMT2 and its silencing concomitantly reduces ASNS and SHMT2 expression (Supplementary Fig. 5B), thus we cannot distinguish whether the rescue of proliferation is due to suppression of ATF4 itself or its target genes. Then we generated cells with silenced ATF4 and overexpressed ASNS to decouple ASNS expression from ATF4 activity. Under asparagine-limited conditions with NDUFB9-knockout, silencing ATF4 and its target genes (except for ASNS) failed to restore cell proliferation, indicating that ATF4 and its target genes (except for ASNS) were not responsible for the growth inhibition (Fig.5B-C). ASNS overexpression alone did not significantly rescue proliferation rate (Supplementary Fig. 5D-E). However, silencing ASNS alone rescued the proliferation inhibition, pinpointing ASNS as the key factor mediating growth inhibition in NDUFB9-knockout cells under asparagine-limited (6 μM ASN) conditions (Fig.5D-E).

Fig. 5 — A Western blot analysis of ATF4, ASNS, and SHMT2 protein levels in MDA-MB-231 (left) and HCC1806 (right) cells transduced with sgCtr, sgNDUFB9#1, or sgNDUFB9#2 and grown in CSF-like medium (6 μM ASN) or CSF-like medium supplemented with asparagine to match its plasma concentration (50 μM ASN). B Western blot of ATF4, ASNS, and SHMT2 protein level in control (siCtr and lenti-vec Ctr) or ATF4-silencing-ASNS-overexpressing (siATF4 and ASNS o/e) MDA-MB-231(left) and HCC1806(right) cells, with or without NDUFB9-knockout. C Relative proliferation rates of cells shown in (B). n (# of independent experiments) = 4 for each group. D Western blot of ATF4, ASNS, and SHMT2 protein levels in control (shCtr) or ASNS-silencing (shASNS) MDA-MB-231 (left) and HCC1806 (right) cells, with or without NDUFB9-knockout. E Relative proliferation rates of cells shown in (D). n (# of independent experiments) = 4 for each group. F Schematic representation of aspartate metabolism, illustrating upstream (gray) and downstream (pink) pathways. G, I Relative abundance of TCA cycle metabolites, aspartate, asparagine (G), and nucleotides (I) in sgCtr and sgNDUFB9 MDA-MB-231 cells grown in CSF-like medium (6 μM ASN) or CSF-like medium supplemented with asparagine to match its plasma concentration (50 μM ASN). n (# of independent experiments) = 3 for each group. H Relative proliferation rates of cells expressing sgCtr, sgNDUFB9, or sgNDUFB9 with SLC1A3 overexpression. n (# of independent experiments) = 4 for each group. The samples derive from the same experiment but different gels for ATF4, ASNS, another for NDUFB9, SHMT2 were processed in parallel (A, D). The samples derive from the same experiment but different gels for ASNS, NDUFB9, another for SHMT2, ATF4 were processed in parallel (B). P values were determined by one-way ANOVA with Benjamini-Hochberg multiple test correction (C, E, H) or two-sided unpaired t-test (G, I). Data are presented as means ± SEM (C, E, G–I). Representative immunoblots from three independent experiments (A, B, D). CSF, cerebrospinal fluid; ASN, asparagine; o/e, overexpression; TCA, tricarboxylic acid; OAA, oxaloacetate. Source data are provided as a Source Data file.

We hypothesized that NDUFB9-knockout selectively suppresses TNBC growth in an asparagine-limited brain environment because two processes act synergistically: (i) NDUFB9-knockout impairs the TCA cycle and reduces intracellular aspartate, and (ii) the intracranial low-asparagine condition induces compensatory upregulation of ASNS, which further diverts aspartate toward asparagine biosynthesis. Together, these effects deplete aspartate pools and restrict nucleotide biosynthesis. Consistent with this model, metabolomic analysis showed that NDUFB9-knockout reduced TCA intermediates (e.g., malate and fumarate) and aspartate regardless of extracellular asparagine levels (Fig. 5G). However, in asparagine-abundant medium (50 μM ASN), cells could compensate by importing asparagine (Supplementary Fig. 5F–G). By contrast, in asparagine-limited medium (6 μM ASN), exogenous uptake was restricted (Supplementary Fig. 5F–G) and ASNS was upregulated (Fig. 5A), resulting in nucleotide pool reduction (Fig. 5I). Supporting this mechanism, U-[¹³C]-glucose tracing demonstrated that silencing ASNS increased m + 5 to m + 9 labeling in UTP, CTP, ATP, and GTP, indicating that silenced ASNS attenuated the asparagine synthesis pull on aspartate pools and enhanced de novo nucleotide synthesis (Supplementary Fig. 5H). Furthermore, overexpression of the aspartate transporter SLC1A3 rescued the proliferation defect of NDUFB9-knockout cells under asparagine-limited conditions (Fig. 5H; Supplementary Fig. 1D, I–L).

These results indicate that the selective suppression of TNBC outgrowth by NDUFB9-knockout in the brain microenvironment is not driven by differences in intracranial versus extracranial microenvironment aspartate concentrations, but rather by differences in microenvironment asparagine availability. Because aspartate has poor cell permeability, modest variations in extracellular concentration are insufficient to substantially influence intracellular levels. By contrast, asparagine is cell-permeable. Under asparagine-limited condition, feedback upregulation of ASNS diverts aspartate toward asparagine biosynthesis, acting as first hit. Simultaneously, NDUFB9-knockout reduces intracellular aspartate production, acting as second hit. This dual-hit mechanism depletes the intracellular aspartate pool, leading to insufficient substrates for nucleotide biosynthesis and inhibition of TNBC BrMs.

Structural analyses reveal differences between NDUFB9 and other complex I subunits

Subsequently, we investigated whether knockout of other subunits of mitochondrial complex I could also selectively suppress TNBC BrMs. We knocked out additional supernumerary subunits of complex I that were upregulated in TNBC BrMs (NDUFB11, NDUFB9, NDUFB10, NDUFA6, NDUFS5, NDUFAB1, NDUFB7, NDUFA8, NDUFA2, NDUFB3). NDUFB11-knockout inhibited cell proliferation under both asparagine-limited (6 μM ASN) and asparagine-abundant (50 μM ASN) conditions, whereas NDUFB9-knockout selectively inhibited proliferation under asparagine-limited condition (Supplementary Fig. 6A). Knockout of other genes had no impact on proliferation under either asparagine-limited (6 μM ASN) or asparagine-abundant (50 μM ASN) conditions (Supplementary Fig. 6A). Consistently, NDUFB11-knockout significantly reduced both intracranial and extracranial tumor growth (Supplementary Fig. 6B-E), whereas NDUFB9-knockout specifically inhibited BrMs (Fig.2, Supplementary Fig. 2).

Next, we investigated why NDUFB9-knockout selectively suppresses TNBC in the asparagine-limited environment, whereas other genes do not exhibit such specificity. NDUFB9-knockout partially impaired complex I (Supplementary Fig. 3G), and caused a relatively modest reduction in the NAD⁺/NADH ratio and intracellular aspartate levels (Supplementary Fig. 6F–G). These changes alone did not inhibit proliferation in asparagine-abundant conditions (Supplementary Fig. 6A). However, in asparagine-limited conditions, compensatory ASNS upregulation diverted additional aspartate toward asparagine biosynthesis, thereby synergistically restraining nucleotide biosynthesis and inhibiting proliferation (Fig. 5). In contrast, NDUFB11-knockout completely abolished complex I (Supplementary Fig. 3G), causing a much more severe depletion of NAD⁺/NADH and aspartate than NDUFB9-knockout (Supplementary Fig. 6F–G). NDUFB11-knockout inhibited proliferation under both asparagine-limited (6 μM ASN) and asparagine-abundant (50 μM ASN) conditions (Supplementary Fig. 6 A). Knockout of other subunits had little effect on the NAD⁺/NADH ratio or aspartate levels (Supplementary Fig. 6F–G). Together, these findings explain why only NDUFB9-knockout selectively suppresses TNBC under the asparagine-limited brain microenvironment.

Then, we performed structural analyses to further investigate why different supernumerary subunits exert distinct effects on complex I. The ETC complexes consist of complex I (CI), CII, CIII, CIV, and CV, with most CI forming supercomplexes with CIII and CIV^27,28. Structural analyses (PDB:5XTH) suggested that NDUFB9 may facilitate interactions between complex I and complex III within supercomplex (Supplementary Fig. 6I). Consistent with the structural analyses, NDUFB9-knockout reduced the abundance of supercomplexes (Supplementary Fig. 6K). Moreover, in NDUFB9-knockout cells, we re-expressed either NDUFB9 or an NDUFB9^H12A mutant, in which a histidine predicted to contact complex III was substituted with alanine. We found that re-expression of NDUFB9 increased supercomplex abundance, whereas expression of the NDUFB9^H12A mutant did not (Supplementary Fig. 6L). These results are consistent with a contributory role for NDUFB9 in promoting CI–CIII interactions within supercomplexes. Since supercomplexes have been proposed to regulate the stability of individual complexes and protect them from proteolytic degradation²⁹, we hypothesized that NDUFB9-knockout may first disrupt supercomplexes and the resulting unassembled complex I is unstable and degraded by mitochondrial proteases. Consistent with this hypothesis, NDUFB9-knockout reduced the abundance of complex I (Supplementary Fig. 3G). Knockdown of mitochondrial protease LONP1 restored complex I abundance in NDUFB9-knockout cells (Supplementary Fig. 3H). Thus, the reduction in complex I abundance observed following NDUFB9-knockout is likely, at least in part, due to proteolytic degradation. Structural analyses (PDB:5XTD) indicate that NDUFB11 forms a strong interface with core subunits of complex I (ΔiG <–20 kcal/mol³⁰; Supplementary Fig. 6J). NDUFB11-knockout completely abolished complex I and the reduction in complex I abundance could not be rescued by LONP1 depletion (Supplementary Fig. 3G-H). Thus, the complete reduction in complex I abundance observed following NDUFB11-knockout is probably caused by direct impairment of complex I. NDUFB10, NDUFA6, NDUFS5, NDUFAB1, NDUFB7, NDUFA8, NDUFA2, and NDUFB3 neither form strong interface with core subunits, nor mediate interaction between complex I and complex III (Supplementary Fig. 6I-J). Together, these data indicate that structural differences among supernumerary subunits might underlie their divergent impacts on complex I.

TNBC BrMs increase aspartate biosynthesis in response to asparagine-limited microenvironment

We then investigated whether TNBC BrMs increase aspartate biosynthesis in response to asparagine-limited microenvironment. Since aspartate is a biosynthetic precursor generated by mitochondrial respiration, we first evaluated whether TNBC cells increase mitochondrial respiration under asparagine-limited microenvironment. We isolated the single-cell cloning (SCC) of MDA-MB-231 and HCC1806 cells to investigate metabolic adaptations to asparagine-limited conditions (Supplementary Fig. 7A). Genes encoding respiratory chain proteins, including those in CI (NDUFB8, NDUFB9), CIII (UQCRC2), CIV (MTCO2), and CV (ATP5A1), were upregulated when SCCs were cultured in asparagine-limited (6 μM ASN) media compared to asparagine-abundant (50 μM ASN) media (Supplementary Fig. 7B-C). The exception was CII (SDHB), whose expression was not significantly affected by the extracellular asparagine concentration (Supplementary Fig. 7B-C). Functional analysis revealed that normalized mitochondrial membrane potential (Supplementary Fig. 7D-G), basal and maximum OCR (Supplementary Fig. 7H-K) were upregulated under low extracellular asparagine conditions. These findings indicate that TNBC cells increased mitochondrial respiration in the asparagine-limited conditions.

We then analyzed FFPE samples of surgically collected primary breast tumors and paired brain oligometastases from five TNBC patients. Tumor tissues were harvested via laser capture microdissection and subjected to proteomics and transcriptomic analyses (Supplementary Data 7-8). The results showed the upregulation of key genes associated with the TCA cycle and ETC in TNBC BrMs (Fig. 6A). Pairwise gene set enrichment analysis (GSEA) of Molecular Signatures Database (MSigDB) Hallmark pathways confirmed that TNBC BrMs significantly upregulated oxidative phosphorylation (OXPHOS), E2F targets, and MTORC1 signaling pathways, while downregulating the epithelial-mesenchymal transition (EMT) pathway at both RNA and protein levels (Fig.6B). IHC analyses confirmed higher expression of NDUFB9 in TNBC BrMs compared to patient-paired primary tumor (n = 10, Fig. 6C-D). Additionally, NDUFB9 expression was significantly elevated in BrMs with TNBC subtypes (n = 10) compared to patient-unpaired extracranial metastases with TNBC subtypes, including liver metastases (n = 8) and lung metastases (n = 10) (Fig. 6E-F). External validation was performed using dataset from Varešlija et al³¹. TNBC BrMs displayed significantly higher gene set variation analysis (GSVA) scores compared to patient-matched primary tumors (Fig.6J), further supporting our findings.

Fig. 6 — A Differential gene analyses between TNBC primary tumors and paired brain oligometastases using proteomics and transcriptomic. n = 5. B Scatter plot of transcriptomic (Y-axis) versus proteomics (X-axis) -log₁₀(FDR) from GSEA analyses. Pathways upregulated (upper right) or downregulated (lower left) in BrMs at both RNA and protein levels are annotated. C, D Representative IHC staining of NDUFB9 in patient-paired TNBC primary tumor and BrMs (C). Quantification by IRS (D). Scale bars, 20 μm. n = 10. E, F Representative IHC staining of NDUFB9 in patient-unpaired brain, liver, and lung metastases all derived from TNBC patients (E). Quantification by IRS (F). Scale bars, 20μm. n = 10(brain mets); 8(liver mets); 10 (lung mets). G–I Experimental process to isolate GFP-positive MDA-MB-231 cells from primary tumors and BrMs by FACS to remove stromal contamination, three weeks after intracardiac or intramammary injection of tumor cells (G). KEGG pathways upregulated or downregulated in BrMs (H). Heatmap of gene expression levels (Z-score) of ETC complexes in primary tumors versus BrMs (I). n = 3. J Comparison of TNBC BrMs and primary tumors using an external clinical dataset. Upper: scatter plot showing the GSVA enrichment scores of the OXPHOS pathway in each sample. Lower: heatmap showing expression levels (Z-score) of OXPHOS genes in primary tumors versus BrMs. n = 5. K–N Schematic of the experimental design (K). Schematic of central carbon metabolism and aspartate generation (L). In vivo metabolite labeling from U-[¹³C]-glucose (M) or U-[¹³C]-glutamine (N) in tumors derived from MDA-MB-231 cells implanted in mammary fat pad or brain parenchyma. n = 3. Created in BioRender. Zhang, J. (2026) https://BioRender.com/akzn9ib. (A, G, K). P values were calculated using one-way ANOVA with Benjamini-Hochberg multiple test correction(F), two-sided paired t-test (D), or unpaired t-test with Benjamini-Hochberg multiple test correction (M, N). Data are presented as means ± SEM (M, N). Box plots show median (central line), upper and lower quartiles (box limits), and minima to maxima (whiskers) (F). GSEA, pairwise gene set enrichment analysis; FDR, false discovery rate; GSVA, gene set variation analysis; TCA, tricarboxylic acid; IHC, immunohistochemistry; IRS, immunoreactive score; CAFs, cancer-associated fibroblasts; Mets, metastases; FACS, fluorescence-activated cell sorting. Source data are provided as a Source Data file.

To eliminate the potential confounding effect of stroma contamination effect when comparing primary tumor and BrMs, pure GFP-positive cancer cells were isolated using fluorescence-activated cell sorting (FACS), and their corresponding transcriptomic profiles were quantified (Fig.6G). This analysis revealed that the respiratory chain proteins was significantly upregulated in isolated brain metastatic cells compared to cancer cells from primary tumors (Fig.6H-I). Further supporting these findings, electron microscopy analysis, which provided sufficient resolution to examine cancer cells alone without stromal interference, showed that mitochondria in BrMs were slimmer and had an increased number of cristae per mitochondrion compared to primary tumors (Supplementary Fig. 7L-Q). These results suggest that the enhanced mitochondrial respiration observed in TNBC BrMs is an intrinsic biological adaptation of cancer cells, rather than an artifact caused by stroma contamination from the brain microenvironment.

To further investigate whether TNBC BrMs increase aspartate production through enhanced mitochondrial respiration, we performed in vivo U-[¹³C]-glutamine and U-[¹³C]-glucose isotope tracing analyses. Mice bearing BrMs or orthotopic breast tumors were infused with U-[¹³C]-glucose or U-[¹³C]-glutamine. BrMs and primary tumors were then collected under the guidance of ex vivo BLI and subjected to metabolomics tests (Fig. 6K). Both BrMs and primary tumors exhibited similar labeling patterns in glycolytic intermediates derived from U-[¹³C]-glucose (Fig. 6L-M). However, BrMs showed a significant increase in labeling of the TCA cycle intermediates, including succinate, fumarate, and malate, indicating upregulated mitochondrial respiration in TNBC BrMs (Fig. 6L-M). Importantly, aspartate, the biosynthetic precursor generated by mitochondrial respiration, was significantly increased (Fig. 6L-M). This pattern indicates that a greater fraction of glycolysis-derived pyruvate is oxidized and funneled into the TCA cycle and driving enhanced aspartate synthesis. In contrast, labeling of oxidative TCA cycle intermediates (m + 4 labeling³²) and reductive TCA cycle intermediates (m + 3 labeling³²) from U-[¹³C]-glutamine were similar in BrMs and primary tumors (Fig. 6L, N). These results suggest that TNBC BrMs display increased production of aspartate generated through glucose-derived mitochondrial respiration.

TNBC exhibits high sensitivity to NDUFB9-knockout due to the imbalance between aspartate upstream substrate supply and downstream biosynthesis demand

We then investigated whether the NDUFB9-knockout could also suppress BrMs in other breast cancer subtypes beyond TNBC. We found that, under asparagine-limited conditions, NDUFB9-knockdown suppressed the proliferation of TNBC cell lines (MDA-MB-231, HCC1806, SUM159PT), while having no significant effect on luminal (T47D and MCF7) or HER2-positive (HCC1954 and SKBR3) cell lines (Supplementary Fig. 8A).

We next investigated why TNBC are more sensitive than luminal and HER2-positive breast cancers. Basal-like is the intrinsic subtype defined by PAM50 profiling, which shows a high degree of overlap with TNBC as determined by IHC. Proteomic analyses of the CPTAC³³ and CBCGA³⁴ datasets revealed that the OXPHOS pathway, particularly genes encoding respiratory chain complex I proteins, was downregulated in basal-like breast tumors compared with other subtypes (Fig.7A-C). Moreover, aspartate aminotransferase (GOT2), which facilitates the mitochondrial synthesis of aspartate from oxaloacetate³⁵, was also downregulated in basal-like tumors (Fig. 7D). However, asparagine, purine nucleotide, pyrimidine nucleotide, and protein biosynthesis, which use aspartate as the direct substrate, were upregulated in basal-like tumors (Fig.7C-D). These findings were further validated using the Cancer Cell Line Encyclopedia (CCLE) dataset³⁶, which showed that mitochondrial respiration pathway and GOT2 (upstream substrate supply) was downregulated, while asparagine, protein, and nucleotide biosynthesis pathway (downstream biosynthesis demand) were upregulated in TNBC compared with other breast cancer subtypes (Fig.7E). This imbalance between aspartate supply and demand may synergize with the NDUFB9-mediated suppression of aspartate synthesis, thereby explaining why TNBC cells are more sensitive to NDUFB9 inhibition than other subtypes.

Fig. 7 — Heatmaps showing the protein abundance (Z-score) of the ETC complexes across PAM50 subtypes in CPTAC (A) and CBCGA datasets (B). C GSEA results for basal-like breast cancer in the CBCGA dataset. D Relative expression of genes between basal-like and other PAM50 subtypes in the CBCGA dataset. The Y-axes denote relative gene expression. E Violin plots compare GSVA enrichment scores of metabolic pathways between TNBC and other subtypes from the CCLE dataset. n(# of cell lines)=30(TNBC); 27(others). F Western blot analysis of protein levels in TNBC compared to HER2-positive/luminal cell lines. G Normalized mitochondrial membrane potentials in TNBC cell lines compared to HER2-positive/luminal cell lines. n (# of independent experiments)=4 for each group. H Heatmap of fold changes (sgNDUFB9/sgCtr) in metabolite levels. Asterisks above each column denote the p-values for TNBC (MDA-MB-231) versus other subtype (MCF7). n = 3 for each group. I Western blot analysis of NDUFB9 protein levels in tumors treated with NDUFB9-targeting or control ASOs. J Schematic of the experimental approach evaluating the effect of NDUFB9-targeting ASO on 4T1 BrMs. Seven days after intracardiac injection of tumor cells, mice were randomized and treated with a single intracerebroventricular (ICV) injection of 200 μg Ctr ASO or NDUFB9-targeting ASO. K Representative BLI showing whole-body tumor burden (upper) and brain-specific tumor burden (lower) in Ctr ASO or NDUFB9-targeting ASO treated groups. L Quantification of brain-specific tumor burden from (K) using BLI. n(# of mice) = 6 for each group. (M) Percentage changes in body weight relative to baseline for mice in (J). P values were determined by two-sided unpaired t-test (H), one-way (G) or two-way ANOVA with Benjamini-Hochberg multiple test correction (M), or two-sided Mann-Whitney test (E, L). Data are presented as the means ± SEM (G, M) and median with interquartile range (L). Box plots show median (central line), upper and lower quartiles (box limits), and minima to maxima (whiskers) (E). Representative immunoblots from three independent experiments (F). BLI, Bioluminescent imaging; GSEA, gene set enrichment analysis. Source data are provided as a Source Data file.

To strengthen our conclusion, we extended our analysis to a panel of luminal (T47D, MCF7), HER2-positive (SKBR3), and TNBC (MDA-MB-231, HCC1806, SUM159PT) cell lines. We found that TNBC cells display lower mitochondrial membrane potential and reduced GOT2 expression (Fig. 7F-G)—indicative of limited aspartate upstream substrate supply—while they exhibit higher levels of the rate-limiting pyrimidine biosynthetic enzyme CAD (and its activating phosphorylation), the rate-limiting purine biosynthetic enzyme PPAT, and asparagine synthetase ASNS, reflecting an elevated aspartate downstream biosynthetic demand (Fig. 7F). NDUFB9-knockout in both MCF7 and MDA-MB-231 cells impaired TCA cycle intermediates (malate, fumarate) and mitochondrial aspartate (Fig. 7H); however, the suppressive effect was more pronounced in MDA-MB-231 (Fig. 7H). Moreover, NDUFB9-knockout reduced the abundance of nucleotide metabolites in MDA-MB-231 but not in MCF7 cells (Fig. 7H). These results collectively show that TNBC exhibits higher sensitivity to NDUFB9 inhibition than other subtypes, at least in part, due to the imbalance between aspartate upstream substrate supply and downstream biosynthesis demand.

Pharmacologic inhibition of NDUFB9 is a therapeutic strategy for TNBC BrMs with reduced toxicity

We next evaluated the therapeutic potential of targeting NDUFB9 in TNBC BrMs. Because no direct inhibitor of NDUFB9 is currently available, we employed IACS-010759^37,38, a BBB-permeable inhibitor that targets the ND1 core subunit of complex I. At clinically tolerable low-dose (10 nM)³⁸ of IACS-010759, IACS-010759 modestly reduced basal OCR, ATP-coupled OCR, maximum OCR, and aspartate production compared to vehicle-treated cells (Supplementary Fig. 8B-D), without significantly affecting total ATP production (Supplementary Fig. 8E), thereby resembling the metabolic phenotype of NDUFB9-knockout. 10 nM IACS-010759 caused proliferation defect under asparagine-limited conditions (Supplementary Fig. 8F). To test in vivo efficacy, we treated mice with either vehicle or 0.3 mg/kg IACS-010759 seven days after intracardially injected with cells, a dose that generates approximately 10 nM plasma concentrations, as demonstrated in a previous study³⁸. Treatment was administered on a 5-days-on/2-days-off schedule. IACS-010759 specifically reduced TNBC BrMs while having no significant effect on extracranial disease, and the inhibitory effect on BrMs was rescued by overexpression of SLC1A3 (Supplementary Fig. 8G-K). No significant weight loss was observed during the treatment period (Supplementary Fig. 8L). However, IACS-010759 treated mice exhibited higher modified neurological severity score (mNSS) than vehicle controls (Supplementary Fig. 8M), confirming a mild neurological deficit, underscoring the need to identify a more tumor-selective target.

We then tested other ETC inhibitors, including rotenone (complex I inhibitor), antimycin A (complex III inhibitor), and oligomycin (complex V inhibitor), all of which exhibited lower IC₅₀ values in asparagine-limited (6 μM ASN) media (Supplementary Fig. 8N). However, due to their off-target effects or suboptimal pharmacokinetic profiles^37,39, these agents were not pursued for in vivo validation.

To identify a more tumor-selective target, we analyzed the public TNBC BrMs scRNA-Seq data (GSE186344) and revealed that brain metastatic TNBC express higher levels of NDUFB9 compared to non-tumor niche cells (astrocytes, B cells, T cells, etc.), while the expression of NDUFB11 between tumor cells and non-tumor cells was not such pronounced (Supplementary Fig. 9A-B). We also analyzed the NDUFB9 and NDUFB11 expression across non-tumor adult human brain cell types (Human Brain Cell Atlas v1.0⁴⁰) and found that both NDUFB9 and NDUFB11 have low cell-type specificity among brain niche cells (Supplementary Fig. 9C-D). These data suggest that NDUFB9 might offer superior tumor selectivity and a safer profile for targeting TNBC BrMs. Consistent with this, treatment with NDUFB9-targeting antisense oligonucleotide (ASO) reduced TNBC BrMs without increasing neurological toxicity (Fig.7I-M; Supplementary Fig. 9E-F). These preclinical results demonstrated that pharmacologic inhibition of NDUFB9 confers a BrMs-specific antitumor effect with reduced toxicity, rendering it a promising therapeutic target.

Discussion

The treatment of TNBC BrMs remains a significant unmet clinical need. This study is the first to analyze matched CSF and plasma samples from patients with TNBC BrMs, offering a detailed depiction of the metabolic microenvironment of TNBC BrMs. Additionally, our study uniquely focused on TNBC patients with brain oligometastases—a rare clinical scenario, yet one that provides clearer insight into brain-specific adaptive mechanisms than cases with multi-organ metastases. To identify potential therapeutic targets, we integrated transcriptomic and proteomics data to depict the discrepancy between TNBC primary tumor and BrMs, and then conducted a high-throughput in vivo CRISPR-Cas9 loss-of-function screen of genes upregulated in TNBC BrMs. Our study found that NDUFB9 was the candidate treatment target of TNBC BrMs. NDUFB9-knockout specifically inhibited TNBC BrMs by impairing the intracranial“post-colonization outgrowth stage”. This selective impairment arises from intracranial–extracranial differences in asparagine availability. Mechanistically, TNBC exhibited an imbalance between aspartate upstream supply and downstream biosynthetic demand. NDUFB9-knockout disrupted mitochondrial complex I and reduced intracellular aspartate, but this alone was insufficient to inhibit TNBC proliferation. Instead, the lower asparagine concentration in the brain microenvironment induced compensatory upregulation of asparagine synthetase, which further diverted aspartate toward asparagine biosynthesis, exhausted the aspartate pool and restricted nucleotide biosynthesis. This dual-hit model identifies NDUFB9 as a metabolic vulnerability specific to BrMs, rather than to extracranial tumors. Finally, our analyses show that TNBC BrMs express higher levels of NDUFB9 compared to non-tumor brain niche cells, and preclinical testing of NDUFB9-targeting ASO achieves BrM-specific efficacy without exacerbating neurological toxicity, underscoring its potential as a therapeutic strategy for TNBC BrMs.

To accurately characterize the discrepancy between primary tumors and TNBC BrMs, we implemented the following key improvements, distinguishing our approach from previous studies^31,41,42. First, only patients with BrMs but no extracranial metastases (brain oligometastases) were included. While challenging to find, brain oligometastases could better help to study the mechanism of brain colonization. Because multi-organ secondary metastases exhibit increased stem cell-like properties and reduced organotropism⁴³, oligometastases are more likely to have characteristics of organ-specific adaptation. Second, primary tumors and BrMs were matched from the same patient, and both were confirmed as TNBC subtype. This approach minimized discrepancy due to subtype conversion or inter-patient variability. Third, integrated transcriptomic and proteomic analyses were performed to reduce the likelihood of false positives and noise interference caused by single-omics approaches. Based on this, our study depicted the discrepancy between TNBC primary tumor and BrMs. The identified discrepancies between primary tumors and BrMs hold two main translational implications: first, discovery of brain-specific adaptations and therapeutic targets. The discrepancy highlights essential traits acquired by tumors from the brain microenvironment that drive outgrowth in the brain. TNBC BrMs were found to upregulate aspartate biosynthesis, increase mTORC1 signaling, upregulate E2F targets, and downregulate epithelial-mesenchymal transition (EMT). NDUFB9 was validated as a druggable, effective target for TNBC BrMs in this study. Other pathways remain open for further analysis. Second, the molecular differences between primary tumors and BrMs may explain discrepancies in treatment response⁴⁴. Historically, limited intracranial efficacy has been attributed to inadequate drug delivery across the BBB and BTB⁴⁵. However, our findings indicate that the upregulation of mitochondrial respiration, a known driver of chemoresistance⁴⁶, may also contribute to the poor intracranial response rates observed with chemotherapies.

We developed the TNBCBM-CK library for in vivo loss-of-function CRISPR screening, enabling the examination of 499 candidate genes in a massively parallel manner. Our screening results have reliability for two reasons: first, cells were injected either intracranially or into the mammary fat pad, ensuring the use of authentic in vivo microenvironments as selection pressures; second, we observed the depletion of sgRNAs targeting HMGCS1 in the brain, which is consistent with HMGCS1’s known role in cholesterol synthesis⁴⁷ and the positive association between cholesterol synthesis and brain metastases¹³, further supporting the robustness of our results. By TNBCBM-CK CRISPR screening, we identified NDUFB9 as a previously unrecognized therapeutic target for TNBC BrMs. Additionally, our TNBCBM-CK screens identified several other targets for TNBC BrMs, including RAN, PTMA, PPP4C, and SLC3A2, which are open for future validations.

Previous studies have shown that NDUFB9 encodes a leucine-tyrosine-arginine-(LYR)-motif-containing subunit of respiratory chain complex I⁴⁸, one of the 30 supernumerary subunits that do not directly participate in catalysis. Thus, it remains unclear whether NDUFB9 is involved in mitochondrial respiration and, if so, how it contributes to this process. In this study, we confirmed that NDUFB9-knockout disrupted respiratory chain complex I, impaired NAD⁺ regeneration, and decreased intracellular aspartate levels by disrupting supercomplex. However, NDUFB9 loss did not inhibit extracranial tumor growth but instead selectively impaired TNBC BrMs. This selectivity was not attributable to differences in aspartate concentrations between intracranial and extracranial microenvironments, as extracellular aspartate has poor cell permeability and modest changes in its levels did not substantially affect intracellular aspartate. Rather, we uncovered a dual-hit mechanism: NDUFB9 loss decreases intracellular aspartate production, while asparagine limitation in the brain microenvironment induces compensatory ASNS upregulation that further diverts aspartate toward asparagine biosynthesis, jointly depleting the aspartate pool and suppressing BrM outgrowth. We further demonstrated that TNBC exhibits a unique imbalance between upstream substrate supply and downstream biosynthetic demand for aspartate, rendering them metabolic vulnerable to NDUFB9 inhibition–induced aspartate depletion compared with other breast cancer subtypes. Finally, we showed that NDUFB9-targeting ASOs achieved BrM-specific efficacy without exacerbating neurological toxicity, underscoring NDUFB9 as a potential therapeutic strategy for TNBC BrMs. Prior study demonstrated that NDUFB9 down-regulation enhances proliferation and migration of breast cancer cells in vitro⁴⁹. However, their study lacks in vivo validation. By contrast, our findings are supported by multiple animal models, showing that NDUFB9-knockout specifically inhibits the post-colonization outgrowth of TNBC brain metastases, thereby providing more physiologically relevant evidence.

Recent studies have investigated the association between mitochondrial ETC function and metastasis. For example, DeBerardinis et al. demonstrated that ETC inhibitor markedly suppresses lung metastasis⁵⁰, and Fiorillo et al. showed that Bedaquiline downregulated ATP5F1C and prevented spontaneous metastasis in vivo⁵¹. However, these studies employed ETC inhibitors at doses exceeding clinically achievable exposures, thereby providing limited guidance for clinical translation⁵². For example, in DeBerardinis’s study, 5 mg/kg IACS-010759 was used, which resulted in ~200–500 nM plasma levels, far exceeding the ~10 nM concentration that was tolerable in patients in phase I clinical trial of IACS-010759³⁸. Likewise, in Fiorillo’s study, 90 μM Bedaquiline was needed to inhibit metastases, far exceeding the 1 μM exposure considered clinically safe⁵³. Thus, although previous studies showed that ETC inhibition can suppress extracranial metastases, the high drug exposures used are not clinically feasible and therefore have limited clinical value. Moreover, previous studies have not clearly addressed why ETC inhibitors reduce metastases. Therefore, three challenges remain to be addressed: First, elucidate why ETC inhibitor could inhibit metastases; Second, identify tumor subtypes with high sensitivity to ETC inhibition so that low, clinically tolerable doses can achieve meaningful antitumor efficacy; Third, discover novel mitochondrial respiration targets that have tumor-selective inhibitory effects and could minimize toxicity. As for the first dilemma, we demonstrate that the antimetastatic activity of ETC inhibition is not a generalizable effect across cancer types or metastatic sites, but in TNBC BrMs specifically arises from suppression of aspartate biosynthesis in synergy with the asparagine-limited brain microenvironment. This dual-hit metabolic restriction explains why ETC inhibition at clinical tolerable dose selectively impairs intracranial rather than extracranial tumor outgrowth, providing a mechanistic framework that previous studies lacked. As for the second dilemma, our study found TNBC exhibit a unique imbalance between upstream substrate supply and downstream biosynthetic demand for aspartate, rendering them relatively sensitive to NDUFB9 inhibition–induced aspartate depletion. As for the third dilemma, we identified NDUFB9 as a therapeutic target. We found that TNBC BrMs express higher levels of NDUFB9 compared to non-tumor niche cells, suggesting superior tumor selectivity. We designed and synthesized NDUFB9-targeting chemically modified gapmer ASOs that hybridize with complementary RNA sequences and mediate ribonuclease H (RNase H) cleavage of the target transcripts. In our in vivo preclinical model, administration of the NDUFB9-targeting ASO reduced brain metastatic burden without causing observable neurological deficits, demonstrating a lower toxicity profile compared with the IACS-010759. ASOs have already progressed from bench to clinic as viable therapeutic agents⁵⁴. These results suggest that intracranial delivery of NDUFB9-targeting ASOs by Ommaya reservoir holds promise for future clinical applications.

Previous study⁵⁵ reported that ASNS upregulation promotes tumor metastases. In contrast, we observed that the combination of NDUFB9-knockout and ASNS upregulation synergistically inhibits TNBC BrMs. Although these observations might seem contradictory, they are reconcilable considering distinct metastatic stages. Previous research indicated that ASNS upregulation enhances the migratory and invasive capabilities of tumor cells specifically during the initial metastatic “seeding” stage, but does not affect the subsequent “post-colonization outgrowth” stage, thus promoting metastasis non-specifically to various distant sites. Our study showed that under asparagine-abundant extracranial microenvironment, NDUFB9-knockout did not upregulate ASNS (Fig.5A). Thus, NDUFB9-knockout did not affect the seeding potential of tumor cells, in agreement with prior findings. Interestingly, our study found that, upon colonization of the asparagine-limited intracranial microenvironment, the combination of NDUFB9-knockout and ASNS upregulation synergistically reduced aspartate and impaired de novo nucleotide biosynthesis, thereby inhibiting the post-colonization outgrowth of TNBC BrMs. Moreover, previous studies proposed that strategies reducing ASNS may impede metastases at the initial “seeding stage”. However, such approaches have limited clinical value for patients already presenting with established brain metastases, which remains a significant therapeutic challenge. Our study differs from previous work by specifically addressing the treatment of established brain metastases and offering potential for clinical translation.

The brain’s distinctive anatomical structures and metabolic constraints differ markedly from those of extracranial organs, compelling breast cancer-derived BrMs to undergo adaptations to thrive in brain microenvironment^56,57. The survival of cancer cells in the brain often hinges on their ability to synthesize specific metabolites that are scarce within brain microenvironment. For instance, BrMs have been shown to increase serine synthesis to cope with the serine and glycine-limitations in the brain environment. Inhibition of 3-phosphoglycerate dehydrogenase (PHGDH), a key enzyme in serine biosynthesis, has been observed to suppress intracranial tumor growth²⁰. Additionally, breast cancer and melanoma BrMs have adapted to utilize acetate as an alternative energy substrate under conditions of limited glucose availability, facilitating their survival in the brain⁵⁸. Moreover, given the brain’s significantly lower levels of triacylglycerols (TAGs) compared to extracranial tissues, perturbation in lipid-metabolism have been exploited to inhibit the growth of breast cancer BrMs¹³. Despite these discoveries, translating such findings into clinically applicable therapies remains a challenge. A deeper understanding of the metabolites available in the brain microenvironment and those that are critical for tumor growth could pave the way for therapeutic strategies for BrMs. Historically, studies have explored the brain’s metabolic microenvironment based on the CSF from normal individuals^59,60. However, in the context of BrMs, the BTB and BBB become more permissive⁶¹, potentially altering metabolite levels in the brain⁶². Therefore, CSF from healthy individuals may not accurately reflect the metabolic microenvironment faced by BrMs. Our study addresses this gap by collecting matched CSF and plasma from patients with TNBC BrMs, offering a more precise depiction of the metabolic environment of TNBC BrMs. We identified several metabolites that are lacking in the brain microenvironment, such as serine, proline, asparagine, and methionine, alongside metabolites that are abundant, like inositol. These findings lay the groundwork for the discovery of more metabolic vulnerabilities in TNBC BrMs in future studies. By understanding how these metabolites affect tumor behavior in the unique brain environment, researchers can target these metabolic pathways to develop more effective therapies.

Methods

This study was approved by the Research Ethical Committee of Fudan University Shanghai Cancer Center (FUSCC, No.1612167-18). Written informed consents were obtained from all participating patients.

Cell lines and cell culture

The MDA-MB-231, MCF7, HCC1954, SKBR3, T47D, 4T1, and HCC1806 cell lines were obtained from the American Type Culture Collection (ATCC), while the SUM159PT cell line was obtained from BioIVT. The HEK-293T cell line was obtained from the National Collection of Authenticated Cell Cultures (Shanghai, China). The MDA-MB-231, MCF7, HCC1954, SKBR3, T47D, SUM159PT, HCC1806, and 4T1 cell lines were cultured in RPMI1640 medium supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin, unless otherwise stated. The HEK-293T cell line was maintained in DMEM with 10% FBS. All the cell lines were cultured in a humidified incubator with 5% CO₂ at 37 °C following standard protocols. Cell line authentication was confirmed via short-tandem repeat (STR) analysis and all cultures were verified to be mycoplasma-negative.

Clinical samples

As for inclusion criteria, five female TNBC patients with BrMs but no extracranial metastases (brain oligometastases) were included. CSF and paired plasma samples were collected from patients with TNBC BrMs. Formalin-fixed paraffin-embedded (FFPE) slides of surgically collected primary breast tumors and paired brain oligometastase were obtained from TNBC patients. The triple-negative status of estrogen receptor (ER), progesterone receptor (PgR), and human epidermal growth factor 2 (HER2) of primary tumor and metastatic lesions was verified by the Department of Pathology, FUSCC. Clinical data, including age, tumor histology, neo/adjuvant chemotherapy regimen, BrM-free survival interval, and survival interval post BrMs were shown in Supplementary Data 1.

Animals

Six- to seven-week-old female BALB/c nude mice were obtained from Shanghai Jihui Laboratory Animal Feeding Co., Ltd. All mice were housed under 12-hour light/dark cycles and ambient temperature (20–22 °C) and humidity (60 ± 10%) in the laboratory animal center of FUSCC, with free access to standard rodent diet and water in individually ventilated cages in a specific pathogen-free facility. All animal experiments were conducted in accordance with the protocol approved by the Research Ethical Committee of FUSCC. Regular monitoring was performed to ensure the health and welfare of the animals: screening for key rodent pathogens (e.g., Sendai virus, PVM, Hantavirus), opportunistic bacteria (e.g., Salmonella, Pseudomonas aeruginosa, Staphylococcus aureus), and ectoparasites and endoparasites (e.g., helminths, flagellates). To monitor the growth of metastases, mice were intraperitoneally injected with 150 mg/kg D-luciferin, followed by anesthesia. Tumor growth and metastases were tracked by bioluminescent imaging (BLI) using the IVIS Spectrum imaging system (PerkinElmer). BLI signal intensity was quantified by Living Image software (PerkinElmer). Mice were euthanized if they lost more than 10% of their body weight. Tumor volumes were measured using calipers and calculated using the following formula: 0.5×length×width². The maximal tumor diameter permitted is 20 mm, and the maximal tumor size was not exceeded.

Mouse brain metastases model by intracardiac injection

All procedures were conducted under general anesthesia. Mice were placed in a supine position on a flat surface, with their forelimbs and hindlimbs gently extended and secured using tape. The skin was sterilized with 70% ethanol, and an injection point was identified ~1–2 mm to the left of a mark between the xyphoid process and the jugular notch. A syringe loaded with cell suspension was positioned vertically at the injection point. A total of 100 μl of tumor cell suspension was injected slowly over one minute. Following the injection, the mice were transferred to a temperature-controlled environment to ensure complete recovery.

Stereotactic brain injections of breast cancer cells

All procedures were conducted under general anesthesia. The mouse head was positioned in a stereotaxic apparatus. After sterilizing the scalp with 70% ethanol, the scalp was incised from the midline to expose the skull. Then a cranial hole was drilled 1 mm posterior and 2.0 mm lateral to the Bregma. A total of 5 μl of tumor cell suspension was injected through the cranial hole using a 10 μl syringe, with the needle inserted to a depth of 2.5 mm. After intracranial injection, the syringe was slowly withdrawn over a 3 min period. Following the procedure, mice were placed in a temperature-controlled environment to ensure complete recovery.

TNBCBM-CK sgRNA CRISPR screening library design

To construct the TNBCBM-CK sgRNA library, we selected 499 genes that were up-regulated (fold change > 1.5) in TNBC BrMs, as identified through RNA-seq and proteomics overlap analysis (Fig.1A). The sgRNA sequences were obtained from https://www.addgene.org/pooled-library/zhang-human-gecko-v2/. For each gene, six distinct sgRNAs were selected. The final TNBCBM-CK library consisted of a total of 3420 sgRNAs, including 2988 sgRNAs targeting the 499 genes of interest and 432 nontargeting control sgRNAs (Supplementary Data 2).

Oligonucleotide pool synthesis and TNBCBM-CK sgRNA library cloning

The oligonucleotide pool of the TNBCBM-CK sgRNA library was synthesized by Synbio Technologies and subsequently amplified by PCR using Q5^® High-Fidelity DNA Polymerase (NEB). The amplified oligonucleotides were then Gibson assembled into the lentiGuide-Puro backbone (Addgene #52963) to generate the TNBCBM-CK sgRNA library plasmids. To avoid loss of sgRNA representation, we performed electroporation to achieve a 200×library coverage, with more than 200 colonies per sgRNA obtained on LB agar plates.

TNBCBM-CK sgRNA library lentivirus production and transduction

Lentivirus of the TNBCBM-CK sgRNA library was produced by transfecting HEK-293T cells with the following plasmids: pMD2.G (Addgene), psPAX2 (Addgene), the TNBCBM-CK sgRNA library plasmids, and polyethyleneimine (PEI) transfection reagent. The medium was collected 48 h post-transfection, filtered through a 0.45 μm membrane to remove cell debris, and used for subsequent transduction. MDA-MB-231 cells, stably expressing lentiCas9-Blast (Addgene #52962), were seeded on 10 cm plate and transduced with the TNBCBM-CK sgRNA library lentivirus at a multiplicity of infection (MOI) of 0.3, ensuring that most cells received only one sgRNA. Forty-eight h after lentivirus transduction, MDA-MB-231 cells were selected with puromycin (1 μg/ml) to isolate successfully transduced cells.

CRISPR screening and MAGeCK analysis

Cas9-expressing MDA-MB-231 cells transduced with the TNBCBM-CK library were subjected to three different conditions to create varying selection pressures: intracranial injection (BrMs group), mammary fat pad injection (orthotopic tumors group), or in-vitro culture (in-vitro group) (Fig.1A). Each group consisted of five mice, and sufficient cells were injected into mouse to ensure >100-fold coverage of all sgRNAs in the TNBCBM-CK library. Eighteen days after injection, genomic DNA was extracted from the in vitro cells, orthotopic tumors, and BrMs using the QIAamp DNA Mini Kit (Qiagen). Nested PCR amplification of the genomic DNA was performed using Q5^® High-Fidelity DNA Polymerase (NEB). A total of 11.3 μg of input genomic DNA was used to achieve 500×library coverage. In the first step of nested PCR, the sgRNA regions were amplified, followed by the addition of Illumina sequencing adaptors in the second step of nested PCR. The PCR products were purified using beads (Beckman) and subsequently sequenced on the Illumina next-generation sequencing platform. The primers used in the nested PCR are listed in supplementary Table 1.

The raw FASTQ files were analyzed using the model-based analysis of genome-wide CRISPR-Cas9 knockout (MAGeCK) algorithm. MAGeCK first maps the sequencing data to the sgRNA library templates and normalizes the sgRNA read counts, adjusting for sequencing depth. The MAGeCK robust rank aggregation (RRA) method was then applied to identify sgRNAs and genes that were differentially selected between the two conditions. MAGeCK RRA ranks sgRNAs based on P values calculated from the negative binomial model and identifies gene hits using a modified RRA algorithm. Genes with an RRA score <0.01 in the BrMs group and RRA score > 0.1 in the orthotopic tumor group were considered BrMs-specific target genes. Genes with RRA score > 0.1 in BrMs group and RRA score <0.01 in orthotopic tumor group were identified as orthotopic tumor-specific target genes.

Animal studies

Mouse BrM models were established through intracardiac or stereotactic brain injections. For intracranial injections, 1 × 10⁵ MDA-MB-231 cells suspended in 5 μl PBS were injected. For intracardiac injections, 1 × 10⁵ MDA-MB-231 cells or 1 × 10⁴ 4T1 cells suspended in 100 μl PBS were injected into the left ventricle. As for the lung metastases model, 2 × 10⁵ MDA-MB-231 cells were intravenously injected via the lateral tail vein. As for the bone metastases model, 2 × 10⁵ MDA-MB-231 cells were injected into the left ventricle. As for the liver metastases model, 2 × 10⁵ MDA-MB-231 cells were injected via the portal vein. Neurological deficits were assessed using a modified neurological severity score (mNSS), which contains motor tests (including flexion of forelimb, flexion of hindlimb, and head movement, scored 0–6), beam balance tests (scored 0–6), and reflexes absent and abnormal movements (scored 0 to 2). Cumulative scores of 1–4, 5–9, and 10–14 correspond to mild, moderate, or severe neurological deficit, respectively. As for IACS-010759 experiments, mice were treated with 0.3 mg/kg IACS-010759 dissolved in 0.5% methyl cellulose. Control mice received vehicle alone. Compound or vehicle was administered by oral gavage once per day on a 5-day-on, 2-day-off schedule. Mice were euthanized by CO₂ asphyxiation at predetermined time points or if they showed signs of neurological distress. For the orthotopic tumor model, 2 × 10⁶ MDA-MB-231 cells were injected into the fourth mammary fat pad.

Analysis of mitochondrial complexes

As for blue-native (BN)-PAGE assay, mitochondria were isolated by a cell mitochondria isolation kit (Beyotime) and solubilized with digitonin (6.0 g/g) or Dodecylmaltoside (2.5 g/g) for 10 min. The samples were centrifuged for 30 min at 20,000 × g, and the supernatants were retained. Coomassie blue G-250 was added to achieve a detergent/dye ratio of 8 (g/g). The samples were loaded onto the 4%-13% acrylamide gradient gels, and BN-PAGE electrophoresis was conducted. The proteins were transferred to PVDF membranes, which were blocked with 5% BSA. Membranes were incubated with the primary antibodies overnight at 4 °C: anti-NDUFB8 (1:10,000, ab192878, Abcam) for complex I, anti-SDHB (1:50,000, ab175225, Abcam) for complex II, anti-UQCRC2 (1:2000, ab203832, Abcam) for complex III, anti-MTCO2 (1:2000, ab198286, Abcam) for complex IV, and anti-ATP5A1 (1:5000, ab176569, Abcam) for complex V. The membranes were incubated with the secondary antibody (Proteintech) and protein levels were quantified using Amersham Imager 600 (Cytiva).

High resolution models of complex I (PDB:5XTD) and supercomplex (PDB:5XTH) were downloaded from the Protein Data Bank. Solvation free energy gain (ΔiG) was calculated using the PISA (Proteins, Interfaces, Structures and Assemblies). Images (Supplementary Fig. 4H-I) were generated using the PyMOL Molecular Graphics System, version 3.1.4.1.

Collection of paired cerebrospinal fluid and plasma of patients

Cerebrospinal fluid (CSF) and paired plasma samples were collected from patients with TNBC BrMs, following the protocol approved by the Research Ethical Committee of FUSCC. Informed consent was obtained from all patients. The collected plasma and CSF samples were immediately stored in liquid nitrogen. For mouse CSF isolation, the procedure was conducted as previously reported⁶³. CSF samples with blood contamination were discarded. Blood was collected from the mice, centrifuged for 10 min at 800 × g to isolate plasma, and then both plasma and CSF samples were stored in liquid nitrogen.

Microdissection of patient-matched primary breast tumors and BrMs

FFPE samples of patient-matched primary breast tumors and BrMs were collected from five TNBC patients with brain oligometastases. Patient clinical information is provided in Supplementary Data 1. Tumor areas were identified by two experienced breast pathologists, using digital high-resolution images of hematoxylin-eosin-stained FFPE slides. Tumors tissues were isolated by laser capture microdissection from 10 μm thick parallel FFPE slides.

Western blot

Protein was extracted by cell lysis buffer (Beyotime), supplemented with phosphatase inhibitor and protease inhibitors. The supernatants of the lysates were collected, and protein concentration was measured by the BCA assay kit (Solarbio). Protein lysates were then normalized and mixed with 5×SDS-PAGE loading buffer, followed by denaturation for 5 min at 99 °C. Equal amounts of protein were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto polyvinylidene fluoride (PVDF) membranes (Millipore). The PVDF membranes were blocked with 5% BSA or milk and incubated with the primary antibodies overnight at 4 °C: anti-NDUFB9 (1:1000, ab200198, RRID: AB_3668935, Abcam), anti-NDUFB8 (1:10,000, ab192878, RRID:AB_2847808, Abcam), anti-SDHB (1:50,000, ab175225, RRID:AB_2904585, Abcam), anti-UQCRC2 (1:2000, ab203832, RRID:AB_2797139, Abcam), anti-ATP5A1 (1:5000, ab176569, RRID:AB_2801536, Abcam), anti-MTCO2 (1:2000, ab198286, RRID:AB_2861364, Abcam), anti-beta Actin (1:5000, 20536-1-AP, RRID:AB_10700003, Proteintech), anti-ATF4 (1:1000, #11815, RRID:AB_2616025, Cell Signaling), anti-ASNS (1:1000, 14681-1-AP, RRID:AB_2060119, Proteintech), anti-SHMT2 (1:2000, 11099-1-AP, RRID:AB_2188452, Proteintech), anti-GOT2 (1:8000, 14800-1-AP, Proteintech), anti-PPAT (1:1000 15401-1-AP, Proteintech), anti-phospho-S1859 CAD (1:500, #70307, Cell Signaling), anti-CAD (1:1000, #93925, Cell Signaling), anti-phospho-T389 S6K (1:1000, #9234, RRID:AB_2269803, Cell Signaling), and anti-S6K (1:1000, #2708, RRID:AB_390722, Cell Signaling). The membranes were then incubated with HRP-conjugated secondary antibody (Proteintech), and protein expression was visualized and quantified using the Amersham Imager 600 (Cytiva).

Immunofluorescence and immunohistochemical staining

FFPE slides were first deparaffinized in xylene and rehydrated through a series of ethanol washes (100%, 90%, and 70%). Antigen retrieval was conducted by placing slides in citric acid buffer (Solarbio) for 20 min at 95 °C. The slides were then treated with 3% H₂O₂ to inactivate the endogenous peroxidase, followed by incubation with 3% BSA to block the nonspecific sites. For immunohistochemical staining, the slides were incubated with primary antibodies (NDUFB9, 1:300, ab200198, RRID: AB_3668935, Abcam) at 4 °C overnight. After primary antibody incubation, the slides were incubated with secondary antibodies at room temperature for one hour and then processed by a DAB detection kit (GeneTech). The staining intensity was graded by immunoreactive score (IRS), calculated by multiplying the intensity score (0-3) by the proportion of positive cells (0, negative; 1, <10%; 2, 10%-50%; 3, 51%-80%; 4, >80%). The IRS scores were evaluated by two experienced pathologists who were blinded to the grouping information. For immunofluorescence, the slides were incubated with primary antibodies and secondary antibodies. The images were acquired using a confocal microscope (Leica).

Oxygen consumption rate (OCR) and ATP rate assay

OCR and real-time ATP production rates were measured using the Agilent Seahorse XFe Analyzer (Agilent Technologies). Cells were seeded at a density of 1 × 10⁴ to 4 × 10⁴ per well in Seahorse XF96 tissue culture microplates and incubated overnight in a 5% CO₂ incubator. Before measurement, the cells were equilibrated for one hour in serum-free XF assay medium in a non-CO₂ incubator. OCR was quantified at baseline and throughout sequential injections the following compounds: 1.5 μM oligomycin, 1 μM carbonyl cyanide-4 (trifluoromethory) phenylhydrazone (FCCP), and 0.5 μM rotenone/antimycin A. For ATP rate measurements, only oligomycin (1.5 μM) and rotenone/antimycin A (0.5 μM) were used. Seahorse data were normalized by protein concentration, which was determined by the BCA assay.

Cell death assay

Trypsinized cells and supernatants were collected and incubated in binding buffer containing propidium iodide (PI) and Annexin V (Solarbio) for 15 min. Then cells were analyzed by fluorescence-activated cell sorting (FACS) within one hour. Cells were classified into four groups based on their staining profile: necrotic cells (PI⁺ and Annexin V^-); late apoptosis cells (PI⁺ and Annexin V⁺); early apoptosis cells (PI^- and Annexin V⁺); and live cells (PI^- and Annexin V^-).

Electron microscopy

Cells or tissues were first fixed in 2.5% glutaraldehyde overnight, followed by one-hour fixation in 1% osmium tetroxide. The samples were then stained with 2% uranyl acetate overnight. After dehydration, samples were embedded in epoxide and sectioned to 70 nm thickness. The sections were then stained with uranylacetate and lead citrate. Images were captured by a JEOL JEM-1230 transmission electron microscope. Mitochondrial circularity and the number of cristae per mitochondrion were quantified in a blinded manner.

Mitochondrial DNA copy number

Total cellular DNA was isolated using the QIAamp DNA Mini Kit (Qiagen) according to the manufacturer’s instructions. The mitochondrial DNA (mtDNA) copy number was determined using the Human mitochondrial DNA monitoring primer set (Takara).

Mitochondrial mass and membrane potential

Cells were stained with 100 nM MitoTracker Red CMXROS (Invitrogen) for 20 min at 37 °C to assess mitochondrial membrane potential, and with 50 nM MitoTracker Green (Invitrogen) for 20 min at 37 °C to assess mitochondrial mass. Flow cytometry analyses were conducted using the CytoFLEX S Flow Cytometer (Beckman Coulter), and data were analyzed by FlowJo software (version 10.8.1).

Co-culture experiments

Primary astrocytes and microglia were isolated⁶⁴. The cerebral cortex from neonatal mouse pups was isolated and dissociated by 0.25% Trypsin (BasalMedia) and 0.1 mg/ml DNase I (Roche) for 15 min at 37 °C. The cells were maintained in DMEM-F12 (1:1) (BasalMedia) containing 10% FBS. To isolate astrocytes, the flasks were shaken overnight to remove microglia. The remaining adherent cells were astrocytes. To isolate microglia, the media were supplemented with 10 ng/ml recombinant mouse M-CSF (Sigma). After 5–7 days, the floating microglia were collected. Cancer-associated fibroblasts (CAFs) were isolated by digesting MDA-MB-231 orthotopic tumors. For co-culture experiments, tumor cells were mixed with microglia, astrocytes, or CAFs at a 1:3 ratio and maintained for 7 days. Green fluorescent protein (GFP)-positive tumor cells were isolated by FACS (Beckman Coulter) from the co-culture and quantified.

Proliferation and cell viability assays

Cell proliferation and viability were measured using an Operetta High-Content Imaging System (PerkinElmer). Briefly, cells were seeded in 24-well plates (Corning) and stained with PI and Hoechst 33342 (both at a final concentration of 1 mg/ml) for 30 min. Hoechst 33342-positive cells were counted to determine the total cell number. Cell viability was calculated by dividing the number of PI-negative cells by the total cell number.

Stable isotope tracing

In vivo isotope tracing was performed⁶⁵. Mice were anesthetized and 27-gauge catheters were placed in the lateral tail vein. Infusions of ¹³C-labeled nutrients began one hour after catheter implantation. For glucose tracing, 2.48 mg/g body mass of ¹³C₆-glucose (Cambridge Isotope Laboratories) was dissolved in 750 μl PBS. A bolus of 125 μl ¹³C₆-glucose was infused over one minute, followed by a continuous infusion of 2.5 μl per minute for three h. For glutamine tracing, 0.5 mg/g body mass of ¹³C₅-glutamine (Cambridge Isotope Laboratories) was dissolved in 1500 μl PBS. A bolus of 150 μl ¹³C₅-glutamine was infused over one minute, followed by a continuous infusion of 2.5 μl per minute for five h. After the infusion, orthotopic tumors and BrMs, tracked by BLI, were isolated and stored in liquid nitrogen. Labeled metabolites were measured⁶⁶. Tissue samples were manually homogenized and suspended in 1 ml 50% methanol, followed by a freeze-thaw procedure. Chloroform (0.4 ml) was added to the samples, which were then centrifugated for 15 min at 18,000 × g at 4 °C. The supernatant was collected and dried. The metabolites were derivatized for gas chromatography-mass spectrometry (GC-MS) analysis (GCMS-QP2010 Ultra, Shimadzu) as follows: First, 70 μl pyridine was added to the dried pellet and the mixture was incubated at 85 °C for 20 min. Then 30 µl of N-tert-butyldimethylsilyl-N-methyltrifluoroacetamide (Sigma) was added, followed by another incubation step at 85 °C for 60 min. The samples were then centrifuged at 18,000 xg for 15 min at 4 °C. The supernatant was collected and transferred into an autosampler vial for GC/MS analysis. A Shimadzu QP-2010 Ultra GC-MS was utilized. Samples (1 µl each) were injected with an injection temperature of 250 °C. The gas chromatograph was programmed with an initial oven temperature of 110 °C held for 4 min, rising to 230 °C at 3 °C/min, then increased to 280 °C at 20 °C /min and maintained for 2 min. The GC flow rate with helium carrier gas was 50 cm/s. Separation was performed using a 20 m x 0.25 mm×0.25 mm Rxi-5ms column. The interface temperature was 300 °C and the (electron impact) ion source temperature was set at 200 °C, with an ionization voltage of 70 V. The mass spectrometer scanned over an m/z range of 50–800, with 1 kV detector. GC/MS data were processed to determine isotope labeling and metabolite abundance. Metabolites exhibiting baseline separated peaks were quantified based on the total ion count peak areas, with calibration performed using standard curves derived from standards analyzed alongside the experimental samples. To evaluate ¹³C labeling, the mass distribution corresponding to specific metabolite fragments was extracted from the relevant chromatographic peaks. For each fragment, the retrieved data contained mass intensities for the lightest isotopomer (M0) and isotopomers with increasing unit mass (M1 to M6). Results were normalized by dividing the sum of M0 to M6, and corrected for the natural abundance of heavy isotopes using matrix-based probabilistic methods.

As for in vitro isotope tracing, cells were cultured with glucose-free medium supplemented with ¹³C₆-glucose (Cambridge Isotope Laboratories) for 6 h. Cells were ground in liquid nitrogen and suspended in 1 ml cold (−40 °C) 50% aqueous methanol. The samples were placed on dry ice for 30 min, followed by thawing on ice. Subsequently, 0.4 ml of chloroform was added, and the mixture was vortexed for 30 s before being centrifuged at 18,000 × g for 15 min at 4 °C. The supernatant was transferred to new 1.5 ml tubes and stored at –80 °C. Water-soluble metabolites were analyzed using the Q Exactive PLUS hybrid quadrupole-orbitrap mass spectrometer (Thermo Scientific), equipped with a hydrophilic interaction chromatography (HILIC). Separation was performed on an XBridge BEH Amide column (150 mm × 2.1 mm, 2.5 μM particle size, Waters, Milford, MA). The gradient was solvent A (95%:5% H2O:acetonitrile with 20 mM ammonium acetate, 20 mM ammonium hydroxide, pH 9.4) and solvent B (100% acetonitrile) 0 min, 90% B; 2 min, 90% B; 3 min, 75%; 7 min, 75% B; 8 min, 70%, 9 min, 70% B; 10 min, 50% B; 12 min, 50% B; 13 min, 25% B; 14 min, 25% B; 16 min, 0% B, 20.5 min, 0% B; 21 min, 90% B; 25 min, 90% B. Chromatographic conditions included a flow rate of 150 μL/min, an injection volume of 5 μL, and a column temperature maintained at 25 °C. The mass spectrometry scans were performed in negative ion mode at a resolution of 140,000 (at m/z 200). The automatic gain control (AGC) target was 1 × 10⁶ and the scan range was m/z 75–1000. Data from isotope labeling experiments were processed using El-MAVEN, incorporating correction for natural isotope abundance.

Metabolites measurement

For CSF and plasma samples, 900 μl acetonitrile-methanol-H₂O solution (4:4:1, containing isotope internal standards) was added to 100 μl samples. For cell samples, three freeze-thaw cycles were performed, followed by the addition of 800 μl acetonitrile-methanol (1:1, containing isotope internal standards) and 200 μl H₂O. After 15 min of sonication, the samples were incubated for 2 h at −40 °C. After centrifugation at 13,200 × g for 15 min, the 800 μl supernatant was collected and dried. The dried samples were re-suspended in 160 μl 60% acetonitrile, centrifuged again, and the 100 μl supernatant was collected for liquid chromatography-mass spectrometry (LC-MS)-based targeted metabolomics analysis. Standard stock solutions of various metabolites were mixed to create a mixed standard solution, which also contained isotopically-labeled internal standards at the same concentration as the sample. These mixed standard solutions were then serially diluted to prepare calibration solutions and generate the standard curve. A correlation coefficient (R²) value exceeding 0.99 for the standard curve demonstrates high linearity and confirms the reliability. LC separation was performed using a Waters H-Class UPLC System with an Atlantis Premier BEH Z-HILIC Column (1.7 μm, 2.1 mm * 150 mm). The mobile phase A consisted of water and acetonitrile (8:2) with 10 mM ammonium acetate, while mobile phase B was composed of water and acetonitrile in a 1:9 ratio, also containing 10 mM ammonium acetate. Both phases were adjusted to pH 9 using aqueous ammonia. The column temperature was set at 40 °C, and samples were injected at a volume of 2 μL with the auto-sampler temperature set at 8 °C. Mass spectrometric detection was carried out using an AB Sciex QTrap 6500+ instrument using multiple reaction monitoring (MRM). The ion source parameters were: IonSpray Voltage = +5000 V/−4500V, Curtain Gas=35 psi, Ion Source Gas 1 = 50 psi, Ion Source Gas 2 = 50 psi, Temperature=400 °C. A time-segmented MRM acquisition method was established based on individual metabolite retention times, and a metabolite-specific ion transition database was constructed. This approach improved detection specificity and enabled the simultaneous identification of low-abundance metabolites and metabolites from multiple classes.

Raw data were analyzed by SCIEX Analyst Work Station Software (1.6.3), and metabolite quantification was processed by BIOTREE Bio Bud (2.1.2). Metabolite concentrations were normalized by sample final volume, concentration factor, and cell number (for cell samples) or sample initial volume (for liquid samples). Missing values were imputed using one-fifth of the minimum detected value for each metabolite. Z-score for each individual sample value was computed as Z= [sample value (x)-mean(μ)]/standard deviation (σ). The pathway enrichment analyses were performed using MetaboAnalyst. Asparagine levels in cell culture media were measured by the Asparagine Fluorescence Assay Kit (BioVision) according to the manufacturer’s instructions.

Isolation of cancer cells from tumors and bulk RNA-seq

Orthotopic and brain metastatic tumors, tracked by BLI, were isolated, minced, and incubated in RPMI medium containing 1 mg/ml collagenase D (Roche), 1 mg/ml collagenase I (Sigma), and 20 μg/ml DNase I (Roche) for 1 h at 37 °C. The digested samples were passed through 70 μm cell strainers to obtain single-cell suspensions. Then, FACS sorting (Beckman Coulter) was used to isolate the GFP-positive cancer cells.

RNA was isolated using the RNeasy Mini Kit (Qiagen). A total of 1 μg RNA was used for library preparation. Poly(A) mRNA was isolated with Oligo(dT) beads, fragmented using high-temperature and divalent cations, reverse-transcribed with random primers. cDNA was synthesized, treated to repair ends and added a dA-tailing, followed by T-A ligation to add adaptors to both ends. DNA was size-selected using clean beads and amplified by PCR using P5 and P7 primers. Libraries were loaded on the Illumina HiSeq instrument for sequencing according to the manufacturer’s instructions.

RNA-seq of patient-matched primary breast tumors and BrMs

Total RNA of FFPE microdissection tissues was isolated by the RNAstorm^TM FFPE kit (Cell Data Science, USA) according to the manufacturer’s protocol. RNA-seq libraries were prepared using the SMART cDNA synthesis technology (Clontech, USA), followed by PCR amplification of purified double-stranded cDNA. The samples were then quality-controlled by Qsep100 (BiOptic, China) and Qubit (Thermo Fisher, USA) before sequencing on the Illumina sequencing platform (Nova). High-quality reads were mapped to the human reference genome (hg38) using STAR (v.2.7.1a), and raw read counts were quantified by featureCounts (v.1.6.3). Differential gene expression was analyzed by DESeq2 paired analysis.

Proteomics of patient-matched primary breast tumors and BrMs

Label-free proteomics of FFPE micro-dissected samples was performed⁶⁷. 5 TNBC primary tumor and 5 paired BrMs samples were lysed in a buffer containing 0.1 M DTT, 0.1 M pH8.0 Tris/HCl, and 1 mM PMSF for one minute on ice, followed by further lysis in 4% SDS at 99 °C for 2–2.5 h. Proteins were precipitated by adding acetone, then digested with trypsin (enzyme-to-substrate ratio of 1:50) for 18–20 h at 37 °C. Peptides were analyzed by a high-performance liquid chromatography system (EASY nLC 1200, Thermo Fisher Scientific) coupled with the Q Exactive HF-X Hybrid Quadrupole-Orbitrap Mass Spectrometer (Thermo Fisher Scientific). Peptides, re-dissolved in solvent-A (0.1% FA in water), were loaded onto a 100-μm-inner-diameter trap column and separated on a 150-μm-inner-diameter column over a 75 min gradient at a constant flow rate (solvent-A: 0.1% FA in water; solvent-B: 0.1% FA in 80% ACN). The eluted peptides were ionized under 2 kV and introduced into the mass spectrometer. Mass spectrometry was performed under the data-dependent acquisition (DDA) mode.

Raw files were processed with the MASCOT search engine and the NCBI human Refseq protein database in the Protein Discovery (Version 2.2). Proteome quantification was performed using the intensity-based absolute quantification (iBAQ) algorithm⁶⁸. All the genes were included when analyzing the suitable candidates for TNBCBM-CK library construction. Genes with fewer than six missing values were included in the differential expression analyses between primary tumors and BrMs. Fraction of total (FOT) was calculated as the normalized protein abundance, defined as the protein’s iBAQ divided by the total iBAQ, and scaled by 10⁷. Missing values were replaced by a half-minimum value, and differential gene analysis was performed using the limma paired-sample analysis⁶⁹.

NAD⁺/NADH levels

Cells were rapidly rinsed three times with ice-cold PBS, lysed in 100 µL of ice-cold buffer (1% dodecyltrimethylammonium bromide in 0.2 N NaOH, 1:1 v/v with PBS), snap-frozen in liquid nitrogen, and stored at −80 °C. To measure NAD⁺, 20 µl samples were mixed with 20 µl lysis buffer and 20 µl 0.4 N HCl and heated at 60 °C for 15 min; for NADH measurement, 20 µl samples were heated at 75 °C for 30 min. Acidic incubation selectively degrades NADH, whereas basic conditions degrade NAD⁺. Then, samples were centrifuged and neutralized with 20 µL of Tris buffer (0.5 M Tris base for NAD⁺ assays; 0.25 M Tris in 0.2 N HCl for NADH assays). The manufacturer’s protocol of NAD/NADH-Glo assay (Promega) was then followed to measure NAD⁺/NADH ratio. As for LbNOX overexpression, cyto-LbNOX (Addgene #75285) and Mito-LbNOX (Addgene #74448) were subcloned into the pCDH-CMV-EF1 vectors.

Pathway enrichment analyses

Gene set enrichment analysis (GSEA) was performed using the GSEA software (v 4.1.0) with the pre-ranked function. The pre-ranked gene list was ranked based on the logFC from differential gene analysis. A total of 1000 permutations were used. The enrichment level of each pathway in each sample was calculated by the gene set variation analysis (GSVA)⁷⁰.

Antisense oligonucleotides

PS-MOE ASOs were purchased from RIBBIO. The ASOs have gapmers with MOE wings and a phosphorothioate (PS) backbone. The gapmer ASOs have a DNA-like central region that directs cleavage of the complementary mRNA by endogenous RNase H and chemically modified wings that promote tighter RNA binding, enhanced stability, and improved cellular uptake. The NDUFB9 target sequence was AGTTTCAGAGCACGCTCACA. ASOs were dissolved in water and diluted in saline before use. For in vivo experiments, seven days after intracardiac injection of tumor cells, mice were randomized and treated with a single intracerebroventricular (ICV) injection of 200 μg Ctr ASO or NDUFB9-targeting ASO using a syringe in isoflurane-anesthetized animals.

External datasets analysis

The RNA-seq data of paired primary breast tumors and BrMs from a previous study³¹ were used as the external validation. Additionally, the FUSCCTNBC dataset, comprising RNA-seq data of 360 primary TNBC samples, was downloaded⁷¹. Protein and metabolite levels across different breast cancer subtypes were collected from CPTAC dataset³³ and CBCGA dataset³⁴. Cell line data for various breast cancer subtypes were retrieved from the CCLE datasets⁷². Heatmaps and survival curves were generated using the Prism 9 software. Moreover, scRNA-seq dataset (GSE186344), which include TNBC brain metastases samples, was curated from Gene Expression Omnibus database. Cell-type annotations and normalized gene expression matrices were extracted directly from the supplemental files.

Gene knockdown and overexpression

For ASNS knockdown, shRNA targeting ASNS or a non-specific scrambled shRNA were cloned into the pLKO.1 neo vector (Addgene). The shASNS lentivirus was produced and used to transduce MDA-MB-231 and HCC1806 cells. Stably transduced cells were selected by G418. Knockdown efficiency was confirmed by Western blot analyses. The shRNA sequences are provided in supplementary Table 1. As for siRNA transfection, the siRNA transfection was conducted using Lipofectamine RNAiMAX (Thermo Fisher) according to the manufacturer’s instruction. For ASNS and SLC1A3 overexpression, we cloned the cDNA into the pCDH-CMV neo vector (System Bioscience). The stably integrated cells were selected by G418. The overexpression efficiency was confirmed by Western blot analyses. To establish doxycycline-inducible shRNA-expressing cells, shCtr or shNDUFB9 were cloned into pTRIPZ. Transduced cells were grown in the presence of puromycin for 72 h, and then expression of NDUFB9 was confirmed after treatment with 2 μg/ml doxycycline followed by Western blot analyses. For in vivo inducible knockdown experiments, mice were given doxycycline hyclate in the drinking water (2 mg/mL) during indicated days. As shown previously, doxycycline can cross the BBB into the brain metastatic lesions⁷³. To knockout NDUFB9, sgRNAs targeting NDUFB9 were cloned into the lentiGuide-Puro vector (Addgene). The sgNDUFB9 lentivirus was produced and used to infect Cas9-expressing cells. Forty-eight h post-infection, the stably integrated cells were selected by 1 μg/ml puromycin. Knockout efficiency was confirmed by Western blot analyses. To rescue NDUFB9 after NDUFB9-knockout, cDNA of NDUFB9 has undergone silent mutagenesis to remove the CRISPR/Cas9 target site. Based on this CRISPR-resistant construct, an NDUFB9^H12A mutant was subsequently generated by introducing an H12A coding substitution (CAT to GCT). The NDUFB9 sgRNA sequences are provided in supplementary Table 1.

Statistics and reproducibility

No statistical method was used to predetermine sample size. No data were excluded from the analyses. Acquisition of patient tumor samples was not randomized. The experiments were randomized. The investigators were not blinded to allocation during experiments and outcome assessment.

Statistical analyses were conducted using Prism 9 or R software (v.4.1.0). Data are shown as mean ± SEM, unless otherwise specified. For comparisons between two groups with normal distribution (assessed using the Shapiro-Wilk test), Student’s t-tests were applied. For comparisons involving three or more groups, one-way ANOVA was used, with corrections for multiple comparisons based on the False Discovery Rate (FDR). The Mann-Whitney test or Kruskal-Wallis test with Benjamini-Hochberg multiple test correction was applied for data without normal distribution. Survival curves were compared using log-rank tests. Two-sided P values < 0.05 were considered statistically significant.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Supplementary Information^{(9.8MB, pdf)}

Peer Review file^{(1.8MB, pdf)}

41467_2026_72927_MOESM3_ESM.pdf^{(55.6KB, pdf)}

Description of Additional Supplementary Files

Supplementary Data 1^{(11.8KB, xlsx)}

Supplementary Data 2^{(128.2KB, xlsx)}

Supplementary Data 3^{(1.1MB, xlsx)}

Supplementary Data 4^{(32.2KB, xlsx)}

Supplementary Data 5^{(28KB, xlsx)}

Supplementary Data 6^{(16KB, xlsx)}

Supplementary Data 7^{(2MB, xlsx)}

Supplementary Data 8^{(170.1KB, xlsx)}

Reporting Summary^{(112.2KB, pdf)}

Source data

Source data^{(5.1MB, xlsx)}

Acknowledgements

We would like to thank the clinical staff in Fudan University Shanghai Cancer Center who participated in this study. Funding: This study was supported by National Natural Science Foundation of China (grant no. 82503430 to MX.L., 82072915 and 82373359 to J.Z., 82473071 to XC.H.); Project of Shanghai Municipal Health Commission (grant no.202140397 to J.Z.); National Key Research and Development Program of China (grant no.2023YFF1205000 to J.Z.); CSCO Cancer Research Fund (grant no.Y-HR2020MS-0298, Y-pierrefabre202102-0066 to J.Z.); Chinese Young Breast Experts Research project (grant no.CYBER-2021-001 to J.Z.); Beijing Science and Technology Innovation Medical Development Foundation Key Project (grant no.KC2022-ZZ-0091-6 to J.Z.).

Author contributions

Conception and design: J.Z., XC.H., and XX.G.; acquisition of data: MX.L., ZX.W., C.Z., YZ.J., T.Z., YX.Y.; analysis and interpretation of data: MX.L., ZX.W., SL.H., and X.H.; writing of the manuscript: MX.L. and ZX.W.; study supervision: J.Z. and XC.H. All authors reviewed the manuscript. MX.L. and ZX.W. contributed equally to this work and share first authorship. The order of authorship was determined based on their overall contributions to the project.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

The RNA-seq and metabolome data generated in this study have been deposited at GSA database with accession numbers PRJCA032856. Proteomics data have been deposited in iProX under accession code PXD073600. Previously published datasets used in this study include CBCGA dataset (PRJCA017539), CPTAC and CCLE datasets from cBioPortal, FUSCCTNBC dataset (NODE: OEP000155), Human Brain Cell Atlas v1.0⁴⁰, TNBC BrMs scRNA-Seq data (GSE186344), and paired primary tumor and BrMs RNA-seq from ref. ³¹ High-resolution models of mitochondrial complex I (PDB: 5XTD) and the respiratory supercomplex (PDB: 5XTH) were downloaded from the Protein Data Bank. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

These authors contributed equally: Mingxi Lin, Zhexu Wen.

Contributor Information

Xiaoxiang Guan, Email: xguan@njmu.edu.cn.

Xichun Hu, Email: huxichun@shca.org.cn.

Jian Zhang, Email: jianz@fudan.edu.cn.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-72927-2.

References

1.Kennecke, H. et al. Metastatic behavior of breast cancer subtypes. J. Clin. Oncol.28, 3271–3277 (2010). [DOI] [PubMed] [Google Scholar]
2.Soffietti, R. et al. Diagnosis and treatment of brain metastases from solid tumors: guidelines from the European Association of Neuro-Oncology (EANO). Neuro Oncol.19, 162–174 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Lee, S. S. et al. Brain metastases in breast cancer: prognostic factors and management. Breast Cancer Res. Treat.111, 523–530 (2008). [DOI] [PubMed] [Google Scholar]
4.Lin, N. U., Bellon, J. R. & Winer, E. P. CNS metastases in breast cancer. J. Clin. Oncol.22, 3608–3617 (2004). [DOI] [PubMed] [Google Scholar]
5.Hurvitz, S. A. et al. Subgroup analyses from the phase 3 ASCENT study of sacituzumab govitecan in metastatic triple-negative breast cancer. NPJ Breast Cancer10, 33 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Lin, M. et al. HER2-low heterogeneity between primary and paired recurrent/metastatic breast cancer: Implications in treatment and prognosis. Cancer130, 851–862 (2024). [DOI] [PubMed] [Google Scholar]
7.Vaz Batista, M. et al. Trastuzumab deruxtecan in patients with previously treated HER2-low advanced breast cancer and active brain metastases: the DEBBRAH trial. ESMO Open9, 103699 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Zhang, L. et al. Microenvironment-induced PTEN loss by exosomal microRNA primes brain metastasis outgrowth. Nature527, 100–104 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Zou, Y. et al. Polyunsaturated fatty acids from astrocytes activate PPARγ signaling in cancer cells to promote brain metastasis. Cancer Discov.9, 1720–1735 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Jin, Y. et al. Targeting polarized phenotype of microglia via IL6/JAK2/STAT3 signaling to reduce NSCLC brain metastasis. Signal Transduct. Target Ther.7, 52 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Hladky, S. B. & Barrand, M. A. Fluid and ion transfer across the blood-brain and blood-cerebrospinal fluid barriers; a comparative account of mechanisms and roles. Fluids Barriers CNS13, 19 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Hladky, S. B. & Barrand, M. A. Elimination of substances from the brain parenchyma: efflux via perivascular pathways and via the blood-brain barrier. Fluids Barriers CNS15, 30 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Jin, X. et al. A metastasis map of human cancer cell lines. Nature588, 331–336 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Kim, A. E., Nieblas-Bedolla, E., de Sauvage, M. A. & Brastianos, P. K. Leveraging translational insights toward precision medicine approaches for brain metastases. Nat. Cancer4, 955–967 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Lorger, M. & Felding-Habermann, B. Capturing changes in the brain microenvironment during initial steps of breast cancer brain metastasis. Am. J. Pathol.176, 2958–2971 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Valiente, M. et al. Serpins promote cancer cell survival and vascular co-option in brain metastasis. Cell156, 1002–1016 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Varkuti, B. H. et al. Neuron-based high-content assay and screen for CNS active mitotherapeutics. Sci. Adv.6, eaaw8702 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Giacomello, M., Pyakurel, A., Glytsou, C. & Scorrano, L. The cell biology of mitochondrial membrane dynamics. Nat. Rev. Mol. Cell Biol.21, 204–224 (2020). [DOI] [PubMed] [Google Scholar]
19.Titov, D. V. et al. Complementation of mitochondrial electron transport chain by manipulation of the NAD+/NADH ratio. Science352, 231–235 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Ngo, B. et al. Limited Environmental Serine and Glycine Confer Brain Metastasis Sensitivity to PHGDH Inhibition. Cancer Discov.10, 1352–1373 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Freret, M. E. & Boire, A. The anatomic basis of leptomeningeal metastasis. J. Exp. Med.221 (2024). [DOI] [PMC free article] [PubMed]
22.Sullivan, L. B. et al. Aspartate is an endogenous metabolic limitation for tumour growth. Nat. Cell Biol.20, 782–788 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Battaglioni, S., Benjamin, D., Wälchli, M., Maier, T. & Hall, M. N. mTOR substrate phosphorylation in growth control. Cell185, 1814–1836 (2022). [DOI] [PubMed] [Google Scholar]
24.Kilberg, M. S., Shan, J. & Su, N. ATF4-dependent transcription mediates signaling of amino acid limitation. Trends Endocrinol. Metab.20, 436–443 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Celardo, I., Lehmann, S., Costa, A. C., Loh, S. H. & Miguel Martins, L. dATF4 regulation of mitochondrial folate-mediated one-carbon metabolism is neuroprotective. Cell Death Differ.24, 638–648 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Wortel, I. M. N., van der Meer, L. T., Kilberg, M. S. & van Leeuwen, F. N. Surviving stress: modulation of ATF4-mediated stress responses in normal and malignant cells. Trends Endocrinol. Metab.28, 794–806 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Schägger, H. & Pfeiffer, K. Supercomplexes in the respiratory chains of yeast and mammalian mitochondria. Embo j.19, 1777–1783 (2000). [DOI] [PMC free article] [PubMed] [Google Scholar]
28.D’Angelo, L. et al. NDUFS3 depletion permits complex I maturation and reveals TMEM126A/OPA7 as an assembly factor binding the ND4-module intermediate. Cell Rep.35, 109002 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Vercellino, I. & Sazanov, L. A. The assembly, regulation and function of the mitochondrial respiratory chain. Nat. Rev. Mol. Cell Biol.23, 141–161 (2022). [DOI] [PubMed] [Google Scholar]
30.Zhu, Z., Zhang, Q. & Sui, Z. Screening of ApDOT1.9 interacting proteins and the potential function of interactor ApSNARE in the rapid growth regulation of Alexandrium pacificum. Mar. Pollut. Bull.209, 117080 (2024). [DOI] [PubMed] [Google Scholar]
31.Varešlija, D. et al. Transcriptome Characterization of Matched Primary Breast and Brain Metastatic Tumors to Detect Novel Actionable Targets. J. Natl. Cancer Inst.111, 388–398 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Mullen, A. R. et al. Reductive carboxylation supports growth in tumour cells with defective mitochondria. Nature481, 385–388 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Krug, K. et al. Proteogenomic landscape of breast cancer tumorigenesis and targeted therapy. Cell183, 1436–1456.e1431 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Jiang, Y. Z. et al. Integrated multiomic profiling of breast cancer in the Chinese population reveals patient stratification and therapeutic vulnerabilities. Nat. Cancer5, 673–690 (2024). [DOI] [PubMed] [Google Scholar]
35.Alkan, H. F. et al. Cytosolic aspartate availability determines cell survival when glutamine is limiting. Cell Metab.28, 706–720.e706 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Li, H. et al. The landscape of cancer cell line metabolism. Nat. Med25, 850–860 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Molina, J. R. et al. An inhibitor of oxidative phosphorylation exploits cancer vulnerability. Nat. Med24, 1036–1046 (2018). [DOI] [PubMed] [Google Scholar]
38.Yap, T. A. et al. Complex I inhibitor of oxidative phosphorylation in advanced solid tumors and acute myeloid leukemia: phase I trials. Nat. Med.29, 115–126 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Sherer, T. B., Kim, J. H., Betarbet, R. & Greenamyre, J. T. Subcutaneous rotenone exposure causes highly selective dopaminergic degeneration and alpha-synuclein aggregation. Exp. Neurol.179, 9–16 (2003). [DOI] [PubMed] [Google Scholar]
40.Siletti, K. et al. Transcriptomic diversity of cell types across the adult human brain. Science382, eadd7046 (2023). [DOI] [PubMed] [Google Scholar]
41.Schulten, H. J. et al. Comprehensive molecular biomarker identification in breast cancer brain metastases. J. Transl. Med15, 269 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Bos, P. D. et al. Genes that mediate breast cancer metastasis to the brain. Nature459, 1005–1009 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Zhang, W. et al. The bone microenvironment invigorates metastatic seeds for further dissemination. Cell184, 2471–2486.e2420 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Brastianos, P. K. et al. Genomic characterization of brain metastases reveals branched evolution and potential therapeutic targets. Cancer Discov.5, 1164–1177 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Lockman, P. R. et al. Heterogeneous blood-tumor barrier permeability determines drug efficacy in experimental brain metastases of breast cancer. Clin. Cancer Res16, 5664–5678 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Lee, K. M. et al. MYC and MCL1 cooperatively promote chemotherapy-resistant breast cancer stem cells via regulation of mitochondrial oxidative phosphorylation. Cell Metab.26, 633–647.e637 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
47.Zhao, L. et al. Lymphatic endothelial-like cells promote glioblastoma stem cell growth through cytokine-driven cholesterol metabolism. Nat. Cancer5, 147–166 (2024). [DOI] [PubMed] [Google Scholar]
48.Wu, M., Gu, J., Guo, R., Huang, Y. & Yang, M. Structure of mammalian respiratory supercomplex I(1)III(2)IV(1). Cell167, 1598–1609.e1510 (2016). [DOI] [PubMed] [Google Scholar]
49.Li, L. D. et al. Down-regulation of NDUFB9 promotes breast cancer cell proliferation, metastasis by mediating mitochondrial metabolism. PLoS One10, e0144441 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
50.Bezwada, D. et al. Mitochondrial complex I promotes kidney cancer metastasis. Nature633, 923–931 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
51.Fiorillo, M., Scatena, C., Naccarato, A. G., Sotgia, F. & Lisanti, M. P. Bedaquiline, an FDA-approved drug, inhibits mitochondrial ATP production and metastasis in vivo, by targeting the gamma subunit (ATP5F1C) of the ATP synthase. Cell Death Differ.28, 2797–2817 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Fischer, G. M. et al. Molecular profiling reveals unique immune and metabolic features of melanoma brain metastases. Cancer Discov.9, 628–645 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Li, J. et al. Safety, efficacy, and serum concentration monitoring of bedaquiline in Chinese patients with multidrug-resistant tuberculosis. Int. J. Infect. Dis.110, 179–186 (2021). [DOI] [PubMed] [Google Scholar]
54.Finkel, R. S. et al. Treatment of infantile-onset spinal muscular atrophy with nusinersen: a phase 2, open-label, dose-escalation study. Lancet388, 3017–3026 (2016). [DOI] [PubMed] [Google Scholar]
55.Chen, W. et al. FAM50A drives breast cancer brain metastasis through interaction with C9ORF78 to enhance ʟ-asparagine production. Sci. Adv.11, eadt3075 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
56.Zhang, L. et al. Blocking immunosuppressive neutrophils deters pY696-EZH2-driven brain metastases. Sci. Transl. Med.12 (2020). [DOI] [PMC free article] [PubMed]
57.Kanaya, N. et al. Gene-edited and -engineered stem cell platform drives immunotherapy for brain metastatic melanomas. Sci. Transl. Med.15, eade8732 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
58.Mashimo, T. et al. Acetate is a bioenergetic substrate for human glioblastoma and brain metastases. Cell159, 1603–1614 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
59.Hagenfeldt, L., Bjerkenstedt, L., Edman, G., Sedvall, G. & Wiesel, F. A. Amino acids in plasma and CSF and monoamine metabolites in CSF: interrelationship in healthy subjects. J. Neurochem.42, 833–837 (1984). [DOI] [PubMed] [Google Scholar]
60.Dolgodilina, E. et al. Brain interstitial fluid glutamine homeostasis is controlled by blood-brain barrier SLC7A5/LAT1 amino acid transporter. J. Cereb. Blood Flow. Metab.36, 1929–1941 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
61.Louveau, A., Harris, T. H. & Kipnis, J. Revisiting the mechanisms of CNS Immune Privilege. Trends Immunol.36, 569–577 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
62.Ferraro, G. B. et al. Fatty acid synthesis is required for breast cancer brain metastasis. Nat. Cancer2, 414–428 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
63.Lim, N. K. et al. An improved method for collection of cerebrospinal fluid from anesthetized mice. J. Vis. Exp. (2018). [DOI] [PMC free article] [PubMed]
64.Wei, Z., Kale, S., El Fatimy, R., Rabinovsky, R. & Krichevsky, A. M. Co-cultures of glioma stem cells and primary neurons, astrocytes, microglia, and endothelial cells for investigation of intercellular communication in the brain. Front Neurosci.13, 361 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
65.Pachnis, P. et al. In vivo isotope tracing reveals a requirement for the electron transport chain in glucose and glutamine metabolism by tumors. Sci. Adv.8, eabn9550 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
66.Nanchen, A., Fuhrer, T. & Sauer, U. Determination of metabolic flux ratios from 13C-experiments and gas chromatography-mass spectrometry data: protocol and principles. Methods Mol. Biol.358, 177–197 (2007). [DOI] [PubMed] [Google Scholar]
67.Li, Y. et al. Proteomic characterization of gastric cancer response to chemotherapy and targeted therapy reveals new therapeutic strategies. Nat. Commun.13, 5723 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
68.Schwanhäusser, B. et al. Global quantification of mammalian gene expression control. Nature473, 337–342 (2011). [DOI] [PubMed] [Google Scholar]
69.Ritchie, M. E. et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res43, e47 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
70.Hänzelmann, S., Castelo, R. & Guinney, J. GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinforma.14, 7 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
71.Jiang, Y. Z. et al. Genomic and transcriptomic landscape of triple-negative breast cancers: subtypes and treatment strategies. Cancer Cell35, 428–440.e425 (2019). [DOI] [PubMed] [Google Scholar]
72.Ghandi, M. et al. Next-generation characterization of the Cancer Cell Line Encyclopedia. Nature569, 503–508 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
73.Chen, Q. et al. Carcinoma-astrocyte gap junctions promote brain metastasis by cGAMP transfer. Nature533, 493–498 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Information^{(9.8MB, pdf)}

Peer Review file^{(1.8MB, pdf)}

41467_2026_72927_MOESM3_ESM.pdf^{(55.6KB, pdf)}

Description of Additional Supplementary Files

Supplementary Data 1^{(11.8KB, xlsx)}

Supplementary Data 2^{(128.2KB, xlsx)}

Supplementary Data 3^{(1.1MB, xlsx)}

Supplementary Data 4^{(32.2KB, xlsx)}

Supplementary Data 5^{(28KB, xlsx)}

Supplementary Data 6^{(16KB, xlsx)}

Supplementary Data 7^{(2MB, xlsx)}

Supplementary Data 8^{(170.1KB, xlsx)}

Reporting Summary^{(112.2KB, pdf)}

Source data^{(5.1MB, xlsx)}

Data Availability Statement

[CR1] 1.Kennecke, H. et al. Metastatic behavior of breast cancer subtypes. J. Clin. Oncol.28, 3271–3277 (2010). [DOI] [PubMed] [Google Scholar]

[CR2] 2.Soffietti, R. et al. Diagnosis and treatment of brain metastases from solid tumors: guidelines from the European Association of Neuro-Oncology (EANO). Neuro Oncol.19, 162–174 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR3] 3.Lee, S. S. et al. Brain metastases in breast cancer: prognostic factors and management. Breast Cancer Res. Treat.111, 523–530 (2008). [DOI] [PubMed] [Google Scholar]

[CR4] 4.Lin, N. U., Bellon, J. R. & Winer, E. P. CNS metastases in breast cancer. J. Clin. Oncol.22, 3608–3617 (2004). [DOI] [PubMed] [Google Scholar]

[CR5] 5.Hurvitz, S. A. et al. Subgroup analyses from the phase 3 ASCENT study of sacituzumab govitecan in metastatic triple-negative breast cancer. NPJ Breast Cancer10, 33 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR6] 6.Lin, M. et al. HER2-low heterogeneity between primary and paired recurrent/metastatic breast cancer: Implications in treatment and prognosis. Cancer130, 851–862 (2024). [DOI] [PubMed] [Google Scholar]

[CR7] 7.Vaz Batista, M. et al. Trastuzumab deruxtecan in patients with previously treated HER2-low advanced breast cancer and active brain metastases: the DEBBRAH trial. ESMO Open9, 103699 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR8] 8.Zhang, L. et al. Microenvironment-induced PTEN loss by exosomal microRNA primes brain metastasis outgrowth. Nature527, 100–104 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR9] 9.Zou, Y. et al. Polyunsaturated fatty acids from astrocytes activate PPARγ signaling in cancer cells to promote brain metastasis. Cancer Discov.9, 1720–1735 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR10] 10.Jin, Y. et al. Targeting polarized phenotype of microglia via IL6/JAK2/STAT3 signaling to reduce NSCLC brain metastasis. Signal Transduct. Target Ther.7, 52 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR11] 11.Hladky, S. B. & Barrand, M. A. Fluid and ion transfer across the blood-brain and blood-cerebrospinal fluid barriers; a comparative account of mechanisms and roles. Fluids Barriers CNS13, 19 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR12] 12.Hladky, S. B. & Barrand, M. A. Elimination of substances from the brain parenchyma: efflux via perivascular pathways and via the blood-brain barrier. Fluids Barriers CNS15, 30 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR13] 13.Jin, X. et al. A metastasis map of human cancer cell lines. Nature588, 331–336 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR14] 14.Kim, A. E., Nieblas-Bedolla, E., de Sauvage, M. A. & Brastianos, P. K. Leveraging translational insights toward precision medicine approaches for brain metastases. Nat. Cancer4, 955–967 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR15] 15.Lorger, M. & Felding-Habermann, B. Capturing changes in the brain microenvironment during initial steps of breast cancer brain metastasis. Am. J. Pathol.176, 2958–2971 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR16] 16.Valiente, M. et al. Serpins promote cancer cell survival and vascular co-option in brain metastasis. Cell156, 1002–1016 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR17] 17.Varkuti, B. H. et al. Neuron-based high-content assay and screen for CNS active mitotherapeutics. Sci. Adv.6, eaaw8702 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR18] 18.Giacomello, M., Pyakurel, A., Glytsou, C. & Scorrano, L. The cell biology of mitochondrial membrane dynamics. Nat. Rev. Mol. Cell Biol.21, 204–224 (2020). [DOI] [PubMed] [Google Scholar]

[CR19] 19.Titov, D. V. et al. Complementation of mitochondrial electron transport chain by manipulation of the NAD+/NADH ratio. Science352, 231–235 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR20] 20.Ngo, B. et al. Limited Environmental Serine and Glycine Confer Brain Metastasis Sensitivity to PHGDH Inhibition. Cancer Discov.10, 1352–1373 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR21] 21.Freret, M. E. & Boire, A. The anatomic basis of leptomeningeal metastasis. J. Exp. Med.221 (2024). [DOI] [PMC free article] [PubMed]

[CR22] 22.Sullivan, L. B. et al. Aspartate is an endogenous metabolic limitation for tumour growth. Nat. Cell Biol.20, 782–788 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR23] 23.Battaglioni, S., Benjamin, D., Wälchli, M., Maier, T. & Hall, M. N. mTOR substrate phosphorylation in growth control. Cell185, 1814–1836 (2022). [DOI] [PubMed] [Google Scholar]

[CR24] 24.Kilberg, M. S., Shan, J. & Su, N. ATF4-dependent transcription mediates signaling of amino acid limitation. Trends Endocrinol. Metab.20, 436–443 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR25] 25.Celardo, I., Lehmann, S., Costa, A. C., Loh, S. H. & Miguel Martins, L. dATF4 regulation of mitochondrial folate-mediated one-carbon metabolism is neuroprotective. Cell Death Differ.24, 638–648 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR26] 26.Wortel, I. M. N., van der Meer, L. T., Kilberg, M. S. & van Leeuwen, F. N. Surviving stress: modulation of ATF4-mediated stress responses in normal and malignant cells. Trends Endocrinol. Metab.28, 794–806 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR27] 27.Schägger, H. & Pfeiffer, K. Supercomplexes in the respiratory chains of yeast and mammalian mitochondria. Embo j.19, 1777–1783 (2000). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR28] 28.D’Angelo, L. et al. NDUFS3 depletion permits complex I maturation and reveals TMEM126A/OPA7 as an assembly factor binding the ND4-module intermediate. Cell Rep.35, 109002 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR29] 29.Vercellino, I. & Sazanov, L. A. The assembly, regulation and function of the mitochondrial respiratory chain. Nat. Rev. Mol. Cell Biol.23, 141–161 (2022). [DOI] [PubMed] [Google Scholar]

[CR30] 30.Zhu, Z., Zhang, Q. & Sui, Z. Screening of ApDOT1.9 interacting proteins and the potential function of interactor ApSNARE in the rapid growth regulation of Alexandrium pacificum. Mar. Pollut. Bull.209, 117080 (2024). [DOI] [PubMed] [Google Scholar]

[CR31] 31.Varešlija, D. et al. Transcriptome Characterization of Matched Primary Breast and Brain Metastatic Tumors to Detect Novel Actionable Targets. J. Natl. Cancer Inst.111, 388–398 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR32] 32.Mullen, A. R. et al. Reductive carboxylation supports growth in tumour cells with defective mitochondria. Nature481, 385–388 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR33] 33.Krug, K. et al. Proteogenomic landscape of breast cancer tumorigenesis and targeted therapy. Cell183, 1436–1456.e1431 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR34] 34.Jiang, Y. Z. et al. Integrated multiomic profiling of breast cancer in the Chinese population reveals patient stratification and therapeutic vulnerabilities. Nat. Cancer5, 673–690 (2024). [DOI] [PubMed] [Google Scholar]

[CR35] 35.Alkan, H. F. et al. Cytosolic aspartate availability determines cell survival when glutamine is limiting. Cell Metab.28, 706–720.e706 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR36] 36.Li, H. et al. The landscape of cancer cell line metabolism. Nat. Med25, 850–860 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR37] 37.Molina, J. R. et al. An inhibitor of oxidative phosphorylation exploits cancer vulnerability. Nat. Med24, 1036–1046 (2018). [DOI] [PubMed] [Google Scholar]

[CR38] 38.Yap, T. A. et al. Complex I inhibitor of oxidative phosphorylation in advanced solid tumors and acute myeloid leukemia: phase I trials. Nat. Med.29, 115–126 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR39] 39.Sherer, T. B., Kim, J. H., Betarbet, R. & Greenamyre, J. T. Subcutaneous rotenone exposure causes highly selective dopaminergic degeneration and alpha-synuclein aggregation. Exp. Neurol.179, 9–16 (2003). [DOI] [PubMed] [Google Scholar]

[CR40] 40.Siletti, K. et al. Transcriptomic diversity of cell types across the adult human brain. Science382, eadd7046 (2023). [DOI] [PubMed] [Google Scholar]

[CR41] 41.Schulten, H. J. et al. Comprehensive molecular biomarker identification in breast cancer brain metastases. J. Transl. Med15, 269 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR42] 42.Bos, P. D. et al. Genes that mediate breast cancer metastasis to the brain. Nature459, 1005–1009 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR43] 43.Zhang, W. et al. The bone microenvironment invigorates metastatic seeds for further dissemination. Cell184, 2471–2486.e2420 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR44] 44.Brastianos, P. K. et al. Genomic characterization of brain metastases reveals branched evolution and potential therapeutic targets. Cancer Discov.5, 1164–1177 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR45] 45.Lockman, P. R. et al. Heterogeneous blood-tumor barrier permeability determines drug efficacy in experimental brain metastases of breast cancer. Clin. Cancer Res16, 5664–5678 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR46] 46.Lee, K. M. et al. MYC and MCL1 cooperatively promote chemotherapy-resistant breast cancer stem cells via regulation of mitochondrial oxidative phosphorylation. Cell Metab.26, 633–647.e637 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR47] 47.Zhao, L. et al. Lymphatic endothelial-like cells promote glioblastoma stem cell growth through cytokine-driven cholesterol metabolism. Nat. Cancer5, 147–166 (2024). [DOI] [PubMed] [Google Scholar]

[CR48] 48.Wu, M., Gu, J., Guo, R., Huang, Y. & Yang, M. Structure of mammalian respiratory supercomplex I(1)III(2)IV(1). Cell167, 1598–1609.e1510 (2016). [DOI] [PubMed] [Google Scholar]

[CR49] 49.Li, L. D. et al. Down-regulation of NDUFB9 promotes breast cancer cell proliferation, metastasis by mediating mitochondrial metabolism. PLoS One10, e0144441 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR50] 50.Bezwada, D. et al. Mitochondrial complex I promotes kidney cancer metastasis. Nature633, 923–931 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR51] 51.Fiorillo, M., Scatena, C., Naccarato, A. G., Sotgia, F. & Lisanti, M. P. Bedaquiline, an FDA-approved drug, inhibits mitochondrial ATP production and metastasis in vivo, by targeting the gamma subunit (ATP5F1C) of the ATP synthase. Cell Death Differ.28, 2797–2817 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR52] 52.Fischer, G. M. et al. Molecular profiling reveals unique immune and metabolic features of melanoma brain metastases. Cancer Discov.9, 628–645 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR53] 53.Li, J. et al. Safety, efficacy, and serum concentration monitoring of bedaquiline in Chinese patients with multidrug-resistant tuberculosis. Int. J. Infect. Dis.110, 179–186 (2021). [DOI] [PubMed] [Google Scholar]

[CR54] 54.Finkel, R. S. et al. Treatment of infantile-onset spinal muscular atrophy with nusinersen: a phase 2, open-label, dose-escalation study. Lancet388, 3017–3026 (2016). [DOI] [PubMed] [Google Scholar]

[CR55] 55.Chen, W. et al. FAM50A drives breast cancer brain metastasis through interaction with C9ORF78 to enhance ʟ-asparagine production. Sci. Adv.11, eadt3075 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR56] 56.Zhang, L. et al. Blocking immunosuppressive neutrophils deters pY696-EZH2-driven brain metastases. Sci. Transl. Med.12 (2020). [DOI] [PMC free article] [PubMed]

[CR57] 57.Kanaya, N. et al. Gene-edited and -engineered stem cell platform drives immunotherapy for brain metastatic melanomas. Sci. Transl. Med.15, eade8732 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR58] 58.Mashimo, T. et al. Acetate is a bioenergetic substrate for human glioblastoma and brain metastases. Cell159, 1603–1614 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR59] 59.Hagenfeldt, L., Bjerkenstedt, L., Edman, G., Sedvall, G. & Wiesel, F. A. Amino acids in plasma and CSF and monoamine metabolites in CSF: interrelationship in healthy subjects. J. Neurochem.42, 833–837 (1984). [DOI] [PubMed] [Google Scholar]

[CR60] 60.Dolgodilina, E. et al. Brain interstitial fluid glutamine homeostasis is controlled by blood-brain barrier SLC7A5/LAT1 amino acid transporter. J. Cereb. Blood Flow. Metab.36, 1929–1941 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR61] 61.Louveau, A., Harris, T. H. & Kipnis, J. Revisiting the mechanisms of CNS Immune Privilege. Trends Immunol.36, 569–577 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR62] 62.Ferraro, G. B. et al. Fatty acid synthesis is required for breast cancer brain metastasis. Nat. Cancer2, 414–428 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR63] 63.Lim, N. K. et al. An improved method for collection of cerebrospinal fluid from anesthetized mice. J. Vis. Exp. (2018). [DOI] [PMC free article] [PubMed]

[CR64] 64.Wei, Z., Kale, S., El Fatimy, R., Rabinovsky, R. & Krichevsky, A. M. Co-cultures of glioma stem cells and primary neurons, astrocytes, microglia, and endothelial cells for investigation of intercellular communication in the brain. Front Neurosci.13, 361 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR65] 65.Pachnis, P. et al. In vivo isotope tracing reveals a requirement for the electron transport chain in glucose and glutamine metabolism by tumors. Sci. Adv.8, eabn9550 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR66] 66.Nanchen, A., Fuhrer, T. & Sauer, U. Determination of metabolic flux ratios from 13C-experiments and gas chromatography-mass spectrometry data: protocol and principles. Methods Mol. Biol.358, 177–197 (2007). [DOI] [PubMed] [Google Scholar]

[CR67] 67.Li, Y. et al. Proteomic characterization of gastric cancer response to chemotherapy and targeted therapy reveals new therapeutic strategies. Nat. Commun.13, 5723 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR68] 68.Schwanhäusser, B. et al. Global quantification of mammalian gene expression control. Nature473, 337–342 (2011). [DOI] [PubMed] [Google Scholar]

[CR69] 69.Ritchie, M. E. et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res43, e47 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR70] 70.Hänzelmann, S., Castelo, R. & Guinney, J. GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinforma.14, 7 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR71] 71.Jiang, Y. Z. et al. Genomic and transcriptomic landscape of triple-negative breast cancers: subtypes and treatment strategies. Cancer Cell35, 428–440.e425 (2019). [DOI] [PubMed] [Google Scholar]

[CR72] 72.Ghandi, M. et al. Next-generation characterization of the Cancer Cell Line Encyclopedia. Nature569, 503–508 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR73] 73.Chen, Q. et al. Carcinoma-astrocyte gap junctions promote brain metastasis by cGAMP transfer. Nature533, 493–498 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Complex I protein NDUFB9 is a metabolic vulnerability in triple negative breast cancer brain metastases

Mingxi Lin

Zhexu Wen

Cheng Zeng

Yizi Jin

Teng Zhou

Yuxin Yan

Shenglin Huang

Xin Hu

Xiaoxiang Guan

Xichun Hu

Jian Zhang

Abstract

Introduction

Results

High-throughput in vivo loss-of-function CRISPR screens identify potential treatment targets of TNBC BrMs

Fig. 1. In vivo high-throughput loss-of-function CRISPR screens with TNBCBM-CK library identify the candidate treatment targets of TNBC BrMs.

NDUFB9-knockout significantly inhibits TNBC BrMs

Fig. 2. NDUFB9-knockout reduces TNBC post-colonization growth in the brain.

NDUFB9-knockout impairs TNBC post-colonization outgrowth in the brain

NDUFB9-knockout specifically inhibits TNBC BrMs without affecting extracranial tumors

NDUFB9-knockout disrupts respiratory chain complex I and reduces intracellular aspartate levels

Fig. 3. NDUFB9-knockout disrupts respiratory chain function and reduces intracellular aspartate levels.

Comprehensively metabolic microenvironment atlas of TNBC brain metastases

Fig. 4. Comprehensively metabolic atlas reveals the brain as an asparagine-limited microenvironment.

NDUFB9-knockout selectively impairs TNBC BrMs due to intracranial–extracranial differences in asparagine rather than aspartate availability

The upregulation of asparagine synthetase induced by asparagine-limited brain environment, coupled with aspartate reduction induced by NDUFB9-knockout, synergistically restrains nucleotide biosynthesis

Fig. 5. NDUFB9-knockout and low extracellular asparagine synergistically restrain nucleotide de novo biosynthesis and inhibit TNBC proliferation.

Structural analyses reveal differences between NDUFB9 and other complex I subunits

TNBC BrMs increase aspartate biosynthesis in response to asparagine-limited microenvironment

Fig. 6. TNBC BrMs increase aspartate biosynthesis in response to asparagine-limited microenvironment.

TNBC exhibits high sensitivity to NDUFB9-knockout due to the imbalance between aspartate upstream substrate supply and downstream biosynthesis demand

Fig. 7. Pharmacologic inhibition of NDUFB9 is an effective therapeutic strategy for TNBC BrMs.

Pharmacologic inhibition of NDUFB9 is a therapeutic strategy for TNBC BrMs with reduced toxicity

Discussion

Methods

Cell lines and cell culture

Clinical samples

Animals

Mouse brain metastases model by intracardiac injection

Stereotactic brain injections of breast cancer cells

TNBCBM-CK sgRNA CRISPR screening library design

Oligonucleotide pool synthesis and TNBCBM-CK sgRNA library cloning

TNBCBM-CK sgRNA library lentivirus production and transduction

CRISPR screening and MAGeCK analysis

Animal studies

Analysis of mitochondrial complexes

Collection of paired cerebrospinal fluid and plasma of patients

Microdissection of patient-matched primary breast tumors and BrMs

Western blot

Immunofluorescence and immunohistochemical staining

Oxygen consumption rate (OCR) and ATP rate assay

Cell death assay

Electron microscopy

Mitochondrial DNA copy number

Mitochondrial mass and membrane potential

Co-culture experiments

Proliferation and cell viability assays

Stable isotope tracing

Metabolites measurement

Isolation of cancer cells from tumors and bulk RNA-seq

RNA-seq of patient-matched primary breast tumors and BrMs

Proteomics of patient-matched primary breast tumors and BrMs

NAD⁺/NADH levels

Pathway enrichment analyses

Antisense oligonucleotides

External datasets analysis

Gene knockdown and overexpression

Statistics and reproducibility

Reporting summary

Supplementary information

Source data

Acknowledgements

Author contributions

Peer review

Peer review information

Data availability

Competing interests

Footnotes