2.1 BDH1 is strongly associated with poor prognosis in patients with LUAD

We assessed the Cancer Genome Atlas (TCGA) dataset to characterize BDH1 expression in LUAD and lung squamous cell carcinoma (LUSC) tissues compared with adjacent normal tissues. BDH1 mRNA expression levels were significantly upregulated in TCGA–LUAD and TCGA–LUSC, relative to adjacent normal tissues (Figure 1A). The relationship between BDH1 expression levels and survival in patients with LUAD and LUSC was analyzed using Kaplan–Meier (KM) (http://kmplot.com/analysis/). The results revealed significantly lower overall survival (OS) (p = 0.017) and first progression (FP) (p = 0.027) in the high BDH1 expression group in LUAD, as opposed to the low expression group. Conversely, in LUSC, patients with lower BDH1 expression exhibited poorer OS than those with higher BDH1 expression, although this was not statistically significant (Figure 1B). Based on these findings, we chose to further investigate the expression and role of BDH1 in LUAD. To confirm BDH1 expression levels in LUAD, we assessed BDH1 protein levels in 30 pairs of human clinical LUAD and adjacent normal tissues using immunohistochemistry (IHC). This revealed a significant increase in BDH1 expression in LUAD tissues, compared with adjacent normal tissues (Figure 1C). These results suggest that the upregulated expression of BDH1 in LUAD may be carcinogenic and strongly correlated with survival in patients with LUAD.

2.2 BDH1 promotes the malignant phenotype of LUAD cells

To delve deeper into the role of BDH1 in LUAD, we selected six lung cancer cell lines and HBE cells to measure BDH1 expression levels. Our findings demonstrated high endogenous expression of BDH1 in several lung cancer cell lines (Figure S1A). Given the relatively low expression of BDH1 in A549 cells, we stably introduced exogenous overexpression of BDH1 expression in this cell. Additionally, because of higher expression levels of BDH1 in PC9 and SPCA-1 cells, we employed three sgRNAs (Table 1) to establish stable knockout of BDH1 in these cell lines. Subsequently, western blotting (WB) confirmed a significant increase in BDH1 expression in s stable overexpression cell lines (Figure 2A), with approximately 100% knockout efficiency achieved for BDH1 KO#1 and BDH1 KO#2 in the knockout cell lines (Figures 2A and S1B). We examined the proliferative capacity of cells overexpressing or lacking BDH1 using CCK-8 and clone formation assays. The results indicated that cell proliferation activity was substantially heightened in the BDH1 overexpression group compared with the vector control group, while it was markedly reduced in the BDH1 knockout group relative to the control group (Figures 2B–C and S1C–D). The overexpression of BDH1 led to the emergence of more stem cell-like characteristics in the tumor sphere-forming assay, whereas fewer stem cell-like characteristics were observed in the BDH1 knockout group (Figures 2D and S1E).

We conducted further assessments of cellular migration and invasion abilities using Transwell migration, Transwell invasion, and wound healing assays. These experiments revealed that BDH1-overexpressing A549 cells exhibited heightened migration and invasion capabilities, while BDH1 knockout in PC9 and SPCA-1 cells led to the inhibition of these processes (Figures 2E–F and S1F–G). Additionally, quantitative reverse transcription-polymerase chain reaction (qRT-PCR) demonstrated significantly reduced expression levels of epithelial markers (ZO-1 and E-cadherin) and elevated expression levels of mesenchymal markers (ZEB1 and vimentin) in the BDH1 overexpression group compared with the vector group. Conversely, the BDH1 KO#2 group showed a contrasting trend (Figures 2G and S1H). These findings collectively suggest that BDH1 promotes the malignant phenotype of LUAD cells, influencing their proliferation, stem cell-like characteristics, migration, and invasion.

2.3 BDH1 inhibits the expression of LRRC31 to affect LUAD progression

To gain deeper insights into the potential molecular mechanisms of BDH1 in LUAD, we employed RNA sequencing (RNA-seq) to analyze the downstream target genes of BDH1. The volcano plots illustrated the presentation of 313 differentially expressed genes (DEGs) in BDH1-knockout PC9 cells versus control cells (Figure S2A; |log₂FC| > 2, P.adj < 0.001). Subsequently, univariate Cox regression analysis was performed on BDH1 and the DEGs with |log₂FC| > 4 and P.adj < 0.001 to identify 15 prognostic-related genes (|HR| > 1, p < 0.01) (Figure S2B). Following this, we performed LASSO analysis on these 15 prognosis-related genes. The results indicated that bias was minimized with a model containing nine genes (Figure S2C). Therefore, we selected the following candidate genes for constructing the prognostic model: BDH1, PRSS12, PTPN13, LRRC31, ETV1, PPP1R9A, CACHD1, CD33, and PABPC4L (Figure S2D). An equation was formulated to calculate the risk score for each patient: risk scores = −0.128 × ETV1 − 0.169 × CACHD1 − 0.058 × PTPN13 − 0.114 × LRRC31 − 0.012 × PRSS12 − 0.118 × CD33 − 0.323 × PABPC4L − 0.044 × PPP1R9A + 0.149 × BDH1. We employed GSE72094 (n = 349) as the training cohort and TCGA–LUAD (n = 375) as the validation cohort for the analysis of our prognostic model. Within the training cohort, patients were categorized into high- and low-risk groups based on their median risk scores. We found that patients in the high-risk group exhibited markedly lower OS than those in the low-risk group (Figures S2E–F). The reliability of our prediction model was confirmed through risk score-based time-dependent receiver operating characteristics (ROC) curve analysis, which yielded area under the curve (AUC) values of 0.75, 0.75, and 0.78 at 1, 3, and 5 years, respectively (Figure S2G). These results underscore the robust predictive efficacy of our prognostic model for anticipating the survival of patients with LUAD. Furthermore, we stratified the samples into three groups based on the clinical stage of patients with LUAD (stage I, II, and III+IV) and observed that stage I exhibited the lowest risk score (Figure S2H). In the validation cohort, we observed that the results of the heat map visualization, KM curve, and clinical staging risk score distribution mirrored those of the training cohort. The AUC of the ROC curve reached 0.72, 0.69, and 0.70 at 1, 3, and 5 years, respectively, further validating the external applicability of our model (Figures S2I–L). These findings demonstrate that our model can swiftly identify high-risk groups associated with BDH1 expression, enabling a timely assessment of the patient's future disease risk.

To further understand the molecular mechanisms driving BDH1-mediated promotion of LUAD progression, we conducted qRT-PCR analysis to evaluate the expression levels of key genes following overexpression or knockout of BDH1. The results showed that overexpression of BDH1 led to a 0.515 decrease in LRRC31 expression, while CD33, PTPN13, and CACHD1 exhibited decreases of 0.269, 0.202, and 0.172, respectively. Among the eight target genes investigated in this study for their association with BDH1 overexpression, LRRC31 displayed the most significant fold reduction in expression levels (Figure S3A). LRRC31, characterized by its nine leucine-rich repeats, is induced by interleukin 13 to regulate esophageal epithelial barrier function in eosinophilic esophagitis.³⁰ In addition, LRRC31 has been identified as a master regulator of the nonhomologous end-joining mechanism and plays a central role in breast cancer radiotherapy.³¹ However, its expression and role in LUAD remain unclear. Patients in GSE72094 and TCGA–LUAD were stratified into high-expression and low-expression groups based on the median expression of LRRC31. KM survival analysis indicated that the high-expression group exhibited markedly improved OS compared with the low-expression group, signifying a favorable prognosis (Figure S3B). We employed qRT-PCR and WB techniques to assess the expression of LRRC31 in various lung cancer cell lines, revealing that LRRC31 was expressed at lower levels across different lung cancer cell lines (Figures S3C–D). Additionally, IHC found lower expression of LRRC31 in LUAD tissues compared with adjacent normal tissues (Figure 3A). Subsequently, we conducted a correlation analysis of BDH1 and LRRC31 expression in LUAD tissues, revealing a negative correlation between BDH1 and LRRC31 in the LUAD tissue (R = 0.432, p = 0.018) (Figure S3E). qRT-PCR and WB demonstrated that both the mRNA and protein expression of LRRC31 decreased in the BDH1-overexpressing group compared with the vector cells, but increased in the BDH1 KO#2 group (Figures S3A and 3B). These findings strongly suggest that LRRC31 may function as a downstream target gene regulated by BDH1.

To elucidate the role of LRRC31 in BDH1-mediated promotion of LUAD, we successfully achieved overexpression of LRRC31 in A549 Vector/BDH1 OE cells, while knockdown of LRRC31 was accomplished in PC9 BDH1 KO#2 cells (Figures 3C and S4A). Through a battery of assays including CCK-8, clone formation, tumor sphere formation, Transwell migration, Transwell invasion, and wound healing, we demonstrated that overexpression of LRRC31 significantly suppressed the promotion of cell proliferation, stem cell-like characteristics, migration, and invasion induced by BDH1 overexpression. Conversely, the knockdown of LRRC31 reversed the inhibitory effects of BDH1 knockout on these cellular processes (Figures 3D–H and S4B–F). These findings strongly suggest that LRRC31 plays a pivotal role in BDH1 regulation of LUAD.

2.4 BDH1 affects LUAD progression by regulating Kbhb modification

In ketone body metabolism, BDH1 is the primary rate-limiting enzyme responsible for converting AcAc and BHB.¹⁰ Given the ability of BHB to mediate the occurrence of Kbhb,¹⁹ we investigated BDH1's effect on Kbhb. WB revealed that overexpression of BDH1 led to a downregulation of Kbhb levels, and vice versa (Figure 4A). Subsequently, A549 Vector/BDH1 OE cells were exposed to varying concentrations of Na-BHB. The results demonstrated that Na-BHB was capable of reversing the attenuating effect of BDH1 overexpression on Kbhb levels, with Kbhb levels increasing in tandem with Na-BHB concentration (Figure 4B). These findings suggest that BDH1/BHB can influence Kbhb modification in LUAD cells. To further explore the potential effect of BHB on LUAD, we conducted a series of functional experiments. We observed that A549 cells overexpressing BDH1 exhibited a significant decrease in cell proliferation, stem cell-like characteristics, as well as migration and invasion abilities after Na-BHB treatment (Figures 4C–G). This indicates that Na-BHB exerts an effective tumor-suppressive effect. Given the reduction in Kbhb levels after BDH1 overexpression, the subsequent rise in Kbhb with the addition of Na-BHB, and the noted tumor-inhibitory effect of Na-BHB, it can be inferred that BDH1 contributes to LUAD progression through Kbhb modification.

2.5 H3K9bhb is involved in LRRC31 regulation by BDH1 in LUAD

A previous study has reported that Kbhb can participate in various physiological processes, including gene expression regulation.²⁰ As BDH1 can regulate the occurrence of Kbhb modifications and LRRC31 is a downstream target gene of BDH1, we sought to understand whether BDH1 mediates LRRC31 expression through Kbhb. We observed an increase in LRRC31 expression after Na-BHB treatment of A549 Vector/BDH1 OE and PC9 Cas9 Control/BDH1 KO#2 cells (Figures 5A–B), suggesting that BDH1 regulates LRRC31 expression through Kbhb. Literature has previously reported that H3K9bhb can be enriched in the gene TSS to promote transcription.¹⁹ To confirm whether BDH1 affects the enrichment of H3K9bhb in the TSS of LRRC31, we performed a chromatin immunoprecipitation (ChIP) analysis. The results showed that H3K9bhb was less enriched in the TSS of LRRC31 in A549 cells overexpressing BDH1 than in vector cells, while it was more enriched in the TSS of LRRC31 in PC9 cells with BDH1 knockout compared with control cells (Figure 5C). Furthermore, the enrichment of H3K9bhb in the TSS of LRRC31 was higher after Na-BHB treatment (Figure 5C). In conclusion, our findings demonstrate that BDH1 regulates LRRC31 expression in a H3K9bhb-dependent manner.

2.6 BDH1 promotes the growth of LUAD transplantation tumors

Through in vivo experiments, A549 Vector/BDH1 OE or PC9 Cas9 Control/BDH1 KO cells were subcutaneously injected into nude mice to establish xenograft models. The body weight of the nude mice was continuously monitored, and it was observed that the body weight of all five groups of nude mice steadily increased over 39 days following the injection of tumor cells, with no significant differences between the groups (Figure 6A). Compared with the vector group, the BDH1 overexpression group exhibited significantly larger tumors within the first 39 days after tumor cell injection, whereas a contrasting trend was observed in the BDH1 knockout group (Figure 6B). After 39 days of tumor cell injection, tumors from all five groups of nude mice were collected for size and weight comparison. As depicted in Figure 6C, we noted a substantial increase in both size and weight of tumors in the BDH1 overexpression group. Conversely, the tumor size and weight were significantly reduced in the BDH1 KO group. Subsequently, we conducted paraffin fixation and sectioning of the collected tumors. IHC was performed to detect the expression of BDH1 and LRRC31 in the tumors. The results revealed a significant increase in positive staining for BDH1 in the BDH1 overexpression group, along with a notable reduction in positive LRRC31 staining. In the BDH1 knockout group, BDH1 staining was predominantly negative, while positive LRRC31 staining was significantly increased (Figure 6D). This further validates the efficacy of BDH1 overexpression and knockout, as well as the suppressive influence of BDH1 on LRRC31. Collectively, the above data strongly indicate that BDH1 assumes a crucial oncogenic role in vivo, actively promoting the growth of LUAD cell xenografts.

2.7 Pimozide and crizotinib targeting BDH1 can synergistically inhibit the proliferation of BDH1-high expressing LUAD cells

At present, effective inhibitors targeting BDH1 in LUAD have not been validated. To identify potential BDH1 inhibitors, we conducted a receptor-based virtual screening utilizing AutoDock Vina, employing small molecule compounds from the ZINC database as ligands. The top 5 inhibitors with the highest docking scores were identified as pimozide, lumacaftor, nebivolol, eltrombopag, and crizotinib (Table S1). To analyze the interaction between BDH1 and the small molecule compounds with the top five docking scores, molecular dynamics (MD) simulations were performed using Gromacs. Root mean square deviation and root mean square fluctuation were computed based on the simulation trajectories. The results indicated that the carbon skeleton of BDH1, when docked with pimozide or crizotinib, exhibited acceptable fluctuations throughout the simulation, maintaining equilibrium until the end of the process. Furthermore, the small molecule ligands (pimozide or crizotinib) did not diffuse away from their initial binding sites and their conformations remained stable within 0.1 nm (Figures S5A–B), confirming the reliability of the molecular docking and MD simulation outcomes. Additionally, in both the BDH1–pimozide and BDH1–crizotinib complexes, a specific region (atoms 3500−4500) displayed increased flexibility. In both complexes, there was less fluctuation at the C-terminus relative to the N-terminal residues (Figures S5C–D). These findings suggest that BDH1 can form stable interactions with pimozide or crizotinib. Upon visualizing BDH1–pimozide complexes and BDH1–crizotinib complexes using Pymol, it became evident that the binding site of pimozide or crizotinib to BDH1 is adjacent to the site where the substrate BHB binds to BDH1. Additionally, the BDH1–pimozide complex forms two hydrogen bonds through active residues SER194A and LYS212A, while the BDH1–crizotinib complex forms seven hydrogen bonds involving reactive residues SER65A, PHE67A, GLY68A, ASN144A, and GLY239A (Figures 7A and Table S2–S7). These results strongly suggest that pimozide and crizotinib may serve as potential inhibitors of BDH1.

We exposed BDH1 highly expressed PC9 and SPCA-1 cells to pimozide or crizotinib for 48 h and determined their respective IC₅₀ values. The IC50 of pimozide for PC9 and SPCA-1 was 22.59 and 17.26 μM, respectively. Crizotinib exhibited IC₅₀ values of 3.25 μM for PC9 and 1.39 μM for SPCA-1 (Figure 7B). Upon treating LUAD cells with pimozide or crizotinib, we observed a gradual downregulation of BDH1 expression with increasing treatment time or concentration, indicating a significant inhibitory effect of these drugs on BDH1 expression (Figure 7C). Furthermore, the treatment of PC9 and SPCA-1 cells with pimozide or crizotinib for 48 h led to an upregulation of LRRC31 and H3K9bhb expression compared with the control group (Figures S6A–B). ChIP analysis unveiled a higher enrichment of H3K9bhb in the LRRC31 TSS following treatment with pimozide or crizotinib (Figure S6C), implying that these compounds exert inhibitory effects on the BDH1/H3K9bhb/LRRC31 pathway.

Crizotinib is a US FDA-approved first-line agent for the treatment of patients with locally advanced or metastatic NSCLC who are ALK-positive.³² Therefore, we sought to determine whether pimozide could synergize with crizotinib to enhance the cytotoxic effect on LUAD cells. PC9 and SPCA-1 cells were treated with fixed concentrations of crizotinib (PC9: 4 μM, SPCA-1: 2 μM) in combination with different concentrations of pimozide for 48 h. Subsequently, we calculated the cell survival and combination index. The drug combination index was consistently lower than “1” at various concentrations of pimozide in combination with crizotinib, indicating a synergistic effect between crizotinib and pimozide in the treatment of LUAD cells with high expression of BDH1 (Figure 7D). We further treated PC9 with a combination of 20 μM pimozide and 4 μM crizotinib, and SPCA-1 with a combination of 20 μM pimozide and 2 μM crizotinib. The results demonstrated a significantly reduced cell survival rate and a markedly increased proliferation inhibition rate after the combined treatment compared with single-drug treatment (Figure 7E). In conclusion, we have identified that pimozide and crizotinib target BDH1 and work synergistically to inhibit the growth of BDH1-high expressing LUAD cells.

5.1 Data collection

Gene expression data along with corresponding clinical information for LUAD and LUSC were obtained from the TCGA database (https://portal.gdc.cancer.gov). RNA-seq data and clinical follow-up details for patients with LUAD in the GSE72094 dataset were accessed from the Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo). After excluding patients with one or more unavailable clinical characteristics, all datasets were included in the subsequent analyses (Table S8).

5.2 IHC analysis

Biopsies of 30 pairs of LUAD tissues and adjacent normal tissues were provided and verified by the Department of Pathology at Xiangya Hospital of Central South University (Table S8). IHC was performed following a previously described method.⁵² The antibodies used were BDH1 (67448-1-Ig; Proteintech), LRRC31 (CSB-PA744274LA01HU; CUSABIO), and horseradish peroxidase (HRP)-labeled secondary antibodies (mouse: G1214, rabbit: G1213; Servicebio). Subsequently, images were captured using a ZEISS microscope (Germany) and analyzed by determining the average optical density of the slices using ImagePro Plus 6.0.

5.3 Cell culture, plasmids, and sgRNA

Human embryonic kidney 293T cells (American Type Culture Collection [ATCC]: CRL-3216) and lung cancer cell lines A549 (ATCC: CCL-185), H358 (ATCC: CRL-5807), HCC827 (ATCC: CRL-2868), and H520 (ATCC: HTB-182) were obtained from ATCC. HBE cells and lung cancer cell lines PC9 and SPCA-1 were acquired from the Institute of Oncology at Central South University. All cells were cultured in a medium supplemented with 10% fetal bovine serum, with specific mediums used for different cell lines. DMEM/F12 (1:1) (Gibco) was utilized for A549 cells, DMEM (Gibco) was used for 293T and HBE cell lines, and RPMI 1640 (Gibco) was used for H358, HCC827, H520, PC9, and SPCA-1 cell lines. All cells were maintained in a cell culture incubator at 37°C with 5% CO_2.

The BDH1 overexpression plasmid, sourced from WZ Biosciences (China), was constructed by inserting BDH1 cDNA into the pCDH-CMV-MCS-EF1-Puro vector. The sgRNA vector targeting BDH1 and the corresponding control were procured from General Biol (China). The specific sgRNA oligo targeting BDH1 used in this study is outlined in Table 1.

Similarly, the LRRC31 overexpression plasmid was obtained from the MiaoLing Plasmid Platform (China) and was generated by inserting LRRC31 cDNA into the pLV3-CMV-EF1-Neo vector. The shRNA vector targeting LRRC31 and the control were purchased from GeneChem (China). The shRNA sequences targeting LRRC31 used in this research are detailed in Table S9. Plasmid transfection was executed employing ExFect transfection reagent (Vazyme, China) by the manufacturer's instructions. Viral solutions were collected for cell infection, followed by screening for stable expression colonies using puromycin or G418.

5.4 Quantitative reverse transcription-polymerase chain reaction

The experimental protocol was conducted as previously described,⁵³ and the primers for this experiment were sourced from Biotech. The sequences for all primers are indicated in Table S10.

5.5 Western blotting

WB was carried out following established procedures.⁵⁴ The following antibodies were utilized: β-actin antibody (AF7018; Affinity), BDH1 antibody (67448-1-Ig; Proteintech), LRRC31 antibody (CSB-PA744274LA01HU; CUSABIO), Kbhb antibody (PTM-1201; PTMBIO), H3 antibody (17168-1-AP; Proteintech), Tubulin antibody (AF7010; Affinity), H3K9bhb antibody (PTM-1250; PTMBIO), goat antirabbit IgG (H+L) HRP (S0001; Affinity), and goat antimouse IgG (H+L) HRP (S0002; Affinity).

5.6 CCK-8 assay, colony formation assay, tumor sphere formation assay, wound healing assay, Transwell migration, and invasion assay

These methodologies have been comprehensively detailed in previous studies, and in this study, they were performed in line with established protocols.^53-56

5.7 RNA sequencing

Total RNA from PC9 BDH1 KO#2 and control cells was collected and subjected to RNA sequencing, which was performed by Majorbio Biopharmaceutical Technology (China).⁵³

5.8 Prognostic model construction and validation

Screening for DEGs between the PC9 BDH1 KO#2 cells and the control was performed using the R package “DEGseq” (|log₂FC| > 2, P.adj < 0.001). Subsequently, the R package “pheatmap” was utilized to generate a volcano plot for visualizing the results of the DEGs. Furthermore, univariate Cox regression analysis of the DEGs was conducted to evaluate their predictive significance for patient OS. Candidate genes were selected through LASSO analysis for constructing a prognostic risk model. Risk scores for each sample were computed using the following formula: risk scores = sum (expression of each candidate gene × corresponding LASSO regression coefficient).⁵³ The GSE72094 dataset served as the training cohort, while TCGA–LUAD was employed as the validation cohort for analyzing the prognostic model. Prognostic heat maps were generated using the R packages “ggrisk” and “pheatmap” to illustrate the relationship between the prognostic model and the gene expression within the model. Subsequently, KM survival curves were constructed using the R package “survival”. The “pROC” R package was employed to carry out time-dependent ROC analysis and compute the AUC values for 1, 3, and 5 years. Additionally, the “ggplot2” R package was utilized to present the risk scores of samples grouped by clinical stage in a violin plot.

5.9 Chromatin immunoprecipitation

ChIP experiments were conducted essentially as previously described.⁵² Antibody–protein complexes were captured using pre-blocked protein G magnetic beads (10004D, Thermo Fisher Scientific, USA) and subsequently immunoprecipitated with either IgG (AC005; Abclonal) or H3K9bhb (PTM-1250; PTMBIO). The primer sequences employed in this assay are listed in Table S10.

5.10 Nude mice xenograft lung tumor model

Four-week-old female BALB/c nude mice were procured from Hunan SJA Laboratory Animal Company (China). Animal experiments were conducted with the approval of the Biomedical Research Ethics Committee of Hunan Normal University. Subsequently, BDH1-overexpressing A549 or BDH1-knockout PC9 cells, along with their corresponding control cells, were subcutaneously injected into the axilla of each nude mice (2 × 106 cells per nude mice). Thereafter, the tumor volume and body weight of the nude mice were measured every 3 days. After stripping the tumors and recording their weights, the tumor tissues were fixed and subjected to IHC staining.

5.11 Virtual screening and MD

The crystal structures of BDH1 were retrieved from the UniProt database (https://www.uniprot.org), while the small molecule compounds were obtained from the ZINC database (https://zinc15.docking.org/). AutoDock Vina was employed to add hydrogen atoms to both BDH1 and the small molecule compounds. This process optimized hydrogen bonding and minimized binding energy.⁵⁷ Subsequently, MD simulations of the BDH1–compound complexes were performed using Gromacs.⁵⁸ Considering the balance between simulation accuracy and computational resources, MD simulations were conducted for a duration of 50 ns. Finally, the complexes of BDH1 with the compounds were visualized using Pymol.⁵⁹

5.12 Statistical analysis

The experiments were replicated at least three times, except for those involving nude mice. Data are presented as mean ± standard deviation. Student's t-test was applied for comparisons between groups. Data analysis was performed using R (version 4.2.0) or GraphPad Prism 9.0, and p values less than 0.05 were deemed statistically significant (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001).