SOX1 acts as a tumor hypnotist rendering nasopharyngeal carcinoma cells refractory to chemotherapy

SOX1, a well-known tumor suppressor, delays malignant progression in most cancer types. However, high expression of SOX1 in late-stage head and neck squamous cell carcinoma leads to poor prognosis. In this study, we show that SOX1 induces nasopharyngeal carcinoma (NPC) cells to enter a quiescent state. Using a model that mimics therapeutic resistance and tumor recurrence, a subpopulation of SOX1-induced NPC cells is refractory to paclitaxel, a cell cycle-specific chemotherapy drug. These cells maintain a quiescent state with decreased translational activity and down-regulated cell growth potential. However, once SOX1 expression is decreased, the NPC cells recover and enter a proliferative state. The chemotherapy resistance induced by SOX1 can not pass to next generation, as the cells that undergo re-proliferation become sensitive to paclitaxel again. Moreover, SOX1 directly binds to the promoter region of the MYC gene, leading to transcriptional suppression. When switching to a paclitaxel-free culture environment, the cells with decreased levels of SOX1 re-express MYC, resulting in increased abundance of proliferative cancer cells. Our study presents an evolutionary trade-off between tumor growth and chemoresistance orchestrated by SOX1-MYC in NPC. Basing on the dynamic role of SOX1 in different stages of cancer development, SOX1 would be regarded as a “tumor hypnotist”.


Introduction
Nasopharyngeal carcinoma (NPC) is endemic in Southern China and Southeast Asian countries. Concurrent chemoradiotherapy is one of the main treatment methods for NPC. Although combined chemoradiotherapy yields satisfactory survival rates, chemoresistance presents a major obstacle to the cure of patients with recurrent NPC. About 10-20% of patients experience recurrence after the rst treatment [1], while 70-80% of the recurrent NPC were locally advanced [2]. Therefore, it is of great signi cance to gain a deep understanding of the mechanism of NPC chemoresistance.
At present, it is generally believed that quiescent cancer cells (QCCs) are the "culprit" of NPC recurrence [3]. QCC can escape drug damage, which is conducive to its long-term retaining and rapid proliferation in a speci c environment [4]. QCC also leads to metastasis and complicated treatment implications [5]. To resolve this challenging clinical problem, lots of studies focus on the initiation and termination mechanisms of quiescence [6].
The induction of QCC depends on intrinsic as well as extracellular signals. Researchers use several intracellular markers to detect the quiescent state of cancer cells, such as Ki-67, c-Myc, Cyclin D, p27Kip1, and Rb, while most models in vitro mimic QCCs through changing tumor micro-environment, such as nutrient deprivation, hypoxia, contact inhibition, or anticancer treatment [7]. Whether there are intrinsic factors that dominate QCCs in a stressfree environment is still unknown. Due to the lack of suitable experimental models, the development of studies on QCCs is relatively limited.
SOX1 is a gene that encodes a transcription factor and functions primarily in neurogenesis. SOX1 acted as a tumor suppressor in most non-brain cancers, including NPC that we previously reported [8]. In the current study, we report a de novo model that mimics QCC by overexpression of the intrinsic factor SOX1 in NPC cells. The terminology "tumor suppressor" gives the impression that it's a disadvantage factor for tumor cells. However, some of them additionally play an important role in cancer survival under therapeutical interventions. Due to the dynamic roles in different tumor micro-environment, we suggest rede ning these genes as "tumor hypnotists".

Materials And Methods
In silico analysis Datasets of RNA expression for various human cancers were acquired from the pan-cancer atlas based on TCGA (The Cancer Genome Atlas) database (https://gdc.cancer.gov/about-data/publications/pancanatlas). The mutation characteristics of SOX1 and clinicopathological features in different tumors were explored via the cBioPortal database (https://www.cbioportal.org) [9]. Survival analysis was performed by the online website tool GEPIA (Gene Expression Pro ling Interactive Analysis, http://gepia.cancer-pku.cn/index.html) [10]. Values of SOX1 expression were extracted and violin plot was illustrated by GraphPad Prism software (version 9.0.0).

Plasmid constructs
The plasmids encoding human SOX1 and its mutant were generated by PCR ampli cation and subcloned into the pLVX-TRE3G expression vector using the ClonExpress II One Step Cloning Kit or ClonExpress MultiS One Step Cloning Kit (Vazyme, C112-02 or C113-02) according to the manufacturer's instructions, which was named "pLVX-TRE-SOX1". The primers used for gene cloning are listed in our previous work [11]. Lentiviral production, Infections, and Cell line generation Lentivirus was produced by transient transfection by Lipofectamine 2000 (Invitrogen, 11668019) in human embryonic kidney (HEK) 293T cells using a second-generation lentiviral vector system. All virus-containing medium was mixed with 8 µg/ml polybrene (Sigma-Aldrich, H9268). Viruses produced from pLVX-Tet3G plasmid were used to infect wild-type CNE2 or HONE1 cells to construct HONE1-Tet-On or CNE2-Tet-On stable cell lines. Cells were selected in 1 mg/ml G418 for at least 2 weeks. Subsequently, HONE1-Tet-On and CNE2-Tet-On cell lines were infected by viruses produced from pLVX-TRE-SOX1 plasmids and selected by 2 µg/ml puromycin for 6 days.

Cell culture and SOX1 induction
The HONE1 and CNE2 cell lines were obtained from Dr. Chao-Nan Qian (Sun Yat-sen University, Guangzhou, China).
Wild-type cell lines and their lentiviral-infected stable cell lines were all maintained in RPMI 1640 (Gibco, 11875119) supplemented with 10% fetal bovine serum (HyClone, SH30070.03). The cells were incubated at 37°C in a humidi ed chamber containing 5% CO 2 . HONE1 TRE-SOX1 and CNE2 TRE-SOX1 cell lines were treated with 1 µg/ml doxycycline to induce overexpression of SOX1. The decreased expression of SOX1 was controlled by renewal of fresh culture medium without doxycycline.
RNA-Sequencing (RNA-Seq) and Pre-processed data Total RNA was extracted from cells using HiPure total RNA Mini Kit (Magen, R4111-03) according to the manufacturer's instructions. RNA-Seq data generation and normalization were performed on an Illumina HiSeq™ PE150 system by the Novogene Bioinformatics Technology Co., Ltd. (Beijing, China). The raw data was processed using "trim_galore", "hisat2", "samtools", and "featureCounts" software in the Linux system. RNA-Seq expression datasets were loaded into R software (version 4.1.0) and evaluated for principal component analysis (PCA) or differential expression analysis through "DESeq2" package. Volcano diagrams were illustrated by "ggplot2" package.
Gene set enrichment analysis (GSEA) GSEA was performed using Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways gene sets from the KEGG Pathway Database (https://www.genome.jp/kegg/ pathway.html) or Hallmark gene sets from the Human Molecular Signatures Database (MSigDB, https://www.gsea-msigdb.org/gsea/msigdb/human/collections.jsp). The ranked genes by fold change were loaded into R software (version 4.1.0) and analyzed with "gseKEGG" function in "clusterPro ler" package [12]. Finally, a list of gene set ranks with information on normalized enrichment scores (NES), P-value, and FDR-q-value were obtained. GSEA and ridge plots of the selected gene sets were illustrated by "gseaplot2" and "ridgeplot" functions in "clusterPro ler" package, respectively. KEGG pathway graphs were rendered by "pathview" package.

Cell lysis, Protein concentration and Western blot (WB) analysis
Cells were digested, counted, and lysed on ice in RIPA buffer (Beyotime, P0013B) with cocktail of protease inhibitors (TargetMol, C0001). Protein concentration was determined by Bradford protein assay, and BSA was used as a standard substance. Equal amounts of cell extracts were subjected to electrophoresis in 10% SDS-PAGE gels and then transferred to 0.45 µm PVDF membranes (Millipore, IPVH00010) for antibody blotting. Horseradish peroxidase-conjugated goat anti-mouse or goat anti-rabbit IgG (Pierce, 31430 or 31460) was used as a secondary antibody. Proteins were visualized with Omni-EC™ enhanced pico light chemiluminescence kit (EpiZyme, SQ101) using a ChemiDoc MP imager (Bio-Rad). The original western blots have been provided in Supplemental Material les.

BrdU incorporation assay and Flow cytometric analysis
All cell samples were seeded on the same day into 6-well tissue culture plates (Jet Bio l, TCP010006) and incubated for 4 days. To avoid cell-cell-contact-induced inhibition of cell proliferation, cells were passaged to a new 6-well plate at a cell density of 20% con uency on day 2. BrdU labeling and staining were performed using FITC-BrdU cell proliferation detection kit (Keygen Biotech, KGA319-1) according to the manufacturer's instructions. Cells were subjected to ow cytometry (CytoFLEX, Beckman Coulter) to analyze the percentage of BrdU-positive cells (FITC: ex 494 nm/ em 520 nm).
Following the incubation period, the culture medium was discarded and the cells were xed with 4% paraformaldehyde (Biosharp, BL539A) for 25 minutes and stained with crystal violet staining solution (Beyotime, C0121) for 30 minutes. Afterward, the cells were slightly washed and air-dried. The plates were imaged using a ChemiDoc MP imager (Bio-Rad) and the average sizes of clone clusters were calculated by ImageJ software (version 1.53c). After the log2 transformation of average sizes, the heatmaps were illustrated by "pheatmap" library in R software (version 4.1.0).

5-ethynyl-2′-deoxyuridine (EdU) incorporation assay and Fluorescence imaging
Cells were seeded onto 96-well tissue culture plates (Jet Bio l, TCP011096) at an initial density of 5 × 10 3 cells/well. "Proliferative" group cells were passaged every other day, while "Quiescence" group cells were pretreated with 1 µg/ml doxycycline for 4 days followed by additional paclitaxel treatment for 7 days. EdU solution was added to each well of the plate on day 6. After 24 hours, EdU staining was performed using K uor488-EdU cell proliferation kit (Keygen Biotech, KGA331-100) according to the manufacturer's instructions. Nuclei were stained with Hoechst 33342 and cells were viewed with a Nikon ECLIPSE Ti2 microscope (Hoechst 33342: ex 350 nm/ em 460 nm; kFluor488-azide: ex 495 nm/ em 520 nm). The percentage of EdU-positive cells in total was calculated by ImageJ software (version 1.53c).

Carboxy uorescein diacetate succinimidyl ester (CFSE) proliferation assay and Flow cytometric analysis
Cells were pretreated with 1 µg/ml doxycycline for 4 days. Then, the cells were resuspended with 5 µM CFSE staining solution and incubated at 37°C for 10 minutes. After labeling, cells were seeded onto 6-well tissue culture plates or 10 cm tissue culture dishes (Jet Bio l, TCD010100) at an initial density of 2 × 10 5 cells/well or 5 × 10 6 cells/dish, respectively. "Proliferation" group cells were passaged every other day. On day 7, cells were subjected to ow cytometry (CytoFLEX, Beckman Coulter) to analyze the intensity of CFSE (CFSE: ex 495 nm/ em 520 nm). The mean uorescence intensity (MFI) of CFSE was calculated by FlowJo software (version 10.5.3).

Cell cycle analysis
Cells were pretreated with 1 µg/ml doxycycline for 4 days. Then, the cells were seeded onto 10 cm tissue culture dishes at an initial density of 5 × 10 6 cells/dish, followed by the schedule.

Results
Tumor suppressor SOX1 contributes to poor prognosis in NPC  Fig. S1A). However, there was less than 5% of mutation, ampli cation, or deep deletion in different tumor samples of the TCGA cohorts ( Supplementary Fig. S1B), revealing that SOX1 might exert its function through RNA/protein alternation. Unexpectedly, patients with an abundance of SOX1 expression in several nonbrain cancer tissues had a poorer prognosis, as compared to those with low levels of SOX1 (Fig. 1A, B). We also found that patients had a good outcome in lower grade glioma with high levels of SOX1 (Fig. 1A, B). Notably, we noticed that patients with "new neoplasm event post initial therapy" had elevated SOX1 expression in original tumors of head and neck squamous cell carcinoma (HNSC), colorectal cancer, and tenosynovial giant cell tumor ( Fig. 1C). Consistently, patients with high expression of SOX1 preferentially experienced "new neoplasm event post initial therapy" in HNSC (Fig. 1D).

SOX1 induces low activity of NPC cells
To deeply understand the con iction that tumor suppressor SOX1 helps HNSC cells to survive after therapy, we compared NPC cells with high versus low expression of SOX1 using RNA-seq analysis ( Supplementary Fig. S2A). Stable clones (HONE1 TRE-SOX1 and CNE2 TRE-SOX1) were established in our previous work [11] and overexpression of SOX1 was induced by doxycycline treatment under the Tet-ON system. Principal component analysis (PCA) analysis showed that within-group distributions of gene expression pro les were well clustered ( Supplementary Fig. S2B). Cells were successfully induced by doxycycline to overexpress SOX1 in "SOX1-High" group, as compared to "SOX1-Low" group ( Fig. 2A, B, Supplementary Fig. S2C). GSEA illustrated that the ribosome pathway was signi cantly enriched in "SOX1-Low" group cells (Fig. 2C, Supplementary Fig. S2D, Supplementary Table S1, S2). The protein RPS3 from small ribosomal subunits and RPL7A from large ribosomal subunits were decreased in "SOX1-High" group cells (Fig. 2B). Reduced protein contents were also observed in "SOX1-High" group cells (Fig. 2C), indicating that SOX1 slowed down the activity of NPC cells. We subsequently used 5-Bromo-2'deoxyuridine (BrdU) to label de novo DNA synthesis in cells (Fig. 2E). Almost all cells in "SOX1-Low" group were labeled at 36 hours, but a small subpopulation of "SOX1-High" group cells (8 ~ 15%) remained BrdU-negative staining (Fig. 2F, G).
According to the above evidence, we created a de novo deduction and hypothesis to explain the con ict. Almost all conclusions about the functions of SOX1 we got were from cell biology assays or tumor transplantation in mice. In these cases, all cancer cells were growing in a stress-free environment and we could imagine that the lesions of tumors also grew well in patients before they came across clinical interventions. In our hypothesis, NPC cells with low expression of SOX1 maintained a state of high proliferation. After radiotherapy and other disadvantageous conditions, SOX1 might be involved in the quiescence of NPC and eventually lead to tumor recurrence. Hence, we preferred to verify SOX1 as a "tumor hypnotist" in our hypothesis (Fig. 2H).

SOX1-induced QCCs are refractory to cell cycle-speci c chemotherapy
To prove the hypothesis, we cultured NPC cells in a stressful environment in which paclitaxel or cisplatin was applied. We rst treated two NPC cell lines (HONE1 and CNE2) with gradient concentrations of paclitaxel, a cell cycle-speci c chemo drug. On day 3, even the lowest concentration (20 nM) of paclitaxel resulted in morphologic changes of all cells, histologically characterized as oating (apoptosis), spherical (G2/M arrest), and giant (cytokinesis failure) phenotypes (Fig. 3A). A pharmacokinetic study showed stable concentration ranges of paclitaxel from 10 to 200 ng/ml in human plasma [21]. Thus, we treated cells with 200 nM (171 ng/ml) paclitaxel for 7 days to completely clear up all proliferative NPC cells (Fig. 3B). Under the intervention of paclitaxel, we observed morphologic differences between cells with high versus low expression of SOX1. On day 3, "SOX1-Low" group showed similar phenotypes with results in Fig. 3A, while "SOX1-High" group resulted in slender and fusiform phenotypes (Fig. 3C). To clear up the retained paclitaxel within cells, the medium was renewed every other day after withdrawal of paclitaxel. After that, doxycycline was not supplied anymore in "SOX1-High" group on day 14 ( Fig. 3B). We found many clonal clustered cells after 6 days withdrawal of doxycycline in "SOX1-High" group, while there were no reactivated cells in "SOX1-Low" group (Fig. 3C). Thus, the QCCs nally recover to a proliferative state under the decreasing expression of SOX1 (Fig. 3D). Colony formation assay also showed distinct clonal clusters in "SOX1-High" group on day 24 (Fig. 3E). In "SOX1-Low" group, proliferative cells went through cell-cycle phase under paclitaxel treatment and they lived no longer than 7 days, mimicking the tumor regression in clinical. In "SOX1-High" group, the QCCs could go through the stress and reactivate, mimicking the tumor recurrence in clinical.
We replaced paclitaxel with a cell-cycle nonspeci c chemo drug (cisplatin) in the same schedule of the model ( Supplementary Fig. S3A). On day 3, there were morphologic changes but no difference between cells with high versus low expression of SOX1 ( Supplementary Fig. S3B). Cisplatin could kill a cell during any phase of the cell cycle, even the QCCs induced by SOX1. There were no recovering cells that could form clonal clusters in the reactivating phase ( Supplementary Fig. S3B). Unlike the model using paclitaxel, "SOX1-High" group could not survive in the end and suffered from the same fate as "SOX1-Low" group ( Supplementary Fig. S3C). The results indicated that SOX1 promoted cells to enter a quiescent state under cell cycle-speci c therapy.
To validate the QCCs, we used 5-ethynyl-2′-deoxyuridine (EdU) to label de novo DNA in cells (Fig. 3F). Almost all nuclei were EdU-positive in "SOX1-Low" group, while only ~ 15% were in "SOX1-High" group (Fig. 3G, H). We also used carboxy uorescein diacetate succinimidyl ester (CFSE) to label cells (Fig. 3F). There still maintained the high intensity of CFSE signal in "SOX1-High" group cells after 7 days (Fig. 3I, J). In summary, these results further proved the quiescent status of cells in our model.

Dynamic expression of SOX1 alters the fate of QCCs
We canceled the "clearance of intracellular paclitaxel" stage to immediately wake up cells after treatment of paclitaxel as the schedule shown (Fig. 4A). However, we couldn't see any clonal clustered cells during the stage of "reactivation of quiescent cancer cells" on day 16 (Fig. 4B). In this situation, the QCCs were waked up too early before paclitaxel within cells was cleared up (Fig. 4C).
Even though the patients have been treated by multiple courses of chemotherapy or radiotherapy, the QCCs stably exist in the body as other normal cells. Here, we elongated the time of the "killing of proliferative cells" stage from 7 days to 14 days (Fig. 4D). After killing all proliferative cells, the recovering cells were observed with arising of clonal cell clusters (Fig. 4E), revealing that the QCCs passed through the long-term severe test (Fig. 4F). Although the patients could be bene ted from long-term treatment, our model indicated that the existence of QCCs became the challenge for developing therapy in clinical.
Many patients suffer from tumor recurrence due to drug resistance. The cells in tumors evolve in a stressful environment and nally adapt to new circumstances. In this case, the inherited drug-resistant cancer cells would grow again unless another therapeutic schedule was performed. To study whether the reactivated NPC cells already acquired the ability of drug resistance for paclitaxel, we planned the following experimental schedule. NPC cells were reactivated the same as before (Fig. 4G). After that, the second induction of SOX1 was performed again to test drug resistance (Fig. 4G). Similarly, all cells in "SOX1-Low" group were dead, while QCCs in "SOX1-High" group survived after paclitaxel treatment on day "X + 14" (Fig. 4H). We decreased the SOX1 level in these QCCs and found that the clonal clustered cells arose again on day "X + 20" (Fig. 4H). These results demonstrated that the survival cells didn't obtain the capacity of paclitaxel resistance. The reactivated cells could be induced into quiescent state again and resisted to the next attack of paclitaxel (Fig. 4I). Not like the small-molecule inhibitors that target a speci c site of molecules, the chemo drugs prefer to kill fast proliferative cells. Hence, these differences determined the diverse mechanisms of drug resistance, and we here focus on the quiescent state of cancer cells under cell cycle related-chemotherapies.

Decreased expression of SOX1 determines the reactivation of QCCs
We elongated the "clearance of intracellular paclitaxel" stage from 7 days to 14 days, in which the cells maintained a high expression level of SOX1 (Fig. 5A). Only if the "reactivation of quiescent cancer cells" stage is more than 6 days, we could nd clonal clustered cells (Fig. 5B). The decreasing expression of SOX1 determined when NPC cells reactivated from quiescence (Fig. 5C).
In addition, we planned to stop the treatment of doxycycline from day 12 to day 17 (Fig. 5D). Colony formation assays also showed that the longer doxycycline was maintained, the later clonal clusters arose (Fig. 5E, F).
Therefore, the QCCs existed as long as SOX1 was highly expressed within NPC cells.

QCCs lose dominant populations in an absence of chemotherapy
Obviously, the SOX1-induced QCCs didn't obtain growth bene ts before paclitaxel treatment. In long-term culture, a proportion of quiescent cells would disappear as the proliferative cells grew and divided. As the cells that overexpressed SOX1 were not all quiescent cells and still growing, we designed the following schedule. We overexpressed SOX1 in NPC cells for at least two weeks before the treatment of paclitaxel (Fig. 6A). Surprisingly, morphologic changes of paclitaxel-treated cells behaved like spherical (G2/M arrest) or giant (cytokinesis failure) phenotypes on day 21 (Fig. 6B). After the withdrawal of doxycycline, there were no reactivated cells on day 34 (Fig. 6B). This method of in nite dilution proved an evolutionary trade-off between proliferative and quiescent NPC cells (Fig. 6C). Not like features of cancer stem cells, the SOX1-induced QCCs were transient but not permanent.
To better understand the difference between proliferative and quiescent subpopulations of NPC cells induced by SOX1, we performed cell cycle analysis (Fig. 6D). The proliferative NPC cells were arrested in the G2/M phase after treatment of paclitaxel, while a crowd of "quiescence" group cells remained in G0/G1 phase (Fig. 6E, F). Similarly, "quiescence" group enriched more than 50% of cells with low DNA/RNA contents (G0 phase) (Fig. 6G, H). Moreover, the IF assay displayed that most of cells (~ 80%) in "quiescence" group were Ki-67-negative (Fig. 6I, J).

SOX1 induces QCCs through downregulating MYC
We performed RNA-seq analysis to explore the expression pro les between these two groups ( Supplementary Fig.   S4A). PCA analysis showed that within-group distributions of gene expression pro les were well clustered ( Supplementary Fig. S4B). The cells with low expression of SOX1 formed the dominant population in "proliferation" group, while "quiescence" group cells highly expressed p27Kip1, an approved marker for quiescent cells (Fig. 7A,   Supplementary Fig. S4C). GSEA demonstrated decreased enrichment of ribosome pathway in "quiescence" group cells (Fig. 7B, Supplementary Fig. S4D, Supplementary Table S3, S4), and reduced expression of RPS3 and RPL7A were veri ed by western blot analysis (Fig. 7C). The hallmark gene sets associated with several cell growth signaling pathways were signi cantly de-enriched in "quiescence" group cells, especially for "MTORC1 signaling", "MYC Targets V1" and "MYC Targets V2" (Fig. S7D, Supplementary Table S5, S6). MYC was reported as a master regulator of ribosome biogenesis [22]. Hence, SOX1 could inhibit protein synthesis and induce QCCs by decreasing c-MYC levels in "quiescence" group cells (Fig. 7C). SOX family proteins share a conserved high-mobility group (HMG) domain, binding to the same core consensus DNA motif (5'-TTGT) [23]. We analyzed de novo and known motifs in SOX1 peaks from the GEO database (Fig. 7E, Supplementary Table S7). There were lots of potential SOX1-binding sequences in the promoter of the MYC gene (Fig. 7F, Supplementary Table S8). SOX1 might recruit transcriptional repressors to halt the expression of MYC. In addition, we mutated two amino acids (Arg53 to Asp and Asn78 to Ala) in HMG domain to destroy the transcriptional function of SOX1, which was veri ed by our previous work [11]. Following the experimental model mimicking QCCs, the mutant form couldn't induce QCCs and clustered colonies didn't arise (Fig. 7G). This evidence indicated that SOX1 induced QCCs through its transcriptional function.

Discussion
In general, the patients with higher expression of tumor suppressors have longer survival times. The loss function of some tumor suppressors, such as RB1, TP53, PTEN, and CDKN2A, contributes to cancer progression and poor prognosis. However, the limited genetic alteration of SOX1 suggests these tumor suppressors have a second mechanism for unfavorable prognosis. Our study rst reveals that the tumor suppressor SOX1 can induce quiescence in individual NPC cells, and demonstrate a second mechanism that the quiescent NPC cells are refractory to chemotherapy. Similarly, the tumor suppressor p21Cip1 was also reported to play a dual role in cancer and correlated with poor prognosis [24]. Therefore, these genes should be classi ed into a de novo description, which we named "tumor hypnotists". Unlike the terminology of tumor suppressors, "tumor hypnotists" neutrally evaluate cancer cells, telling that these genes are not good or bad in an absolute sense for tumors. Our model describes a dynamic process in vitro that mimics patients who experience stages of primary tumor, dormant tumor, reactivated tumor, and relapsed tumor (Fig. 8). During the growth of tumors, cancer cells with low levels of SOX1 have a clear advantage in the population. When the tumors are attacked by chemo drugs, the one with high expression of SOX1 can live longer. After months, years, or even decades, tumor relapse happens when some of these survival cancer cells decrease SOX1 expression.
Our study showed that the NPC cells escape the continued chemotherapy and remain static for a long-term period under overexpression of SOX1. For one thing, the administration of cytotoxic drugs often leads to deleterious effects not only on fast-growing cancer cells but also on normal cycling normal cells, resulting in side effects. Our model demonstrated that the inert cancer cells induced by SOX1, including quiescent and slow cycling cells, can survive for more than 2 weeks under attack of high-dose paclitaxel. They even grow much slower than many healthy cells in human body. For another, it is not clear if cancer will come back after the treatment ends. Our model displayed that QCCs induced by SOX1 are static for at least two weeks without clustered colonies. High expression of SOX1 maintains a long-term hibernation, and the QCCs recover only if decreasing levels of SOX1.
The results provide a possibility that tumor relapse happens in an uncertain time course.
SOX1 expression is mainly restricted to normal brain tissues (thalamus) or glioma. Although SOX1 is low expressed (hardly detected) in bulk non-brain tumors, a small subpopulation of QCCs with high levels of SOX1 plays a grand role under chemotherapy. Immunohistochemistry or single-cell sequencing technologies would provide a possibility to predict recurrence rate by detecting the potential prognosis biomarker SOX1 in individual cancer cells. A few SOX1-positive cancer cells within tumor lesions might represent that traditional chemotherapy could not eliminate all cancer cells for patients, even though they achieve a complete response.
There are three main strategies to target QCCs in tumors so far, but the best therapeutic approach is still unknown due to developing resistance [25]. Our study provides a model that mimics QCC induced by intrinsic factors. Unlike other common models in vitro, our model is independent of extracellular cell growth signaling, providing an ideal method to deeply understand the initiation or termination of quiescence in cancer. It could also help us to thoroughly know the molecular mechanisms of SOX1 in different stages of cancer progression. But limited to the arti cial gene expression system, how SOX1 is induced to express in individual cancer cells within tumors, or how QCCs sense the changing of environment and decrease SOX1 level to reproliferating, is still unknown.

Conclusion
We rede ne the intrinsic transcriptional factor SOX1 as "tumor hypnotist" instead of "tumor suppressor" according to the dynamic roles in tumor development. High expression of SOX1 within individual cancer cells is a potential indicator of chemotherapy resistance in NPC, and it is of great signi cance to propose feasible targeting strategies for tumor recurrence after clinical interventions.

DATA AVAILABILITY
All supporting data are included in the manuscript and supplemental les. Additional data are available upon reasonable request to the corresponding author.