Shikonin Enhances the Antitumor Effect of Cabazitaxel in Prostate Cancer Stem Cells and Reverses Cabazitaxel Resistance by Inhibiting the Expression of ABCG2 and ALDH3A1

Background: Cancer stem cells (CSCs) are a small population among cancer cells, dened as capable of self-renewal, and driving tumor growth, metastasis, and therapeutic relapse. The development of therapeutic strategies to target CSCs is of great importance to prevent tumor metastasis and relapse. Increasing evidence shows that shikonin has inhibiting effects on CSCs. This study was to determine the effect of shikonin on prostate CSCs, and on drug resistant cells. Methods: Sphere formation assay was used to enrich prostate CSCs. The effect of shikonin on viability, proliferation, migration, and invasion was studied. Typical CSCs markers were analyzed by ow cytometry and RT-qPCR. The cytotoxic mechanism of shikonin was analyzed by staining for annexin V, reactive oxygen species (ROS) and mitochondrial membrane potential. To study the effect of shikonin on drug resistant cells a cabazitaxel resistant cell line was established. Results: Shikonin inhibited the viability, proliferation, migration, and invasion of prostate CSCs. Shikonin enhanced the antitumor effect of cabazitaxel, which is a second-line chemotherapeutic drug in advanced prostate cancer. Shikonin induced apoptosis through generating ROS and disrupting the mitochondrial membrane potential. Furthermore, shikonin suppressed the expression of ALDH3A1 and ABCG2 in prostate CSCs, which are two markers related to drug-resistance. When inhibiting the expression of ABCG2 and ALDH3A1, the cabazitaxel resistant cells acquired more sensibility to cabazitaxel. Conclusions: Shikonin enhances the cytotoxic activity of cabazitaxel in prostate CSCs and reverses the cabazitaxel-resistant state. was determined after 48 c Apoptosis was measured by and 7-AAD staining and ow cytometry: shikonin in combination with cabazitaxel enhanced the apoptosis rate in DU145 cells and the corresponding DU145 CSC d Invasion assay: combination of shikonin and cabazitaxel decreased the invasion ability of DU145 cells and DU145 CSCs compared with cabazitaxel or shikonin alone. The results are presented as the means ± SEM of values obtained in three independent experiments. *p < 0.05; **p < 0.01; ***p < 0.001. a ROS generation in DU145, DU145-CSC, PC-3, and PC-3 CSC cells in a concentration dependent manner. b The viability assay showed that ROS generation was rescued by the antioxidant N-acetyl-L-cysteine (NAC) after pretreatment for 4 hours at 1 mM. c The integrity of mitochondrial membrane potential ( ∆ ψm) was evaluated using the JC-1 mitochondrial membrane potential kit. JC-1 dye staining indicated that shikonin dysregulated the mitochondrial membrane potential along with higher concentrations. NAC pretreatment caused the inhibition of mitochondria membrane potential dysregulation. d Cabazitaxel cannot inuence ROS generation. The results are presented as the means ± SEM of values obtained in three independent experiments. *p < 0.05; **p < 0.01; ***p < 0.001. e ABCG2 positive control. The results are presented as the means ± SEM of values obtained in three independent experiments. *p < 0.05; **p < 0.01; ***p < 0.001.

(Invitrogen), 10 ng/ml human recombinant basic broblast growth factor (bFGF, Sigma Aldrich Chemie GmbH, Taufkirchen, Germany), and 10 ng/ml epidermal growth factor (EGF, Sigma Aldrich). The cabazitaxel-resistant DU145 cell line was generated by culturing in presence of cabazitaxel starting at a concentration of 1 nM. When DU145 was resistant to the cabazitaxel at 1 nM, concentration was slowly increased. We cultured DU145 cells with cabazitaxel for at least eight months until a concentration of 6 nM was reached.
Sphere formation assay Sphere formation was conducted to generate the PCSCs. DU145 and PC-3 cells were harvested using StemPro® Accutase® (Life Technologies, Thermo Fisher Scienti c, Waltham, MA, USA) and seeded in 75 cm 2 ultra-low attachment asks (Corning Costar, Amsterdam, The Netherlands) in 10 ml CSCs medium. After seven days, spheres were harvested into a 15 ml tube, centrifuged at 500 g for 4 minutes at room temperature, and dissociated using Accutase for 10 minutes at 37 °C. After centrifugation, cells were ltered through a cell strainer with 40 µM nylon mesh (BD Biosciences, Heidelberg, Germany), counted, and used for different assays. For enrichment of CSCs, the spheres were dissociated and a second round of sphere formation was done.

Drug sensitivity assay
Cell viability was assessed using the CellTiter-Blue Cell Viability Assay (Promega, Mannheim, Germany) according to the manufacturer's instructions. Cells were seeded on a 96-well microtiter plate (1500 cells/well) and treated with different concentrations of shikonin or cabazitaxel (both from Selleckchem, Houston, Texas, USA). After 24 h and 48 h, measurement of uorescence was done at 560 (20) excitation/590(10) emission using the FLUOstar OPTIMA microplate reader (BMG LABTECH, Ortenberg, Germany) and the OPTIMA software version 2.0. The logit regression model was used to calculate the half-maximal inhibitory concentration (IC50).

Cell proliferation assay
Proliferation was evaluated using the CellTiter 96 AQ ueous One Solution Cell Proliferation Assay (Promega GmbH, Mannheim, Germany) according to the manufacturer's instructions. Cells were seeded in 96-well microtiter plates at 5 × 10 3 and treated with the drugs. After 24 h and 48 h, the reagent was added directly to the wells, the plates incubated for three hours and then the absorbance was read at 490 nm with the Emax precision microplate reader (MWG Biotech, Ebersberg, Germany).

Apoptosis assay
A total of 4 × 10 5 cells were seeded and treated with or without drugs. After ve days, cells were harvested and resuspended in 100 µl Annexin V Binding Buffer. Five microliter APC-Annexin V and 5 µl 7aminoactinomycin D (both from BD Biosciences) were added. Following incubation for 15 minutes at room temperature 100 µl Annexin V Binding Buffer were added again and the cells measured using the FACSCalibur (Becton Dickinson, San Jose, CA, USA). A minimum of 1 × 10 4 cell events were recorded per sample. BD CellQuest software (version 4.0.2) was used for data acquisition, and data were analysed using FlowJo software (version 9.9.5; Tree Star Inc., Ashland, OR, USA). The Annexin V positive cells were considered as apoptotic cells.

Migration assay
For cell migration the scratch wound healing assay was applied using 24-well culture plates with small 2well silicone inserts per well building a cell free gap of 500 µm (ibidi GmbH, Martinsried, Germany). Cells were diluted to a concentration of 4 ⋅10 5 cells/ml, seeded in a volume of 70 µl into each culture insert, and incubated at 37 °C until a con uent cell monolayer was achieved. Drugs were added after the inserts were removed with sterile tweezers. Pictures were taken under a microscope at different time points, 0, 3,6,9,21,24,27, and 30 h. The proportion of gap covered by cells at each time was determined and the data were analyzed with the web-based Automated Cellular Analysis System (ACAS, MetaVì Labs, Bottrop, Germany) using FastTrack AI image analysis algorithms.

Invasion assay
Cell invasion was done using the Boyden Chamber system. The transwell inserts in 24-well plates (8.0 µm pores; Falcon, Corning, New York, NY, USA) were coated with growth factor reduced Matrigel Basement Matrix (Corning; 30 µg/100 µl/insert) and incubated for at least 4 hours at room temperature. Afterwards 30,000 cells were seeded in 250 µl serum-free DMEM medium with and without drugs onto the Matrigel.
In the lower chamber, 750 µl DMEM with 10% FCS was added. After 48 h incubation, the upper surface of the transwell membrane was gently wiped with a moistened cotton swab to remove cells that had not migrated. Cells, which had moved through the Matrigel to the lower surface of the membrane were xed with 4% paraformaldehyde for 5 minutes and then stained with 1% crystal violet for 1 minute. After drying, cells were counted (three elds per insert; Fiji ImageJ software [25]).

Determination of aldehyde dehydrogenase
Active aldehyde dehydrogenase (ALDH) was measured using the ALDEFLUOR kit according to the manufacturer's instructions (StemCell Technologies, Grenoble, France). Brightly uorescent ALDHexpressing cells were detected in the green uorescence FL1 channel (520-540 nm) of the FACSCalibur.
The control group included N,N-diethylaminobenzaldehyde (DEAB) at 15 µM, which is an inhibitor of ALDH activity. BD CellQuest software was used for data acquisition, and data were analysed using FlowJo software.
Measurement of ABCG2 using ow cytometry Spheres were dissociated using Accutase and seeded in 6-well low attachment plates with or without shikonin. Twenty-four hours later, cells were stained with APC-conjugated ABCG2 monoclonal antibody (clone 5D3/CD338, BD Biosciences). LIVE/DEAD® Fixable Blue Dead Cell Stain Kit (Life Technologies) was used to determine the viability and exclude dead cells during analysis. All measurements were accomplished using the LSRII ow cytometer (BD Biosciences). Data were processed using FlowJo software.
RT-qPCR RNA was isolated using the RNeasy Mini-Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. A total of 1 µg RNA was then used for reverse transcription according to the kit instructions (Promega GmbH). Reverse transcription polymerase chain reaction was performed with FastStart Essential DNA Green Master kit (Roche, Penzberg, Germany) and Light Cycler® 96 (Roche). The reaction was set up with 2 µl cDNA template, 1 µl forward primer, 1 µl reverse primer, 5 µl FastStart Essential DNA Green Master reagent mix, and 1 µl PCR grade water. For ampli cation, a hot start with 95 °C for 10 minutes, followed by 40 cycles of denaturation at 95 °C for 10 seconds, annealing at 60 °C for 10 seconds, and extension at 72 °C for 10 seconds were used. Then a melting process with 95 °C for 10 seconds, 65 °C for 60 seconds and 97 °C for 1 second was performed. Data were analyzed by the Lightcycler® 96 software (1.1 version). The relative expression was calculated using the 2 −ΔΔCt method.
The transcription level of GAPDH and ACTB was used as an internal control. Primers were listed in the Additional le 1 Table S1.

Confocal immuno uorescence microscopy
Fifty thousand cells per well were seeded in 16-well chambered coverslips (Molecular Probes, Thermo Fisher Scienti c) and cultured for 24 h. Then cells were xed with 4% paraformaldehyde for 10 minutes, permeabilized with 0.1% Triton X-100 for 5 minutes, and blocked in 3% BSA for 1 h. Afterwards cells were incubated with the primary rabbit polyclonal ALDH3A1 antibody (Abcam, Cambridge, UK) at 5 µg/ml, and mouse monoclonal ABCG2 antibody (clone BXP-21, Abcam) at a 1/50 dilution overnight at 4 ℃. After a three time washing step, the cells were incubated with the secondary antibody goat anti-rabbit IgG H&L (Alexa Fluor® 647) preadsorbed (Abcam), and goat anti-mouse IgG H&L (Alexa Fluor® 488, Abcam) used in a dilution of 1:200 at room temperature for 1 h in the dark. Cells were then washed three times with PBS and incubated with Molecular Probes® NucBlue® Fixed Cell Stain ReadyProbes® reagent (DAPI special formulation, Thermo Fisher Scienti c) for 20 minutes at room temperature in the dark. Ibidi mounting medium (ibidi GmbH) was dropped to cover the cell layer, followed by a coverslip. The uorescence intensity was measured by confocal Leica SP5 for DAPI at 365 nm excitation and 420 nm emission wavelengths. The negative control was a secondary-only assay to demonstrate low non-speci c binding of the secondary antibody.

Measurement of reactive oxygen species (ROS)
Cells were seeded in a 96-well microplate (25,000/well) and cultured overnight. Then, 100 µl 25 µM 2',7'dichloro uorescein diacetate (DCFDA, Abcam) was added for 45 minutes at 37 ℃ in the dark. After washing three times, cells were treated with shikonin for 6 h. ROS was measured on the FLUOstar OPTIMA microplate reader at 485 nm excitation/535 nm emission. ROS generation was rescued by the antioxidant N-acetyl-L-cysteine (NAC) with pretreatment for 4 hours at 1 mM.

Measurement of mitochondrial membrane potential
The integrity of mitochondrial membrane potential (∆ψm) was evaluated using the JC-1 mitochondrial membrane potential kit (Abcam). Brie y, cells were seeded and treated with shikonin for 48 h. Then, cells were stained with 20 µM JC-1 (tetraethylbenzimidazolylcarbocyanine iodide) for 10 minutes at 37 ℃, washed once and measured on the FLUOstar OPTIMA microplate reader. With normal mitochondrial function (high ∆ψm) the dye aggregates yielding a red to orange emission (590 ± 17.5 nm), injured mitochondria are depolarized (low ∆ψm) and leads to monomer formation yielding green emission (530 ± 15 nm).Signals were recorded in a monomer form at Excitation/Emission equaling to 485 ± 10/590 nm, and in an aggregate form at Excitation/Emission equaling to 485 ± 10/520 nm. A ratio between the monomer and aggregate was obtained and plotted.

Downregulation of ABCG2 and ALDH3A1
Two inhibitors of ABCG2 and ALDH3A1 were used namely Ko143 (Tocris Biosciences, Bio-Techne GmbH, Wiesbaden-Neuenstadt, Germany) at 1 µM, and CB29 (Sigma Aldrich) at 32 µM, respectively. Small interfering RNAs (siRNAs) were also used to silence ABCG2 and ALDH3A1, which were designed by Ambion (Life Technologies, Thermo Fisher Scienti c). The Silencer® Select siRNAs sequences are shown in the Additional le 2 Table S2. Silencer® Select negative control and Silencer® Select GAPDH positive control were also used. For transfection the Lipofectamine RNAiMAX Reagent (Invitrogen, Life Technologies, Thermo Fisher Scienti c) was applied as described in the manufacturer's instruction.

Statistical methods
All experiments were independently performed at least three times. The analyses were carried out using SPSS version 25.0 (IBM, Armonk, NY). All numerical data were expressed in the form of the average of the values (mean), plus the standard error of the mean (SEM). Data from two different groups were compared using the Mann-Whitney U test. The Pearson product-moment correlation coe cient was calculated to determine the correlation between ABCG2 and ALDH3A1 in the confocal immuno uorescence microscopy. All p-values were 2-sided, and p<0.05 was considered signi cant.

Shikonin inhibits viability and proliferation of PCSCs
In the rst part of the study, we generated PCSCs using sphere formation assay over 2-3 generations. The PCSCs in the ultra-low attachment asks with serum-free medium formed non-adherent free-oating spheres, while differentiated adherent prostate cancer cells shrank and died [26]. Figure 1a shows the generation of the spheres within 10 days. DU145 and PC-3 and the corresponding PCSCs were treated with different concentrations of shikonin (0, 1.5, 3, 6, 12, 24, and 48 µM) for 24 h and 48 h, see Additional le 3 Fig. S1. The results of the drug sensitivity assay show that shikonin inhibited the cell viability in a dose-dependent manner (Fig. 1b). The half-maximal inhibitory concentration (IC50) analyzed by logit regression model using the SPSS software mounted around 0.75 µM for DU145, 4 µM for DU145 CSCs, 5 µM for PC-3, and 7 µM for PC-3 CSCs. Using different concentrations ( 0.5x IC50, IC50, and 2x IC50) an inhibiting effect of shikonin on the proliferation rate was also observed ( Fig. 1c and Additional le 3 Fig.  S1). Results showed that DU145 CSCs and PC-3 CSCs were much less sensitive to shikonin than the adherent DU145 and PC-3 cells. It supported the point that cancer stem cells were a special cell population like drug-resistant cells.

Shikonin inhibits migration and invasion of PCa cells
Since the migration and invasion ability are two important biological characteristics, we evaluated the effect of shikonin on migration and invasion of DU145, PC-3 and their corresponding sphere cells.
Migration was evaluated using the scratch wound healing assay. After treatment with different concentrations of shikonin pictures were taken at different time points. Shikonin inhibited the migration of both the adherent cells and sphere cells to the center of the gap (Fig. 2a-b and Additional le 4 Fig. S2). Invasion ability was determined using the Boyden chamber system. Cells were treated again with different concentrations of shikonin (0.5 × IC50, 1 × IC50, and 2 × IC50) for 48 hours. The numbers of invaded cells were decreased signi cantly in the adherent PCa cell lines and even in the corresponding PCSCs, (Fig. 2c-d). It can be concluded that shikonin was able to inhibit migration and invasion remarkably in PCa cells.  Fig. S4) in DU145 and DU145 CSCs compared to cabazitaxel alone. The effect was more pronounced in the combination treatment, even at a lower concentration of cabazitaxel (< IC50). Similarly, the number of apoptotic cells was higher following treatment with cabazitaxel plus shikonin (Fig. 3c). Also in the invasion assay a signi cant higher inhibition was observed (Fig. 3d). Altogether, shikonin sensitizes the antitumor effect of cabazitaxel in both DU145 and DU145-CSCs.

Shikonin induces apoptosis through ROS-mitochondria membrane potential apoptosis pathways in PCa cell lines and PCSCs
The effect of shikonin on the apoptosis rate was also tested in the cell line PC-3 and the corresponding PC-3 CSCs. Results showed that shikonin dramatically induced apoptotic death both in the PCa cell lines DU145 and PC-3 and the corresponding PCSCs (see Additional le 7 Fig. S5).
Reactive oxygen species (ROS) generation is an essential effect of shikonin during cell death in several types of cancers [27][28][29][30]. To study the role of shikonin in prostate cancer cells we used the cell permeable reagent 2',7' -dichloro uorescein diacetate (DCFDA), a uorogenic dye that measures ROS activity within the cell. After diffusion into the cell, DCFDA is deacetylated by cellular esterases to a nonuorescent compound, which is later then oxidized by ROS into 2', 7' -dichloro uorescein (DCF), a highly uorescent substance. Treatment with shikonin for 6 hours dramatically enhanced ROS generation in the PCa cell lines and the corresponding PCSCs (Fig. 4a). When the cells were pretreated with the ROS scavenger, N-acetyl-L-cysteine (NAC), at 1 mM for 4 h the effect of shikonin was signi cantly reduced as shown in the viability assay, which proves that shikonin induces apoptosis via ROS generation (Fig. 4b).
Then we conducted staining with tetraethylbenzimidazolylcarbocyanine iodide (JC-1) to investigate if shikonin disrupted the mitochondrial membrane potential of PCa cells and PCSCs. According to the kit instructions due to low mitochondrial membrane potential, JC-1 is predominantly a monomer that yields green uorescence with emission of 530 ± 15 nm. At high concentrations (due to high mitochondrial membrane potential), the JC-1 aggregates yielding a red to orange colored uorescence (590 ± 17.5 nm) Both monomer and aggregate forms were measured, and a ratio between the two measurements was obtained and plotted. Exposure to shikonin for 48 h dramatically reduced aggregates uorescence, indicating that shikonin induced apoptosis through ROS formation and triggered the disruption of the mitochondrial membrane potential. NAC reversed mitochondrial dysfunction and veri ed our conclusion (Fig. 4c). Besides this, cabazitaxel alone cannot in uence ROS generation in DU145 cells (Fig. 4d). It suggested that cabazitaxel cannot interrupt the antitumor effect of shikonin.
Shikonin inhibits the expression of ABCG2 and ALDH3A1 in PCSCs.
There are different cancer stem cell markers known for prostate cancer driving drug-resistance, and cancer relapse as for example ABCG2, ALDH, CD44, CXCR4 [7]. Figure 5a and Additional le 8 Fig. S6 demonstrates that shikonin inhibited the expression of both ALDH and ABCG2 (CD338) shown by ow cytometry. Similarly, RT-qPCR showed that shikonin inhibited the expression of ALDH3A1 and ABCG2 (Fig. 5b). In some publications it was reported that increased expression of the drug transporter gene ABCG2, was detected in the ALDH + cell population [31,32]. Inhibition of ALDH1A1 activity decreased the expression of ABCG2 and reduced chemotherapy resistance in ovarian cancer cell lines [33].To explore if shikonin downregulates the expression of ABCG2 through blocking the expression of ALDH3A1, we silenced ALDH3A1 using siRNA for 48 h and tested ABCG2 expression by RT-qPCR. The results showed that downregulation of ALDH3A1 could not in uence the expression of ABCG2 (Fig. 5d). However, in the confocal uorescence microscopy it was observed that ALDH3A1 was overlapped with ABCG2 in DU145 CSCs (Fig. 5c). Both the Pearson correlation coe cient and the overlap coe cient were 0.9 analyzed by the colocalization nder of image J software (Fig. 5d). It can be suggested that ALDH3A1 and ABCG2 were at least co-expressed in DU145 CSCs. To investigate the relation between ABCG2 and ALDH3A1, we silenced the ABCG2. After downregulating ABCG2, the expression of ALDH3A1 decreased (Fig. 5e). This implies that shikonin can block the expression of ALDH3A1 by inhibiting ABCG2.
Shikonin reverses the resistant-state of caba-DU145 through inhibiting the expression of ABCG2 and ALDH3A1 In clinical practice, cabazitaxel has proved to be effective for docetaxel-resistant CRPC patients [24]. When the CRPC patients become resistant to cabazitaxel, there are few therapeutic options left. According to our results, ABCG2 and ALDH3A1 are two primary cancer stem cell markers that can drive drug-resistance and their expression can be regulated by shikonin. Then we wanted to know if shikonin can abolish the cabazitaxel-resistant state by inhibiting the expression of ABCG2 and ALDH3A1. We generated a cabazitaxel-resistant DU145 cell line (caba-DU145) by gradually increasing the dose of cabazitaxel over eight months (see Additional le 9 Fig. S7). Then the caba-DU145 cells were pretreated with the ABCG2 inhibitor Ko143 at 1 µM. After ve days, the cell viability of caba-DU145 response to cabazitaxel was determined. Results showed that after treatment with Ko143 the cells were more sensitive to cabazitaxel compared to the untreated cells (Fig. 6a). The same was observed using the proliferation assay (Fig. 6b). Also, treatment with the ALDH3A1 inhibitor CB29, showed a higher sensitivity of the cells to cabazitaxel compared to the untreated cells (Fig. 6c). The same effect was observed again using the proliferation assay (Fig. 6d). Furthermore, inhibiting ALDH3A1 and ABCG2 using CB29 and Ko143, decreased the apoptosis rate of caba-DU145 cells (Fig. 6e). As we expected, using low concentration of shikonin at 0.375 µM in combination with cabazitaxel at 3 nM increased the apoptotic rate compared to cabazitaxel alone in the caba-DU145 cells (Fig. 6f).
Our results suggested that shikonin not only enhances the antitumor effect of cabazitaxel but also reverses the cabazitaxel-resistant state by inhibiting the expression of ABCG2 and ALDH3A1, and ABCG2 suppresses the expression of ALDH3A1.

Discussion
Prostate cancer is a highly malignant carcinoma, due to its late detection and therapeutic-resistant caused relapse. Although androgen deprivation treatment is effective at the beginning of advanced PCa [34], half of those cases become resistant to androgen deprivation treatment [35]. The taxane docetaxel is the rst-line agent for CRPC, but most patients acquire resistance and disease relapse. Then cabazitaxel, a second-line taxane, is used for patients who developed resistance to docetaxel [23]. It is urgent to nd some new agents for the treatment of CRPC, and to reveal the mechanisms of cabazitaxel resistance.
Recently, it was proven that shikonin shows antitumor effects in human glioblastoma stem cells [36], and prostate cancer [19,21,37]. In our study shikonin targeted not only normal prostate cancer cell lines but also PCSCs, which are responsible for drug resistance [5]. Several mechanisms have been discussed for drug resistance in CSCs. First, the CSCs stay in the resting phase of the cell cycle, losing sensitivity to chemotherapeutic drugs because the majority of agents mainly target cells in the proliferating phase. Second, CSCs express e ux transporters mostly belonging to the ATP-binding cassette gene family (ABC transporters). These transporters pump chemotherapeutic drugs out of the CSCs [8]. Also, the hypoxic CSCs microenvironment is one leading cause of drug resistance. This microenvironment protects CSCs to hide in a quiescent state in the tissue and avoid the attack of drugs [38,39]. Since shikonin attacks the CSCs, we guessed that it also inhibits the cabazitaxel resistance. We tested different cancer stem cell markers in RT-qPCR, including CD44, TROP2, ALDH1A1, ALDH2, ALDH3A1, and ABCG2 ( Fig. 5b and Additional le 10 Fig. S8), and found that shikonin downregulated the expression of ABCG2 and ALDH3A1 in PCSCs. In addition, blocking of ABCG2 reduced ALDH3A1 expression and can regulate those pathways reversing the resistant state.
In our studies shikonin attacks PCSCs by inhibiting proliferation, migration, invasion, and by promoting apoptosis. It was further demonstrated that apoptosis induced by shikonin in prostate cancer stem cells was activated through the ROS-mitochondria membrane potential pathway as earlier discussed for normal PCa cell lines [40].
ABCG2 is one of the ABC transporters, which drives CSCs resistance through e ux drugs. The ABCG2 + cancer cells initially called side population cells based on ow cytometry results, which can e ux most agents and prevent treatment success [12]. Inhibition of ABCG2 increased the chemotherapeutic agents' sensitivity in breast cancer cells [41], and ovarian carcinoma cells [42].
Cancer cells with high ALDH expression display elevated tumorigenicity capacity, and have a critical role in regulating self-renewal and differentiation. Therefore, high ALDH activity is known as a characteristic feature of CSCs [43]. ALDH3A1, which is one of the ALDH isoforms, is highly expressed in PCSCs and its expression is related to PCa progression [44]. A dramatic elevation of ALDH3A1 is observed in DU145 cellderived lung metastasis, metastatic lymph node, and bone metastasis compared to local xenograft tumors. ALDH3A1 highly expressed in DU145 spheres contrasted with adherent DU145 cells [44].
Inhibition of ALDH3A1 increased the drug-sensitivity of lung adenocarcinoma, glioblastoma cell lines [45], and head and neck squamous cell carcinoma [46].
Shikonin downregulated both ABCG2 and ALDH3A1 in PCSCs, which were likely to correlate to cabazitaxel resistance, and ABCG2 can inhibit the expression of ALDH3A1. When blocking ABCG2 and ALDH3A1, the caba-DU145 cells are anew sensitive to cabazitaxel. Altogether, we concluded that shikonin can reverse the cabazitaxel-resistant state by blocking the expression of ABCG2 and ALDH3A1, and ALDH3A1 themselves can be inhibited by suppressing ABCG2.
From the literature it can be concluded, that the antitumor attack of shikonin will be exerted via blocking the PI3K/AKT signaling pathway [47,48]. Also, this pathway regulates the expression of ABCG2 and confers resistance to chemotherapy. The blockage of the PI3K/AKT pathway enhances the sensitivity to chemotherapeutic agents [49,50]. Therefore, we supposed that shikonin reversed the drug-resistant state through modulating the PI3K/AKT-ABCG2-ALDH3A1 axis in PCSCs.
Therefore, shikonin can sensitize both PCa cells and PCSCs may lead to a better therapeutic option in the future. However, a further preclinical in vivo study with shikonin and cabazitaxel is necessary to verify these in vitro ndings. Also, the results indicate that some potential inhibitors of ABCG2 and ALDH3A1 can be applied to patients with chemotherapy resistance, and further researches are needed.

Conclusions
In summary, our results reveal for the rst time that shikonin attacks PCSCs through ROS-mitochondria membrane potential apoptosis pathways, and cabazitaxel cannot in uence this process. Shikonin reverses the cabazitaxel-resistant state by inhibiting the expression of ABCG2 and ALDH3A1 or by inhibiting ALDH3A1 through suppressing ABCG2 (Fig. 6g). The combination of shikonin and cabazitaxel has a synergistic effect. It enables the use of a lower dose of cabazitaxel and might be a new therapeutic option for CRPC with perhaps lower side effects. Declarations Acknowledgments Not applicable.

Funding
The study was supported by the LMU clinic. The author Lili Wang is funded by the China scholarship council.

Availability of data and materials
All data generated or analysed during this study are included in this published article and its supplementary information les.
Authors' contributions LW performed the experiments, analysed the data and was responsible for writing the manuscript, BS provided technical support, CL discussed the experimental approaches. AB was involved in analyzing the data. HP was responsible for the conception and design of the study, analyzing the data and writing the manuscript. All authors read and approved the nal manuscript.

Competing interests
The authors declare that they have no competing interests.

Consent for publication
Not applicable Ethics approval and consent to participate Not applicable shikonin inhibited the proliferation of DU145, PC-3, and their corresponding PCSCs, shown for 24, 48, and 72 hours. The results are presented as the means ± SEM of values obtained in three independent experiments. *p < 0.05; **p < 0.01; ***p < 0.001.

Figure 2
Shikonin inhibits migration and invasion in PCa cells and PCSCs. a and b Scratch wound healing assay.
Plates with a cell free gap of 500 m were used (ibidi GmbH): shikonin inhibited the migration ability of The proliferation was determined after 48 hours. c Apoptosis was measured by Annexin IV and 7-AAD staining and ow cytometry: shikonin in combination with cabazitaxel enhanced the apoptosis rate in DU145 cells and the corresponding DU145 CSC d Invasion assay: combination of shikonin and cabazitaxel decreased the invasion ability of DU145 cells and DU145 CSCs compared with cabazitaxel or shikonin alone. The results are presented as the means ± SEM of values obtained in three independent experiments. *p < 0.05; **p < 0.01; ***p < 0.001.  Shikonin inhibits the expression of ABCG2 and ALDH3A1 in PCSCs. a ALDEFLUOR kit was used to measure the expression of ALDH in DU145 and DU145 CSCs. As suspected DU145 CSCs expressed more ALDH than DU145 cells. The ALDH inhibitor N,N-diethylaminobenzaldehyde (DEAB)was used as control. CSCs. Both Pearson's correlation and overlap coe cients between ALDH3A1 and ABCG2 were 0.9. d RT-qPCR analysis: silencing ALDH3A1 by speci c siRNA cannot interrupt the expression of ABCG2. GAPDH siRNA was used as positive control. e Silencing ABCG2 by speci c siRNA decreased the expression of ALDH3A1. GAPDH siRNA was used as positive control. The results are presented as the means ± SEM of values obtained in three independent experiments. *p < 0.05; **p < 0.01; ***p < 0.001. Figure 6