Tumor stromal topography promotes chemoresistance in migrating breast cancer cell clusters

Abstract

Multicellular clustering provides cancer cells with survival advantages and facilitates metastasis. At the tumor migration front, cancer cell clusters are surrounded by an aligned stromal topography. It remains unknown whether aligned stromal topography regulates the resistance of migrating cancer cell clusters to therapeutics. Using a hybrid nanopatterned model to characterize breast cancer cell clusters at the migration front with aligned stromal topography, we demonstrate that topography-induced migrating cancer cell clusters exhibit upregulated cytochrome P450 family 1 (CYP1) drug metabolism and downregulated glycolysis gene signatures, which correlates with unfavorable prognosis. Screening on approved oncology drugs shows that cancer cell clusters on aligned stromal topography are more resistant to diverse chemotherapeutics. Full-dose drug testings further indicate that topography induces drug resistance of hormone receptor-positive breast cancer cell clusters to doxorubicin and tamoxifen and triple-negative breast cancer cell clusters to doxorubicin by activating the aryl hydrocarbon receptor (AhR)/CYP1 pathways. Inhibiting the AhR/CYP1 pathway restores reactive oxygen species-mediated drug sensitivity to migrating cancer cell clusters, suggesting a plausible therapeutic direction for preventing metastatic recurrence.

Introduction

Metastasis is the major driver of morbidity and mortality in cancer patients and can be seeded by the migration and systemic circulation of clusters of cancer cells [1]. Multicellular clusters formed by tumor budding and collective migration have increased survival in circulation relative to individual cancer cells and a 50-fold higher probability of colonizing distant organs [2]. These cancer cell clusters break away from the main tumor and are evident in histology as small tumor nests surrounded by a fibrillar stroma [3]. Their presence predicts poor prognosis in patients [4]. For example, morphometric scoring of tumor budding has been applied in clinical practice to identify patients with a higher risk of tumor relapse and a greater need for adjuvant therapy [5, 6].

During metastasis, cancer cells, especially multicellular clusters, undergo distinct survival adaptations compared to cancer cells in the tumor core [7]. Signaling network rewiring, such as shifts in metabolic dependency [8], epithelial-mesenchymal transition [9], and cytoskeletal reorganization [10], protect disseminating cancer cells from succumbing to environmental stress. In addition, disseminating as clusters shields vulnerable cancer cells after detaching from their main tumor. For example, decreased expression of immune cell-activating ligands on clustered cells makes them more resistant to immunosurveillance than single cells [11]. Furthermore, cluster organization promotes cancer cell survival in an E-cadherin dependent fashion [12]. Drugs that dissociate clusters into individual cells significantly suppress metastasis [13], indicating that cancer cell clustering is a critical yet incompletely understood step in their response to therapeutics.

The tumor microenvironment plays a vital role in regulating cancer cell migration toward metastasis [3]. Migrating cancer cells remodel their surrounding stromal fibers, which in turn both induces and guides cancer cell migration to create an ominous cycle toward metastatic progression [13]. Remodeled tumor stroma characterized by dense radially aligned fibers along the direction of cancer cell migration [13] has been observed in human cancer [14] and both in vitro [15] and in vivo [16] tumor models. Aligned stromal topography is frequently observed in surrounding migrating cancer cell clusters at the tumor border [14, 17] and correlates with unfavorable prognosis in both invasive and preinvasive tumors [18]. The existence of aligned stromal topography even prior to cancer cell migration and its predicting disease progression in breast ductal carcinoma in situ further justify the importance of aligned topography in cancer cell migration [18]. Our previous study indicates that aligned stromal topography is an independent factor in promoting cancer cell cluster migration by a 3D model in which tumor organoids readily contact radially aligned collagen fibers [19]. Moreover, the migratory phenotype of patient-derived cancer cells on a nanopatterned model, which recapitulates aligned stromal topography, also predicts tumor recurrence [20]. However, it remains unknown whether and how aligned topography drives molecular alterations in cancer cell clusters to promote cancer cell survival.

To further understand the molecular mechanism induced by aligned stromal topography in cancer cell clusters, we engineer a nanopatterned tumor model with highly controlled transitional topography recapitulating the stromal topographical transition from the tumor center to the migration front with aligned topography. We showed that aligned topography promotes cancer cell cluster migration and enhances chemoresistance through the antioxidant functions of the aryl hydrocarbon receptor (AhR)/cytochrome P450 family 1 (CYP1) pathway activated by mechanotransduction. Importantly, our findings suggest a potential strategy of combining AhR/CYP1 inhibitors with chemotherapy to target migrating cancer cell clusters and prevent metastatic recurrence.

Materials and Methods

Fabrication of a cancer cell migration model with biomimetic stromal topography

We designed a polyurethane acrylate (PUA) (Norland Optical Adhesive 76, Norland, USA) nanopatterned substrate that has 800 nm wide grooves to recapitulate the mean spacing between tumor stromal fibril bundles [21]. The substrates were generated using our established capillary force lithography technique [22]. In brief, we spread PUA prepolymer between a nanopatterned mold and the bottom of a multi-well plate. After UV curing, the nanopatterned mold was peeled off to form a high-fidelity nanopatterned substrate. The fidelity of the substrates has been validated using a scanning electron microscope in our previous study [22]. For unpatterned substrates, the same fabrication protocol was applied, except an unpatterned flat mold was used.

To allow a standardized analysis of cell migration, we initially confined the cell monolayer in a cell seeding area by customized polydimethylsiloxane (PDMS) blocks. Before cell seeding, the substrates were functionalized with 0.05 mg/ml collagen I (High Concentration Rat Tail, Corning, USA) at 37 °C for one hour. Next, PDMS blocks were attached to the substrates to allow a 2 mm-wide gap area for cell seeding. Cells were seeded at 100,000 cells/cm². After a confluent cell monolayer formed overnight, PDMS blocks were removed to initiate cell migration (Figure 1A).

Figure 1. Aligned parallel topography promotes cancer cell cluster migration.

(A) Overview and (B) computational morphometric analysis of nanopatterned cancer cell migration model. A cell monolayer was initially confined by a PDMS block. After the block was removed, cells started to migrate on the nanopatterned substrate with aligned nanogrooves recapitulating tumor stromal topography. After a 5-day migration on substrates with or without nanopatterns, the migration pattern was analyzed using a customized pipeline to obtain parameters at both cell cluster and single-cell levels. Representative fluorescent and segmented images of T47D cell sheets after a 5-day migration show more cell clusters detached from the main cell sheet on a parallel nanopatterned than unpatterned substrate. (C) Cell cluster number, (D) shape index of cell sheet (1/circularity), (E) average cell number in a cluster, and (F) cell aspect ratio on nanopatterned (NP) and unpatterned (UnP) substrates. (C-E) Data represent the mean ± SEM ($\mathrm{n}=8$ for T47D; $\mathrm{n}=4$ for MCF7). * p < 0.05; ** p < 0.01; **** p < 0.0001 by one-way ANOVA. (F) Box and whisker plot with median, min, and max. **** p < 0.0001 by one-way ANOVA.

Cell culture

MCF7, T47D, and MDAMB231 human breast cancer cell lines were obtained from the American Type Culture Collection (USA). MCF7 and MDAMB231 cells were maintained in Dulbecco’s Modified Eagle’s Medium. T47D cells were maintained in Roswell Park Memorial Cancer Institute 1640 medium, respectively (GIBCO, USA). All media were supplemented with 10% fetal bovine serum (Thermo Fisher Scientific, USA) and 1% penicillin-streptomycin (10,000 U/mL) (Thermo Fisher Scientific, USA). Cells were incubated at 37 °C in a 5% CO₂ humidified atmosphere. Cell authentication and mycoplasma test were performed.

Computational morphometric analysis of cell migration

After a 5-day migration, cells were stained with CellTracker Red CMTPX Dye (Thermo Fisher Scientific, USA) to label cytoplasm and Hoechst 33342 solution (Thermo Fisher Scientific, USA) to label nuclei. Then, cells were fixed with 4% paraformaldehyde (Thermo Fisher Scientific, USA) and imaged by a widefield microscope (Nikon TiE inverted widefield microscope, Japan). We established a computational morphometric analysis of cell migration using a customized CellProfiler pipeline (Broad Institute, USA) for image segmentation and quantification at both tissue and single-cell levels.

At the tissue level, our pipeline is comprised of thresholding by minimum cross entropy method, converting images to objects, and filtering objects according to their surface areas to recognize main cell sheets and migrating cell clusters. Migrating cell clusters were defined as objects without continuous connection with the main cell sheet in segmented images. At the single-cell level, our pipeline involved identifying primary objects from the nucleus by Otsu threshold, identifying secondary objects from cytoplasm by the propagation method, and filtering to exclude objects with an area larger or smaller than regular cell size. Single cells were then masked to be classified into cells of the main cell sheet or cells of a migrating cell cluster. The parameters of segmented images from both tissue and single-cell levels were then measured for quantitative analysis. A shape index of the cell sheet representing the complexity of the cell sheet border was defined as the reciprocal of circularity.

$$\mathrm{S}\mathrm{h}\mathrm{a}\mathrm{p}\mathrm{e}\mathrm{i}\mathrm{n}\mathrm{d}\mathrm{e}\mathrm{x}\mathrm{o}\mathrm{f}\mathrm{t}\mathrm{h}\mathrm{e}\mathrm{c}\mathrm{e}\mathrm{l}\mathrm{l}\mathrm{s}\mathrm{h}\mathrm{e}\mathrm{e}\mathrm{t}=\frac{1}{\text{circularity}}=\frac{{\text{perimeter}}^2}{4\mathrm{\pi }\times \text{area}}$$

Hybrid unpatterned-nanopatterned transitional model to isolate cell subpopulations from topography-induced cell migration

The hybrid transitional model was designed to have an unpatterned cell seeding area to mimic the tumor center without aligned stroma topography and a parallel nanopatterned area to mimic aligned stromal topography at the tumor migration front (Figure 2A and Supplementary Figure 2A). Cells were seeded and confined in the unpatterned area. After a confluent cell monolayer was formed, the PDMS confining block was removed to allow cells to migrate into the nanopatterned area (Figure 2A). After a 10-day migration, cells from the cell sheet center in the unpatterned area, the cell sheet edge in the nanopatterned area, and migrating cells in the nanopatterned area were isolated by microscope motor stage-controlled cutting (Supplementary Figure 2B). The borderlines between subareas were cut by a vinyl cutting blade (Cricut, USA) while controlling the movement of the motor stage. Cells from subareas, along with their underneath substrates, were immersed in RNAprotect Cell Reagent (Qiagen, USA). RNA was then extracted using the RNeasy Plus Micro Kit (Qiagen, USA). RNA integrity was determined using an Agilent 2200 TapeStation (Agilent Technologies, USA) prior to RNA sequencing.

Figure 2. Cytochrome P450 family 1 (CYP1) drug metabolism-related genes are upregulated in the topography-induced migrating cancer cell clusters.

(A) A hybrid transitional platform to separate cell subpopulations by topography-induced cancer cell migration. The unpatterned area simulates the tumor center in vivo without aligned topography, while the nanopatterned area recapitulates aligned topography at the tumor edge. A white arrow shows the orientation of aligned topography. Migrating cells detached from cell sheet after 10-day migration. Red rectangles label multicellular clusters and purple rectangles label single cells. Only attached cells are labeled. Microscope motor stage-controlled cutting was applied to separate cells from subareas. A bulk RNAseq was performed to identify gene signatures of migrating cells on the nanopatterned area compared to the confluent cells in the cell sheet center on the unpatterned area. The 20 upregulated and 127 downregulated genes were identified (Log[Fold change (FC)] threshold = 0.585 and false discovery rate (FDR) 5%). (B) The volcano plot shows differentially expressed genes between cells from migrating cells and confluent center. (C) The mRNA expression of representative up- or down-regulated genes was validated using qPCR in T47D and MCF7 cells. Fold changes are migrating cells on the nanopatterned area compared to the confluent center on the unpatterned area. Data represent the mean ± SEM ($\mathrm{n}=3$). * p < 0.05; ** p < 0.01; *** p < 0.001; **** p < 0.0001 by Student’s t-test for qPCR data. (D) Gene ontology enrichment analysis identifies CYP1 and the related xenobiotic metabolism as upregulated pathways and glycolysis as downregulated pathways of migrating cells. (E) Representative images of MCF7 cells cultured at low cell density (single cell), medium cell density (cell cluster), and high cell density (cell sheet) on unpatterned (UnP) or nanopatterned (NP) substrates. CYP1A1 and CYP1A2 RNA expression levels of (F) T47D and (G) MCF7 on nanopatterned versus unpatterned substrates. (F and G) Data represent the mean ± SEM ($\mathrm{n}=3$). * p < 0.05 by t-test between nanopatterned and unpatterned group at the same cell density.

RNA sequencing (RNAseq)

RNAseq libraries were prepared using the SMART-Seq v4 Ultra Low Input RNA Kit (Takara Bio, USA). The libraries were then sequenced employing a paired-end, 50-base read length sequencing strategy on an Illumina HiSeq 2500 (Illumina, USA). First, the low-quality reads were filtered. Reads passing the filtering were then aligned to the human genome (hg38) using STAR v2.5.2a in the 2-pass mapping mode [23]. Counts of each gene were generated using Subread featureCounts v1.6.0 [24], and genes that did not have at least one count per million in at least two samples were removed. A generalized linear model was applied to normalize data. A quasi-likelihood F-test method pairing samples by batch was used for significance testing in edgeR v3.20.9 [25]. Differentially expressed genes were defined by a cutoff of |log 2 (fold change) | ≥ 0.585 and false discovery rate (FDR) < 5%. We performed gene ontology enrichment analysis on the significantly differentially expressed genes using goseq v1.30.0 [26]. Up- and down-regulated genes were input separately against a concise set of gene ontology biological process terms from Gene Ontology (GO) Biological Processes [27]. We also performed a systems-level pathway enrichment analysis on Metascape, an integrated portal of OMICs datasets [28]. To identify upstream regulators, we performed transcription factor enrichment analysis on TRRUST v2 transcription factor reference database [29] by the Enrichr enrichment analysis tool [30].

Reverse transcription quantitative real-time PCR (RT-qPCR)

RNA was harvested using RNAprotect Cell Reagent (Qiagen, USA) and extracted by RNeasy Plus Micro Kit (Qiagen, USA). The extracted total RNAs were quantitated using a Nanodrop 2000 (Thermo Fisher Scientific, USA). Then, the extracted mRNAs were reverse-transcribed to cDNAs using the SuperScript IV First-Strand Synthesis System (Thermo Fisher Scientific, USA). cDNAs were quantified by real-time PCR with primers (Thermo Fisher Scientific, USA) and SsoAdvanced Universal SYBR Green Supermix (BioRad, USA). The primers used are listed in Supplementary Table 1.

Survival analysis of a gene signature of topography-induced cell cluster migration

We first evaluated the prognostic significance of the top 20 up- and 20 down-regulated genes from our RNAseq result in The Cancer Genome Atlas (TCGA) breast cancer dataset (version 09-15-2017) using a Cox univariate hazard model. The gene signature score representing topography-induced cell cluster migration was then generated by calculating the sum of the mRNA expression level of prognostically significant upregulated genes minus the sum of the mRNA expression level of the prognostically significant downregulated genes. Median values were set as the cutoffs to stratify patients into high or low score groups. Finally, the power of prognostic prediction of the gene signature score was evaluated on the TCGA breast cancer dataset and a GEO breast cancer dataset, GSE19536 [31] by Kaplan–Meier survival analysis. The same algorithm was applied to analyze the prognostic significance of a gene signature score least correlated with the topography-induced cell cluster migration. Analyses were performed using SPSS 25 (IBM, USA).

Approved oncology drugs (AOD) screening

To test the effects of stromal topography on cancer cell clusters’ response to oncology drugs, we utilized PUA 96-well plates (96-Well NanoSurface Plate, Curi Bio, USA) for high throughput drug screening (Figure 4A). Cancer cells were then seeded at a density of 2 × 10⁴ cells/cm² to form cell clusters on nanopatterned or unpatterned 96-well plates using a multidrop reagent dispenser (Multidrop Combi Reagent Dispenser, Thermo Fisher Scientific, USA) and incubated for one day before adding a drug library of approved oncology agents. The Approved Oncology Drugs set (AOD IX) containing 147 FDA-approved anticancer drugs was obtained from the National Cancer Institute. Each drug was distributed and diluted into appropriate wells of the nanopatterned or unpatterned 96-well plates by an automated simultaneous pipettor (CyBi-well 96-Channel Simultaneous Pipettor, CyBio, Germany) to yield the final concentration of 1 μM in the culture medium. Eight negative control (DMSO) wells and eight positive control (doxorubicin) wells were included in each plate. After 72-hour drug treatment, cell viability was assessed using PrestoBlue HS Cell Viability Reagent (Thermo Fisher Scientific, USA) on a microplate reader (CLARIOstar Plus, BMG Labtech, Germany).

Figure 4. Cancer cell clusters on aligned topography exhibit drug resistance to chemotherapy.

(A) Schematic representation and (B) study design flowchart of the Approved Oncology Drug library (NCI AOD IX) screening on 96-well nanopatterned and unpatterned platforms. Drugs are classified based on whether the cells are sensitive to or not. Only drugs to which cells are sensitive to are included in assessing topography-induced drug resistance. (C) Drug response profile heatmap shows the differences in cell viability between nanopatterned and unpatterned substrates after drug treatment. Only the drugs that T47D or MCF7 cells are sensitive to are listed. Drugs with a higher Z-score of cell viability percentage change indicate topography-induced drug resistance. (D) Cancer cell clusters on aligned topography show significantly increased resistance to chemotherapy than targeted therapy drugs. Data represent the mean ($\mathrm{n}=3$). The $p$ values are determined by the paired t-test.

The drug screening algorithm includes quality control, classifying drugs based on the sensitivity of cells to drugs, determining topography-induced drug resistance, ranking drug hits by topography-induced resistance, and full dosage validation of top drug hits (Figure 4B). Each plate was first evaluated for quality control using the robust strictly standardized mean difference (SSMD’) criteria [32].

$${\mathrm{S}\mathrm{S}\mathrm{M}\mathrm{D}}^{\prime }=\frac{{\mathrm{X}}_{\mathrm{P}}-{\mathrm{X}}_{\mathrm{N}}}{1.4826\sqrt{{\mathrm{S}}_{\mathrm{P}}^2+{\mathrm{S}}_{\mathrm{N}}^2}}$$

where ${\mathrm{X}}_{\mathrm{P}},{\mathrm{X}}_{\mathrm{N}},{\mathrm{S}}_{\mathrm{P}}$, and ${\mathrm{S}}_{\mathrm{N}}$ are the medians and median absolute deviations of the positive and negative controls on unpatterned plates. A cutoff value of ${\mathrm{S}\mathrm{S}\mathrm{M}\mathrm{D}}^{\mathrm{’}}\le -2$ was applied. Then a control-based normalization of signals on unpatterned plates was used to classify drugs according to the cell sensitivity to drugs. The percentage of control (POC) represents the normalized cell viability after drug treatment.

$$\mathrm{N}\mathrm{o}\mathrm{r}\mathrm{m}\mathrm{a}\mathrm{l}\mathrm{i}\mathrm{z}\mathrm{e}\mathrm{d}\mathrm{c}\mathrm{e}\mathrm{l}\mathrm{l}\mathrm{v}\mathrm{i}\mathrm{a}\mathrm{b}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{y}(\mathrm{P}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e}\mathrm{o}\mathrm{f}\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l},\mathrm{P}\mathrm{O}\mathrm{C})=\frac{{\mathrm{X}}_{\mathrm{i}}}{{\mathrm{\mu }}_{\mathrm{N}}}$$

where ${\mathrm{X}}_{\mathrm{i}}$ is the raw measurement of drug $\mathrm{i}$ and ${\mathrm{\mu }}_{\mathrm{N}}$ is the mean of the negative controls on the same unpatterned plate. All drugs were ranked by their Z-score of normalized cell viability.

We then tested if cancer cells acquire a topography-induced resistance to drugs on nanopatterned substrates. Drugs that the cells are not sensitive to are excluded from testing the hypothesis of topography-induced drug resistance because T47D or MCF7 cells are intrinsically not sensitive to those drugs. Cell viability percentage change represents the difference in normalized cell viability between nanopatterned (NP) and unpatterned (UnP) plates.

$$\mathrm{C}\mathrm{e}\mathrm{l}\mathrm{l}\mathrm{v}\mathrm{i}\mathrm{a}\mathrm{b}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{y}\mathrm{p}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{g}\mathrm{e}\mathrm{c}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{e}\mathrm{s}=\frac{{\mathrm{P}\mathrm{O}\mathrm{C}}_{\mathrm{N}\mathrm{P}}-{\mathrm{P}\mathrm{O}\mathrm{C}}_{\mathrm{U}\mathrm{n}\mathrm{P}}}{{\mathrm{P}\mathrm{O}\mathrm{C}}_{\mathrm{U}\mathrm{n}\mathrm{P}}}\times 100$$

A drug with a Z-score of cell viability percentage change above 0 displays topography-induced drug resistance. All 147 drugs were ranked by their Z-score of relative cell viability to select the drug hits most related to topography-induced drug resistance. The top three drugs and other standard therapeutic agents for breast cancer were then tested by an 11-dose cytotoxicity assay with a 3-fold serial dilution from 100 μM to validate the drug resistance induced by topography. Drugs, including doxorubicin hydrochloride (Sigma-Aldrich, USA), idarubicin hydrochloride (Sigma-Aldrich, USA), temozolomide (Sigma-Aldrich, USA), tamoxifen citrate (Sigma-Aldrich, USA), palbociclib (Sigma-Aldrich, USA), and everolimus (Sigma-Aldrich, USA) were used.

Lentiviral-based shRNA transduction and inhibitors

Lentiviral particles expressing human AHR shRNA and a non-target shRNA (Mission shRNA, Sigma-Aldrich, USA) with a multiplicity of infection of 10 were applied to knockdown gene expression. Cells were infected with the lentiviral particles in the presence of 8 μg/mL polybrene (Sigma-Aldrich, USA). After a 16-hour infection, media were replaced with fresh 1 μg/ml puromycin-containing media (Sigma-Aldrich, USA). The AHR shRNA (TRCN000024528) sequence was CCGGATCCACAGTCAGCCATAATAACTCGAGTTATTATGGCTGAC-TGTGGATTTTTTG, and the non-target shRNA sequence (negative control) was CCGGGCGCGATAGCGCTAATAATTTCTCGAGAAATTATTAGCGCTATCGCGCTTTT.

An AhR inhibitor, CH223191 (PeproTech, USA), at 10 μM [33] or a CYP1 inhibitor, Alizarin (MilliporeSigma, USA), at 10 μM [34] was used to block the AhR/CYP1 signaling pathway. An actin polymerization inhibitor, latrunculin A, at 500 nM [35] was applied to block the mechanotransduction from aligned topography. Cells were treated with or without the inhibitors together with the chemotherapy or hormone therapy agents for 72 hours.

Immunofluorescent staining

Cells were fixed with 4% paraformaldehyde (Thermo Fisher Scientific, USA) and permeabilized by 0.2% Triton X-100. Then, the cells were incubated with AhR (MA1-513, 1:100, Thermo Fisher Scientific, USA) and CYP1A1 (13241–1-AP, 1:500, Proteintech, USA) antibodies at 4°C overnight. Cells were then incubated with secondary antibodies, goat anti-rabbit IgG (H+L) Alexa Fluor 488 and goat anti-mouse IgG (H+L) Alexa Fluor 594 (Invitrogen, USA), at room temperature for 1 hour and DAPI (Thermo Fisher Scientific, USA) for 15 minutes, then imaged by a spinning disk confocal microscope (Nikon TiE inverted widefield microscope and Yokogawa W1 spinning disk, Japan).

Reactive oxygen species (ROS) detection assay

The oxidative stress level in cells was detected by CellROX Green Reagent (Thermo Fisher Scientific, USA). CellROX at a final concentration of 5 μM and CellTracker Red CMTPX Dye (Thermo Fisher Scientific, USA) at a final concentration 1 μM were incubated at 37 °C for 30 minutes then washed away with PBS once before microscope imaging. A customized ImageJ macro was used to quantify the corrected total cell fluorescence (CTCF).

$$\mathrm{C}\mathrm{T}\mathrm{C}\mathrm{F}=\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{g}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{d}\mathrm{d}\mathrm{e}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{y}-(\mathrm{c}\mathrm{e}\mathrm{l}\mathrm{l}\mathrm{a}\mathrm{r}\mathrm{e}\mathrm{a}\times \mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}\mathrm{f}\mathrm{l}\mathrm{u}\mathrm{o}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{c}\mathrm{e}\mathrm{o}\mathrm{f}\mathrm{b}\mathrm{a}\mathrm{c}\mathrm{k}\mathrm{g}\mathrm{r}\mathrm{o}\mathrm{u}\mathrm{n}\mathrm{d})$$

The area of each cell was first segmented by CellTracker images, and the segmented mask of each cell was applied to measure the integrated density of ROS staining in each cell. The segmented mask of cells was then inverted to measure the mean fluorescence of the background.

Statistical analysis

All results were generated from at least three independent experiments. Data were analyzed using GraphPad Prism 9 (GraphPad Software, USA) unless noted. One-way ANOVA was used to compare differences among more than two groups. Pearson’s correlation coefficient was applied for correlations between groups. The extra-sum-of-squares F test was applied to compare IC₅₀ values of dose-response curve fits. Statistical differences were considered significant at $\mathrm{p}<0.05$.

Results

Aligned parallel topography recapitulating tumor stromal structure promotes cancer cell cluster migration

To recapitulate the aligned tumor stromal topography of invasive breast cancer [14], we fabricated a nanopatterned substrate model containing 800-nm wide parallel nanogrooves (Figure 1A). In order to mimic cancer cells budding and detaching from a tumor mass containing tightly packed cancer cells, we first confined T47D and MCF7 breast cancer cells on the nanopatterned substrate to form a confluent cell sheet by cell-cell adhesion. We applied T47D and MCF7 cells to represent hormone receptor-positive breast cancer, the most common molecular subtype of breast cancer with a risk of late recurrence, suggesting the importance of understanding how cancer cells survive during migration [36]. After releasing the confinement, cancer cells started to migrate along the parallel nanogrooves. After a 5-day migration, cell sheets were visualized by cytoplasmic and nuclear staining and segmented at both tissue and single-cell levels (Figure 1B).

At the tissue level, migrating cell clusters were distinguished by their discontinuity from the main cell sheet. Morphometric analysis showed a significant increase in the number of cell clusters on nanopatterned substrates (Figure 1C and Supplementary Movie 1 and 2). The shape index of the cell sheet, which equals the reciprocal of circularity (Figure 1D), was higher on nanopatterned topography since the cell sheet perimeter was higher with no change in surface area (Supplementary Figure 1A). This indicated a more complex cell sheet border on nanopatterned topography. Likewise, time-lapse imaging of MCF7 cells on nanopatterned topography showed an increased number and surface area of cell clusters and a greater distance between cell clusters and the main cell sheet (Supplementary Figure 1D–G and Supplementary Movie 3).

At the single-cell level, cells were classified into subpopulations according to their location (Figure 1B). Increased cell number in cell clusters on nanopatterned substrates (Figure 1E) showed that aligned topography promotes cancer cell migration as multicellular clusters. With the increase in cell sheet width and no change in cell sheet surface area, the cell density on nanopatterned substrates was significantly lower than on unpatterned substrates (Supplementary Figure 1A). In addition, cells in cell clusters on nanopatterned substrates had a higher cell and nuclear aspect ratio than cells in main cell sheets and on unpatterned substrates (Figure 1F and Supplementary Figure 1B and 1C). Our morphometric analyses showed that aligned topography induced the dissociation of a cell sheet into multicellular clusters during migration, recapitulating the tumor microarchitecture of human breast cancer.

Topography-induced migrating cancer cell clusters display an upregulated CYP1-related drug metabolism pathway

To define the transcriptomic signature of topography-induced migrating cancer cell clusters, we developed a hybrid unpatterned-nanopatterned transitional model mimicking in vivo tumors in which stromal fiber alignment occurs at the tumor-stroma interface, not in the tumor core (Figure 2A and Supplementary Figure 2A). T47D breast cancer cells were seeded and confined in the unpatterned area to form a confluent monolayer. Then, after the removal of confining blocks, cells started to migrate collectively into the parallel nanopatterned area. After 10-day migration, we observed finger-like projections at the edge and cell clusters fully detached from the main cell sheet in the nanopatterned area (Figure 2A). We isolated cell subpopulations from the cell sheet center in the unpatterned area (around 100,000 cells), the cell sheet edge in the nanopatterned area (around 80,000 cells), and migrating cells on the nanopatterned area (around 10,000 cells) by microscope motor stage-controlled cutting (Figure 2A and Supplementary Figure 2B), then performed bulk RNAseq to identify gene signatures and signaling pathways specific to each cell subpopulation.

The principal component analysis showed that the gene expression profiles of migrating cells in nanopatterned areas and the cell sheet center in the unpatterned area were clearly separated (Supplementary Figure 3A). On the other hand, the gene expression profiles of the cell sheet edge in nanopatterned areas were between the cell sheet center and migrating cells and close to each other (Supplementary Figure 3A). Furthermore, the differential expression analysis showed no differentially expressed genes between the edge and center groups (Supplementary Figure 3B). Therefore, we focused on comparing migrating cells in nanopatterned areas and the cell sheet center in the unpatterned area. We identified 20 upregulated and 127 downregulated genes in migrating cells in nanopatterned areas compared to the cell sheet center in unpatterned areas (Figure 2B). Key upregulated genes in the cytochrome P450 family 1 included CYP1A1 (6.40-fold), CYP1A2 (10.03-fold), and CYP1B1 (1.76-fold). The most significantly downregulated changes were glycolysis-related genes, LDHA (2.34-fold), PDK1 (3.42-fold), and ALDOC (3.25-fold), and hypoxia-related genes, PPFIA4 (77.74-fold), CA9 (25.61-fold), NDRG1 (5.29-fold), and SPAG4 (5.37-fold). We validated the representative differentially expressed genes by qPCR in both T47D and MCF7 cells (Figure 2C). In addition, the differentially expressed genes between migrating cells and cell sheet centers, such as upregulated CYP1 and downregulated PPFIA4, were also observed between migrating cells and cell sheet edge, indicating a sequential change during the topography-induced cell migration (Supplementary Figure 3C). Because of the limited differentially expressed gene number, 8 upregulated and 23 downregulated genes, between migrating cells and the cell sheet edge, we again focused on comparing migrating cells and the cell sheet center for the functional annotation analyses. The results showed that CYP1-modulated pathways, including drug and xenobiotic metabolism, were the key upregulated pathways, while glycolysis and hypoxia were the main downregulated pathways (Figure 2D and Supplementary Figure 4A).

On the other hand, we did not observe an epithelial-mesenchymal transition (EMT) either by the EMT score [37] of our RNAseq data (Supplementary Figure 4B) or by qPCR of VIM, TWIST, and FN1 mRNA expressions in T47D cells (Supplementary Figure 4C), suggesting that the elongation of cell morphology on a nanopatterned substrate is more likely due to cytoskeleton reorganization induced by contact guidance without triggering an EMT [38]. In addition, no difference in the expression of CDH1, CDH2, and EPCAM suggested that cell-cell adhesion did not change during the topography-induced cell cluster migration (Supplementary Figure 4D).

Next, we examined whether stromal topography is an independent factor or confounded by cell density with respect to inducing the differential expression of the CYP1- and glycolysis-related genes. To test this, we cultured T47D or MCF7 cells on aligned nanopatterned or unpatterned substrates for 5 days at low (3,000 cells/cm²), medium (20,000 cells/cm²) and high (100,000 cells/cm²) seeding densities to form single cells, multicellular clusters, or confluent monolayers, respectively (Figure 2E). Directly seeding and culturing cells on the substrates instead of cell migration from a cell sheet also helped to exclude the possibility that the identified transcriptomic signatures were from a selection of highly motile cells or developed during cell migration. We cultured cells on the substrates for 5 days but not 10 days as in the cell migration assay because, in the cell migration assay, cell cluster detaching from the main cell sheet only became apparent after day 3 (Supplementary Figure 1E). Short-term culture of 5 days also kept the desired cell density to evaluate the role of cell density in topography-induced gene expression changes. The results showed a significantly increased RNA expression of CYP1A1 and CYP1A2 in multicellular clusters on nanopatterned compared to those on unpatterned substrates by qPCR (Figure 2F and 2G) and increased CYP1A1 protein expression by western blot (Supplementary Figure 5A). On the other hand, no significant topography-induced CYP1A1 and CYP1A2 upregulation were found in single cells or cells in a confluent monolayer. The expression of CYP1A1 and CYP1A2 in multicellular clusters on nanopatterned versus unpatterned substrates was higher by 73-fold and 632-fold in T47D cells, and 128-fold and 11-fold in MCF7 cells. In contrast, no significant decrease was observed in glycolysis-related genes, LDHA and PDK1, suggesting that topography does not downregulate the glycolysis genes independently from cell density (Supplementary Figure 5B). Our results demonstrate that migrating cancer cell clusters on aligned topography are distinguishable from confluent cells by upregulation of CYP1 via the synergistic effect of stromal topography and cell density.

The gene signature of topography-induced cell cluster migration predicts poor prognosis in breast cancer patients

We next sought to determine if the gene signature representing topography-induced cell cluster migration has a prognostic value. We first evaluated the top 20 up- and 20 down-regulated genes from our RNAseq results for their prognostic potential (Figure 3A). The Cox univariate hazard model analysis of the identified genes in the TCGA breast cancer dataset showed that six upregulated genes (CYP1A1, IL1R1, RHOBTB3, ID1, CLEC3A, and ID3) were correlated with a higher risk of death, and one upregulated gene (TMEM86A) was correlated with a lower risk of death. For the top 20 downregulated genes, five genes (SPAG4, TFF3, CDC20B, TIMP1, and SLC16A3) were correlated with a lower risk of death, and three genes (NDRG1, BNIP3, and PYGL) were correlated with a higher risk of death. To examine the prognostic effect of genes unbiasedly representing topography-induced cell cluster migration, we grouped the genes according to their up- or down-regulation by topography (Figure 3B). A 15-gene signature score was calculated by the expression level sum of seven upregulated genes with prognostic significance minus the expression level sum of eight downregulated genes with prognostic significance. The median value of the score was applied as the cutoff to stratify breast cancer patients into high or low score groups. The patients with a higher score, whose tumors shared a gene expression profile with topography-induced cell cluster migration, had a poor prognosis in the TCGA breast cancer dataset $(p=0.000049)$ (Figure 3C) and a GSE breast cancer dataset $(p=0.0461)$ (Supplementary Figure 6). Moreover, in molecular subtypes of breast cancer, similar results are observed in hormone receptor-positive patients $(p=0.0020)$ (Figure 3D) and triple-negative patients $(p=0.0485)$ (Figure 3F), suggesting topography may affect the response of hormone receptor-positive and triple-negative breast cancer cells to some of their standard therapeutic agents, such as chemotherapies, hormone therapies, or targeted therapies. Therefore, in the following drug testing, we focused on evaluating whether and how topography induces drug resistance of breast cancer cells to clinically used oncology drugs. On the other hand, a non-statistically significant trend in prognostic value in HER2-positive patients $(p=0.2224)$ may be due to a limited patient number (Figure 3E).

Figure 3. The gene signature of topography-induced cell cluster migration predicts poor prognosis in breast cancer patients.

(A) Hazard ratios (HR) of the top 20 up- and 20 down-regulated genes of topography-induced cell cluster migration in TCGA breast cancer dataset by Cox univariate hazard model. (B) The principle of generating a 15-gene signature score representing topography-induced cell cluster migration. Kaplan–Meier survival analysis of the 15-gene signature score in (C) TCGA breast cancer dataset and in (D) hormone receptor-positive, (E) HER2 positive, and (F) triple-negative breast cancer from the TCGA database. The p values are determined by the logrank test.

To verify that the prognostic significance of our gene signature is not a result of stochastic events, we applied the same algorithm to establish a 12-gene signature score from the bottom 20 up- or 20 down-regulated genes in our RNAseq data to represent gene profiles least correlated with topography-induced cell cluster migration (Supplementary Figure 7A and 7B). The 12-gene signature score least correlated with topography-induced cell cluster migration had no prognostic significance in breast cancer patients (Supplementary Figure 7C). Our survival analysis suggests that our topographical model is reliable for identifying transcriptomic changes that bear prognostic value in cancer patients.

Aligned topography enhances drug resistance of breast cancer cell clusters to chemotherapy and hormone therapy but not targeted therapy

To phenotypically evaluate topography-induced drug metabolism, we examined the effects on breast cancer cell viability with or without aligned topography of drugs from the NCI Approved Oncology Drug IX library containing 147 FDA-approved cancer therapeutic agents (Figure 4A). T47D or MCF7 breast cancer cells were seeded at a medium density to form multicellular clusters on nanopatterned or unpatterned 96-well plates and then treated with the Approved Oncology Drug library to assess how the drug response is affected by aligned topography. The analysis algorithm includes quality control, drug classification based on the intrinsic drug sensitivity of cells, topography-induced drug resistance determined by relative cell viability, hit ranking, and validation of top hits (Figure 4B). For quality control, we only included experiment batches with a robust $\mathrm{S}\mathrm{S}\mathrm{M}\mathrm{D}\le -2$. This indicates that the cell viability of positive control (doxorubicin) is at least two times the median absolute deviation lower than the negative control (DMSO) on the unpatterned plates.

Before determining whether topography induces drug resistance in cancer cells, we first excluded the possible confounding factors that may affect the analysis. First, because cell proliferation rate may affect drug sensitivity, we tested if topography influences cell proliferation. The result showed no significant difference in cell viability between negative controls on nanopatterned and unpatterned plates at the end of drug testing (Supplementary Figure 8A). Likewise, no difference in cell numbers growing on nanopatterned and unpatterned plates for 10 days (Supplementary Figure 8B). This excluded the possibility that topography-induced drug resistance was due to differences in baseline cell proliferation rates. Then, we excluded drugs that cells are intrinsically resistant to, which did not induce higher cell death than the negative control on unpatterned substrates, since these drugs could not be used to assess the level of cell viability rescued by topography. To classify the drugs to which cells are intrinsically sensitive or resistant, we performed an independent viability assay of cells after drug treatment on unpatterned substrates for T47D and MCF7, respectively. The drugs the cells were intrinsically sensitive to were defined by a Z-score of normalized cell viability below 0, which indicated higher cell death than the negative control (Figure 4B). In the subsequent analysis determining whether topography induces drug resistance, only the drugs that cells are intrinsically sensitive to would be included.

To evaluate whether topography induces drug resistance, we analyzed T47D and MCF7 cell clusters treated with the drugs that cells are intrinsically sensitive to and showed higher viability on nanopatterned than unpatterned plates (Figure 4C). Interestingly, the topography-induced drug resistance was more significant with chemotherapy than with targeted therapy (Figure 4C and 4D). In contrast, T47D and MCF7 cell clusters did not show further topography-induced resistance to the drugs that cells are intrinsically resistant to (Supplementary Figure 8C). The results were consistent in T47D and MCF7 cell lines (cosine similarity ~ 0.73) (Supplementary Figure 8D) and between independent experiments (cosine similarity 0.76, 0.67, 0.66 between three independent experiments of T47D cells) (Supplementary Figure 9).

Doxorubicin, one of the chemotherapy agents for high-risk hormone receptor-positive breast cancer patients such as rapid progression or high tumor burden [39], was identified as the top drug to which T47D and MCF7 cells exhibited the most topography-induced drug resistance (Figure 5A). To validate the drug screening results, we performed a full-dose cytotoxicity assay on the top three drugs (doxorubicin, idarubicin, and temozolomide) and observed significantly higher IC₅₀ against cancer cell clusters on nanopatterned than on unpatterned plates (Figure 5B). Hormone therapies such as tamoxifen are the standard treatment of hormone receptor-positive breast cancers. Although tamoxifen did not show as large an increase in topography-induced drug resistance as doxorubicin in single-dose drug screening, a significantly higher IC₅₀ against cancer cell clusters on nanopatterned than on unpatterned plates was found in a full-dose cytotoxicity assay (Figure 5C). In addition, the high enrichment of upregulated genes in topography-induced cell cluster migration in estrogen receptor-positive breast cancer with tumor recurrence after tamoxifen therapy suggested that the sensitivity of hormone receptor-positive breast cancer cell clusters to tamoxifen is affected by topography (Supplementary Figure 10A).

Figure 5. Aligned topography induces resistance of breast cancer cell clusters to chemotherapy and hormone therapy but not targeted therapy.

(A) Doxorubicin, idarubicin, and temozolomide are the top drug hits most related to the topography-induced drug resistance in T47D and MCF7 cell clusters. This scatter plot is a condensed form of supplementary figure 10C, which shows all 147 drugs. Increases in the IC₅₀ of (B) chemotherapy (doxorubicin, idarubicin, and temozolomide), (C) hormone therapy (tamoxifen), but not (D) targeted therapies (palbociclib and everolimus), against T47D cell clusters on nanopatterned (NP) compared to unpatterned (UnP) substrates. (E) Increases in the IC₅₀ of doxorubicin against triple-negative breast cancer cell clusters, MDAMB231, on nanopatterned compared to unpatterned substrates. (B-E) Data represent the mean ± SEM ($\mathrm{n}=3$). The $p$ values are determined by extra-sum-of-squares F test to compare IC₅₀ values of dose-response curve fits between groups.

On the other hand, standard targeted therapies used in hormone receptor-positive breast cancer, such as CDK4/6 inhibitor, palbociclib, and mTOR inhibitor, everolimus did not show a significantly higher IC₅₀ against T47D cancer cell clusters on nanopatterned than on unpatterned plates (Figure 5D), suggesting that they remained effective in killing cancer cells even with aligned topography. Likewise, the drug screening results showed that topography-induced drug resistance did not affect several targeted therapy drugs (Supplementary Figure 11). For example, panobinostat, which was effective in killing T47D cancer cells on both unpatterned and nanopatterned substrates, is a histone deacetylase inhibitor tested in a clinical trial to restore drug sensitivity in hormone therapy-resistant breast cancer [40]. Thus, our results may also help identify drugs that remain effective in treating breast cancer cell clusters.

Furthermore, we examined if the topography-induced chemoresistance is also evident in triple-negative breast cancer, as its treatment relies mainly on chemotherapy. Doxorubicin showed a significantly higher IC₅₀ against MDAMB231 breast cancer cell clusters on nanopatterned than on unpatterned plates (Figure 5E). In addition, a significant correlation was observed between the expression levels between upregulated genes in topography-induced cell cluster migration and those in paclitaxel-resistant MDAMB231 breast cancer cells (Supplementary Figure 10B). These findings support that topography induces chemoresistance in breast cancer cells regardless of their hormone receptor status.

Aligned topography activates AhR/CYP1 signaling by actin-regulated mechanotransduction to promote chemoresistance

Next, to elucidate the mechanisms of topography-induced chemoresistance, we performed transcription factor enrichment analysis for the upregulated genes from our RNAseq data. Aryl hydrocarbon receptor (AhR) is predicted as a top upstream regulator (Figure 6A), followed by Aryl Hydrocarbon Receptor Nuclear Translocator (ARNT), the heterodimerization partner of AhR to bind to the promoter and activate downstream genes [41]. While AhR is usually activated by xenobiotics [41], mechanical cues from cell microenvironments such as fluid shear stress have been shown to translocate AhR to nuclei and upregulate downstream CYP1 signaling [42]. Our results showed higher nuclear AhR and cytoplasmic CYP1A1 protein expression in T47D cell clusters on aligned nanopatterned substrates compared to unpatterned substrates, indicating AhR/CYP1 signaling activation by aligned topography (Figure 6B). Then, we tested the dependency of topography-induced chemoresistance on AhR/CYP1 signaling by genetic and chemical perturbation. Knockdown of AhR significantly decreased the expression of its downstream cytochrome P450 family 1 genes CYP1A1, CYP1A2, and CYP1B1 (Supplementary Figure 12). It restored the sensitivity of T47D cell clusters to doxorubicin on aligned topography $(p<0.0001)$ but not on unpatterned substrates $(p=0.496)$ (Figure 6C). The CYP1 inhibitor, Alizarin, also reversed the resistance to doxorubicin in T47D cell clusters induced by aligned topography $(p=0.0007)$ (Figure 6D). Likewise, the CYP1 inhibitor, Alizarin, restores the sensitivity of T47D cell clusters to tamoxifen (Figure 6E) and MDAMB231 cell clusters to doxorubicin (Figure 6F) on nanopatterned but not unpatterned substrates. These findings reveal AhR/CYP1-dependent topography-induced chemoresistance as a new candidate mechanism of drug resistance in cancer.

Figure 6. Inhibiting AhR/CYP1 signaling rescues the chemosensitivity of cancer cell clusters on aligned topography.

(A) Transcription factor analysis shows AhR as an upstream regulator of topography-induced chemoresistance. (B) Immunofluorescent staining shows increases in AhR nuclear expression and CYP1A1 cytoplasm expression of T47D cells on aligned topography. A white arrow shows the orientation of aligned topography. Inhibiting AhR/CYP1 signaling by (C) an shRNA targeting AHR (shAHR) or (D) CYP1 inhibitor (CYP1i) at 10 μM rescues the sensitivity of T47D cells to doxorubicin. CYP1 inhibitor 10 μM rescues (E) the sensitivity of T47D cells to tamoxifen and (F) MDAMB231 cells to doxorubicin. (C-F) Data represent the mean ± SEM ($\mathrm{n}=3$). * p < 0.05; *** p < 0.001; **** p < 0.0001 by one-way ANOVA.

Mechanotransduction through actin and focal adhesion transmits the effect of aligned topography on cells to alter their morphology and guide migration [21]. Therefore, we hypothesized that blocking mechanotransduction inhibits topography-induced chemoresistance. Latrunculin has been shown to change cell morphology and migration by preventing actin polymerization and focal adhesion activation [35, 43, 44]. Our results showed that treating MDAMB231 cell clusters with latrunculin A significantly decreased the cell aspect ratio on nanopatterned substrates but less significantly on unpatterned substrates (Figure 7A and 7B) and a trend of reducing cell number in a cell cluster (Figure 7C). Furthermore, latrunculin A treatment diminished the upregulation of AhR and CYP1A1 expression induced by aligned topography (Figure 7D). Next, we evaluated the effect of latrunculin A on topography-induced chemoresistance and showed latrunculin A treatment sensitized cell clusters to doxorubicin on nanopatterned but not unpatterned substrates (Figure 7E). Our findings suggest that aligned topography upregulates the AhR/CYP1 pathway by mechanotransduction to promote chemoresistance.

Figure 7. Interrupting actin-regulated mechanotransduction blocks AhR/CYP1 signaling and diminishes topography-induced chemoresistance of cancer cell clusters.

(A) Representative images, (B) cell aspect ratio, and (C) average cell number in a cluster of MDAMB231 cells treated with or without actin polymerization inhibitor (actin i) latrunculin A (Lat A) at 500 nM for 72 hours on unpatterned and nanopatterned substrates. A white arrow shows the orientation of aligned topography. (D) Immunofluorescent staining shows decreased AhR and CYP1A1 expression of MDAMB231 cells on aligned topography after being treated with latrunculin A at 500 nM for 72 hours. A white arrow shows the orientation of aligned topography. (E) Cell viability of MDAMB231 cells treated with doxorubicin at 0.05 μM and with or without actin polymerization inhibitor (actin i) latrunculin A (Lat A) at 500 nM for 72 hours. (B) Box and whisker plot with median, min, and max. **** p < 0.0001 by one-way ANOVA. Data are from three independent experiments. (C and E) Data represent the mean ± SEM ($\mathrm{n}=3$). * p < 0.05; ** p < 0.01 by one-way ANOVA.

AhR/CYP1 signaling regulates topography-induced chemoresistance by reducing oxidative stress

Our drug screening results revealed that topography broadly increases the resistance of cancer cell clusters to diverse classes of chemotherapeutic agents (Figure 4C). One of the cytotoxic mechanisms of chemotherapeutic agents is to generate overwhelming oxidative stress beyond the survival threshold of cancer cells [45]. Hydrogen peroxide biosynthesis was one of the significantly upregulated pathways in topography-induced cell cluster migration (Figure 2D), suggesting that topography-induced chemoresistance may relate to reactive oxygen species (ROS) dysregulation. In addition, the differential therapeutic sensitivity of cancer cells on substrates with or without aligned topography was greatest for doxorubicin, a drug known to promote cancer cell death by increasing ROS levels [46]. Furthermore, the AhR/CYP1 pathway has been shown to have antioxidative functions by downregulating intracellular ROS levels [47]. Therefore, we hypothesized that the topography-induced chemoresistance mechanism is based on increased resistance to oxidative stress. To test this hypothesis, we evaluated the ROS level of cancer cell clusters generated by doxorubicin treatment with or without the effect of topography. Our results revealed that, after doxorubicin treatment, ROS levels of cells were significantly lower on nanopatterned than on unpatterned substrates (Figure 8A and 8B). Furthermore, inhibitors of AhR or CYP1 restored ROS levels of cells on aligned topography, suggesting that AhR/CYP1 mediates topography-induced chemoresistance through its antioxidant effect (Figure 8A and 8B). In summary, aligned tumor stromal topography promotes chemoresistance in migrating cancer cell clusters through the antioxidant functions of AhR/CYP1 signaling. Our findings support a model in which blocking AhR/CYP1 signaling rescues the oxidative stress generated by chemotherapy and thus inhibits the survival advantages of migrating cancer cell clusters (Figure 8C).

Figure 8. Aligned topography induces chemoresistance of breast cancer cell clusters by antioxidant functions of AhR/CYP1 pathway.

(A) Representative images and (B) fluorescent intensity of reactive oxygen species (ROS) levels in T47D cell clusters indicate that aligned topography represses doxorubicin-induced ROS. Combined treatments of AhR or CYP1 inhibitors (AhRi or CYP1i) at 10 μM with doxorubicin restore the ROS levels generated by doxorubicin at 1 μM. Data represent the mean ± SEM ($\mathrm{n}=3$). ** p < 0.01 and **** p < 0.0001 by one-way ANOVA. (C) A schematic representing the mechanism of topography-induced chemoresistance. Aligned topography of tumor stroma facilitates multicellular cluster migration. Cancer cell clusters on aligned topography upregulate the AhR/CYP1 signaling pathway to reduce chemotherapy-induced ROS, resulting in chemoresistance. Conversely, inhibiting the AhR/CYP1 signaling pathway blocks the antioxidant effect triggered by aligned topography and reverses the chemoresistance of cancer cells.

Discussion

In the current study, we have applied a novel topography-induced cell migration model to provide evidence for a link between stromal topography and drug resistance in cancer. It is known that the mechanical characteristics of the stroma, such as stiffness and shear stress, can promote chemoresistance [48, 49]. For example, the doxorubicin resistance of breast cancer cells can be enhanced by stiffer stroma via YAP [50] or by shear stress through the STAT3/NANOG pathway [51]. AhR, which is usually activated by xenobiotics [41], can also be triggered by fluid shear stress [52] or loss of cell-cell contact [53]. Our data indicate that topography induces chemoresistance through the AhR/CYP1 pathway, activated by actin-regulated mechanotransduction. Topography-induced chemoresistance through AhR in our study highlights a new function of AhR in mechanotransduction.

CYP1, the downstream effector of topography-induced chemoresistance, was overexpressed in early metastatic events [54]. Cytochrome P450 and oxidative phosphorylation were significantly upregulated in circulating tumor cells (CTCs) compared to primary tumors [54]. CTCs with an upregulated cytochrome P450 pathway were more resistant to chemotherapy than primary tumors [54]. Notably, inhibiting CYP1 proteins can have both chemosensitizing and cardioprotective effects [55]. Inhibiting CYP1 proteins could make chemotherapies, such as doxorubicin, more effective in eliminating migrating cancer cells throughout the body. As metastatic recurrence drives breast cancer mortality, combinations of CYP1 inhibition and chemotherapy in the adjuvant setting could potentially improve clinical outcomes. Moreover, applying the tumor stromal topography to classify cancers according to the fiber alignment [56] can identify patients most likely to benefit from combining AhR/CYP1 inhibitors with chemotherapy. In our study, the highly controlled topographical model excluded confounding factors other than topography in the tumor microenvironment in upregulating AhR/CYP1 signaling, thus determining topography-induced chemoresistance. Other functions of AhR/CYP1, such as cell proliferation and stemness in tumor progression, have been widely investigated in vivo [41, 57], which is outside the scope of the current study. However, a follow-up in vivo study to further develop treatment strategies targeting the antioxidant functions of AhR/CYP1 to enhance chemotherapeutic effects will significantly benefit cancer therapy.

The nanopatterned topographical model in the present study, together with the 3D collagen model in our recent research [19], have demonstrated the aligned stromal topography as a critical mechanism for promoting cancer cell cluster migration. Aside from the stromal topography we have investigated, other biochemical and biophysical factors have been reported to cause cancer cell cluster migration [58]. Molecular reprogramming hijacks developmental processes such as partial epithelial-mesenchymal transition (EMT) or retained cell adhesion molecules and promotes tumors to migrate as multicellular clusters [59, 60]. Partial EMT, not complete EMT, is more frequently observed in cancer cell clusters at the tumor border [61]. Partial EMT broadens the window of co-expressing epithelial and mesenchymal markers [60]. Retained epithelial markers such as E-cadherin in partial EMT help to keep cancer cells together as a cluster. Similarly, cancer cells that are not completely de-differentiated highly express adhesion molecules to form cancer cell clusters. For example, cancer cell clusters expressing Keratin 14 [62] and plakoglobin [2] collectively migrate in the surrounding stroma and circulation. In contrast, topography-induced migrating cancer cell clusters in the present study do not display an EMT or upregulated adhesion molecules, suggesting that aligned stromal topography is a distinct mechanism to induce cancer cell cluster migration.

The current study applied a highly controlled nanopatterned model to recapitulate the effect of pre-existing aligned stromal topography on cancer cell clusters. In addition, the high-fidelity topography of the nanopatterned substrate, even as a multiwell platform, enables us to perform high-throughput drug screening. However, since cancer cells and extracellular stroma fibers interact mutually to remodel the stromal topography, it would be valuable to evaluate the topography-induced chemoresistance using in vivo and 3D models, such as our recent 3D dual topographical model, where each tumor organoid is in contact with radially aligned collagen I fibers on one side and circumferentially oriented fibers on the other side [19]. On the other hand, in this study, we examined an FDA-approved oncology drug library containing standard therapeutic agents in treating cancer patients to demonstrate the clinical relevance of topography-induced chemoresistance. Future studies to screen inhibitors or small molecules which can block the AhR/CYP1 pathway on our high throughput nanopatterned platform and 3D dual topographical model would help identify potential therapeutic agents targeting cancer cell clusters.

By developing a highly controlled topographical mimetic in vitro model, our study reveals the multifaceted role of stromal topography in tumor progression. The aligned stromal topography promotes migration and chemoresistance of breast cancer cell clusters. Inhibition of topography-induced antioxidant functions of AhR/CYP1 signaling may serve as a therapeutic strategy to overcome chemoresistance and prevent metastasis.

Data availability

The data underlying figures in this article are available from the corresponding author upon reasonable request.