To assess the effect of hypoxia on CAFs, we first isolated fibroblasts from breast cancer tissue biopsy from a patient with ER ​+ ​PR ​+ ​HER2-breast cancer, and treated a monolayer of CAFs with 1% O₂ for 48 ​h. Molecular O₂ has extremely low solubility in aqueous solutions like tissue culture media,¹³ and can take many hours to equilibrate from ambient O₂ atmosphere. Because we also aimed at observing the effect of oscillations in hypoxia, we cultured the cells on O2 permeable silicone surface, so that ambient O₂ is immediately available to the cells from the bottom of the plate (Fig. 1B). RNA was isolated immediately after hypoxic conditioning, sequenced, and analyzed for differential expression of genes between normoxia and hypoxia.

Principal Component Analysis showed a large separation in gene expression for normoxia and hypoxia samples (Fig. 1C). To assess the broad contours of gene expression changes, we identified the key gene ontologies (GOs) activated by hypoxia. We found many GOs significantly activated in hypoxia were related to regulation of extracellular matrix assembly, and organization, actin cytoskeletal reorganization, tissue migration, and wound healing (Fig. 1D). In addition, we also found activation of glycolytic processes and circadian regulation, which are well known to be regulated by HIF-1.¹⁴ Other notable GOs were those associated with cell-matrix adhesion, including integrin activation, cell-matrix adhesion, and cell-substrate junction organization. Crucially, non-canonical Wnt signaling was downregulated by hypoxia in breast CAFs. Overall, hypoxia appeared to directly regulate many matricellular and biomechanical pathways associated with wound healing, and stromal response to cancer dissemination (Fig. 1D). We selected the genes prominently contributing to activation of these pathways from the top activating sets, finding key genes responsible for fibroblast activation, wound healing, and matrix interaction, as well as those related to glycolysis (Fig. 1E). These included genes encoding key components of the Transforming Growth Factor beta (TGFβ) pathway, involved in myofibroblast transformation of fibroblasts, including transcriptional co-regulator Smad3, TGFβ-1 ligand itself, and TGFBI encoding the TGFβ -inducible protein. In addition, we found other key genes associated with CAFs interaction with matrix to be increased by hypoxia, including FAP encoding Fibroblast-activating Protein, SDC4 encoding Syndecan-4, MUC1 encoding mucin-1, HMGA2 which is an independent marker for stromal activation,¹⁵ ITGA5 encoding Integrin a5 subunit responsible for enhanced tumor-stroma interaction, and CD44, a membrane glycoprotein involved in ECM regulation of TGFβ pathway. While hypoxia increased expression of several genes regulating CAF-matrix interaction, those downregulating matrix degradation were also increased, including SERPINE1, PLAT, and PLAU, key inhibitors of the matrix metalloproteinases. Furthermore, SOX9, which is a key regulator of fibroblast activation was also increased in hypoxia, as well as PDGFC.^16,17 In addition to the genes associated with fibroblast activation, other key genes were involved in the glycolytic shift in metabolism, including PFKP encoding a member of the phosphofructosekinase family, as well as ENO1 and ENO2 encoding enolases, HK2 encoding hexokinase-2, and PGK1 encoding phosphoglyceratekinase. Hypoxia also resulted in reduced expression in FZD1 encoding Frizzled-1, SMURF2 which is a Smad specific ubiquitin ligase and is an inhibitor of fibrosis (Fig. 1E).

Upstream TFs predicted to be regulating the differentially expressed genes in hypoxia included HIF1A, TP53, EPAS1 which encodes for Hif-2a, as well as Smads, which regulate TGFb signaling (Fig. 1F). In addition, we also found other nuclear regulators like Nucleophosmin (NPM1), and Nuclear Protein 1 (NUPR1), as well as XBP1 which regulates unfolded protein folding response. Hypoxia induced gene expression also indicated inhibition in activity of Smad7, which acts as an inhibitory TF in TGFb pathway, SREBF1, and SREBF2 encoding for sterol regulatory element binding TF 1, and 2, as well as MYOD1 encoding for myogenic differentiation factor 1. Overall, transcriptomic analysis strongly suggested that hypoxia resulted in increased fibroblast activation in breast CAFs, which was surprisingly not accompanied by myofibroblast transformation, evidenced by reduced MYOD activity.^18,19

To further explore the specific genes and pathways associated with hypoxia induced transformation of CAFs, we performed Gene Set Enrichment Analysis (GSEA) for the differentially expressed genes (DEGs) between hypoxia and normoxia in CAFs. GSEA is a non-parametric statistical test to assess if a given set of DEGs are preferentially enriched in a given functionally meaningful gene-set vs in random sets, and normalized to the size of the gene-set.²⁰ GSEA analysis confirmed our initial finding that hypoxia can induce a broadly pro-fibroblast activation program in CAFs, typically mediated by activation of TGFβ signaling, and characterized by increased production of extracellular matrix, and reorganization of cytoskeletal organization. We found key gene ontologies to be significantly enriched by DEGs in hypoxia including cell-matrix adhesion, connective tissue development, and ECM assembly with high normalized enrichment scores (NES). Leading edge genes of these pathways revealed several key genes associated with the fibroblast transformation of CAFs in hypoxia.

Many genes encoding ligands and receptors of TGFβ signaling including ACVR1C, ACVR2B, BMP6, BMP7, NODAL, as well as SMAD3, and SMAD7 which are the endpoint transcriptional regulators of TGFβ signaling. In addition, surprisingly we also found many genes encoding the various subunits of inhibins, including INHBA/B/C/E in hypoxia. Typically inhibins and activins can act oppositely to regulate TGFβ signaling, but many inhibin subunits interact with both inhibins and activins.^18,20 Hypoxia therefore seem to not just activate TGFβ signaling, but also its regulators, including the negative ones, suggesting a more nuanced regulation of myofibroblast activation (Fig. 2A). This was also observed in the upregulation of GREM1, which encodes Gremlin-1, a secreted ligand negatively regulating TGFβ activation (Fig. 2B). As mentioned, both ECM assembly, and cell matrix adhesion were enriched by DEGs in hypoxia (Fig. 2B-C). Leading edge genes included ECM molecules including those encoding for fibronectin (FN1), and collagens (COL13A1, COL1A2), Laminin (LAMB2, and LAMB4), Nidogen (NID2), as well as hyaluronon (HAS1, HAS3) as well as CD44, which can act as a receptor for hyaluronans. Interestingly, we also found increased expression for PKD1 which encodes polycystin-1, a transmembrane Ca²⁺ ion channel, as well as SLC9A1 encoding a Na⁺/H⁺ exchanger. pH dysregulation can alter cancer manifestation, and more and more evidence are building on pro-tumor functionalization of CAFs by pH.^21,22

Leading edge genes revealed in GSEA enrichment of “connective tissue development” (Fig. 2D) included SOX9, a key developmental TF which has been implicated in fibroblast activation and fibrosis,²³ KLF7 a TF involved in osteoclast differentiation²⁴ and has been implicated in cancers,^25,26 HMGA2 which is induced by Fibroblast Growth Factor 2 and is therefore associated with fibroblast activation,²⁷ and has been reported to be an independent stromal marker for pancreatic adenocarcinoma. In addition to the gene ontologies associated with wound healing response and fibroblast activation, we also found that hypoxia resulted in enrichment of actin cytoskeletal reorganization. Surprisingly, among the leading edge genes, we did not find many myosin chain encoding genes except MYH9 which increased in hypoxia (Fig. 2E). Other notable genes were ARAP1 which encodes an adapter protein with several signaling docking domains including PH1, RhoGAP, and Ankyrin repeats, and has been considered as a signaling hub to converge Rho and Arf signaling.²⁸ RAP2A and RALA encoding members of the RAS oncogene family, BCAR1 which encodes for Breast Cancer Anti-Estrogen Resistance Protein 1, and KIT, a member of the protooncogene KIT family. Interestingly, we also found increase in CLDN3 which encodes for Claudin-3, a tight junction protein, which could contribute to the collective migration of fibroblasts.²⁹

CAFs are now considered as key mediators of cancer dissemination. However, the debate is yet not settled whether CAFs universally promote, or resist cancer invasion. Fibroblast activation is a physiological response to the wounding caused by basal lamina disruption by a growing tumor, and is aimed at containing and resolving the wound. Indeed, stromal invasibility to cancer invasion is a regulated and selected phenomenon, which we have shown is a secondary outcome of the differential control of placentation during pregnancy. At some stage during the cancer progression, fibroblasts are transformed to be assistive in cancer invasion. We have shown in pancreatic ductal cell carcinoma, that this transition is attained at the onset of metastasis, wherein the CAFs transform from a homogeneous population with a resistive potential to a heterogeneous stromal population which either does not resist cancer invasion, or actively promote it. Here, we asked how does hypoxia, which is a common microenvironmental variable in tumors, regulate stromal invasibility in breast cancers.

Stromal invasion by cancer, unlike cancer migration, is a slow process to measure experimentally. We, therefore, have developed an assay to measure stromal invasion in an accelerated manner by using an underlying anisotropic nanopatterned matrix to align cellular actomyosin assemblage in the same direction. The assay, called ANSIA (Accelerated Nanopatterned Stromal Invasion Assay) allows live analysis of the phenomenon of stromal invasion at the interface of monolayers of stromal fibroblasts and fluorescently labeled cancer cells, with the interface arranged orthogonal to the underlying nanofibers using stencil-based cell patterning (Fig. 3A). Using ANSIA, we measured breast CAF's resistance to MDA-MB-231 invasion in normoxia and hypoxia.

To quantitatively measure the stromal invasion in hypoxia, we measured the difference between the spatial positioning of the initial CAF-cancer interface, and followed the changes in the interface for 24 ​h (Fig. 3B). The difference in the aerial advancement of MDA-MB-231 cells was normalized to the initial length of interface. The normalized extent of invasion measured showed that hypoxia could significantly increase stromal invasion, suggesting that the gene expression changes induced by hypoxia reduce stromal resistance to cancer invasion (Fig. 3C).

As hypoxia had a large effect on stromal invasibility, we asked if fluctuations in hypoxia could mitigate the hypoxia-induced vulnerability to cancer invasion. As oxygen availability is intermittent in many intra-tumor regions, it results in fluctuations in hypoxia. Oscillatory hypoxia could be construed as an intermediate biological signal, averaging the normoxic and hypoxic response of the cell. We had previously found that emergent oscillations in HIF-1 activity can allow cancer cells to escape hypoxia driven inhibition of cell proliferation. This phenomenon suggested an intermediate response to hypoxia and normoxia. However, we had surprisingly found that many genes responded quite differently to oscillatory hypoxia, sometimes in the opposite direction to stable hypoxia, suggesting that gene expression by oscillatory hypoxia is not trivially predicted to be intermediate in between hypoxia and normoxia. Although oscillatory hypoxia has been studied in cancer cells, its effect on CAFs has not yet been elucidated.

We therefore subjected CAFs to an oscillatory or intermittent hypoxic dose using a dynamic tri-gas controller on a tissue culture plate with an O2 permeable bottom surface (Fig. 4A). Cells were isolated immediately after the experiment during an oxidation cycle, RNA isolated and sequenced. A broad analysis suggested that the most significant changes between oscillatory and stable hypoxia were related to reversal in the HIF-1 induced glycolytic metabolism. Kegg signaling pathway analysis highlighted TCA cycle, glutathione metabolism, oxidative phosphorylation, as well as fatty acid metabolism (Fig. 4B). However, we also found that many of the GOs which had increased in hypoxia showed a reversal in oscillatory hypoxia (Fig. 4B). These included TGFb signaling, and focal adhesion, as well as ECM-Receptor interaction. These results were in accordance to the default expectations that oscillatory input should average the high and low inputs, with an intermediate “sub-hypoxic” downstream response.

However, GSEA analysis of GOs also showed specific ontologies which were activated in oscillatory hypoxia (vs stable hypoxia) related to actin network formation in cells (Fig. 4C). These included gene-sets related to actin modification, actin cross-link formation, cortical cytoskeletal organization, and regulation of Arp2/3 based complex. The leading edge genes of these pathways also highlighted genes which had an “oscillatory specific” effect on expression, suggesting that oscillatory hypoxia caused changes in CAF actin cytoskeletal organization differently from hypoxia (Fig. 4D). These genes included LMOD1 encoding Leiomodin1 and TMOD1/2/3 encoding tropomodulin isoforms, both regulating myosin contractility, several actin binding proteins including TWF1 encoding twinfilin, PARVA encoding Parvin, and ANLN encoding anilin. In addition, we found many other actin capping proteins encoded by genes including CORO1B encoding Coronin-1B, ARPC5 and ARPC3 encoding Arp2/3 subunit. Overall, our data showed that while oscillatory hypoxia reduced TGFb activation and actomyosin maturation, it also resulted in the increase of actin crosslink formation and maturation. We therefore sought to investigate the phenotypic effect of these dichotomous changes.

Our data showed that while hypoxia caused fibroblast activation in breast CAFs, accompanied by upregulation of TGFb signaling, increased matrix production, and integrin mediated matrix adhesion. Oscillations in hypoxia resulted in partial reversal of these pathways expectedly, but also surprisingly caused increased activation in actin crosslinking and maturation. These data suggested that oscillatory hypoxia may be differently sensed by CAFs than stable hypoxia in terms of their actomyosin crosslinking, likely contributing to increase in contractile machinery maturation. Furthermore, considering that while TGFb signaling was reversed, and actin crosslinking was increased, oscillation may produce increased heterogeneity in the mechanical response in CAFs.

We therefore functionally tested the effect of stable and oscillatory hypoxia on contractile force generation in CAFs using Traction Force Microscopy (TFM). Using fluorescent beads embedded in a labile hydrogel with a known Young's modulus, which served as the substrate for CAFs, we measured the distribution of force generation at subcellular resolution after subjecting CAFs to 48 ​h of normoxia, hypoxia, or oscillatory hypoxia (Fig. 5A). As cells form matrix-integrin adhesion bonds, which is coupled to the intracellular actomyosin assembly, the contractile force generation is transmitted to the substrate, which could be measured by the displacement of fluorescent beads using microscopy.

When observed over 30 different cells, we found that although no statistical differences were found between the means for each condition (Fig. 5B), observing the distribution of traction force generated per cell, we surprisingly found that oscillatory hypoxia increased the percentage of cells with high traction force generation (∼30%) (Fig. 5C). Does stromal fibroblast contractility directly regulate their resistance to invasion from carcinomas? Transforming growth factor beta (TGFβ) is the primary cytokine involved in myogenic transformation of stroma observed in cancers. As CAFs are a heterogeneous group of cells, we treated BJ fibroblasts, a cell line derived from skin, with 25 ​ng/ml TGFβ for 24 ​h and tested their potential to the relevant A375 melanoma invasion. We found that TGFb treatment significantly enhanced stromal invasibility (Fig. 5D). We then asked if the stromal invasibility was altered in oscillatory hypoxia. ANSIA based stromal invasion assay showed that oscillatory hypoxia further decreased resistance of CAFs to invasion, or transformed CAFs towards a more invasible phenotype even more than stable hypoxia (Fig. 5E-F).

Overall, our data show that oscillations in hypoxic stimuli could further enhance stromal invasibility, but likely through a different mechanism than hypoxia alone, by increasing the variability in CAF contractile force generation. Coupled with our previous observation that while hypoxia increased fibroblast activation programs, it also decreased TF activity associated with myofibroblast transformation, and that oscillation in hypoxia increased actin networking and maturation, our data suggests that regions with unstable hypoxia may increase CAF's assistance to breast cancer dissemination.

Fresh breast cancer tissue was obtained from Biorepository department at UConn Health Center (19-23700ptA) which was collected from a patient undergoing surgery and placed in sterile RPMI. CAFs were isolated using the ex-plant method. Briefly, the tissue was cut into small pieces (approximately 1 mm3) and placed in tissue culture plates containing DMEM/F12 supplemented with 10% FBS and 1% penicillin-streptomycin. The plates were then incubated at 37 ​°C in a humidified incubator with 5% CO2. After 48 ​h, the CAFs migrated out of the tissue pieces and adhered to the plates. The tissue pieces were then removed and the CAFs were cultured for further experiments. The CAFs were maintained in DMEM supplemented with 10% FBS, 1% penicillin-streptomycin and 10 ​ng/ml basic fibroblast growth factor (bFGF).

In vitro oxygen perfusion system from CoyLabs was used to culture cells in 20% O2, 1%O2, or a cycle of changing O2 levels between the two. Immunoblots for HIF-1a was used to confirm each cycle that HIF-1a abundance was increased in hypoxia (Proteintech #20960-1-AP), as described previously.⁶

Invasive Carcinoma - Histologic Type: Invasive lobular carcinoma ER ​+ ​PR ​+ ​HER2-

Histologic Type Comments: Invasive lobular carcinoma predominantly shows a solid architectural pattern; tumor cells are negative for E-cadherin. Glandular (Acinar)/Tubular Differentiation Score 3.

Nuclear Pleomorphism, Score 2 | Mitotic Rate: Score 1 (<=3 mitoses per mm2) | Overall Grade: Grade 2 (scores of 6 or 7) | Tumor Focality; Single focus of invasive carcinoma.

The raw RNAseq reads were aligned to the GRCh38 human genome and transcriptome using hisat2,³⁰ the SAM file was sorted using samtools,³¹ and reads aligning to each coding gene were counted using featurecounts.³² The read counts were subsequently analyzed using DESeq2³³ on the R platform to give gene wise log fold change ratios and p-values. The p-values were corrected for multiple testing using the Benjamini-Hochberg method³⁴ (i.e., false discovery rates). Gene set enrichment analysis for Gene Ontology (GO)³⁵ and KEGG³⁶ pathways was performed using the Kolmogorov-Smirnov test in R using the fgsea package.³⁵ All GO analysis was performed using the biological process ontologies. For all heatmaps, read counts were first transformed to transcripts per million (TPM values), and then z-scores calculated by normalizing each gene independently across the samples shown in each figure. Transcription factor activation scores were calculated using Ingenuity Pathway Analysis from the.

Cell patterning was performed using a PDMS stencil fabricated using a stereolithographic plastic mold. The stencil was placed on the substrate after washing with isopropanol and drying with N2 steam. The device was kept in a vacuum chamber to remove air bubbles under the stencil. Cancer cells were labeled with CellTracker™ Green CMFDA dye (Thermo Fisher Scientific) following the manufacture's guideline. Briefly, MDA-DB-231 cells are culture to 90% confluence and washed once with PBS. Then, cells were incubated with 1uM of dye prepared in HBSS or PBS for 20 ​min. Cells were then detached and washed for at least 3 times in 50 ​ml tubes with PBS. Afterwards, cells were seeded at a density of 5 ​× ​105 cells and attached to the gas permeable plate overnight. The stencil was removed carefully using blunt-end tweezers. The unlabeled stromal cells were seeded at a density of 5 ​× ​105 to fill and attach that area covered by the stencil before. The unattached cells were washed off after 5 ​h of incubation. The plate was mounted on the microscopy stage. Microscopy was performed on Zeiss Observer Z1 microscope with Colibri.2 LED-based fluorescence and Apotome.2 sectional illumination using a 20 ​× ​objective.

Cancer cells labeled with green tracker were imaged at t ​= ​0 and t ​= ​24 ​h. The area occupied by the cancer cells were measured by using Region of Interest (ROI) panel in the Fiji software. ROIs were measured by tracing of boundaries with the freehand tool module in the software. The normalized extent of invasion was calculated by dividing total Area by the length of the initial cancer-stroma interface.

Traction force gels were fabricated using protocols previously described.³⁷ Briefly, coverslips for gel attachment were cleaned with ethanol and sonication, treated with air plasma, and activated with 0.5% glutaraldehyde and 0.5% (3-Aminopropyl)triethoxysilane (Sigma Aldrich). Coverslips for beads coating were treated with air plasma and coated with 0.01% poly-l-lysine (Sigma Aldrich) before coated with red carboxylate-modified microspheres with diameters of 0.1 um (Thermo Fisher). Gel precursor solution containing 7.5% acrylamide and 0.15% bis-acrylamide was degassed for 30 ​min and mixed with 0.1% tetramethylethylenediamine and 0.1% ammonium persulfate before sandwiched between silane-activated coverslips and bead-coated coverslips for 20 ​min. After peeling off the bead-coated coverslips, the resulting traction force gels were coated with 30 ug/ml collagen type I using sulfo-SANPAH (Thermo Fisher) overnight at 4 ​°C. Gels were sterilized under UV for at least 2 ​h before cell seeding. Images containing microbeads location before and after cells trypsinization were recorded using Zeiss Observer A1 microscope. Traction forces were calculated following protocols previously described.³⁸