Fat promotes growth and invasion in a 3D microfluidic tumor model of triple-negative breast cancer

Abstract

Diet influences the levels of small molecules that circulate in plasma and interstitial fluid, altering the biochemical composition of the tumor microenvironment (TME). These circulating nutrients have been associated with how tumors grow and respond to treatment, but it remains difficult to parse their direct effects on cancer cells. Here, we combine a three-dimensional (3D) microfluidic tumor model with physiologically relevant culture media to investigate how concentrations of circulating nutrients influence tumor growth, cancer cell invasion, and overall tumor metabolism. Human triple-negative breast cancer cells cultured in 2D under media conditions mimicking five different dietary states show no observable differences in proliferation or morphology. Nonetheless, those exposed to high-fat conditions exhibit increased metabolic activity and upregulate genes associated with motility and extracellular matrix remodeling. In the 3D microfluidic model, high-fat conditions accelerate tumor growth and invasion and induce the formation of hollow cavities. Surprisingly, the presence of these cavities does not correlate with an increase in apoptosis or ferroptosis. Instead, RNA-sequencing analysis revealed that high-fat conditions induce the expression of MMP1, consistent with cavitation via cell invasion. Mimicking the interstitial flow of nutrients within the TME can thus be used to identify novel connections between metabolic states and tumor phenotype.

NOMENCLATURE

3D

Three-dimensional

DEG

Differentially expressed gene

ECAR

Extracellular acidification rate

ECM

Extracellular matrix

EMT

Epithelial-to-mesenchymal transition

HPLM

Human plasma-like medium

MMP1

Matrix metalloproteinase 1

MPC

Mitochondrial pyruvate carrier

OCR

Oxygen consumption rate

OXPHOS

Oxidative phosphorylation

TME

Tumor microenvironment

TNBC

Triple-negative breast cancer

INTRODUCTION

Tumor progression is often accelerated in patients with obesity.^1,2 Although obesity is most commonly associated with an increase in adipose tissue,³ the condition is also associated with metabolic syndrome and alterations in the levels of nutrients and metabolites that circulate in plasma and interstitial fluid.^4,5 The presence of metabolites in the tumor microenvironment (TME) is likely to affect tumor progression, as cancer cells can use them to generate energy via glycolysis, oxidative phosphorylation (OXPHOS), and fatty acid oxidation.^6,7 However, it has been difficult to parse the relative effects of insulin, glucose, ketones, fat, and other circulating nutrients associated with obesity because these often change together in vivo.

One approach to disentangle the relative effects of circulating nutrients on tumor progression is to test the effects of dietary conditions in patients and mice, which has also been challenging. Clinical trials often overlook individual factors such as diet,⁸ with limited studies focused on insulin-related dietary factors or changes in carbohydrate intake after breast cancer diagnosis.⁹ A larger number of studies have been conducted using mouse models of cancer and have suggested a potential for dietary conditions to impact overall survival.^10–14 Specifically, high-fat diets have been found to accelerate breast cancer progression in tumor-bearing mice,¹⁵ regardless of obesity status.¹⁶ In contrast, ketogenic diets have been found to reduce the growth of breast tumors,¹⁷ but paradoxically enhance the metastatic capacity of breast cancer cells.¹⁸ The effectiveness of a ketogenic diet may vary by mouse strain and sex,¹⁹ highlighting the need for a more detailed understanding of the synergy between circulating nutrients and tumor progression. Diet also affects other features of the TME, including the immune system, microbiome, and adipose tissue, all of which likely contribute to the interpretation of its effects on tumor progression. Because of this complexity, isolating the direct impact of circulating nutrients and metabolites on tumor cells remains challenging.

The TME is characterized by a continuous flow of interstitial fluid that transports small molecules to and from cancer cells.^20,21 This convective transport adjusts the availability and concentration of nutrients, impacting tumor behavior and responsiveness to treatments. Because interstitial flow can alter nutrient gradients, drug delivery,²² and cancer cell migration,²³ it plays a central role in how tumors interact with their metabolic surroundings. Consequently, accurately replicating the flow of interstitial fluid within the TME is essential for modeling tumor responses to dietary interventions. In mouse models, it is challenging to alter individual components within a diet and to track the phenotype of an individual tumor as it grows.^24,25 In contrast, culture models allow for one to examine the effects of specific metabolites and nutrients on human cancer cells over time, with higher throughput than animal models.²⁶ However, culture models also present their own limitations. Standard culture methods, such as spheroids embedded in collagen, rely upon the intermittent delivery of nutrients. As a result, nutrient concentration in these models is neither consistently maintained nor accurately controlled, failing to replicate the continuous, circulating nutrient flow observed in vivo. As such, accurately studying the effects of circulating nutrients in culture requires a tumor model that replicates the continuous flow of interstitial fluid.

Here, we took advantage of human plasma-like media (HPLM)²⁷ and supplemented it with insulin, glucose, ketones, or fat to mimic the circulating nutrients characteristic of five different dietary conditions: baseline, postprandial, diabetic, ketogenic, and hyperlipidemic, respectively. Under 2D culture, MDA-MB-231 human triple-negative breast cancer (TNBC) cells exhibited similar morphologies and proliferation rates across the media conditions. However, RNA-sequencing analysis revealed differences in gene expression, with high-fat conditions upregulating genes associated with motility and extracellular matrix (ECM) remodeling. These transcriptional changes were accompanied by elevated metabolic activity in high-fat media, despite mass spectrometry analysis showing largely similar profiles of secreted metabolites across all conditions. We next mimicked the continuous delivery of circulating nutrients by culturing cells as three-dimensional (3D) tumor-like aggregates (“tumors”) in a microfluidic system that generated interstitial fluid flow comparable to that observed in vivo.^28–35 Under high-fat conditions, tumors exhibited accelerated growth and invasion, without appreciable differences in proliferation compared to baseline HPLM. Notably, high-fat conditions also led to the formation of hollow cavities within the tumors. These structures did not correlate with increased rates of apoptosis or ferroptosis but were instead associated with elevated expression of MMP1, which we hypothesize promoted ECM degradation and subsequent cancer cell dispersal into the surrounding microenvironment.

RESULTS

MDA-MB-231 cells show similar morphologies and proliferation rates when cultured in 2D in media mimicking different dietary states

Dysregulated cellular metabolism is one of the hallmarks of cancer.³⁶ Metabolism can be influenced by the circulating nutrients and metabolites available in the tissue microenvironment,²⁴ which are controlled in part through diet. To study the effects of circulating nutrients on cancer cell behavior, we took advantage of HPLM, which is formulated to contain polar metabolites and salts at concentrations observed in human plasma (Table S1).²⁷ Using HPLM, we generated five different media conditions: baseline, high insulin (5 ng/ml), high glucose (15 mM), high ketones (2 mM β-hydroxybutyrate), and high fat (4% lipid-rich albumin). The latter is a physiological mixture that contains both fatty acids and cholesterol (Table S2), which have been shown to affect tumor growth in vivo.^37,38 These five media conditions mimic important aspects of fasting, feeding, diabetes, ketosis, and hyperlipidemia.

The motility of breast cancer cells increases in response to elevated extracellular fluid viscosity, which promotes actin remodeling at the leading edge of the cell.³⁹ We therefore began by measuring the viscosities of the five different media formulations to ensure that any observed effects on cell morphology could be attributed to differences in biochemical composition rather than differences in viscosity. We found the viscosity of baseline HPLM at room temperature to be 1.06 ± 0.02 cP, which is approximately 3.2% higher than that of DMEM supplemented with fetal bovine serum (FBS) [Fig. S1(A)]. We also found that the high-fat condition had a slightly higher viscosity (∼17.5%) than baseline, consistent with in vivo observations, as high-fat diets are known to increase the viscosity of plasma in patients.⁴⁰ Nonetheless, the viscosities of all the media conditions were below the thresholds previously found to alter cell motility (>3 cP at 37 °C).³⁹

We then proceeded to establish cultures of MDA-MB-231 human breast cancer cells in each of the different media formulations [Fig. 1(a)]. To determine whether the media conditions altered cellular morphology in 2D, we measured projected cellular and nuclear area and circularity. This analysis revealed that the projected cell area and circularity remained consistent across all media conditions [Figs. 1(b) and 1(c)], with cells displaying a spindle-like morphology [Fig. 1(a)]. Cells in the high-ketone media exhibited a slight increase in nuclear area [Fig. 1(d)], while those in the high-fat media exhibited a slight decrease in nuclear circularity [Fig. 1(e)]. Consistent with the spindle-like morphology, immunofluorescence analysis revealed that cells in all conditions maintained expression of the mesenchymal marker, vimentin [Fig. S1(B)].⁴¹ To determine whether the media conditions differentially affected cell proliferation, we carried out EdU analysis. This analysis revealed a uniform rate of proliferation, with ∼40% of cells incorporating EdU over a 2-h pulse, irrespective of the media in which the cells were cultured [Fig. 1(f)]. These findings suggest that the overall mesenchymal phenotype of MDA-MB-231 cells remains stable across the different media formulations.

FIG. 1.

MDA-MB-231 human breast carcinoma cells show similar morphologies and rates of proliferation in media that mimic different circulating nutrient conditions. (a) Phase-contrast images of MDA-MB-231 cells cultured in 2D under different media conditions; scale bars = 100 μm. Graphs of (b) projected cell area, (c) cell circularity, (d) nuclear area, and (e) nuclear circularity for MDA-MB-231 cells cultured in 2D under different media conditions, pooled over n = 3 independent experiments. Shown are the mean and range. (f) Graph showing percentage of EdU⁺ MDA-MB-231 cells cultured in 2D under different media conditions, pooled over n = 3 independent experiments. Cell segmentation was performed using Cellpose.⁷⁴ Shown are the mean and range. ^*p < 0.05; ^**p < 0.01; ns, not significant.

Since the five media conditions resulted in no appreciable differences in cell morphology or proliferation, we turned to RNA-sequencing as an unbiased approach to uncover possible differences in gene expression. This analysis revealed significant transcriptional changes upon transitioning cells from DMEM to HPLM [Fig. S2(A)]. Directly comparing the baseline HPLM condition with DMEM revealed 1139 genes that were differentially expressed, including 292 genes upregulated and 847 genes downregulated by culture in HPLM [Figs. S2(B) and S2(C)]. This shift in gene expression suggests that cells undergo metabolic, structural, and signaling adaptations when transitioned to the more physiologically relevant biochemical microenvironment of HPLM [Fig. S3(A); supplementary text].

Our RNA-sequencing analysis also revealed differences in gene expression between cells cultured in media mimicking the different dietary conditions, despite the fact that we observed no differences in morphology or proliferation. Specifically, principal component analysis (PCA) of the 1000 most variable genes revealed that cells cultured under the high-fat condition displayed a transcriptional signature distinct from those under the other conditions [Fig. 2(a)]. A heatmap of differentially expressed genes (DEGs) further highlighted gene-expression changes induced by the high-fat media [Fig. 2(b)]. To assess condition-specific transcriptional changes, we separately compared each media formulation to baseline HPLM [Fig. 2(c) and Figs. S3(B)–S3(D); supplementary text]. Consistent with our PCA analysis, we found that the high-fat condition resulted in the most pronounced transcriptional changes relative to baseline [Fig. 2(c)]. Cells cultured under high-fat conditions altered the expression of several genes associated with fatty acids and lipid metabolism [Fig. 2(c)]. Specifically, we observed upregulation of PDK4, which encodes pyruvate dehydrogenase kinase 4, an enzyme that phosphorylates and inactivates the pyruvate dehydrogenase complex, thereby shifting cellular metabolism from glucose oxidation toward fatty acid utilization.⁴² Similarly, we observed increased expression of ANGPTL4, which encodes a secreted glycoprotein induced by fatty acids that inhibits lipoprotein lipase, regulating lipid metabolism and glucose homeostasis.⁴³ Conversely, high-fat conditions led to the downregulation of SCD and PCSK9, both of which are involved in lipid metabolism, which is consistent with the abundance of lipids in this media condition, potentially reducing the need for de novo lipid synthesis and cholesterol regulation.

FIG. 2.

Transcriptomic analysis reveals gene-expression patterns specific to different circulating nutrient conditions. (a) Principal component analysis (PCA) of the 1000 most variable genes in cells cultured under five different media conditions. (b) Heatmap displaying hierarchical clustering of differentially expressed genes across the five media conditions. Each row represents a gene, and each column represents a condition. Red and blue indicate upregulated and downregulated genes, respectively, relative to the mean expression across conditions. (c) Volcano plot comparing gene expression between cells cultured under baseline and high-fat conditions. Genes with a fold change ≥2 or ≤−2 and an adjusted p-value <0.05 are highlighted in red (upregulated) and blue (downregulated), respectively. (d) Over-representation analysis with the 40 most significant GO terms of differentially expressed genes in cells cultured under high-fat compared to baseline condition. Each dot represents a biological process, with shading representing statistical significance (−log₁₀ adjusted p-value).

Consistently, gene ontology (GO) analysis revealed significant downregulation of GO terms related to lipid biosynthesis in cells cultured under high-fat conditions as compared to baseline [Fig. 2(d)]. Additionally, GO terms associated with cellular migration and cell adhesion were upregulated by culture in high-fat media. These findings suggest that the circulating nutrients and metabolites characteristic of a high-fat diet might promote transcriptional changes that alter metabolism and invasive potential.

High-fat conditions alter metabolic activity of MDA-MB-231 cells

To determine whether the different media conditions led to alterations in energy metabolism, we measured oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) using a Seahorse extracellular flux analyzer [Figs. S4(A) and S4(B)]. We found that the basal OCR of cells cultured under DMEM was ∼250 pmol/min/10⁵ cells [Fig. S4(C)], consistent with previous reports of MDA-MB-231 cells in their conventional (non-physiological) media.^44,45 Cells cultured under baseline, high-insulin, high-glucose, and high-ketone conditions had an essentially identical basal OCR of ∼250 pmol/min/10⁵ cells [Fig. 3(a)], similar to that of cells cultured in DMEM. In contrast, cells cultured in high-fat media had a basal OCR of ∼350 pmol/min/10⁵ cells [Fig. 3(a)], which is ∼40% higher than those cultured in the other HPLM conditions, suggesting increased mitochondrial activity. However, we found no differences in the basal ECAR values of cells cultured in DMEM [Fig. S4(D)] or in the different HPLM formulations [Fig. 3(b)]. Consistently, measurements of glucose concentration in conditioned media revealed that cells depleted glucose at similar rates [Fig. S4(E)]. These observations suggest that high-fat conditions increase oxygen consumption but have no effect on glycolysis in MDA-MB-231 cells.

FIG. 3.

Cells cultured under high-fat conditions exhibit elevated metabolic activity. Graphs showing (a) basal oxygen consumption rate (OCR), (b) basal extracellular acidification rate (ECAR), and (c) mitochondrial and glycolytic ATP production rates (mitochondrial production rates: baseline 42.7%, high insulin 36.6%, high glucose 43%, high ketone 36%, and high fat 58.6%; glycolytic production rates: baseline 57.3%, high insulin 63.4%, high glucose 57%, high ketone 64%, and high fat 41.4%). Graphs showing (d) acute response to UK5099 and (e) maximal respiration after being exposed to UK5099. Shown are the mean and range for five independent experiments, each with three to four biological replicates. ^*p < 0.05; ^**p < 0.01; ^****p < 0.0001; ns, not significant.

To test the validity of this hypothesis, we calculated the fraction of ATP produced through mitochondrial oxidative phosphorylation (OXPHOS) and glycolysis using the OCR and ECAR measurements. In normal cells, glucose is primarily metabolized through OXPHOS to produce ATP; however, cancer cells can shift their metabolism to favor glycolysis even in the presence of oxygen, a phenomenon known as the Warburg effect.⁴⁶ This metabolic reprogramming allows cancer cells to rapidly generate ATP and biosynthetic precursors needed for proliferation. Our measurements revealed that in MDA-MB-231 cells cultured in DMEM, ∼52% of the ATP is generated through glycolysis [Fig. S4(F)]. When cultured in HPLM, glycolytic ATP production increased to 57% under baseline conditions, further rising to 64% under high-ketone conditions. In contrast, glycolytic ATP production decreased to ∼41% under high-fat conditions [Fig. 3(c)]. These findings indicate that MDA-MB-231 cells rely on glycolysis when cultured in HPLM but shift toward a greater dependence on OXPHOS under high-fat conditions.

We next assessed mitochondrial oxidation of glucose and pyruvate to further understand the metabolic reprogramming that might be induced by the different media conditions. To test whether the mitochondria are using glucose/pyruvate for respiration, we treated cells with UK5099, which inhibits the mitochondrial pyruvate carrier (MPC) to block the transport of pyruvate from the cytoplasm into the mitochondrial matrix, thus preventing its use in OXPHOS.⁴⁷ When cultured in DMEM, MDA-MB-231 cells responded to UK5099 by decreasing their OCR by ∼33% [Fig. S4(G)], consistent with previous observations of these cells under similar conditions.^44,48 Surprisingly, however, MDA-MB-231 cells showed no discernible response to UK5099 when cultured in HPLM [Fig. 3(d)]. These data suggest that in physiologically relevant media conditions, mitochondrial function is independent of pyruvate transport via MPC. Nonetheless, the maximal respiration of cells cultured in high-fat media was higher than that of cells in the other media [Fig. 3(e) and Fig. S4(H)]. High-fat conditions thus appear to enhance the capacity for OXPHOS, potentially due to an increased availability of fatty acids as a fuel source.

High-fat conditions accelerate tumor growth and invasion

In contrast to the static microenvironment of 2D culture systems, cells in vivo are exposed to the flow of interstitial fluid, which circulates the nutrients that surround them. To replicate these interstitial fluid dynamics, we used an engineered microfluidic tumor model^30,32,33 and investigated the effects of the different media conditions on tumor growth and cancer cell invasion. This 3D culture model consists of a solid, packed aggregate of cancer cells (which we refer to as the “tumor”) embedded within a type I collagen gel [Fig. 4(a)]. Each tumor is initially the same size and geometry, with dimensions that approximate those of a human breast duct, allowing for reproducible comparisons across samples. After 2 days of culture, a hydrostatic pressure difference is applied to induce interstitial fluid flow at a rate similar to that of interstitial fluid observed in vivo.^30,49 Over time, the tumors grow, expanding in diameter and length, and cells invade into the surrounding collagen.

FIG. 4.

Tumors under high-fat conditions exhibit elevated rates of invasion. (a) Schematic of procedure for generating tumors cultured under interstitial fluid flow in microfluidic devices. (b) Phase-contrast images of tumors cultured under different media conditions taken on different days after seeding, showing the cutoff point for segmentation (see Methods). Arrowheads point to example invasions; scale bars = 50 μm. (c) Kaplan–Meier plot of invasion from tumors cultured under different media conditions. Results are from 9 to 12 tumors in each condition, pooled across n = 3 independent experiments. ^**p < 0.01; ^***p < 0.001.

To investigate how circulating nutrients affect tumor metabolism, we performed mass spectrometry on conditioned media collected from engineered tumors cultured under the five different media conditions. We observed a consistent increase in the levels of several metabolites in each conditioned medium relative to its respective unconditioned control (Fig. S5). Specifically, the conditioned media all contained increased levels of lactate and pyruvate. Similarly, we observed an increase in the levels of glutamate, suggesting enhanced glutaminolysis, a metabolic adaptation that helps sustain citric acid cycle activity by replenishing intermediates through increased glutamine metabolism,⁵⁰ and which is often observed in aggressive breast cancer cells. These findings are consistent with both the Warburg effect and our Seahorse data [Fig. 3(c)]. The conditioned media were also depleted of hypoxanthine (Fig. S5), which may reflect an upregulation of the purine-salvage pathway to efficiently synthesize nucleotides.⁵¹ Nonetheless, we surprisingly observed no differences in relative levels between the different media conditions.

To determine whether circulating nutrients influence functional behaviors such as tumor invasion, we monitored the tumors twice per day and noted when cells first invaded into the surrounding collagen. By day 4, ∼40% of tumors in the baseline condition and ∼44% of the tumors in the high-insulin condition had invaded [Fig. 4(b)]. Similarly, ∼30% of tumors in the high-glucose condition and ∼33% of tumors in the high-ketone condition had invaded by this timepoint. In contrast, the high-fat condition induced an accelerated rate of invasion, with 100% of these tumors invading by day 3.5. This observation is consistent with our gene ontology analysis, which showed that culture in high-fat media leads to enrichment of GO terms related to cell adhesion and migration [Fig. 2(d)]. We continued monitoring the samples until day 6, by which time nearly all tumors across all conditions had invaded into the surrounding collagen.

We also assessed whether circulating nutrients influence overall tumor growth by using a custom image-analysis pipeline to quantify different aspects of tumor morphology²⁹ [Fig. 5(a)]. We found that tumors cultured under high-fat conditions were significantly larger than those cultured under the other conditions [Fig. 5(b)]. In contrast, tumors cultured under high-glucose conditions were significantly smaller than those under baseline, high-ketone, or high-fat conditions [Fig. 5(b)]. When we excluded the area occupied by invasions and segmented out the dense core of the tumors [Fig. 5(c)], we found that the tumor core metrics were roughly the same across all five conditions [Figs. 5(d)–5(f)]. However, the tumor cores under high-glucose conditions were significantly smaller and narrower than those under both the high-insulin and high-fat conditions [Figs. 5(d) and 5(f)]. Additionally, we found that tumors under high-fat conditions had invasions that were significantly larger and longer than the other tumors [Figs. 5(g) and 5(h)], consistent with their accelerated rate of invasion [Figs. 4(b) and 4(c)]. Overall, the high-fat conditions led to larger and more invasive tumors than the baseline, high-insulin, high-glucose, or high-ketone conditions.

FIG. 5.

Tumors under high-fat conditions are larger in size. (a) Outlines of the segmented regions of tumors under each media condition, using images taken on day 4. Each outline represents a different tumor within that condition. The outlines have been aligned at the cutoff point for segmentation and centered on this cutoff line on the left. (b) Graph of the area of the total tumor on day 4 after seeding across different media conditions. (c) Example of segmentation of the tumor core (yellow), width of the tumor core, and length of the tumor core. Graphs of the (d) area, (e) length, and (f) width of the tumor core across different media conditions. Graphs of the (g) area of invasions and (h) maximum length of invasion in different media conditions. Shown are the mean and range for 8–9 tumors for each condition across n = 3 independent experiments. ^*p < 0.05; ^**p < 0.01; ^***p < 0.001; ^****p < 0.0001; ns, not significant.

High-fat conditions do not affect proliferation

Even though we did not detect a difference in proliferation across the media conditions in cells cultured in 2D, the elevated growth and invasion that we observed in tumors under high-fat conditions could still result from an increase in proliferation of the cancer cells in the 3D microenvironment under interstitial fluid flow. To test this possibility, we conducted EdU analysis on the engineered tumors [Fig. 6(a)]. Despite the differences in tumor size and rates of invasion between tumors cultured under baseline and high-fat conditions, we found a similar number of cells [Fig. 6(b)] and rate of proliferation [Fig. 6(c)] across most conditions at the end of the experiment. The number of cells in tumors cultured under the high-glucose condition was lower than other conditions, consistent with their smaller tumor area [Fig. 5(b)]. Our RNA-sequencing analysis similarly revealed that the media conditions did not appreciably affect the expression of genes associated with proliferation [Fig. S6(A)]. We therefore conclude that the increased size of tumors under high-fat conditions cannot be attributed to an increased rate of proliferation.

FIG. 6.

Tumors under high-fat conditions do not exhibit increased rates of proliferation. (a) Maximum-intensity z-projections of stacks of confocal images of EdU incorporation (yellow) in tumors under different media conditions, counterstained with Hoechst (blue) to visualize nuclei; scale bars = 50 μm. Graphs of the (b) total number of nuclei and (c) percentage of cells that incorporated EdU (EdU⁺ cells) in tumors under different media conditions. Shown are the mean and range for 9–12 tumors in each media condition, pooled across n = 3 independent experiments. ^*p < 0.05; ^**p < 0.01; ^***p < 0.001.

High-fat conditions lead to the formation of cavities within tumors

Given that the larger tumors observed under high-fat conditions are not linked to an increase in proliferation or number of cells, we investigated other potential causes. Upon examining z-stacks of nuclear stains, we observed that the centers of tumors under high-fat conditions lacked nuclei at the end of the experiment, while those in the baseline condition remained essentially solid [Fig. 7(a)]. To confirm the presence of hollow regions in tumors under high-fat conditions and determine when they formed, we examined hematoxylin and eosin (H&E) stained tumors that were fixed and sectioned at different days of culture. This staining confirmed that hollow, cell-free cavities formed within tumors under high-fat conditions by day 6, while the baseline controls retained a solid core [Fig. 7(b)].

FIG. 7.

Tumors in high-fat conditions develop hollow regions that correlate with MMP1 expression but not apoptosis or ferroptosis. (a) A z-slice from the center of Hoechst-stained tumors in baseline (top) and high-fat (bottom) conditions on day 6; scale bars = 50 μm. (b) Images of H&E staining of tumor sections on days 2, 4, and 6 in baseline (top) and high-fat (bottom) conditions; scale bars = 100 μm. (c) Representative images of tumors stained for cleaved caspase-3 (magenta), lipid peroxidation (green), and Hoechst (white) on days 3, 4, and 5 in baseline (top) and high-fat (bottom) conditions; scale bars = 50 μm. Graphs showing quantification of (d) total nuclei, (e) percentage of cleaved caspase-3⁺ cells (%cc3+), and (f) percentage of lipid peroxidation⁺ cells in tumors over time. Shown are the mean and range for 6–7 tumors in each media condition, pooled across n = 3 independent experiments. (g) Expression of MMP1 measured by RNAscope in tumor sections from baseline and high-fat conditions at day 6 of culture; scale bars = 100 μm. ^*p < 0.05; ^**p < 0.01; ^***p < 0.001; ns, not significant.

We hypothesized that the hollow regions might form because of cell death, as nutrient deprivation is known to induce apoptosis in the center of tumors.⁵² To test this hypothesis, we fixed tumors at different time points and stained for both nuclei and cleaved caspase-3 as a marker of apoptosis [Fig. 7(c)]. We found that the total number of nuclei increased over time in both baseline and high-fat conditions [Fig. 7(d)], with no significant differences in cell number between the two conditions at any time point. Similarly, we observed no significant differences in the percentage of apoptotic cells between baseline and high-fat conditions [Fig. 7(e)]. Consistently, RNA-sequencing analysis revealed that cells under both conditions expressed low levels of apoptosis-related genes [Fig. S6(B)]. These results suggest that hollowing of tumors under high-fat conditions occurs independently of apoptosis.

Although the circulating nutrient conditions did not result in differences in apoptosis, high-fat conditions might promote a different form of cell death. Specifically, an increase in the availability of fatty acids can enhance oxidative stress and promote the accumulation of peroxidized lipids within cells, which can lead to nonapoptotic cell death via ferroptosis.⁵³ To determine whether tumor hollowing resulted from ferroptosis, we measured lipid peroxidation in tumors cultured under baseline and high-fat conditions [Fig. 7(c)]. Surprisingly, quantification of lipid peroxidation showed no significant differences between baseline and high-fat conditions [Fig. 7(f)]. Since ferroptosis is an iron-dependent form of cell death, we also performed Prussian blue staining to assess iron accumulation within the tumors, which revealed neither detectable iron accumulation nor significant differences between baseline and high-fat conditions (Fig. S7). Consistent with the staining data, RNA-sequencing analysis showed that ferroptosis-related genes are expressed at low levels under both baseline and high-fat conditions [Fig. S6(C)]. We therefore conclude that hollowing of high-fat tumors occurs independently of ferroptosis.

Our results show that high-fat tumors form hollow regions without increasing cell proliferation or death. The only other physical mechanism that could generate a hollow tumor is enhanced migration of cells away from the tumor core. To examine whether changes in the ECM might contribute to this behavior, we performed multiphoton reflectance imaging to visualize the collagen architecture surrounding tumors [Fig. S8(A)]. These images revealed no apparent differences between the high-fat and baseline conditions, indicating that the increased rate of invasion observed in the high-fat condition is unlikely to result from bundling or cross-linking of the collagen fibers. Nonetheless, our RNA-sequencing data showed an increase in the expression of genes associated with cell motility and ECM degradation, specifically in the high-fat condition [Figs. 2(b) and 2(c)]. The gene most elevated in cells cultured in high-fat media is MMP1, an interstitial collagenase previously found to be upregulated in TNBC and associated with poor prognosis.⁵⁴ We therefore hypothesized that cells invading into the collagen surrounding high-fat tumors would show an increase in expression of MMP1. To test this hypothesis, we conducted fluorescence in situ hybridization (RNAscope) and immunofluorescence analysis. Our results confirmed that MMP1 expression is significantly higher in cells cultured in 2D under high-fat media [Figs. S8(B) and S8(C)], and that MMP1 is elevated at both the transcript and protein levels in tumors cultured under high-fat conditions compared to baseline [Fig. 7(g) and Fig. S8(D)]. Surprisingly, however, MMP1 expression appears uniformly distributed throughout the tumor, with no clear spatial patterning associated with the invasive strands or core. These findings are consistent with the hypothesis that high-fat conditions enhance ECM degradation and invasion by leading to elevated expression of MMP1, thus leading to tumor hollowing.

DISCUSSION

Here, we adapted a microfluidic culture model to study the effects of circulating nutrients on tumor growth and cancer cell invasion. This reductionist approach allowed us to eliminate many of the complexities of in vivo conditions and focus solely on three key components of the TME: the ECM, the tumor, and the biochemical composition of the interstitial fluid. For the latter, we used human plasma-like medium²⁷ to mimic five different circulating nutrient conditions: baseline, high insulin (postprandial), high glucose (diabetic), high ketone (ketogenic), and high fat (hyperlipidemic). RNA-sequencing analysis of cells cultured in HPLM under these five media conditions revealed a broad transcriptional reprogramming compared to those cultured in DMEM, highlighting the importance of using physiologically relevant media when studying the behavior of cancer cells in culture. Notably, cells cultured in the high-fat condition showed a transcriptomic shift, particularly in GO terms related to lipid metabolism and ECM reorganization. The upregulation of GO terms associated with lipid metabolism was expected, as the abundance of lipids in the high-fat media provides an ample energy source, prompting tumor cells to adapt their metabolism accordingly. Consistently, cells under high-fat conditions exhibited an increase in oxidative phosphorylation. The upregulation of GO terms associated with ECM reorganization was unexpected. These changes indicate an increased propensity for tissue remodeling and cell dissemination, which was not obvious in 2D culture. Consistently, however, tumors under high-fat conditions were larger, more invasive, and formed hollow regions over time, despite showing similar rates of proliferation, apoptosis, and ferroptosis as those under other conditions. Hollowing of these tumors correlated with elevated expression of the collagenase MMP1 and increased cell invasion. Our results collectively highlight the impact of circulating nutrients on gene expression, metabolism, growth, and invasion, independent of alterations to the tumor genome.

These findings complement studies investigating the effects of high-fat diets in both cell-culture models and mouse models of cancer. In previous work, we found that adipose cells surrounding engineered tumors enhance the aggressiveness of MDA-MB-231 cells by secreting soluble factors that directly promote tumor cell invasion and vasculogenic escape.^28,35 Because a high-fat diet elevates lipid availability and promotes adipose tissue expansion, it can alter the composition, metabolic activity, and secretory profile of adipocytes, which in turn affect the TME through lipid transfer. In a mouse model comparing different diets, the largest tumors formed under a high-fat and moderate-carbohydrate diet,⁵⁵ which reflects a similar increase in lipid availability to that reproduced by our high-fat media condition. In addition, the elevated metabolic capacity that we observed in the high-fat condition might contribute to the observed increase in mitochondrial activity and could play a role in the aggressive behavior often associated with high-fat dietary states in cancer progression.^56,57

Our results are consistent with emerging evidence that suggests that high-fat conditions can increase the expression of MMP1 in cancer cells, including breast cancer. For example, MDA-MB-231 cells showed no changes in proliferation but exhibited increased migration and elevated expression of MMP1 when exposed to lipid-rich microenvironments.⁵⁸ A palmitate-rich microenvironment was recently found to promote metastasis in breast cancer by activating NF-κB signaling,⁵⁹ a pathway previously shown to regulate MMP1 expression.⁶⁰ High concentrations of oleic acid were shown to promote MMP1 expression and drive greater migration and invasion in endometrial cancer cells.⁶¹ Although current evidence linking high-fat conditions to the upregulation of MMP1 is limited, our findings provide direct support for this association in MDA-MB-231 TNBC cells. Future studies will be needed to determine whether this response generalizes across additional breast cancer subtypes and other tumor types. Furthermore, it will be important to determine which fatty acids are inducing MMP1 and whether similar transcriptional responses occur in other cancer cells and mechanical microenvironments in culture or in vivo.

Several studies have suggested that elevated levels of glucose accelerate cancer cell proliferation⁶² and promote migration,^63,64 or that a ketogenic diet delays tumor growth.^12,17,65 However, we did not observe these effects in our high-glucose and high-ketone conditions, either in 2D or 3D under interstitial fluid flow. Instead, we found that tumors comprised of MDA-MB-231 cells cultured under high-glucose conditions were smaller in overall size. It is important to note that our model is a simplified system that does not incorporate key components of the in vivo microenvironment, such as stromal cells, immune interactions, and systemic metabolic effects. These missing elements likely influence how cancer cells respond to circulating nutrients. For instance, the cytokine IL-22, which plays a key role in regulating immune responses and maintaining mucosal barrier integrity, was found to be reduced in both obese leptin receptor-deficient mice and in mice with hyperglycemia induced by a high-fat diet, compared to healthy control mice.⁶⁶ Therefore, our findings suggest that, at least for the TNBC cell line used here, the high-insulin, high-glucose, and high-ketone conditions may not directly influence the cancer cells themselves but might instead impact other features of the TME that are absent from our model. This possibility highlights the strength of using microfluidic models to disentangle cancer cell-intrinsic responses from those mediated by other components of the TME.

Previous studies have shown that MDA-MB-231 cells exhibit distinct gene-expression profiles depending on the composition of the culture medium⁶⁷ and when they are exposed to fluid flow.⁶⁸ While DMEM remains a standard medium for cell culture, our RNA-sequencing analysis revealed that transitioning MDA-MB-231 cells into HPLM led to broad transcriptional reprogramming, with significant changes across several GO terms, including those related to metabolism, invasion, stress response, and tumor suppression. This observation suggests that standard culture media may obscure biologically relevant behaviors by failing to replicate the nutrient and ion composition of human plasma. Therefore, it is important to emphasize that combining a physiologically relevant biochemical microenvironment with interstitial fluid flow, both often overlooked in conventional models, is useful for more comprehensively assessing tumor cell behavior.

METHODS

Cell culture

MDA-MB-231 human breast cancer cells (ATCC) were maintained in a humidified incubator at 37 °C and 5% CO₂ in one of six different media conditions: DMEM, baseline, high insulin, high glucose, high ketone, or high fat. Specifically, cells were cultured in either DMEM/F12 medium (Hyclone) supplemented with 10% fetal bovine serum (FBS, R&D Systems) and 0.5 mg/ml gentamicin (Gibco) or human plasma-like medium (HPLM; Table S1)²⁷ supplemented with 10% FBS dialyzed by ultrafiltration (Sigma-Aldrich), 0.5 mg/ml gentamicin (Gibco), and the following aqueous-formulated additives: 5 ng/ml bovine pancreas insulin (Sigma-Aldrich; high insulin); 10 mM D-glucose (Sigma-Aldrich; high glucose); 2 mM DL-β-hydroxybutyric acid sodium salt (BHB; Sigma-Aldrich; high ketone); or 4% (wt./vol.) lipid-rich bovine serum albumin (BSA, AlbuMAX™ I, Gibco; high fat). The high-fat condition contained 80 μM alpha-linolenic acid, 90 μM linoleic acid, 300 μM oleic acid, 290 μM stearic acid, and 300 μM palmitic acid (Table S2). Cells were passaged at a 1:3–6 ratio and used before passage 25. MDA-MB-231 cells cultured in baseline HPLM were tested for authentication by short tandem repeat (STR) profiling using the ATCC Cell Line Authentication Service. The STR profiles were compared to the reference profiles provided by ATCC, confirming the identity of the cell line.

Viscosity measurements

The viscosity of each media formulation was measured at room temperature using an Ostwald viscometer (Ace Glass, Inc.). Viscosity ( $\eta$) was calculated using the equation $\eta ={\eta }_0\frac{\rho t}{{\rho }_0{t}_0}$, where ${\eta }_0=1$ cP is the viscosity of water at room temperature, $\rho$ and ${\rho }_0$ are the densities of the medium and water, respectively, and $t$ and $t_0$ are the times of outflow of the medium and water, respectively. The measurements were performed three times using three independently prepared media samples. The uncertainty for each measurement was determined using the propagation of errors using the equation ${\sigma }_{\eta }=\eta \sqrt{{(\frac{{\sigma }_{\rho }}{\rho })}^2{+(\frac{{\sigma }_{{\rho }_0}}{{\rho }_0})}^2{+(\frac{{\sigma }_t}{t})}^2{+(\frac{{\sigma }_{t_0}}{t_0})}^2}$.

Microfluidic tumor model

Microtumors comprised of MDA-MB-231 cells were generated using a microfluidic approach, as described previously.^30,32,33 Briefly, a silicone mold was used to create channels within polydimethylsiloxane (PDMS), which were exposed to UV/ozone and bonded to a glass-bottomed tissue-culture dish. The channels were then immediately coated with 1 mg/ml poly-D-lysine hydrobromide (Sigma-Aldrich). Acupuncture needles (Austin Medical; 120 μm in diameter) were coated with 1% BSA in phosphate-buffered saline (PBS) and positioned within one of the PDMS channels. A 4 mg/ml solution of type I collagen (Koken) was prepared using 10 N NaOH, 10× Hank's Balanced Salt Solution (HBSS; Gibco), Milli-Q water, and HPLM, resulting in an aqueous solution of collagen with a pH of 9.5. The collagen solution was introduced into the PDMS channels and incubated at 37 °C for 20 min to form a gel. The needles were subsequently removed to yield two empty cavities, one on each side of the gel. A suspension of MDA-MB-231 cells supplemented with 100 μg/ml DNase I (Roche) was added to fill the cavity on one side of the gel, and then the excess cells were washed away. Samples were cultured for 48 h, after which the flow of interstitial fluid was initiated, as previously described.^30,32,33 Tumors were monitored for invasions, and conditioned medium was collected from the downstream well every 12 h on days 3.5–5.5 for mass spectrometry analysis, while fresh medium was added to the upstream well.

The permeability of the collagen gel was calculated using Darcy's Law $Q=\frac{\kappa A}{\mu L}\Delta P$, where $Q$ is the volumetric flow rate of the culture medium, $\kappa$ is the permeability of the collagen gel, $A$ is the cross-sectional area of the collagen gel, $\mu$ is the viscosity of the culture medium, $L$ is the length of the gel, and $\Delta P$ is the pressure difference across the gel. This approach revealed that the permeability of the collagen gel was 0.100 ± 0.0076 μm², resulting in an interstitial fluid flow rate of 5–8 μl/h, which is on the same order of magnitude as previously reported values.^30,31 This range corresponds to interstitial flow rates observed in vivo,⁴⁹ indicating that our experimental system reproduces both the physical properties of the ECM and the transport dynamics of interstitial fluid.⁶⁹

Immunofluorescence analysis

To assess morphology and phenotype in 2D culture, cells were seeded at 4000 cells per well in a 96-well plate and cultured for 72 h under each of the five different media conditions. Samples were fixed for 15 min in 4% paraformaldehyde (PFA) in PBS and then washed three times with PBS. Fixed samples were incubated for 1 h at room temperature in blocking buffer comprised of 20% BSA and 10% normal goat serum (NGS) in PBS supplemented with 0.3% Triton X-100 (PBST). Blocked samples were washed with PBST and incubated overnight with primary antibodies against E-cadherin (Cell Signaling, 24E10 rabbit mAb #3195) and vimentin (Sigma-Aldrich, clone LN-6 mouse V2258), or MMP1 (Abcam, EP1247Y rabbit ab52631), diluted 1:200 in blocking buffer, followed by additional washing with PBST. Samples were incubated in the dark for 1 h at room temperature with Alexa Fluor-conjugated secondary antibodies and phalloidin (Thermo Fisher Scientific) diluted 1:200 in blocking buffer. Samples were washed twice with PBS, counterstained with Hoechst 33342 (Invitrogen) diluted 1:1000 in PBS for 15 min, and then washed twice with PBS.

To evaluate MMP1 expression in tumor sections by immunofluorescence staining, slides with tumor sections were placed in staining racks at room temperature for 30 min. Tumor sections were fixed again in 4% PFA in PBS for 15 min and washed three times with Tris-buffered saline supplemented with 0.5% Triton X-100 (TBST). Fixed samples were incubated for 1 h at room temperature in BlockAid blocking solution (Invitrogen). Blocked samples were incubated overnight with a primary antibody against MMP1 (Abcam, EP1247Y rabbit ab52631), diluted 1:50 in blocking solution, followed by additional washing with TBST. Samples were incubated in the dark for 1 h at room temperature with Alexa Fluor-conjugated secondary antibody diluted 1:200 in blocking solution. Samples were washed three times with TBST, counterstained with Hoechst 33342 (Invitrogen) diluted 1:1000 in PBS for 15 min, and mounted under a coverslip using Fluoromount Aqueous Mounting Medium (Sigma-Aldrich).

Proliferation assays were conducted using the Click-iT EdU Alexa Fluor 555 Kit (Thermo Fisher Scientific). Samples were exposed to EdU for 2 h and then immediately fixed with 4% PFA in PBS before applying the Click-iT cocktail reagent. 2D samples were then counterstained with Hoechst 33342 diluted 1:1000 in PBS. Engineered tumor samples were incubated overnight at 4 °C with Hoechst 33342 diluted 1:1000 in 1% BSA in PBST. Subsequently, the tumor samples were incubated for 30 min in 25% glycerol in PBS, then 50% glycerol in PBS, and then transferred into 75% glycerol in PBS and stored at 4 °C for future imaging.

Lipid peroxidation was assessed using the Click-iT Lipid Peroxidation Imaging Kit—Alexa Fluor 488 (Thermo Fisher Scientific). Tumors were exposed to linoleamide alkyne for 2 h, washed three times with PBS, fixed with 4% PFA in PBS, and then washed three times with PBS. Samples were incubated in 0.1% PBST for 1 h, blocked with 1% BSA in PBS for 1 h, and then washed twice with PBS. Subsequently, tumors were incubated in Click-iT reaction cocktail for 1 h and washed twice with 1% BSA in PBS and twice with PBS. To label apoptotic cells, tumors were first incubated in blocking buffer comprised of 10% NGS in 0.1% PBST overnight at 4 °C. Blocked tumors were incubated overnight with primary antibody against cleaved caspase-3 (Cell Signaling, Asp175 rabbit 9661) diluted 1:400 in blocking buffer. The following day, samples were washed four times with 0.1% PBST and incubated overnight at 4 °C in Alexa Fluor 647-conjugated secondary antibody (Invitrogen, goat anti-rabbit A21245) diluted 1:200 in blocking buffer. Samples were washed four times with 0.1% PBST and counterstained with Hoechst 33342 diluted 1:2000 in PBS for 15 min, washed once with PBS, and transferred into 75% glycerol in PBS using the same procedure as described above.

Analysis of energy metabolism

The oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) of cells in the different media conditions were measured using a Seahorse XFe24 Analyzer (Agilent). The Seahorse XF Glucose/Pyruvate Oxidation Stress Test Kit (Agilent) was used according to the manufacturer's protocols. Cells were seeded at a density of 100 000 cells/well in a Seahorse XF24 V28 PS Cell Culture Microplate (Agilent) in their corresponding media conditions 1 day before measurement. One hour before the measurements, the media was replaced with HPLM without phenol red and bicarbonate, adjusted to a pH of 7.4, for all five media conditions. For the cells cultured in DMEM, the media was replaced with DMEM (Agilent) supplemented with 10 mM glucose, 1 mM pyruvate, and 2 mM glutamine. The plates were incubated at 37 °C for 1 h prior to measurements. OCR and ECAR were measured every 8 min during basal respiration and after sequential injection of the relevant pathway inhibitors: 2 μM UK5099, 1.5 μM oligomycin, 1.0 μM carbonyl cyanide-4 (trifluoromethoxy) phenylhydrazone (FCCP), and 0.5 μM rotenone/antimycin A. The Open Lab software portfolio was used to calculate all values presented in Fig. 3 and Fig. S4.

Bulk RNA-sequencing analysis

Cells were cultured in six different media conditions in a 6-well tissue-culture plate at a density of 200 000 cells/well. After 72 h, RNA was extracted and purified using an RNeasy Mini Kit (Qiagen) according to the manufacturer's protocols. The extracted RNA was processed by the Princeton University Genomics Core Facility. RNA quality was evaluated using an Agilent Bioanalyzer. RNA-Seq directional library preparation was performed on an Apollo 324 Robot, and the samples were run in a single lane on a NovaSeq SP 100 nucleotide Flowcell v1.5 with a read depth of ∼650–800 × 10⁶ reads.

Reads were aligned to the human reference genome (GRCh38.p14) using STAR, followed by transcriptome quantification using HTSeq-Count. Raw read counts were normalized using the estimateSizeFactors function in the DESeq2 package.⁷⁰ Normalized counts were transformed into regularized log for variance stabilization before performing principal component analysis (PCA) and Z-score heatmaps. Differentially expressed genes (DEGs) were identified using Wald tests, followed by shrinkage of log fold change using the apeglm method to avoid inflation of fold changes of genes with low counts.^70,71 Significant DEGs had a Benjamini–Hochberg-adjusted p-value <0.05 and an absolute log₂ fold change >2. Over-representation analysis (ORA) in gene ontology (GO) terms of biological processes was conducted separately for the downregulated and upregulated significant DEGs of the high-fat condition in reference to the baseline condition using the gprofiler2 package.⁷² Data are available from GEO (GSE302974).

Sectioning tumors

Tumor samples were fixed as described above, placed in a disposable cryomold, and fully surrounded with Optimal Cutting Temperature (OCT) compound to ensure complete coverage. Cryomolds containing tumors were rapidly frozen using dry ice. Once frozen, cryomolds were stored at −80 °C until sectioning. Frozen tumors were cut into ∼50-μm-thick sections using a CM3050S cryostat. After sectioning, tumor sections were placed onto Fisherbrand Superfrost Plus microscope slides (Fisher Scientific) and stored at −80 °C until use.

Hematoxylin and eosin (H&E) staining

H&E staining was performed using a modified version of a previously described protocol.⁷³ Slides with tumor sections were placed in staining racks at room temperature for 30 min. Tumor sections were fixed again in 4% PFA in PBS for 10 min. Fixed sections were washed twice with PBS and stained in Harris Hematoxylin solution for 2 min. Stained samples were rinsed with one to two dips in acid alcohol. Sections were washed three times in Milli-Q water, stained with bluing solution, and washed three times in Milli-Q water. Sections were dehydrated by sequential immersion in 75%, 80%, and 95% alcohol. Dehydrated sections were stained with Eosin Y for 20 s, and then washed by immersing once in 95% alcohol and then three times in absolute alcohol. Sections were cleared in two successive xylene substitute (Electron Microscopy Sciences) baths. Finally, sections were mounted under a coverslip with acrylic mounting media (Sigma-Aldrich).

Prussian blue staining

For Prussian blue staining, slides with tumor sections were placed in staining racks at room temperature for 30 min. Tumor sections were fixed again in 4% PFA in PBS for 10 min and then washed twice with PBS. Sections were stained for 3 min using a working iron-stain solution prepared by mixing equal volumes of potassium ferrocyanide and hydrochloric acid (Iron Stain Kit, Prussian Blue stain ab150674). Sections were then washed three times in Milli-Q water, followed by staining with nuclear fast red. Samples were then washed and dehydrated by sequential immersion in 75%, 80%, and 95% alcohol. Finally, sections were mounted under a coverslip with acrylic mounting media.

Fluorescence in situ hybridization

Fluorescence in situ hybridization was performed using the standard RNAscope Multiplex Fluorescent V2 Assay (Advanced Cell Diagnostics) protocol for fixed-frozen samples. The probe was for Homo sapiens MMP1 (channel 1, 1057851), and the fluorophore was Opal 650 (Akoya Biosciences, FP1496001KT).

Imaging

Engineered tumors were imaged daily using a Nikon Ti-U microscope equipped with a digital CMOS camera (Hamamatsu ORCA-Flash 4.0 V3) using a 10× air objective. Fluorescent samples were imaged using a Nikon Eclipse Ti2 microscope equipped with a spinning disk confocal (X-light V2, Crest Optics) and an ORCA charge-coupled device camera (Hamamatsu, Japan). Images of 2D samples were acquired using either a Plan Fluor 20× 0.45 NA or Plan Fluor 40× 1.3 NA oil objective. Confocal stacks of tumors (200 μm deep) were acquired using a Plan Fluor 20× 0.95 NA water objective with a step size of 2 μm. Tumors stained for cleaved caspase-3 and lipid peroxidation were imaged on a Nikon Ti-E microscope attached to a Maico MEMS confocal unit (Hamamatsu, Japan) using a Fluor 40× 1.3 NA oil objective. To visualize collagen architecture, we performed multiphoton reflectance imaging of fixed samples on a Nikon AX-R multiphoton microscope attached to a Nikon Ti2 inverted microscope using a Plan Fluor 20× 0.95 NA water objective and 900-nm excitation.

H&E and Prussian blue-stained sections were imaged using a Leica M205 stereoscope equipped with an MC170 HD camera (25195014 DOC2) using a Leica PLANAPO 2.0× objective (M-series). RNAscope for MMP1 mRNA and immunofluorescence staining for MMP1 protein in tumor sections were imaged on a Nikon Ti-E microscope attached to a Maico MEMS confocal unit using a Plan Fluor 40× 1.3 NA oil objective.

Image analysis

To quantify cell and nuclear morphological features in 2D, cells were segmented using Cellpose,⁷⁴ and measurements were obtained with ImageJ⁷⁵ measurement tools. To quantify proliferation of cells cultured in 2D, the number of EdU-positive cells and total number of nuclei were counted in ImageJ using the StarDist plugin.⁷⁷ The percentage of EdU-positive cells was given by the ratio of EdU-positive cells to the total number of nuclei for all images of each condition in each replicate. To quantify proliferation of cells in the 3D engineered tumors, confocal stacks were imported into Imaris (Oxford Instruments). The number of EdU-positive cells and total number of nuclei were counted using the Spots function, with subsequent analysis limited to the initial 200 μm of the tumor and anything extending beyond that point along its axial length.

Tumor morphology was analyzed using a custom image-analysis pipeline.²⁹ Briefly, a cutoff point for each tumor was delineated as 200 μm into the tumor in the direction of the interstitial flow at the time of seeding. The total tumor and tumor core that extended beyond this cutoff point were segmented from images acquired on day 4 of culture. The tumor core was defined as the dense region of the tumor excluding any invasions. The area of invasions was calculated as the area of the total tumor remaining after the tumor core was dilated by 10 μm to account for small invadopodia extending from the tumor core and then subtracted from the area of the total tumor.

Glucose uptake assay

MDA-MB-231 cells were seeded at 200 000 cells/well in a 6-well plate. After 72 h, media was collected and stored at −20 °C. Media collected from cell-free wells were used as a control and stored at −20 °C. Glucose uptake was measured using the Glucose Assay Kit (Abcam ab65333) according to the manufacturer's instructions.

Mass spectrometry

Conditioned medium was harvested from the downstream well of the microfluidic device beginning on day 3.5 after seeding and continuing through day 5.5. The control medium was harvested by adding 150 μl of media into a 96-well plate at the same time as fresh media was added to the microtumors. This plate was then placed in the same incubator as the tumors, and the media was collected at the same time as the conditioned media was collected from the tumors. Media collected from tumors in each condition were combined and stored at 4 °C until the end of the experiment. The samples were then preserved at −20 °C. On the day prior to mass spectrometry, samples were thawed, diluted 1:29 in methanol (Fisher Chemical), and stored at −80 °C overnight. The next day, samples were centrifuged at 19 930g for 20 min at 4 °C and the supernatant was collected.

Metabolites were separated by hydrophilic interaction liquid chromatography (HILIC) with an XBridge BEH amide column (2.1 × 150 mm², 2.5-μm particle size; Waters, 186006724). The column temperature was set at 25 °C. Solvent A was 95:5 (vol.:vol.) H₂O:acetonitrile (with 20 mM ammonium acetate, 20 mM ammonium hydroxide, pH 9.4). Solvent B was acetonitrile. Flow rate was 0.15 ml/min. The LC gradient was 0–2 min, 90% B; 3–7 min, 75% B; 8–9 min, 70% B; 10–12 min, 50% B; 12–14 min, 25% B; 16–20.5 min, 0.5% B; and 21–25 min, 90%. MS analysis was performed on an Orbitrap Exploris 480 mass spectrometer. A full scan was performed in negative mode, at the m/z of 70–1000. The automatic gain control (AGC) target was 1000%, and the maximum injection time was 500 ms. The orbitrap resolution was 180 000. Raw mass spectrometry data were converted to mzXML format by MSConvert (ProteoWizard). Peak-picking was done on El Maven (v0.8.0, Elucidata). Peak-intensity values for the conditioned media corresponding to each media composition were normalized to their respective unconditioned control media using the equation $\frac{{PI}_{\mathit{\text{cond}}}-{PI}_{\mathit{\text{cntl}}}}{{PI}_{\mathit{\text{cntl}}}}$, where ${PI}_{\mathit{\text{cond}}}$ is the peak intensity of the conditioned media, and ${PI}_{\mathit{\text{cntl}}}$ is the peak intensity of the control media.

Statistical analysis

All statistical analyses were performed using GraphPad Prism. The metrics for 2D and 3D morphology, OCR and ECAR values, the percentage of EdU-positive cells, the total number of nuclei, the percentage of cleaved caspase-3 positive, and the percentage of lipid peroxidation in 3D were compared using one-way ANOVA with Tukey's multiple comparison test. When the result of Bartlett's test was significant (p < 0.05), the Brown–Forsythe ANOVA test was performed and followed by Dunnett's T3 multiple comparisons test. Otherwise, the one-way ANOVA test and Tukey's multiple comparisons test were performed. The metrics for maximum OCR—comparing DMEM control and UK5099—were analyzed using an unpaired t-test. When comparing the maximum distances of invasion, the Bartlett's test and Brown–Forsythe ANOVA test could not be performed, so the one-way ANOVA test and Tukey's multiple comparisons test were used. ATP production rates were compared using two-way ANOVA. The rates of invasion were compared using the log-rank (Mantel–Cox) test. Significance for all comparisons was considered p < 0.05, and p-values were adjusted for multiple comparisons.

SUPPLEMENTARY MATERIAL

See the supplementary material for the supplementary text, supplementary references, Tables S1 and S2, and Figs. S1–S8.

AUTHOR DECLARATIONS

Conflict of Interest

The authors have no conflicts to disclose.

Ethics Approval

Ethics approval is not required.

Author Contributions

Maryam Kohram, Carolina Trenado-Yuste, and Molly C. Brennan-Smith contributed equally to this paper.

Maryam Kohram: Conceptualization (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – original draft (equal). Carolina Trenado-Yuste: Conceptualization (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – original draft (equal). Molly C. Brennan-Smith: Conceptualization (equal); Investigation (equal); Methodology (equal). Evelyn S. Navarro Salazar: Formal analysis (equal). Pengfei Zhang: Formal analysis (equal). Jasmine E. Hao: Investigation (equal). Xincheng Xu: Investigation (equal). Bharvi Chavre: Investigation (equal). William Oh: Investigation (equal). Sherry X. Zhang: Investigation (equal). Susan E. Leggett: Conceptualization (equal); Investigation (equal); Methodology (equal). Rolf-Peter Ryseck: Resources (equal). Joshua D. Rabinowitz: Conceptualization (equal); Funding acquisition (equal); Supervision (equal); Writing – review & editing (equal). Celeste M. Nelson: Conceptualization (equal); Funding acquisition (equal); Methodology (equal); Project administration (equal); Supervision (equal); Writing – review & editing (equal).

DATA AVAILABILITY

The RNA-seq data that support the findings of this study have been deposited at GEO under accession number GSE302974 and are available as of the date of publication, Ref. 76.