Modeling the Effect of the Metastatic Microenvironment on Phenotypes Conferred by Estrogen Receptor Mutations Using a Human Liver Microphysiological System

Reciprocal coevolution of tumors and their microenvironments underlies disease progression, yet intrinsic limitations of patient-derived xenografts and simpler cell-based models present challenges towards a deeper understanding of these intercellular communication networks. To help overcome these barriers and complement existing models, we have developed a human microphysiological system (MPS) model of the human liver acinus, a common metastatic site, and have applied this system to estrogen receptor (ER)+ breast cancer. In addition to their hallmark constitutive (but ER-dependent) growth phenotype, different ESR1 missense mutations, prominently observed during estrogen deprivation therapy, confer distinct estrogen-enhanced growth and drug resistant phenotypes not evident under cell autonomous conditions. Under low molecular oxygen within the physiological range (~5–20%) of the normal liver acinus, the estrogen-enhanced growth phenotypes are lost, a dependency not observed in monoculture. In contrast, the constitutive growth phenotypes are invariant within this range of molecular oxygen suggesting that ESR1 mutations confer a growth advantage not only during estrogen deprivation but also at lower oxygen levels. We discuss the prospects and limitations of implementing human MPS, especially in conjunction with in situ single cell hyperplexed computational pathology platforms, to identify biomarkers mechanistically linked to disease progression that inform optimal therapeutic strategies for patients.


Supplemental Materials and Methods
Lentiviral transduction of hepatocytes and confocal imaging to examine the spatial relationship between MCF7 cells and hepatocytes. The hepatocyte biosensor construct pCT-mito-GFP was obtained from Systems Biosciences (Mountain View, CA) and contains a mitochondrial localization tag derived from cytochrome C that is C-terminally fused to GFP. Lentiviral particles were produced and transductions performed as previously described 1,2 . To examine the spatial relationship between hepatocytes and mCherry-tagged ESR1-expressing mutant cells, LAMPS models were set up as described in the Materials and Methods with the additional step of hepatocyte transduction. Briefly, after thaw, hepatocytes were incubated for 2h in lentivral supernatant generated from pCT-mito-GFP diluted to an MOI of 5 and containing 8µg/mL polybrene (EMD Millipore). Following this incubation, hepatocytes were gently pelleted, resuspeded, and seeded into LAMPS models as described in the Materials and Methods. Hepatocytes reached their maximum transduction efficiency (~50%) after 72h and GFP expression remained stable throughout the experimental time course.
Once models were assembled and flow (15 µl/h) established, images were collected with a Nikon 20x (0.45 NA) objective using the IN Cell Analyzer 6000 (GE Healthcare) in confocal mode using the 488 nm (GFP) and 561 nm (mCherry) lasers and associated emission filter sets with the aperture set to 1 airy unit. Z-stacks were collected after days 1, 9, and 17 of the experimental time course. During imaging, 35 z-slices were acquired for each field with 3µm spacing between slices (105 µm distance). For each ESR1 clone, 3 individual LAMPS models were imaged, acquiring 8 fields per device. Image stacks were then imported into ImageJ (FIJI) to generate 3D renderings and orthogonal view images to describe the spatial relationship between hepatocytes and MCF7 cells.
Fluorescence intensity in individual z-planes was analyzed in ImageJ and intensity values were normalized to the highest peak intensity in each plane of the stack and were plotted for each individual z-slice. For statistical analysis comparing fluorescence within the z-stack between hepatocytes and MCF7 cells, a Kruskal-Wallis rank sum test was used to assess significant differences in intensity throughout the z-stack. For each ESR1 clone 3 individual LAMPS models were evaluated (8 fields/device). For WT, Y537S, and D538G LAMPS models P-values of 0.23, 0.31, and 0.27 were obtained, respectively.
The slopes obtained for each of the cell lines varied by < 10% and were given by the equation for each line: (WT: y = 6.8x + 39284; Y537S: y = 6.5x + 22421; D538G: y = 7.0x + 35231). A one-way ANOVA test performed comparing the fluorescence intensities for ESR1 clones at the indicate cell densities. The results of this analysis produce P-values > 0.2 for each cell density, indicating that there is no statistical difference in the average fluorescence intensity between the WT, Y537S, and D538G clones at any of the plating densities examined.

Sample collection and albumin measurement
Efflux media collection and albumin measurements were performed as previously described 2,4 using an enzyme-linked immunosorbent assay (ELISA) (Bethyl Laboratories, Montgomery, TX) according to the manufacturer's specifications in a 96well plate format.

Drug binding/recovery in PDMS-containing LAMPS device
To assess the drug binding capability of the polydimethylsiloxane (PDMS)-containing LAMPS device for compounds used in these studies (β-estradiol, fulvestrant, AZD9496, and doxorubicin), we used perfusion flow tests and mass spectrometry analysis of efflux to determine the overall effective concentration of each compound as previously described 2,5 . These studies provided the basis for the use of the monoculture and coculture models for our initial drug testing studies described in the Results section.  For each ESR1-expressing clone, the peak fluorescence values are found within a similar z-range as the hepatocytes (D-F), indicating that MCF7 cells are growing within the hepatocyte layer, with some of the Y537S and the D538G also growing slightly above the hepatocyte layer as indicated in the images (B, C -Day 16) and the shoulders around Z=65-85 µm on the graphs (E, F). A Kruskal-Wallis rank sum test was used to assess significant differences in intensity throughout the z-stack between hepatocytes and each ESR1 clone, and no significant differences were observed. For each ESR1 clone 3 individual LAMPS models were evaluated (8 fields/device). For WT, Y537S, and D538G LAMPS models P-values of 0.23, 0.31, and 0.27 were obtained, respectively. Z-stacks were acquired on Days 1, 9, and 17. Scale bar; 200 µm.  Together, these results demonstrate a phenotypic switch in estrogen-dependent growth that depends upon TME composition.  co-culture models (D-F) as previously described 2,4 . Results are shown as the mean ± SD from 3 LAMPS devices or plate wells. Albumin output of LAMPS devices maintained at zone 1 (12-15%) was higher than that observed in LAMPS models maintained at zone 3 (3-6%) oxygen tensions. For co-culture models, a similar trend is observed for plates maintained at zone 1 (20%) and zone 3 (5%) oxygen tensions, but the overall amount of albumin output is lower than in LAMPS devices, consistent with previous work 2,4 .
C.   To assess the drug binding capability of the polydimethylsiloxane (PDMS)containing LAMPS device for compounds used in these studies (βestradiol, fulvestrant, AZD9496, and doxorubicin), we used perfusion flow tests and mass spectrometry analysis of efflux to determine the overall effective concentration of each compound as previously described 5 . Cellfree LAMPS devices were coated with collagen (200 µg/mL)/fibronectin (100 µg/mL) and washed with PBS prior to incubation with hepatocyte maintenance media containing either 5 nM estradiol or 2 µM drug (fulvestrant, AZD9496, or doxorubicin) -/+ the indicated carrier protein at 0.1 mg/mL. For estradiol, sex hormone binding globulin (SHBG) was used as a carrier while low-density lipoprotein (LDL) was used as a carrier for fulvestrant, AZD9496, and doxorubicin. LAMPS devices were incubated for 72 h and the amount of estradiol or drug present in the efflux media compared to the initial input was determined by mass spectrometry and expressed as a percentage of the starting input media in the presence or absence of carrier protein.