Regulation of p65 nuclear localization and chromatin states by compressive force

Abstract

The tumor microenvironment (TME) is a dynamic ecosystem, that evolves with the developing tumor to support its growth and metastasis. A key aspect of TME evolution is the recruitment of stromal fibroblasts, carried out via the release of various tumor signals including tumor necrosis factor (TNFα). These tumor signals in turn alter the mechanical properties of the TME as the tumor grows. Because of the important role of stromal cells in supporting tumor progression, new therapies aim to target stromal fibroblasts. However, these therapies have been unsuccessful in part due to the limited understanding of cross-talk between chemical and altered mechanical signaling within stromal fibroblasts. To investigate this, we designed a coculture assay with YFP-TNFα releasing spheroids embedded within collagen gels alongside fibroblasts to mimic the stromal response within the TME. This resulted in the nuclear translocation of p65 in the stromal fibroblasts which was further intensified by the addition of compressive stress. The combination of mechanical and chemical signals led to cytoskeletal disruption and induced a distinct chromatin state in the stromal fibroblasts. These results highlight the important cross-talk between cytokine signaling and mechanical forces on stromal cells within the TME and facilitate the development of a better spheroid model for therapeutic interventions.

• Tumor progression is supported by the recruitment of stromal cells via various tumor signals such as TNFα which alters the mechanical properties of the TME, but the cross-talk of mechanical and chemical signaling is poorly understood.

• The authors found that prolonged exposure to TNFα and compressive stress led to cytoskeletal disruption, induced enhanced inflammation, and a distinct chromatin state in the stromal fibroblasts embedded in the three-dimensional collagen matrix.

• These results highlight the important cross-talk between cytokine signaling and mechanical forces on stromal cells within the TME and facilitate the development of a better organoid model for therapeutic interventions.

INTRODUCTION

In the past several decades, our understanding of tumorigenesis has fundamentally shifted from viewing cancer as simply an accumulation of genetic mutations to acknowledging the complex and interconnected roles of genetic perturbations, mechanical stresses, and chemical signaling in promoting tumor development. While some of these factors originate from the cancer cells themselves, many of the factors involved in cancer progression are now thought to originate from the tumor microenvironment (TME) (Saraswathibhatla et al., 2023). The TME is the dynamic ecosystem surrounding the tumor, which evolves alongside the growing tumor to support its development and eventual metastasis. Surprisingly, many of the cells that support tumor development are stromal cells. Stromal cells are functionally and morphologically heterogeneous groups of cells that form the major infrastructure of connective tissues (Schuster et al., 2021). Stromal cells are primarily responsible for tissue repair during wound healing and inflammation, but upon receiving cancer signals from the TME, they acquire hallmarks of cancer such as sustained proliferation, resistance to cell death, and promotion of metastasis (Fouad and Aanei, 2017; Eiro et al., 2018; Cruceriu et al., 2020). This recruitment of stromal cells to support the growing tumor is key to tumor survival. Recruited stromal cells have been shown to contribute to chemotherapy-resistant populations of cells and have been shown to mediate tumor–immune interactions, rendering strategies such as chemotherapy and immunotherapy less effective (Feng et al., 2022).

One of the largest classes of stromal cells is fibroblasts. Tumor recruitment of fibroblasts is facilitated by the release of various tumor signals including tumor necrosis factor (TNFα), a proinflammatory cytokine, into the TME causing malignancy (Kalluri, 2016). TNFα signaling is ubiquitous in all stages of cancer progression, causing DNA damage to stromal cells and inducing a range of matrix metalloproteinase cytokines and chemokines that promote tumor development (Balkwill, 2006). This chemical signaling occurs within an altered extracellular matrix (Stylianopoulos et al., 2012), which is characterized by tension at the periphery of the tumor. This significantly alters the mechanical properties of the embedded stromal cells (Li and Wang, 2020; Venkatachalapathy et al., 2021). Fibroblasts transmit these mechanical signals through cytoskeletal networks to the nucleus to alter the cellular gene expression programs at play (Iyer et al., 2012). These mechanical signaling pathways have been shown not only to induce a wide range of functional changes but also to elicit a differential inflammatory response to TNFα in two-dimensional (2D) culture conditions (Mitra et al., 2017). Therefore, it is clear that the cross-talk between TNFα-mediated chemical signaling and altered extra cellular matrix (ECM) mechanical signaling provides a basis for tumor progression within the TME. However, thus far our insights are mostly limited to understanding each of these signaling modalities independently and are primarily based on studies performed in 2D.

Because of the growing research that shows that the TME is intimately involved in cancer progression, many therapies are being developed to target the TME. For example, anti-angiogenesis therapy, checkpoint inhibitors, and matrix modifying enzymes are among the many treatments being investigated to alter the ECM chemically and mechanically (Emambux et al., 2018; Bejarano et al., 2021; Cox, 2021). While these therapies have promise, none have yet to be fully effective due in large part to our gap in knowledge about how the three-dimensional (3D) mechanical environment interplays with chemical recruitment signaling to activate tumor progression within the TME (Shojaei, 2012; Discher et al., 2017; Emambux et al., 2018; Cox, 2021). Without exploring potential treatments in a physiologically relevant, 3D TME, the interplay between the stromal response to chemical and mechanical signaling that are both present in the TME cannot be captured and thus TME-targeting treatments cannot be well characterized and understood.

In this paper, we present a novel spheroid model that captures the cross-talk between TNFα-releasing cancer spheroids and compressive force–induced mechanical signaling on stromal fibroblasts within the TME. We designed a heterologous protein expression assay, expressing the novel construct of YFP-TNFα designed for extended release of soluble TNFα. The release of TNFα by cancer spheroid analogues into a 3D collagen gel TME causes p65 nuclear translocation in stromal fibroblasts. This TNFα-induced p65 nuclear translocation is intensified in the presence of compressive stresses, also leading to cytoskeletal disruption. Importantly, we show that these environmental cues lead to distinct chromatin states of mechanochemically activated fibroblasts. These distinct states reveal that there are two populations of stromal fibroblasts within the TME: one that primarily responds to TNFα signaling and another that primarily responds to mechanical signaling. These findings underscore the importance of understanding the heterogeneous response of cells within the TME and could be used to guide therapeutic design.

RESULTS

Heterologous protein expression assay yields viable TNFα secretion and stromal signaling response

This study aimed to understand the multidimensional nuclear response of naive/healthy stroma exposed to TNFα in a mechanically relevant TME over a prolonged period. To produce a TME with prolonged TNFα signaling, we designed a heterologous protein expression assay in which Human Embryonic Kidney cell line (HEK293) spheroids were transiently transfected with a novel TNFα construct that was optimized for extended release. The construct consisted of the soluble domain of TNFα and a YFP fusion protein to allow visualization of the secreted TNFα. For our construct design, we used the available structural information of the extracellular domain of the human 55 kd TNF receptor in complex with human TNFβ molecule to ensure its accessibility for receptor binding (Banner et al., 1993). Evident from the structure, the soluble domain of TNFα forms a homotrimer (PDB:1TNF) (Eck and Sprang, 1989) that has a conical receptor binding domain (Figure 1A). To avoid interference with receptor binding, the YFP fusion protein was added at the terminal end, located on the wider end of the binding domain (Figure 1, A and B). To further compensate for the elongated structure of the TNF receptor, extending further toward the wider region of the conical binding site of TNFα, we introduced an additional glycine-serine flexible linker into our construct (Figure 1B). To allow direct secretion, the domain boundaries of the construct contain the soluble domain of TNFα without the membrane anchor, and we included the IFNalpha2 (human interferon alpha 2 signal peptide) signal peptide with Cytomegalovirus (CMV) promoter for enhanced TNFα secretion (Román et al., 2016). In addition, we added a Rhinovirus 3C protease cleavage site to optionally remove the YFP. For detection of the fusion protein or cleaved-off YFP, we included a FLAG-tag at the C-terminus of YFP upstream of a human rhinovirus 3C protease cleavage site.

FIGURE 1:

Heterologous protein expression assay yields viable TNFα secretion and stromal signaling response. (A) Structure of the receptor binding domain of TNFα (left), and the extracellular domain of the human 55 kd TNF receptor in complex with human TNFβ (right). (B) Design of the YFP-TNFα construct. (C) Schematic of the heterologous protein expression assay. HMF3A cells are plated on a glass substrate and treated with CM from either control HEK293 spheroids (transfected with mCherry) or YFP TNFα secreting HEK293 spheroids (transfected with the YFP-TNFα construct). (D) Representative fluorescence images of p65 cellular localizations (red) and nuclei (blue) of 2D plated HMF3A cells (top), HMF3A cells treated with the CM from the control spheroids (middle), and HMF3A cells treated with CM from the YFP TNFα secreting spheroids (bottom). Inset: Enlarged images showing single cell p65 cellular localization. (E) Violin plot showing the relative p65 localization (Nuclear Intensity/Cytoplasmic Intensity) for 2D plated HMF3A cells, HMF3A cells treated with CM from control spheroids, and HMF3A cells treated with CM from the YFP TNFα secreting spheroids. ANOVA, p ≤ 0.0294. Cell cultures = 6 samples per condition, biological replicates = 3, n_total = 1638, n_total = 558, n_{Control − Spheroid − CM} = 427, n_{YTS − CM} = 653. All scale bars are 50 µm. p values (Mann–Whitney U test/Wilcoxon rank-sum test): **** ≤ 0.0001, *** ≤ 0.001, ** ≤ 0.01, * ≤ 0.05.

We transfected HEK293 cells with the novel YFP-TNFα construct and grew the cells on rectangular fibronectin micropattern for 24 h to form spheroids. Both the YFP-TNFα transfected spheroids and the mCherry transfected spheroids were grown in normal growth media, and the media was collected after 24 h (“conditioned media,” CM). Interestingly, we observed that these YFP-TNFα transfected spheroids (225 ± 50 µm) were found to be 4.5 ± 0.5 times smaller than spheroids transfected with a carrier control, mCherry when cultured for 96 h using the hanging drop method (Supplemental Figure S1, B and C). We did not observe significant size differences between the spheroids treated with the carrier control and those transfected with the YFP-TNFα construct grown on the micropatterns (Supplemental Figure S1, A and C). To quantify the release of YFP-TNFα, we varied the seeded cell density and observed a corresponding increase in YFP-specific fluorescence (527 nm) in the CM of cells transfected with the YFP-TNFα construct (Supplemental Figure S1D). To test the biological functionality of the TNFα secreted by our assay, we treated 2D plated human mammary fibroblasts (HMF3A) with CM collected from both the YFP TNFα secreting spheroids and control spheroids (Figure 1C). When HMF3A cells were treated with CM from control spheroids, p65 was localized within the cytoplasm (Figure 1, D and E). However, upon stimulation with CM from the YFP TNFα secreting spheroids, the HMF3A cells show a statistically significant nuclear translocation of p65 (Figure 1, D and E). Therefore, our YFP TNFα secreting spheroids transfected with our novel YFP-TNFα construct did indeed secrete TNFα for a prolonged period (24 h) and this secreted TNFα produced a signaling response in naive fibroblasts. We therefore can utilize our YFP TNFα secreting spheroids to investigate the stromal response to persistent TNFα signaling in the TME.

Heterologous protein expression assay captures TNFα mediated p65 nuclear translocation in fibroblasts embedded in 3D collagen matrix

To characterize the effect of TNFα signaling from our YFP TNFα secreting spheroids on stromal fibroblasts embedded within 3D TME, we developed a 3D coculture assay (TME) of YFP-TNFα transfected HEK293 spheroids (tumor analogue) and HMF3A cells (stromal analogue). We seeded both the HEK293 spheroids and HMF3A cells in a collagen matrix and cocultured the cells for 24 h (Figure 2A). This allowed the HMF3A stromal cells to persistently experience TNFα signaling from the YFP TNFα secreting spheroids within a 3D TME. As expected, there was a marked translocation of p65 to the nucleus of the HMF3A cells cocultured with the YFP TNFα secreting spheroids (Figure 2, B and C). This translocation is significantly higher than the nuclear p65 ratio of stromal cells embedded in collagen matrices with control spheroids (Figure 2C). We also find significant degradation of IκB in the cytoplasm when compared with cells cultured with control HEK293 spheroids (Figure 2, B and D). Therefore, within a 3D, physiologically relevant TME, we show that our YFP TNFα secreting spheroids persistently release TNFα and capture the hallmarks of TNFα signaling within the TME. However, the nuclear translocation of p65 is relatively subdued in the 3D collagen matrix (Figure 2, B and C). To place the cytokine signaling in the context of TME, we next study the effect of compressive forces on the NFκB signaling in stromal fibroblasts.

FIGURE 2:

Heterologous protein expression assay captures TNFα-mediated activation in fibroblasts embedded in 3D collagen matrix. (A) Schematic representing the coculture of HMF3A cells with HEK293 cells expressing YFP-TNFα fusion protein. (B) Representative fluorescence images stained for p65 (red), IκB (magenta), YFP-TNFα (green), and DNA (blue). Images show the coculture of HMF3A stromal cells (outside dashed line) and the HEK293 spheroids (inside dashed line) both for untransfected control HEK293 spheroids (top) and HEK293 spheroids transfected with our YFP-TNFα construct (bottom). Inset: Enlarged images showing single cell p65 cellular localization. (C) Violin plot showing the relative p65 localization (Nuclear Intensity/Cytoplasmic Intensity) for HMF3A cells cocultured with control spheroids and HMF3A cells cocultured with YFP TNFα secreting spheroids, ANOVA, p ≤ 6.71e-06. (D) Violin plot showing the relative expression of IkB for HMF3A cells cocultured with control spheroids and HMF3A cells cocultured with YFP TNFα secreting spheroids, ANOVA, p ≤ 0.00149. All scale bars are 50 µm. Cell cultures = 8 gels per condition, biological replicates = 4, n_total = 473, n_{Control − Sph. − CC} = 208, n_{YTS − CC} = 265. p values (Mann–Whitney U test/Wilcoxon rank-sum test): **** ≤ 0.0001, *** ≤ 0.001, ** ≤ 0.01, * ≤ 0.05.

Compressive stress in the TME leads to enhanced p65 translocation in fibroblasts

Next, we utilized our heterologous protein expression assay to explore the mechanochemical cross-talk between TNFα-mediated p65 nuclear translocation in healthy stroma in the context of compressive stress. Previous studies have characterized mechanical stresses within the TME and shown that stromal cells at the periphery of growing tumors can experience a maximum mechanical load of up to 800 Pa (Stylianopoulos et al., 2012; Tse et al., 2012). Under the influence of mechanical stimuli, cells exposed to chemical signals have distinct inflammatory responses (Mitra et al., 2017; Venkatachalapathy et al., 2021). We applied a compressive load on top of the collagen matrices to generate compressive stress of ≈200 Pa on the HMF3A/HEK293 spheroid cocultures for 60 min (Figure 3A). Before applying the compressive load, the stromal fibroblasts cocultured with YFP TNFα secreting spheroids displayed significantly more nuclear p65 than fibroblasts cocultured with control spheroids, as shown above (Figure 3, B and C). Upon compressive load, there was a significant increase in p65 translocation to the nuclei of both the stromal cells cocultured with YFP TNFα secreting spheroids and of those cocultured with control spheroids (Figure 3, B and C). We did not find a strong distance and orientation dependence in p65 nuclear translocation as shown in Supplemental Figure S8, A and B because the mean collagen pore size is significantly larger than the molecular size of TNFα (Carey et al., 2017). We also find no evidence of significant cell death in the presence of TNFα, and compressive forces as shown in Supplemental Figure S9, A–C. To rule out differences in TNFα concentrations produced by YFP-TNFα secreting spheroids due size differences, and compressive forces, we treated the cells with the CM containing endogenously produced YFP-TNFα for 24 h. Under compressive forces and TNFα treatment, the fibroblasts embedded in the 3D collagen matrix show the most p65 nuclear translocation (Figure 4, A and C). We also find a significant increase in p65 nuclear translocation in the fibroblasts under compressive forces (Figure 4, A and C). To validate the consistent concentration of endogenously produced YFP-TNFα in the media across multiple experimental runs, we measured the fluorescent intensity of YFP TNFα containing CM. We find that intensity values across different experimental runs are consistent as shown in Figure 4B. We repeated the above experiments by treating the cells with exogenously produced 20 ng/ml TNFα. We found that both the endogenous and exogenously produced TNFα show a similar trend (Figure 4, A and C).

FIGURE 3:

Under compression, TNFα induces enhanced p65 nuclear translocation in healthy stroma. (A) Schematic representing the compression assay. HMF3A cells are cocultured with HEK293 spheroids in collagen matrices. A compressive load is placed on top of the collagen matrices to produce mechanical stress within the matrix. (B) Representative fluorescence images of p65 (red), actin (green) and nuclei (blue) for HMF3A cells (outside dashed line) cocultured with HEK293 spheroids (inside dashed line). Images are shown for HMF3A cells cocultured with control spheroids with and without compression (top two rows), and HMF3A cells cocultured with YFP TNFα secreting spheroids with and without compression (bottom two rows). (C) Violin plot showing the relative p65 localization (Nuclear intensity/Cytoplasmic Intensity) of the HMF3A cells in the cocultures, ANOVA, p ≤ 0.0032. Cell cultures = 4 gels per condition, biological replicates = 2, n_total = 1816, n_{Cont. − Sph. − CC(−)Comp.} = 408, n_{Cont. − Sph. − CC (+) Comp.} = 323, n_{YTS − CC(−)Comp.} = 420, n_{YTS − CC (+) Comp.} = 664. All scale bars are 50 µm. p values (Mann–Whitney U test/Wilcoxon rank-sum test): **** ≤ 0.0001, *** ≤ 0.001, ** ≤ 0.01, * ≤ 0.05.

FIGURE 4:

(A) Representative fluorescence images of p65 (red) and nuclei (blue) for HMF3A cells. Images are shown for HMF3A cells treated with and without compression (left two columns), and HMF3A cells treated with endogenously produced and exogenously added TNFα with and without compression (right two columns). (B) The violin plot represents the fluorescent YFP intensity in the CM. Cell cultures = 8 samples per condition, biological replicates = 4. p values (independent t test): **** ≤ 0.0001, *** ≤ 0.001, ** ≤ 0.01, * ≤ 0.05. (C) Box plot showing the relative p65 localization (Nuclear intensity/Cytoplasmic Intensity) of the HMF3A cells in the cocultures, two rows). Cell cultures = 8 gels per condition, biological replicates = 4, n_total =16054, n_{Cont. − Ex.(−)Comp.} = 2677, n_{Cont. − End.(−)Comp.} = 2069, n_{Cont. − Ex (+) Comp.} = 1452, n_{Cont. − End.(+) Comp.} = 1395, n_{YFP − TNFα − End(−)Comp.} = 3053, n_{TNFα − Ex.(−)Comp.} = 1793, n_{TNFα − Ex. (+) Comp.} = 1650, n_{YFP − TNFα − End. (+) Comp.} = 1975. p-values (Wilcoxon rank-sum test): **** ≤ 0.0001, *** ≤ 0.001, ** ≤ 0.01, * ≤ 0.05. The fold change of gene expression of identified targets (D) NFKB1, (E) BCL10, (F) IL6, (G) IL1B. Cell cultures 6 gels per condition, biological replicates = 3, Error bars represent s.e.m. Scale bars: (a) 100 µm, p values (unpaired t test): ≤ **** ≤ 0.0001, *** ≤ 0.001, ** ≤ 0.01, * ≤ 0.05.

To validate the downstream effect of p65 nuclear translocation, we carried out gene expression analysis using qRT-PCR for some of the targets of p65 that were identified in the previous work (Mitra et al., 2017). We find significant changes in the gene expression of cells embedded in the 3D culture condition where cells were treated with endogenously produced YFP-TNFα and compressive forces (Figure 4, D–G). These results imply that the p65 nuclear localization has a differential effect on transcriptional output. We also note that the cells under the compressive forces and in the presence of TNFα in the coculture show a greater degree of Actin depolymerization (Supplemental Figures S3A and S7A). These observations are in line with what has been previously observed in 2D cell culture studies under different geometrical settings (Mitra et al., 2017). Particularly under compressive forces, actin depolymerization results in HDAC3 translocation, chromatin remodeling, and transcriptional quiescence (Damodaran et al., 2018). We also find a reduction in nuclear localization of MKL 1 in the cells undergoing compression and exposed to exogenous TNFα as shown in Supplemental Figure S10, A and C. These findings are in line with the previous studies (Mitra et al., 2017). However, TNFα mediated actin depolymerization leads to IκB degradation followed by p65 nuclear translocation (Goldblum et al., 1993; Mitra et al., 2017). We also see a marked decrease in polarity given by the overall decrease in cellular aspect ratio (along the polarity axis) under compressive forces when compared with control and coculture with YFP-TNFα transfected spheroids as shown in Supplemental Figure S3B. Such geometrical changes are hallmarks from TNFα-mediated transition of healthy stromal cells to myofibroblast precursor cells (Schuster et al., 2021). Overall, these results indicate differential response of p65 nuclear translocation in response to chemical and mechanical signals.

Chromatin organization captures the functional state of mechanochemically activated fibroblasts

While TNFα and mechanical forces have both been shown to induce inflammation in the TME separately (Mitra et al., 2017), it is still unknown how the combination of these inflammation-inducing modalities affects the functional state of the stroma. Therefore, we investigated how the chromatin structure of the cells experiencing TNFα signaling and/or compressive force adapts within the TME to respond to these combinatorial stresses. We analyzed chromatin structures by quantifying over 140 handcrafted features of the nuclear morphology and chromatin intensity profiles of single-cell fluorescence images of the stromal fibroblasts, as described previously (Venkatachalapathy et al., 2021) (Figure 5A). These descriptive features underpin the functional state of the cell (Yang et al., 2021). We used these descriptive features to train a 4-layer feedforward neural network (FNN) to learn the nuclear morphometric and chromatin organizational changes induced by the different mechanochemical perturbations (Materials and Methods, Figure 5B). These mechanochemically-induced organizational changes were so distinct that the FNN could identify which cells (at a single-cell level) had received which treatment with 85% accuracy based only on the nuclear morphometry and chromatin organizational distributions (Figure 5F). When we repeated our analysis and evaluated the model's classification accuracy using a leave-one-out cross-validation method, we found it to be significantly higher than random chance for each category as shown in Supplemental Figure S11. This indicates that, on the single-cell level, stromal fibroblasts undergo large-scale changes in chromatin organization in response to chemical signaling, mechanical signaling, and the combination of the two which produces distinct functional changes in the cells experiencing each perturbation. In the presence of TNFα and compressive force, we find an overall increase in the definition of heterochromatin and euchromatin domains given by the increase in entropy under the perturbation conditions (Supplemental Figure S4B). We also found an increase in the size of heterochromatin domain boundaries in the presence of TNFα and a decrease in heterochromatin domain size in the presence of compressive forces and compression in the presence of the TNFα. We additionally characterized these organizational changes by analyzing the 32-dimensional output layer of FNN, by reducing it to 3D using t-SNE (Van der Maaten and Hinton, 2008) (Figure 5, C–E). Evident from the latent space, the cells form three clusters (Figure 5D). The first cluster is primarily comprised of control cells and also contains a small population of cells that underwent compression (Figure 5, C and G). The second cluster is primarily comprised of cells that experienced TNFα signaling from the TME and includes both those cells that experienced TNFα signaling alone and also those that experienced TNFα signaling and compression (Figure 5, C and G). Finally, the third cluster is primarily comprised of cells that experienced compressive stress within the TME and includes both those cells that experienced compressive stress alone and also those that experienced TNFα signaling and compression (Figure 5, C and G). Therefore, while each of these perturbations induces distinct chromatin states (Figure 5F), it is clear that these states can be broadly clustered into cells that are unaffected by treatment (Cluster 1), cells that are primarily affected by chemical signaling (Cluster 2), and cells which are primarily affected by mechanical signaling (Cluster 3). Thus, even for cells that received the same perturbation, the response to this perturbation is heterogeneous throughout the TME. This is also reflected in the heterogeneity of the cells undergoing each perturbation. While cells experiencing either TNFα signaling or compressive force alone are primarily contained within Cluster 2 or Cluster 3, respectively, cells experiencing both chemical and mechanical signaling have the largest degree of heterogeneity in chromatin state and are split between Clusters 2 and 3 (Figure 5, C and G). Notably, these clusters are additionally separated by the amount of p65 nuclear translocation within the cells (Figure 5, E and H). Cluster 3 has a significantly higher amount of p65 translocation into the nucleus than Cluster 2, notably in the cells exposed to TNFα and compressive force (Supplemental Figure S5). Therefore, the change in chromatin state is in part a reaction to the degree of p65 nuclear translocation.

FIGURE 5:

Chromatin states underpin stromal heterogeneity in mechanochemically induced activation. (A) Representative images of nuclei from corresponding treatment conditions. (B) Schematic representation of the FNN used to predict treatment conditions based on the input of handcrafted chromatin and nuclear features quantified from single-cell images. (C) t-SNE dimensionality reduction of the output of the FNN. The t-SNE plot is a 3D representation of the chromatin state (≥100 dimensions) based on the output of the FNN. (D) Clustering of the t-SNE chromatin representation. (E) Chromatin states are colored by the corresponding p65 nuclear to cytoplasmic ratio. Warmer colors represent more nuclear p65 translocation. (F) Confusion matrix representing the classification precision of the FNN at distinguishing treatment conditions based on single-cell chromatin states. (G) Composition of the clusters. The heatmap shows the number of cells in each treatment group that is represented in each identified cluster. (H) Box plot representing p65 nuclear to cytoplasmic localization corresponding to individual clusters. p values (Mann–Whitney U test/Wilcoxon rank-sum test): **** ≤ 0.0001, *** ≤ 0.001, ** ≤ 0.01, * ≤ 0.05.

Altogether, these results suggest that signaling within the TME causes marked changes in chromatin state as the stromal cells surrounding the growing tumor adapt to external mechanical and chemical cues. As the cells experiencing these cues have distinct changes in their chromatin states, which would cause functional changes within the cells, this analysis highlights the need to investigate signaling within the TME in a multidimensional way.

DISCUSSION

Herein, we present a novel TME assay in which stromal fibroblasts are embedded in a collagen matrix with YFP-TNFα secreting spheroids, which are transfected with a fusion protein construct to ensure extended expression of TNFα into the TME. Previous studies have pointed out that the growing tumor develops internal stresses due to cell proliferation and exerts compressive stresses on stromal cells located at its periphery (Stylianopoulos et al., 2012) which have been shown to induce invasive phenotypes in 2D cultures (Tse et al., 2012). The effect of cytokine signaling has also been shown to alter matrix stiffness and induce invasive phenotype (Provenzano et al., 2009). In our previous studies, we found that cells constrained by different 2D geometries exhibit differential gene expression in response to cytokine signaling (Mitra et al., 2017). However, the cellular and nuclear response to the microenvironmental signals has not been fully explored (Kalukula et al., 2022). Expounding further in this study, we used our heterologous protein expression assay to examine the response of healthy fibroblasts in the presence of compressive stresses under the persistent TNFα mediated p65 nuclear translocation in 3D TME. With this assay, we show that the spheroids actively release TNFα in such a way that it causes increased p65 nuclear translocation in stromal fibroblasts both in 2D and 3D. This is amplified by applying a compressive load to the matrix, simulating the mechanical pressure experienced by stromal cells at the periphery of the tumor. This combinatorial addition of chemical and mechanical signals not only increases the activation of the stromal fibroblasts but also rearranges the chromatin structure contained within these cells. We see that, within the same TME, the chromatin states of the stromal fibroblasts either appear to be primarily responding to the chemical signals or primarily responding to the mechanical signals. Therefore, we show that the response of the stromal cells to a uniform signal within a TME is nonuniform. In conclusion, our results highlight the interplay between mechanical forces and NFκB signaling in determining the functional changes in stromal cells. Specifically, compressive mechanical forces generated by growing tumors were found to promote malignancy and activation within the TME. To study the role of cytokine signaling in this context, we employed HEK293 cells transfected with a fusion protein construct to reliably produce high concentrations of pure, biologically active YFP-TNFα cytokine (Bierig et al., 2020). Although coculturing cell types from the same tissue origin could provide more physiologically relevant observations, as shown previously (Venkatachalapathy et al., 2020), such an approach is often complicated by the diverse and heterogeneous cytokine profiles of many cancer cell lines (Chen et al., 2020). Overall, this study highlights how mechanical stress and cytokine signaling contribute to shaping the TME, and it emphasizes the potential of identifying cells with specific stress responses for therapeutic intervention or to inform the design of new therapeutic strategies.

MATERIALS AND METHODS

Request a protocol through Bio-protocol

Cell culture

HMF3A fibroblast cells (Cat No: T0153, Applied Biological Materials, Inc.) and HEK293 cells (Cat No: CRL-1573, ATCC) were cultured in high glucose DMEM (Life Technologies, Life Technologies) supplemented with 10% (vol/vol) FBS (GIBCO, Life Technologies) and 1% PenStrep at 37°C in 5% CO₂ in a T25 cc flask. HMF3A cell line used for these experiments was within the range of 6–12 passage numbers. Later passages (≥12) of the HMF3A cell line were found to be unsuitable for 3D coculture experiments as we found that the high passage number HMF3A cells were significantly less proliferative, spread less, and showed decreased mobility. HEK293 cells were transfected (Lipofectamine 3000, Invitrogen) with novel YFP-TNFα plasmid according to the manufacturer's protocol in 6-well cell culture plates (Sarstedt). For the CM experiments, transiently transfected HEK293 cells were initially washed with 1 × DPBS and cultured with fresh DMEM for 24 h. For the plasmid control, we collected the CM from the HEK293 cells transfected with mCherry plasmid. CM was aspirated from the 6-well plates and filtered through a 0.2 µm syringe filter to remove cellular debris. For spheroid coculture experiments, HEK293 cells were trypsinized and resuspended 24 h posttransfection for the spheroid formation. Transiently transfected HEK293 spheroids were grown on a rectangular micropattern of area 1800 µm² for 24 h. We used the HEK293 spheroids transiently transfected with mCherry as a control.

Micropatterning and spheroid formation

Polydimethylsiloxane (PDMS) stamps were prepared by mixing PDMS elastomer (SYLGARD 184; Dow Corning) with curative precursor in a 1:1 ratio according to the manufacturer's protocol and by molding PDMS in microfabricated silicon wafers. The fibronectin solution (100 µg/ml) was then allowed to adsorb on the surface of each PDMS stamp under sterile conditions. The coated PDMS stamp was gently placed onto the surface of an untreated hydrophobic dish (Ibidi, 35 mm) to print the micropatterns. The surface was then treated with 2 mg/ml Pluronic F-127 (Sigma-Aldrich) for 30 min to passivate regions surrounding the coated surface. Transiently transfected HEK293 cells were seeded on the micropatterns at 37°C in 5% CO₂ and left undisturbed overnight. Unadhered cells were removed the day after, and fluorescent spheroids expressing YFP-TNFα and mCherry (Control) were scraped from the micropatterns.

CM, 3D coculture, and compressive force assay

For the CM experiments, HMF3A cells plated on 2D substrates were treated with the CM for 24 h and fixed with 4% paraformaldehyde (Sigma-Aldrich) in 1 × PBS. For the 3D coculture experiments, HMF3A cells were trypsinized and resuspended in collagen solution and placed in cell culture dishes (IBIDI, 35 mm). We used 0.6E6/ml cells for 1 mg/ml rat tail collagen (Thermo Fisher Scientific, A1048301) neutralized with 0.1 N NaOH concentration for 3D coculture. Scraped spheroids were carefully placed in the collagen gel. HMF3A and transiently transfected HEK293 spheroids were cocultured for 24 h and fixed with 4% paraformaldehyde (Sigma-Aldrich) in 1 × PBS. Compressive Force experiments were performed as described in previous studies (Damodaran et al., 2018, 2019). Briefly, the collagen matrix comprising HMF3A and HEK293 spheroids was covered with 2 mg/ml Pluronic F-127 (Sigma-Aldrich) washed glass coverslip after 24 h of seeding. Paraffin film-coated load rings of 5.1 g were placed on the collagen coculture for 1 h. Cells were then fixed with 4% paraformaldehyde (Sigma-Aldrich) in 1 × PBS. We have calculated the compressive force considering the effect of buoyancy and gravity, assuming that the compressive force is uniformly applied over the collagen gel.

For all the experiments, we observed that the submerged volume, V_sub ≈ V_Coverslip:

Considering, ρ_glass = 2500 kg.m⁻³, ρ_media = 1000 kg.m⁻³ (Poon, 2022), V_Coverslip = 0.3 cm³, and M_load = 0.005 kg:

The area of the collagen gel before the load ranging from 280 to 285 cm², therefore P is in the range from 187.4 to 190.75 Pa

Concentration analysis of Endogenous YFP-TNFα

HEK cells transfected with YFP-TFNα construct were seeded in the cell culture dish for 24 h at the cell density of 5 × 10⁴, 1 × 10⁵, 5 × 10⁵, and 1 × 10⁶ cells. The YFP-TNFα enriched cell culture media was aspirated from the culture dish and centrifuged at 2400 RPM to remove the debris. The 150 µl of media from YFP-TNFα and corresponding control is added to each well in a 96-well plate to validate the expression of TNFα against the control. The relative YFP-TNFα concentration against the control is measured using Pherastar FSX (BMG Labtech) measured using a 485–535 filter and spiral sampling protocol after adjusting the gain corresponding to the well with the highest concentration of YFP-TNFα.

Treatment with endogenous and expogenous TNFα and compressive forces

The cell culture media enriched with endogenous YFP-TNFα from the seeded cell number of 5 × 10⁵ cells is selected for the treatment. The concentration of YFP-TNFα is measured against the corresponding control (Material and Methods). The HMF3A cells cultured in a 3D collagen matrix (Material and Methods) are treated then treated with 1 ml of YFP-TNFα for 24 h and with compressive force (Material and Methods) for 60 min.

Recombinant TNFα (Life Technologies, Catalogue No. PHC3011) was added to cell culture media to create 20 ng/ml solution. The HMF3A cells cultured in a 3D collagen matrix (Material and Methods) are then treated with 1 ml of Exogenous TNFα enriched cell culture media for 24 h and with compressive force (Material and Methods) for 60 min.

Real-time polymerase chain reaction analysis (qRT-PCR)

To investigate the selected gene expression in the cells from the Control (−) Compression, Control (+) Compression, Endogenous YFP-TNFα (−) Compression, Endogenous YFP-TNFα (+) Compression, qRT-PCR was performed. Total Ribonucleic acid (RNA) was isolated from cells embedded in collage gel by using the RNeasy Plus Micro Kit(Qiagen). Cells in 3D collagen-I gel were treated with collagenase for 15 min before RNA isolation. By using an iScript cDNA synthesis kit (Bio-Rad), cDNA was synthesized from total RNA that was isolated. RT-PCR detection was performed by using a SsoFast qPCR kit (Bio-Rad) for 40 cycles in a Bio-Rad CFX96. The relative fold changes in the gene levels were obtained from qRT-PCR data, by using ∆∆Ct methods with respect to glyceraldehyde3-phosphate dehydrogenase (GAPDH) levels. The primer sequences for these selected genes are

• IL1B-F: CTC TCT CCT TTC AGG GCC AA

• IL1B-R: GAG AGG CCT GGC TCA ACA AA

• IL6-F: CCT GAA CCT TCC AAA GAT GGC

• IL6-R: TTC ACC AGG CAA GTC TCC TCA

• NFKB1-F: GGC AGC ACT ACT TCT TGA C

• NFKB1-R: CAG CAA ACA TGG CAG GCT AT

• BCL10-F: GCA GTT GTG AAC CTT TTC CAG A

• BCL10-R: TGG ATG CCC TCA GTT TTT CAG

Immunostaining 3D coculture samples

Gels were rinsed twice with 1 × PBS, followed by fixation using 4% paraformaldehyde (Sigma) for 20 min. Cells were washed and permeabilized with 0.5% Triton-X (Sigma) in 1 × PBS for 30 min. After washing thrice with 1 × PBS, the cells were treated with 10% goat serum in Immunofluorescence (IF) (0.2% triton and 0.2% Tween) wash buffer (blocking solution) for 3 h. This was followed by incubation with required primary antibodies (in blocking buffer) overnight at 4°C. Cells were incubated with primary antibodies for p65 (CST, D14E12) overnight at 4°C. Cells were washed thrice for 15 min with IF wash buffer and incubated with corresponding Alexa Fluor secondary antibody at room temperature for 3 h and washed thrice with IF buffer. Following that samples were incubated with Hoechst-33342 (1 mg/ml; 1:1000; Invitrogen, NucBlue R37605) for 15 min. Filamentous actin (F-actin) was labeled using phalloidin Alexa Fluor 488 (1:1000; Invitrogen, ActinGreen, R37110).

The samples were washed thrice with IF wash buffer and imaged in DPBS.

Confocal imaging and image analysis

The samples were scanned using the Nikon A1 Confocal microscope (Nikon, USA) with a 60 × 1.25 NA oil immersion objective. Stacks of 16-bit grayscale 2D images were obtained with a voxel size of ≈ 300 nm in the XY direction and ≈ 1 µm in the Z direction. Images were analyzed using the custom script written in Matlab (The MathWorks Inc., 2022) and Python (Van Rossum and Drake, 2009) using built-in functions for image processing. Nuclei images were thresholded from the background using imbinarize function in Matlab, Cytoplasmic mask was segmented using the Cellpose (Stringer et al., 2021) library in Python. Morphological properties of nuclear and cellular features were calculated using regionprops3 function in Matlab. The densely clustered with overlapping nuclei cells were segmented using watershed function in with seed points obtained by thresholding distance transform. The oversegmented regions from the watershed algorithm were reconnected using the approach suggested in previous studies (Gamarra et al., 2019). The concave, oversegmented, clustered, and oversaturated nuclei were ignored from the analysis. HC nodes were identified from DNA intensity images based on an intensity and volume threshold: pixels with intensity more than Θ(Inuc) were first binarized and 3D nodes were identified as those objects with volumes of at least 1 µm³:

The feature list used for calculating chromatin and nuclear morphometric features is the same as described previously in Venkatachalapathy et al. 2020 (Venkatachalapathy et al., 2020) briefly described in Supplemental Information. The nuclear to cytoplasmic p65 intensity was calculated using the cellular binary mask obtained using Cellpose (Stringer et al., 2021).

Statistical analysis

All statistical analyses and plotting were carried out in R (R Core Team, 2021), Python (Van Rossum and Drake, 2009), and Python Packages (Virtanen et al., 2020; Pedregosa et al., 2011; Stringer et al., 2021; Waskom, 2021; Harris et al., 2020). For box plots, the box limit represents the 25th to 75th percentile and whiskers 1.5 × interquartile range. Unless stated otherwise in the figure legends, we evaluated the statistical significance of the mean with the Mann–Whitney U-test, performed between a sample of interest and the corresponding control because the distribution of data was found to be skewed in most cases. The total sample sizes for all the experiments are mentioned in the respective figure legends. We described the changes in chromatin states under mechanochemical perturbations, using the nuclear and chromatin morphometric features for each nucleus. We normalized the feature list to obtain the z-score corresponding to the given feature for each nucleus using the fit transform function from sklearn library. We had in total 141 features describing a total of 2681 nuclei in our final dataset. We split the normalized feature list for all nuclei into training and testing sets randomly in the ratio of 3:2 split using the train test split function from sklearn library. We defined the 4-layer feed-forward neural network using keras tensorflow package with rectified linear activation function for input and hidden layers and softmax activation function for output layer, initialized with random numbers. We trained the feed-forward neural network using Adam solver using categorical cross-entropy as loss function using compile function from keras tensorflow package on the testing set for over 100 epochs for a batch size of 32. We tested the accuracy of our model using predict function from keras tensorflow package on the testing set. The confusion matrix after 3-fold cross-validation is plotted as a heatmap. We populated the latent space by predicting the perturbation condition for the entire dataset on the trained feed-forward neural network. We performed dimensionality reduction using the t-SNE method over 1000 iterations using TSNE library with a perplexity value of 30. We performed the k-means clustering on the 3rd layer and removed the clusters that had a population of less than 10. We chose the optimal number of clusters using the elbow method by evaluating within cluster squared sum for various iterations of the number of clusters obtained after K-Means clustering. As shown in Supplemental Figure S6A, the number of clusters as 4 or 5 can both be the optimal choice; however, we chose 5. As shown in Supplemental Figure S6B, both clusters # 2 and 4 were removed as they each contained 3 points. For the baseline performance evaluation of FNN, we also performed the parallel classification using the Random Forest algorithm using the MASS library. The corresponding confusion matrix after 3-fold cross-validation is plotted as a heatmap. We obtained the accuracy score for all the features above 5 and plotted them as an annotation to the heatmap alongside the mean value for each perturbation condition. Statistical data visualization and representation are carried out using ggplot and seaborn library. No samples were excluded from the study and all the data collected is presented in the figures. The controlled experiments were not randomized. The investigators were not blinded.

Abbreviations used:

BCL10-B-cell

lymphoma/leukemia 10

CC

co-culture

CM

conditioned media

CMV

cytomegalovirus

DMEM

dulbecco's modified eagle medium

DNA

deoxyribonucleic acid

ECM

extra cellular matrix

F-actin

filamentous actin

FBS

fetal bovine serum BSA

FNN

feedforward neural network

GAPDH

glyceraldehyde3-phosphate dehydrogenase.

HDAC3

histone deacetylase 3

HEK

human embryonic kidney

HMF3A

human mammary fibroblasts 3A

IF

immunofluorescence

IFNalpha2

human interferon alpha 2 signal peptide

IL1B

interleukin 1 beta

IL6

interleukin 6

MKL

megakaryoblastic leukemia

NFκB

nuclear factor kappa-light-chain-enhancer of activated B cells

PBS

phosphate-buffered saline

PDB

protein data bank

PDMS

polydimethylsiloxane

qRT-PCR

real-time polymerase chain reaction

RNA

ribonucleic acid

TME

tumor microenvironment

TNFα

tumor necrosis factor

t-SNE

t-distributed stochastic neighbor embedding

YFP

yellow fluorescence protein

YTS

YFP TNFα secreting spheroids.