At Entelos, a common refrain was that biologist don’t take the right data when it comes to building mechanistic mathematical models of disease. Yet, the company was reluctant to publish primary research articles showing how existing data are used and what kind of data are needed. Over time, this blanket critique of biologists, to me, seemed a bit hypocritical. So in re-entering academia with a single ticket to play at WVU, I wanted to change that narrative.
Originally my lab focused on how cells integrated signals internally by analyzing data generated by others and, increasingly, my own group using mechanistic mathematical models. While there was a robust community that modeled signaling cascades within cells, two events encouraged me to take a different direction. One was feedback on a proposal where the reviewer mentioned that I should find a scientific space that doesn’t compete with larger groups at more prestigious institutions. The second was emerging literature illustrating the extensive and sustained rounds of mutations that occur during oncogenesis. To me, these events collectively suggested that developing mechanistic models of how malignant cells integrate signals would be a difficult path to take. Around 2009, my lab started to tack a different course.
Within tissues, signaling among different cell types is important for organizing the right response. One cell type that I had previously modeled was the dendritic cell. As scouts of the immune system, this specialized cell senses threats by recognizing molecular patterns associated with pathogens or traumatic cell death through genetically encoded sensors and secretes signals that activate other cells that are adapted to get rid of the threat, like Natural Killer (NK) cells and cytotoxic T lymphocytes (CTL). One of these secreted signals is the cytokine Interleukin-12 (IL12), which was originally called Natural Killer cell stimulating factor.
Given that oncogenesis is an evolutionary process associated with mutation and selection, malignant cells that arise within a tissue may potentially interrupt this communication between dendritic cells and NK cells/CTLs. Towards this aim, we set up a phenotypic screening assay – an assay where you model a cellular behavior and seeing if changing the experimental conditions changes the cellular behavior [1]. If it does, this is followed by identifying the factor responsible for the change in phenotype. The phenotype that we were modeling was a T cell response to IL12 in the production and release of a cytokine that enables CTL cells to recognize their target cells (IFN-gamma). We had also published a quantitative study of a T cell line that responded to IL12 [2]. Experimentally, we tested whether the T cell response to IL12 changed in the presence of a mouse cancer cell line: B16F0. The B16 model and its’ variants are interesting as they have been considered the gold standard for testing immunotherapies in pre-clinical mouse models, as essentially most interventions don’t work in it.
To extract as much information as we could from this experiment that involves two cell lines interacting in a dish, a time-course and multiple read-outs were included in the experimental design. Having a mechanistic model of the T cell response to IL12 helped rule out some competing explanations. Interestingly, the time-course data suggested that the tumor cell was secreting something that was interfering with the T cell response to IL12. Of note, a recent study reported that the anti-tumor response enhanced by immune checkpoint blockade depends on tonic IL12 signaling within the tumor microenvironment [3].
To identify this secreted factor, we also wanted to minimize observational bias and incorporated a mass spectrometry-based proteomic component into the study. The analysis of the proteins secreted by the B16F0 cells suggested a short list of candidates that included a number of secreted proteins and extracellular vesicles. Given a head-to-head comparison of likely potential targets summarized in [4], we focused on the secreted protein Cell Communication Network factor 4 (CCN4 and previously known as Wnt-inducible Signaling Protein – 1 (WISP1)). Extracellular vesicles are also intriguing but the fastest path to translating our findings to the clinic would be to target a secreted protein using an antibody.
As there are only about 500 papers related to CCN4/WISP1 that have been published since 1993, CCN4 biology is still a bit murky. In the study of cancer, one of the first questions that people ask is whether the protein of interest is increased in a tumor relative to normal tissue. In contrast to an earlier time where one would have to gather and analyze tissue samples themselves, recent public repositories of data obtained from tumor and normal tissue facilitate such comparisons. Leveraging these public databases, I found that CCN4 is upregulated in essentially every tumor sample from patients with invasive breast cancer but not in normal mammary tissue [5]. In the context of melanoma, we also found that increased CCN4 expression correlates with a worse outcome [6] and that malignant melanocytes are a source of CCN4 within the tumor [8].
Observing that CCN4 was increased in invasive breast cancer and unregulated at the invasive front of melanoma nests motivated a follow-on study to clarify the association between CCN4 expression and invasion. Using a combination of wet experiments and mechanistic mathematical modeling, we found that disrupting adherens junctions, which occurs during invasion, increases CCN4 expression with an interlocked positive and negative feedback network motif [9]. Interestingly, genetic alterations that activate this pathway were significantly enriched in melanoma compared to random chance [6].
Ok so increased CCN4 is associated with a worse outcome, but what does it do? Well secreted proteins can have two different effects. First they can influence the behavior of the cell that makes it. You can kind of think of this as the out-loud self talk that we sometimes use to encourage ourselves to complete a task – “You can do this!”. Second, they can influence the behavior of other cells that are able to respond to the secreted signal. You can think of this as the cross-talk between different players of the same soccer team as they play an opponent – “Man on, man on”.
In cases where the particular secreted protein is one of the 20% of genes that are well-studied, clarifying these two different effects is a bit easier. In such case, the experimental tools have been refined and you have a strong guess as to what you’re looking for. Unfortunately, CCN4 is not one of the well-studied genes. So we decided to address the self-talk or autocrine signaling first. In the context of melanoma, we published two papers to help clarify how CCN4 altered the behavior of melanoma cells [6,7]. While all cells have the capability to express every gene, access to the DNA within a cell is modified during development to shape what genes are expressed by the cell and, ultimately, the cell’s biological function. As malignant melanocytes and malignant breast cancer cells have very different developmental origins, we would expect that the action of CCN4 on breast cancer cells is likely different than on melanocytes. But we’re currently working on that.
In contrast to the different developmental origins of these two malignancies, the immune system is distributed throughout the body and regulated by similar mechanisms irrespective of the anatomical location. Finding the same mechanism to suppress host immunity at work in different cancers may suggest a convergent evolutionary path to oncogenesis and an interesting therapeutic target. However, identifying cross-talk between different cell types for a relatively understudied gene is difficult. To do this, we took a two-pronged approach, one traditional and one data-driven.
First, we used a traditional approach in cancer immunology: remove the gene of interest in a transplantable tumor cell line, implant the wild-type and knock-out variants of the tumor cell line in immunocompetent mice, and look for changes in tumor growth. If there are differences in tumor growth, you isolate the tumors, digest into single-cell suspensions, and characterize the difference in prevalence of immune cell types using flow cytometry. In a recent paper [8], we report findings using two different mouse melanoma models, where knocking out CCN4 reduced tumor size and increased the prevalence of CTLs and NK cells in the tumor.
One limitation of the traditional cancer immunology approach is that maybe you’re missing something as you have to decide what to measure when you design the experiment. Sure single-cell RNA sequencing can address this point, but it’s really expensive and is a relative new technique. While there is a bit of nuance here, resolution and statistical power are concerns such that while one can generate different clusters, you don’t really know why the clusters are different. Moreover excitement about using single cell sequencing to reveal cell-cell communication was dashed by the fact that these algorithms rely on existing modes of cell-cell interaction. How CCN4 alters cell-cell communication is not known, so that doesn’t help. So in the end, you have two different techniques that largely give the similar information – that CCN4 elicits changes. We went with the less expensive route.
A second limitation is based on using mouse models of cancer, where translating the findings in mice to humans is not clear and a common source of critique. Sure there are some “humanized” mouse models but, to me, these are not a good choice here. If you observe something in these “humanized” mouse models, the results should be interpreted with caution due to potential artifacts of putting two different species together. The chances of going down some scientific rabbit hole to find a dead end are significant and was not a risk I wanted to take.
How about taking actual human data and predicting how increased expression of a gene influences the prevalence and functional orientation of different cell types within a particular tissue, like the skin in the case of melanoma or mammary gland in the case of breast cancer? Given the prevalence of public databases that contain gene expression information obtained from samples of melanoma and breast cancer tissue, we combined two different computational approaches to predict how a gene, like CCN4, alters the network of cells within a tissue during oncogenesis [10]. We validated the predictions using results obtained from our mouse models. This data-driven approach helps motivate focused experiments to clarify CCN4’s role in suppressing anti-tumor immunity.
In summary, we’ve taken an approach informed by my experiences building industrial-scale mechanistic mathematical models of disease. While the initial phenotypic screening study [1] provided preliminary data for a CAREER award from the National Science Foundation and a R01 grant from the National Cancer Institute, additional progress has stalled due to a lack of funding (not that we haven’t tried). Some thoughts on why in a future post.
References
[1] Kulkarni YM, Chambers E, McGray AJ, Ware JS, Bramson JL, Klinke DJ 2nd. A quantitative systems approach to identify paracrine mechanisms that locally suppress immune response to Interleukin-12 in the B16 melanoma model. Integr Biol (Camb). 2012 Aug;4(8):925-36.
[2] Klinke DJ 2nd, Cheng N, Chambers E. Quantifying crosstalk among interferon-γ, interleukin-12, and tumor necrosis factor signaling pathways within a TH1 cell model. Sci Signal. 2012 Apr 17;5(220):ra32.
[3] Garris CS, Arlauckas SP, Kohler RH, Trefny MP, Garren S, Piot C, Engblom C, Pfirschke C, Siwicki M, Gungabeesoon J, Freeman GJ, Warren SE, Ong S, Browning E, Twitty CG, Pierce RH, Le MH, Algazi AP, Daud AI, Pai SI, Zippelius A, Weissleder R, Pittet MJ. Successful Anti-PD-1 Cancer Immunotherapy Requires T Cell-Dendritic Cell Crosstalk Involving the Cytokines IFN-γ and IL–12. Immunity. 2018 Dec 18;49(6):1148-1161.e7.
[4] Pirkey AC, Deng W, Norman D, Razazan A, Klinke DJ 2nd. Head-to-head comparison of CCN4, DNMT3A, PTPN11, and SPARC as suppressors of anti-tumor immunity bioRxiv 2022.04.01.486749.
[5] Klinke DJ 2nd. Induction of Wnt-inducible signaling protein-1 correlates with invasive breast cancer oncogenesis and reduced type 1 cell-mediated cytotoxic immunity: a retrospective study. PLoS Comput Biol. 2014 Jan;10(1):e1003409.
[6] Deng W, Fernandez A, McLaughlin SL, Klinke DJ 2nd. WNT1-inducible signaling pathway protein 1 (WISP1/CCN4) stimulates melanoma invasion and metastasis by promoting the epithelial-mesenchymal transition. J Biol Chem. 2019 Apr 5;294(14):5261-5280.
[7] Deng W, Fernandez A, McLaughlin SL, Klinke DJ 2nd. Cell Communication Network Factor 4 (CCN4/WISP1) Shifts Melanoma Cells from a Fragile Proliferative State to a Resilient Metastatic State. Cell Mol Bioeng. 2019 Oct 17;13(1):45-60.
[8] Fernandez A, Deng W, McLaughlin SL, Pirkey AC, Rellick SL, Razazan A, Klinke DJ 2nd. Cell Communication Network factor 4 promotes tumor-induced immunosuppression in melanoma. EMBO Rep. 2022 Apr 5;23(4):e54127.
[9] Klinke DJ 2nd, Horvath N, Cuppett V, Wu Y, Deng W, Kanj R. Interlocked positive and negative feedback network motifs regulate β-catenin activity in the adherens junction pathway. Mol Biol Cell. 2015 Nov 5;26(22):4135-48.
[10] Klinke DJ 2nd, Fernandez A, Deng W, Razazan A, Latifizadeh H, Pirkey AC. Data-driven learning how oncogenic gene expression locally alters heterocellular networks. Nat Commun. 2022 Apr 13;13(1):1986.
DK Design Blog 