The nature of the tumor microenvironment (TME) influences the ability of tumor-specific T cells to control tumor growth. In this study, we performed an unbiased comparison of the TME of regulatory T-cell (Treg)–replete and Treg-depleted carcinogen-induced tumors, including Treg-depleted responding (regressing) and non-responding (growing) tumors. This analysis revealed an inverse relationship between extracellular matrix (ECM) and T-cell infiltrates where responding tumors were T-cell rich and ECM poor, whereas the converse was observed in non-responder tumors. For this reason, we hypothesized that the ECM acted as a barrier to successful T-cell infiltration and tumor rejection. However, further experiments revealed that this was not the case but instead showed that an effective T-cell response dramatically altered the density of ECM in the TME. Along with loss of ECM and high numbers of infiltrating T cells, responder tumors were distinguished by the development of lymphatic and blood vessel networks with specialized immune function. ECM-rich tumors exhibited a stem cell–like gene expression profile and superior tumor-initiating capacity, whereas such features were absent in responder tumors. Overall, these findings define an extended role for an effective immune response, not just in direct killing of tumor cells but in widescale remodeling of the TME to favor loss of ECM, elimination of cancer stem cells, and propagation of adaptive immunity.
Variable responses to cancer immunotherapies are well documented among and within different cancer types (1, 2). Although the extent of T-cell activation is crucial for treatment success, there is evidence that other features of the tumor microenvironment (TME) act as barriers or enablers of immune attack. The type of stromal cells present, the nature of blood vessels, the composition of the extracellular matrix (ECM), and tumor-intrinsic gene expression pathways may all influence tumor growth directly as well as affect the ability of T cells to infiltrate and/or function effectively (3–8). Conversely, other anatomic features are associated with productive immune responses, namely the presence of specialized vasculature termed high endothelial venules (HEV), which can be induced by immunotherapy and are often associated with development of tertiary lymphoid structures (9).
It remains unclear the extent to which the preexisting TME prevents development of an effective immune response or whether an effective immune response alters the TME. In support of the latter, using a mouse model of carcinogen-induced fibrosarcomas (3, 4), we found that targeting Foxp3+ regulatory T cells (Treg) results in significant control of tumor growth in approximately half of tumor-bearing mice. These mice, termed “responders”, could be distinguished from “nonresponder” mice by presence of intratumoral HEV and a significantly higher number of tumor-infiltrating T cells (TIL; refs. 3, 4). HEVs were not identified in untreated mice, implying that development of HEV was a consequence of successful immunotherapy. Other studies using adoptively transferred T cells or PD-1 blockade have similarly shown that activated T cells can alter tumor blood vessels in a manner that promotes tumor immunity (5, 10).
To explore the concept that T cells modulate the TME beyond tumor blood vessels, we have used a variety of tools to identify features distinguishing the TME of responding and non-responding tumors and their relationship to the TME of untreated tumors. Specifically, we questioned whether the TME dictated responses to immunotherapy or whether the nature of the response to immunotherapy overcame the suppressive conditions of the TME. Overall, our data indicate that a successful response to immunotherapy drives events that ultimately promote tumor regression through removal of cancer stem cells (CSC), TME remodeling and amplification of the adaptive immune response.
Materials and Methods
Mice, induction of tumors, and depletion of Tregs
B6.129(Cg)-Foxp3tm3(DTR/GFP)Ayr/J mice used in this study were gratefully received from Professor Alexander Rudensky (11). Female Balb/c mice were purchased from Charles River Laboratories. Mice were housed in accordance with UK Home Office regulations, and isolator-bred before being housed in filter-top cages for the experiments. Fibrosarcomas were induced in 8- to 15-week-old Foxp3DTR mice by subcutaneous injection with 400 μg of 3-methylcholanthrene carcinogen (MCA) suspended in 100 μL of olive oil into the left hind leg (12). Mice were monitored for up to 18 weeks. Tumor-bearing mice were sacrificed before tumors reached 1.5 cm in diameter (measured with calipers), or if tumors caused discomfort. Upon dissection, tumors were resected avoiding local muscle, other normal tissues and the local popliteal lymph node. Tumor measurements and growth rate calculations were performed as described previously (4). To ensure complete depletion of Tregs, doses of diphtheria toxin (DT) were selected based on dose-optimization performed in a previous study (3). DT, diluted in sterile PBS, was administered by intraperitoneal injection at a dose of 15 μg/kg body weight every other day, after development of a palpable tumor. Tumors were variable sizes at the start of treatment; hence, growth rates and not absolute tumor sizes were compared to evaluate treatment response.
The tumor cell line 4T1, obtained from ATCC in 2014 (CRL-2539, not authenticated in the past year), was cultured for 10 to 14 days before being injected subcutaneously into the mammary fat pad of Balb/c mice. These cells were regularly tested and found to be negative for Mycoplasma infection. A total of 105 cells were injected subcutaneously into the mammary fat pad of Balb/c mice. PI-3065, a small-molecule inhibitor of the PI3K delta subunit provided by Genentech, was administered by oral gavage at a dose of 75 mg/kg, vehicle-treated mice were given an equivalent volume of carrier solution. Mice were dosed daily from day −1 prior to tumor inoculation until experiment termination. Animal studies were conducted with approval of the UK Home Office.
Immunofluorescence staining and microscopy
Five-micron sections of frozen tissue embedded in OCT (KMA-0100–00A, CellPath) were fixed in ice-cold acetone, and nonspecific binding blocked with 2.5% normal horse serum (S-2012–50, VectorLabs). Sections were incubated in primary antibody overnight at 4°C, washed with PBS, incubated in secondary antibody, washed again, counterstained with Hoechst (14533, Sigma), and then mounted in ProLong Gold Antifade mountant (P36930, Thermo Fisher Scientific). Primary antibodies were used as follows: anti-TNC (ab108930, Abcam) 1 μg/mL, anti-CD3 (GA50361–2, Agilent) 2 μg/mL, anti-GZMB (AF1865, R&D Systems) 0.4 μg/mL, anti-CD8 (42–0081–82, Thermo Fisher Scientific) 1 mg/mL, anti-Lumican (AF2745, R&D Systems) 0.4 μg/mL, anti-LYVE-1 (14–0443–82, Thermo Fisher Scientific), anti-CCL21 (AF457, R&D Systems) 5 μg/mL, anti-ZEB1 (HPAO27524, Atlas Antibodies) 0.2 μg/mL, anti-CD206 (AF2535, R&D Systems) 0.3 μg/mL, and anti-MHC class II (NBP1–43312, Novus Biologicals) 20 μg/mL. Sections were imaged using a Zeiss Axioscan.Z1 Slide Scanner and analyzed using Zen software Blue edition. Immune cells were counted per high power field of view, and an average of 10 fields of view was calculated per section.
Images were analyzed for TNC using QuPath software version 0.2.0 milestone 11 (13), quantified using the pixel classifier tool, and results shown in percentage of positive TNC pixels per mm2 of tissue.
TNC mRNA in situ hybridization was measured with RNAscope assay, developed by Advanced Cell Diagnostics (ACD) according to the manufacturer's protocols. Briefly, tumor sections were deparaffinized, incubated with hydrogen peroxide at room temperature for 10 minutes, boiled with target retrieval solution for 15 minutes, and then treated with protease plus at 40°C for 30 minutes. Sections were hybridized with Mm-Tnc probe (465021, ACD), Mm-Ppib probe as positive control (313911, ACD) and DapB probe as negative control (310043, ACD) at 40°C for 2 hours. Hybridization signals were amplified with RNAscope 2.0 HD Reagent Kit-Red (322350, ACD). Sections were imaged using a Zeiss Axioscan.Z1 Slide Scanner and higher power images were captured using a Zeiss Apotome Axio Observer.
Cell line establishment
To establish cell lines derived from MCA-induced tumors, two untreated tumors were dissected, chopped into small pieces, and incubated in a cell culture flask with R10 media [RPMI media (Gibco) supplemented with 10% FCS (Gibco), l-glutamine, and penicillin/streptomycin] at 37°C and 5% CO2. The following day, pieces of tissue were removed, and media were replaced, to exclude cells that did not attach to the flask. Cells were split up to 3 times a week and, once these were expanded, cells were frozen at −80°C, in single vials of 107 cells/mL of freezing media [FCS with 10% DMSO (Sigma)]. Cell lines (MCA-GG1 and MCA-GG2) were grown from each tumor and were left to grow for 3 to 6 passages before being injected into mice.
Lentivirus production and tumor cell injection
Oligonucleotide pairs for shTNC (CACCGCCCTTGGCTGAAATTGATAGCGAACTATCAATTTCAGCCAAGGGC and AAAAGCCCTTGGCTGAAATTGATAGTTCGCTATCAATTTCAGCCAAGGGC) were ligated into pENTR/U6 Gateway system entry vector (K4944–00 and K4945–00, Invitrogen) according to the manufacturer's instructions. A scrambled plasmid was generated as control. Hairpin sequences were verified by transduction of the virus on test cells and knock down levels measured by qPCR. Sequences were then transferred and viral supernatants were generated as described (14). MCA-GG2 cells were resuspended at 106 cells/mL in viral supernatant and plated at 1 mL/well in ultralow attachment 24-well plates. After 16 hours, cells were washed and replated in RPMI1640 medium (Gibco, Thermo Fisher Scientific) supplemented with 10% FBS (Gibco, Thermo Fisher Scientific), and l-glutamine and penicillin/streptomycin. After 48 hours, cells were transferred to normal tissue culture plastic and maintained in growth medium for one week. After this time, cells were trypsinized and flow sorted to isolate GFP+ cells for qPCR analysis. Mice aged 8 to 15 weeks were anesthetized and injected subcutaneously with 0.5 × 106 control or knockdown cells into the left hind leg. Mice were monitored, and tumors were measured as described above.
Quantitative real-time PCR
A sample of each cell line was collected, and cells were disrupted with 500 μL of RLT buffer. RNA was purified with RNeasy Mini Kits (74104, Qiagen), according to the manufacturer's instructions, and quantified using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific). DNase treatment and cDNA synthesis were performed using High-Capacity RNA-to-cDNA Kit (4387406, Applied Biosystems), according to the manufacturer's instructions. qRT-PCR reaction was performed in a final volume of 20 μL, containing 2 μL of cDNA sample, 10 μL of Gene Expression TaqMan Master Mix (4369514, Applied Biosystems), and 1 μL of TaqMan Gene Expression Assay probes. Probes used were mouse β-actin (4352341E, Applied Biosystems), as housekeeping gene, and Tnc (Mm00495662_m1, Applied Biosystems) as target gene. Reactions were performed in triplicate on an Applied Biosystems QuantStudio 3 Real Time Machine. Relative expression of mRNA was calculated using the ΔΔCt method.
Bioinformatic analyses—mouse data
Microarray data was previously published (4). Probe intensity values were corrected by background subtraction using Genome Studio software and subsequently log2 and baseline (median) transformed using Genespring software (Agilent). Differential expression analysis was performed using the R package “Limma” (15), only probes with the highest baseline average were used in heatmaps and later analyses. The “parallelDist” R package was used to perform Manhattan distance analysis of all genes. Heatmaps were created using the pheatmap R package (hierarchical clustering method: ward.D2 using Manhattan distance). All pathway analysis was carried out using the “ReactomePA” R package (16) or gene sets downloaded from the Reactome database with conversion of gene IDs to mouse nomenclature using the “biomaRt” R package (17). Code is available at https://github.com/ALGW71/ImmuneRemodellingECM. Raw microarray data are available at ArrayExpress accession E-MTAB-9351.
CIBERSORT (18) analysis was performed on the prenormalized microarray data using the immune deconvolution package in R (https://github.com/icbi-lab/immunedeconv; ref. 19). The CIBERSORT binary for use in R was obtained from the website: https://cibersort.stanford.edu/ following registration and approval of access (16th April 2020).
Bioinformatic analyses—human data
Level 3 (raw counts, htseq.counts.gz) RNA sequencing (RNA-seq) data, and sample meta data (GDC sample sheet and clinical cart files) were download from The Cancer Genome Atlas (TCGA) GDC portal (https://portal.gdc.cancer.gov/) on October 3, 2019, for 21 cancer types. All datasets were normalized as one matrix within the “DESeq2” Bioconductor package (20) in R and the DESeq2 normalized counts were used for all analyses.
Only data from primary tumors was used for analysis. Patient information (GDC sample sheet - vital status, days to death, days to last follow-up) was used for Kaplan–Meier survival analysis (R packages: “survminer” and “survival”). Data were censored at days-to-last-follow-up or days-to-death and log-rank P values calculated between two curves. Medoids clustering was performed using the “ClusterR” package. Clustering was carried out on expression data, center and scaled by each gene, using the Manhattan distance metric to obtain two clusters. RNA-seq data from distinct cancers were not combined to perform clustering; this ensured gene expression levels from different cancer types were not unfairly compared.
To cluster on cytotoxic T lymphocytes (CTL) alone, a meta gene signature from (21) was used (CTL genes – CD3G, CD3E, CD3D, PTPRC, CD8A; cytotoxicity genes – PRF1, GZMH, GNLY, GZMB, GZMK, GZMA) in each cancer type. Sixteen cancers in which CTL signature was prognostic of survival, based on visual inspection of survival curves, were taken forward for TNC clustering. CTL-high and CTL-low clusters were separated, and a second round of medoids clustering was performed based on expression of TNC resulting in four groups. Several R packages from the “Tidyverse” were used for data wrangling and visualization (22). Code is available at https://github.com/ALGW71/ImmuneRemodellingECM.
Tumorsphere formation assay
Live single cells obtained from untreated, non-responder, and responding tumors were seeded at 104 cells/mL in 96-well ultra-low attachment plates (3474, Corning) and cultured in DMEM/F12 (Thermo Fisher Scientific) with N2 (17502048, Thermo Fisher Scientific), 10 ng/mL EGF (PMG8043, Thermo Fisher Scientific), and 10 ng/mL FGF (PMG0033, Thermo Fisher Scientific). After 7 days, spheres were counted under an inverted phase microscope (Nikon) and passaged. After 14 days, spheres were counted again, under the same conditions.
The statistical difference between pairs of groups was assessed dependent on the normality of the data, through either nonparametric Kruskal–Wallis with Dunn's multiple comparison test or one-way ANOVA with Tukey posttest. A P value <0.05 was considered statistically significant. Statistical analyses were performed using GraphPad Prism software, version 8.4.3.
Expression patterns of T cells and TNC in responders or non-responder tumors
Using Foxp3DTR mouse model to deplete Tregs in the presence of carcinogen-induced fibrosarcomas, we report that the antitumor response following Treg depletion is variable (3, 4). The mouse group defined as “non-responders” have tumors without HEVs and have no significant decrease in tumor growth rates or increase in TILs after Treg depletion compared with untreated mice. Conversely, “responders” have HEVs in their tumors and overall show a significant decrease in tumor growth rate and a significant increase in TILs after Treg depletion (summarized in Fig. 1A and refs. 3 and 4). In this model, both neogenesis of HEV and control of tumor growth is strictly T-cell dependent (4). In these previous studies, the extent of Treg depletion is comparable in all treated mice, ruling out a differential effect of the DT in individual animals (3).
To further examine differences between the tumor types, we compared the transcriptomes of untreated, responder and non-responder tumors. In this analysis, we failed to find any genes that were differentially expressed (statistical test, Padj less than 0.05) between non-responder and untreated tumors (Fig. 1B), while 20 genes were significantly higher in non-responder versus responder tumors (Fig. 1C), of which the gene encoding the ECM protein, tenascin-C (Tnc) was most differentially expressed (Padj = 0.025, logFC = 2.365). TNC, a protein known to have important roles in embryonic development and wound healing, is thought to impede T-cell function and is often associated with poor cancer prognosis, local recurrence, and metastasis (23–26).
When the transcriptomes of responder and non-responder tumors were compared, 74 genes emerged as having significantly higher expression in responder versus non-responder tumors (Fig. 1D). These revealed a significant enrichment of immune pathways (Supplementary Fig. S1A). Cd6 (Padj = 0.019, logFC = −2.189), a T-cell adhesion gene, and Zap70 (0.019, −1.85), Cd3d (0.019, −2.38), Cd3g (0.030, −2.28), and Cd3e (0.020, −2.20), key mediators of TCR signaling, were among the most significant differentially expressed genes (Fig. 1D). Although there was a clear T-cell immune signature present in the differentially expressed genes of responder mice (Supplementary Fig. S1A), no pathways were significantly enriched in the non-responder gene list in which Tnc was most highly expressed. Analysis of gene expression patterns relative to Tnc using a Manhattan distance metric (Supplementary Fig. S2) indicated that Tnc and Zap70 consistently had opposing expression patterns across all samples with the largest Manhattan distance value (21.08).
TNC protein expression was next confirmed by immunofluorescence staining. We found that TNC was more abundant at tumor periphery, presenting a well-organized fibrillar pattern compared to the tumor center (Supplementary Fig. S3A). TNC was abundant in non-responder tumors compared with responders and variable in untreated tumors (Fig. 1E; Supplementary Fig. S3B). RNAscope revealed that Tnc transcripts were clearly more abundant in untreated and non-responder tumors compared with responders (Fig. 1F) and were observed in tumor cells (Fig. 1G). The inverse relationship between TNC and the level of T-cell infiltrate was striking (Fig. 1E) because responding tumors had high numbers of TILs, indicating that successful antitumor immunity was strongly associated with absence of TNC.
TNC expression associated with a CTL signature impacts patient survival
T-cell infiltration of tumors (particularly CTLs; 27) and presence of TNC (25, 28) demonstrate opposing correlations with survival in human cancers. However, aside from the data presented above, evidence of an antagonistic relationship between T cells and TNC has solely been demonstrated in vitro (28). We investigated the relationship between CTL gene signatures and Tnc expression in the context of survival, specifically searching for trends analogous to those revealed by the mouse data. For this purpose, we examined RNA-seq datasets from primary tumors in TCGA, to determine whether there was a link between a favorable CTL gene signature and TNC expression.
We identified 14 cancers where survival was improved with a CTL gene signature (21) in patients' primary tumors (6 statistically significant; Supplementary Fig. S4A). This was performed using an unbiased medoids clustering approach where we first clustered the datasets based on expression of CTL genes (Supplementary Fig. S4A), followed by expression of TNC within each individual cancer (Supplementary Fig. S4B and S4C). This approach ensured that we were comparing expression of CTL genes or TNC within and not across different cancer types (see Materials and Methods). We used this analysis to generate groups of CTL-high and CTL-low tumors (Fig. 2A), which were then separately clustered by TNC expression (Fig. 2B and C), revealing that although TNC expression had no impact on survival within CTL-low tumors (P = 0.48; Fig. 2C), low TNC expression was associated with a highly significant increase in survival within CTL-high tumors (P = 0.00062; Fig. 2B). When we reversed the analysis, clustering TNC high/low first followed by CTL high/low, the same effect was apparent (Supplementary Fig. S5). These data were similar to those obtained in the mouse model above as they clearly indicated that a successful T-cell response was associated with low TNC expression. The key question for the study described herein was whether TNC impinged on CTL activity in patients and mice resulting in poorer survival/outcome, or whether an effective immune response resulted in loss of TNC.
TNC did not impinge on responsiveness to Treg depletion
Previous reports have suggested that TNC can affect T-cell function through multiple mechanisms (24, 29–32). With these in mind and our own mouse data indicating an association between control of tumor growth and loss of TNC (Fig. 1) corroborated by an analysis of TCGA data for multiple cancers (Fig. 2), we hypothesized that TNC prevented T cells from successfully controlling cancer. In the case of the fibrosarcoma mouse model described here, we postulated that high TNC expression impeded development of a successful T-cell response generated after Treg depletion.
We first examined the relationship between TNC and the intratumoral distribution of T cells by identifying and enumerating T cells within high power images of tumor areas with high and low intensity of TNC staining (Supplementary Fig. S6). These comparisons were only possible in untreated and non-responder tumors, due to lack of TNC in responder tumors. Findings indicated that T cells were found in both areas and were even slightly more frequent in TNC high areas (Fig. 3A and B). Granzyme-B (GZMB) expression revealed significantly more GZMB+CD8+ T cells in TNC-high compared with TNC-low areas of non-responder tumors, suggesting that high expression of TNC might be associated with less degranulation of T cells in poorly controlled tumors (Fig. 3C and D).
Given the well-documented role of macrophages in remodeling the TME, we analyzed our microarray data using CIBERSORT to ascertain whether differences in the M0/M1/M2 phenotypes in responders and non-responders are apparent. This showed a strong M0 signature in untreated mice, but no significant differences between M1/M2 phenotypes between responders and non-responders, with a large variation between individual animals (Supplementary Fig. S7A). Tumor sections from non-responder and responder tumors were costained for TNC and CD206 or MHC class II. In agreement with the CIBERSORT findings, these data indicated no significant difference when comparing CD206+ and MHC-II+ cells between the different groups, or between TNC high and low areas (Supplementary Fig. S7B and S7C). Thus, we concluded that the loss of TNC was mainly associated with a robust T-cell response.
To definitively address whether TNC impinges on a successful T-cell response, we generated fibrosarcoma cell-lines from the MCA-induced tumors by in vitro culture of dissociated cells. These cell lines were genetically modified using TNC-targeted shRNA (shTNC 3559) or control scrambled shRNA (shScr) and then injected into Foxp3DTR mice (Fig. 4A). Tumor growth was monitored, and Tregs depleted once tumors were palpable. Although Treg depletion resulted in a reduction in tumor growth rate in both shScr and shTNC tumors (Fig. 4B; Supplementary Fig. S8), TNC expression did not affect tumor growth rate in either Treg-depleted or untreated tumors (Fig. 4B). It was striking that similar to the carcinogen-induced tumors, loss of TNC expression was observed in tumors responding to Treg depletion (Fig. 4C). These findings indicate that although TNC may have impeded T-cell responses, this effect could be overcome after Treg depletion. Moreover, the data pointed to loss of TNC expression being a consequence of an effective immune response.
We used a second, distinct mouse model of Treg-targeted immunotherapy to address whether TNC expression similarly alters as a response to successful immunotherapy. For this purpose, we used a model of triple-negative breast cancer, namely the 4T1 cell line, which recapitulates features of aggressive breast cancer in Balb/c mice. Mice were inoculated with 4T1 cells and treated with PI-3065 previously shown to inactivate Tregs and promote control of tumor growth in treated mice (33). We selected untreated tumors, responder and non-responder treated tumors to compare TNC expression. A similar distribution was observed to the fibrosarcoma model whereby control of tumor growth was associated with a loss of TNC (Fig. 4D).
Alterations in TNC were indicative of ECM changes in responder tumors
The data described above indicated that although TNC did not inhibit development of a successful T-cell response, loss of TNC was a consequence of a successful T-cell response. We next wished to determine whether loss of TNC was indicative of other TME alterations. We first looked at genes with a less stringent Padj < 0.10. When pathway analysis was applied to this list of 123 genes, the pathway “Extracellular Matrix Organization” was the only significantly enriched pathway (Supplementary Fig. S1B; 12 genes, Padj = 2.285e−05). An opposing pattern of expression of genes encoding these proteins was observed in non-responder versus responder tumors (Fig. 5A).
These data indicated that although focusing on TNC enabled us to evaluate the relationship between ECM, T cells, and tumor growth, it was clear that TNC was not the only ECM protein overrepresented in non-responder versus responder tumors. To examine this further, untreated, non-responder and responder tumors were stained with Lumican- and Collagen-specific antibodies (Fig. 5B and C); the staining pattern was as observed for TNC thereby pointing to widescale alterations in the ECM composition in responder versus non-responder tumors.
Previous studies indicate that ECM-rich tumors are served by poor lymphatic networks, due to compression of the lymphatic vessels (33). To determine whether this was true for the tumors here, sections from untreated, non-responder, and responder tumors were stained with LYVE-1- and CCL21-specific antibodies. The difference in the appearance of lymphatic vessels was striking (Fig. 5D). Robust staining was observed only in responder tumors, where lymphatics were characterized by the presence of CCL21 around LYVE-1–positive vessels entirely compatible with improved lymphatic function and potentially with an enhanced ability to guide dendritic cells (DC) to lymphatic vessels (34). Overall, these findings point to widescale alterations to the TME characterized by a widescale loss of ECM and a reciprocal gain in immune function.
Cancer cell gene alterations in responder versus non-responder tumors
As part of the bioinformatic analysis, we mapped the expression pattern of Tnc against all other genes and samples in the microarray analysis using a Manhattan distance metric. In this way, we could rank genes according to how similar or opposing/inverse their expression pattern was against Tnc. As already mentioned, this revealed that Zap70 had the furthest distance (opposing expression pattern) from Tnc. In contrast, those with a similar pattern to Tnc included several cancer-associated (Vegfa, Trib3, Uck2) and stem cell/EMT genes (Ttc3 and Twist2). Given that dense ECM associate with the emergence of aggressive cancer cells with mesenchymal stem-like properties (35, 36) and that TNC has itself been reported as a potential marker of CSCs (37–39), these bioinformatic findings led us to investigate and compare expression of CSC-associated genes (40) in responder and non-responder tumors. The data indicated that CSC genes are overrepresented in non-responder compared with responder tumors (Fig. 6A and B). When tumors were stained for ZEB1, high ZEB1 expression was seen in control and non-responder compared with a dramatic decrease in responder tumors (Fig. 6C). This finding indicated that cancer cells in untreated and non-responder tumors were more stem-like, a finding that was in line with their more rapid in vivo growth compared with responder tumors.
We were consistently successful in generating tumor cell lines from untreated and non-responder mice but not from responder mice, suggesting an intrinsic failure of cells from the latter tumors to self-renew. However, to confirm that CSCs were more abundant in non-responder compared with responder tumors, it was necessary to examine the ability of cells isolated from both tumor types to self-renew. A tumorsphere formation assay was performed by seeding equal numbers of live disaggregated tumor cells from untreated, non-responders, and responder tumors in serum-free, nonadherent conditions (41). With this assay, it is possible to determine the presence of CSCs or progenitor cells, because these are the only cell types capable of surviving and proliferating under such conditions (41). Tumorspheres with a diameter >50 μm were counted following two passages. Data showed that tumorspheres readily formed from untreated and non-responder tumors but not in responders (Fig. 6D and E). These findings indicated that a successful antitumor T-cell response associated with a loss of CSCs. Overall, the data presented in this study showed that a sufficiently robust T-cell response induced multiple changes to the TME, which served to drive elimination of tumor cells while concurrently amplifying the immune response.
In this study, using a mouse model of carcinogen-induced tumors, we detected a relationship between the ECM protein, TNC, and a poor overall response to immunotherapy.
Since there are several reports describing an effect of the ECM on immune cell behavior (7, 26, 30, 42) including a negative effect of TNC on T-cell function and motility (7, 43), we questioned whether TNC hampered the ability of T cells, stimulated following Treg depletion, to effectively access and kill tumor cells (28, 43, 44). Our findings disproved this hypothesis as TNC expression before Treg depletion did not predict response to treatment; indeed, depletion of Tregs resulted in successful elimination of both TNC-high and TNC-low tumor cell lines. We did however find that a robust T-cell response associated with a loss of TNC and other ECM alterations. Our data do not rule out the involvement of other immune cells in driving changes to the ECM. It has been documented that tumor-associated macrophages play a role in altering TNC expression during tumor progression. Deligne and colleagues indicate that CD206+ M2-like macrophages are enriched in TNC-rich tumors and associate with poorer control of tumor growth (45). Whereas ECM deposition in the TME can be influenced by macrophages (46), we have no evidence that this is the case in Treg-depleted, carcinogen-induced fibrosarcomas.
Along with ECM genes, we observed that some stem cell associated and EMT genes were downregulated in responder compared with untreated and non-responder tumors, indicating that an effective immune response associated with a less aggressive cancer cell genotype. Whether the decrease in TNC and CSC gene expression was due to loss of CSCs as a result of immune destruction or whether a strong immune response drives loss of ECM, which then drives changes in cancer cell gene expression patterns is yet to be determined. The latter hypothesis is supported by the demonstration that ECM proteins contribute to tumor progression through increasing tumor tension leading to biomechanical signaling cues that drive expression of CSC/EMT genes (36). In addition, previous studies show that cognate, nonlytic interactions between T cells and cancer cells can promote expression of stem-like genes by cancer cells (47). It is possible that in our model, non-responder tumors may have become more stem-like and aggressive by avoiding lysis by cognate T cells.
The tumor-promoting effects of ECM-mediated corruption of tissue tension are also linked to poorly developed lymphatic vessels, which serve to further increase intratumoral tensile forces (30, 35, 36). We found a striking difference in the appearance of lymphatic vessels in responder versus non-responder tumors in terms of organization, density, and CCL21 expression. These findings supported a scenario whereby loss of ECM associated with improved lymphatic function. This may both relieve intratumoral interstitial pressure and improve T-cell response, through guiding migration of antigen-presenting CCR7+ DCs to draining lymph nodes for priming T cells (34, 48), further facilitating a switch from a predominantly immunosuppressive environment to a predominantly immunostimulatory one.
A picture is emerging that describes the ability of the immune system to initiate and direct remodeling of the TME in a manner that serves to amplify its antitumor activities. Despite the extensive genotypic and phenotypic comparison of responder and non-responder tumors, the mechanisms behind this split remain unknown. Methylcholanthrene is highly mutagenic making it possible that individual tumors exhibit different frequencies of neoantigens that define their immunogenicity. In this scenario, responder tumors may have many neoantigens, while non-responders present fewer neoantigens (49, 50). This hypothesis is supported by work that indicates that the selective expansion of certain T-cell receptors in defined tumor regions correlates with the number of nonsynonymous mutations at the same sites (51). Our findings indicate that if a sufficiently robust T-cell response is induced by immunotherapy, the balance is tipped whereby the ECM is lost, aiding lymphatic vessel function, HEV development and a self-amplifying immune response. This was accompanied by a reversal in CSC-associated gene expression and an intrinsic loss of cancer cell aggression. We propose that the simultaneous occurrence of these events is essential for creating the “perfect storm” required for successful antitumor immunity. This could be enhanced by empowering adoptively transferred T cells with the ability to digest the ECM and/or to improve endothelial cell function, or by combining cancer immunotherapies with antifibrotic treatments or treatments designed to improve the tumor vasculature. By altering the TME in this way, it may be possible to tip the balance in favor of the “perfect storm” even when the T-cell response is suboptimal.
Disclosure of Potential Conflicts of Interest
A. Pires reports grants from Cancer Research UK, Cancer Research Wales, Breast Cancer Now, and The Wellcome Trust and nonfinancial support from Genentech during the conduct of the study. M. Somerville reports grants from Cancer Research Wales during the conduct of the study. J.P. Hindley reports other from Indoor Biotechnologies (employee and director of Indoor Biotechnologies) outside the submitted work. R. French reports grants from Cancer Research UK, Cancer Research Wales, Breast Cancer Now, and The Wellcome Trust and nonfinancial support from Genentech during the conduct of the study. W.J. Watkins reports grants from Cancer Research UK, Cancer Research Wales, Breast Cancer Now, and The Wellcome Trust and nonfinancial support from Genentech during the conduct of the study. A. Gallimore reports grants from Cancer Research UK, Cancer Research Wales, Breast Cancer Now, and The Wellcome Trust and nonfinancial support from Genentech during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.
A. Pires: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. A. Greenshields-Watson: Data curation, formal analysis, methodology, writing–original draft, writing–review and editing. E. Jones: Conceptualization, data curation, investigation, methodology. K. Smart: Resources, methodology, project administration. S.N. Lauder: Resources, investigation, methodology. M. Somerville: Resources, validation, methodology. S. Milutinovic: Data curation, software. H. Kendrick: Resources, investigation, methodology. J.P. Hindley: Data curation, methodology. R. French: Resources, methodology. M.J. Smalley: Resources, methodology. W.J. Watkins: Software, formal analysis. R. Andrews: Software, formal analysis. A. Godkin: Conceptualization, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing. A. Gallimore: Conceptualization, resources, supervision, funding acquisition, visualization, writing–original draft, project administration, writing–review and editing.
This work was supported by a program grant from Cancer Research UK (C16731/A21200), Cancer Research Wales (5272BR), an MRC-funded studentship (to A. Pires), a Breast Cancer Now project grant (PR1377), and a Wellcome Trust collaborator award (grant 209213/Z/17/Z). The authors are grateful to the staff at Cardiff University JBIOS for their continued support and Genentech for provision of PI-3065.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.