Oxidative stress and serum deprivation influence the evolution of newly formed tetraploid cells during tumorigenesis

Significance

Understanding the mechanism that links whole genome doubling (WGD), extra centrosomes, and tumorigenesis to one another is crucial to understanding tumor progression, as many human tumors have a high incidence of WGD and extra centrosomes. Using in vivo and in vitro approaches, we find that newly formed tetraploid (4N) cells have a proliferative disadvantage, due to low nutrient availability. However, a small population of 4N cells is sufficient to promote the recruitment of protumorigenic stromal cells. Moreover, 4N cells are more likely to undergo centrosome overduplication in vivo, which could be driven by an increase in reactive oxygen species common in the tumor microenvironment (TME). Taken together, our work expands how WGD, extra centrosomes, and the TME can promote tumorigenesis.

Abstract

We recently found that newly formed tetraploid (4N) cells in culture quickly lose extra centrosomes after whole genome doubling (WGD). This is inconsistent with the high incidence of centrosome number abnormalities in human cancers and with the observation that 4N cells from mouse tumors carry extra centrosomes, suggesting that centrosome numbers could be affected by certain conditions in the tumor microenvironment (TME). To determine the effect of the TME on the evolution of newly formed 4N cells, we induced WGD in DLD1 colorectal cancer cells and analyzed centrosome and chromosome numbers in mouse tumor samples. We found that the 4N cells displayed a proliferation defect in vivo, that they could enhance the recruitment of stromal cells to the tumor site, and that they were more likely to harbor extra centrosomes compared to 4N cell populations evolved in vitro. Combining a mathematical model that tracks the coevolution of ploidy and centrosome numbers in different cell populations with Bayesian inference, we identified centrosome overduplication as the mechanism underlying the supernumerary centrosome phenotype. Finally, through in vitro evolution experiments, we found that deprivation of growth factors and oxidative stress could explain, respectively, the proliferation defect and the supernumerary centrosomes identified in our in vivo experiments. Overall, our work shows that oxidative stress plays a major role in centrosome overduplication, particularly in 4N cells, suggesting that supernumerary centrosomes and WGD may coexist in certain tumors. Moreover, our findings suggest that the oncogenic effects of WGD could be due, in part, to stromal cell recruitment.

Tetraploidy via whole genome doubling (WGD) occurs in 30 to 40% of all human tumors and is associated with cancer progression and poor prognosis (1–5). Tetraploid (4N) cells are found in premalignant lesions of the cervix and esophagus (6, 7), suggesting that WGD promotes tumorigenesis. Consistent with this, other studies have shown that WGD is usually a clonal (or truncal) event in many breast, colorectal, and lung tumors, where it occurs at different stages of cancer evolution (8–11). In some cases, WGD is only detected in a small fraction of cancer cells (i.e., subclonal), such that tumors harbor mixed populations of WGD-negative and WGD-positive cells (12, 13). WGD contributes to cancer progression by increasing both the incidence and tolerance of genomic alterations (2, 14), allowing WGD-positive cells to accumulate more genetic and functional diversity than WGD-negative cells (4, 5, 15).

WGD leads to doubling of not only the genome but, in many cases, also the number of cellular organelles, including the centrosomes. Accurate cell division is largely dependent upon a cell having two centrosomes during mitosis, ensuring equal chromosome distribution into two daughter cells (16, 17). However, as a result of WGD, newly formed 4N cells inherit an extra centrosome. Centrosome duplication in the ensuing cell cycle will lead to newly formed 4N cells entering mitosis with four centrosomes, which can cause spindle multipolarity and chromosome missegregation during cell division (18–20). One study found that a large fraction of 4N cells recovered from mammary tumors in mice were aneuploid (i.e., having an abnormal number of chromosomes) and possessed extra centrosomes (21). This prompted the idea that centrosome amplification, or the presence of extra centrosomes in a cell, may contribute to tetraploidy-driven tumorigenesis (22). In line with this theory, centrosome amplification is common in human tumors and can promote tumor formation in some contexts (23–27).

We recently found that newly formed 4N cells quickly lose extra centrosomes after WGD in standard cell culture conditions (20). This observation is inconsistent with the high incidence of centrosome number abnormalities in human cancers (23, 24, 28) and in 4N cells recovered from mouse tumors (21), suggesting that centrosome number evolution in 4N cells could be affected by certain conditions in the tumor microenvironment (TME) (17). The poor vasculature in the TME can limit the availability of nutrients and the removal of cellular waste, which can lead to acidosis, hypoxia, and oxidative stress (29)–conditions previously shown to induce centrosome amplification in cancer cells (30–33). However, how environmental stresses affect 4N cells after WGD has not been investigated. Therefore, to determine the effect of the TME on the evolution of newly formed 4N cells, we induced WGD in DLD1 colorectal cancer cells and analyzed centrosome and chromosome numbers in mouse xenograft tumors as well as in nonstandard cell culture conditions meant to mimic the TME.

Results

Tetraploid Cells Recruit Stromal Cells and Reduce Apoptosis within the Tumor Microenvironment.

As opposed to their diploid (2N) counterparts, nontransformed tetraploid (4N) cells have been shown to be tumorigenic (21, 34). On the other hand, tetraploidization does not always increase the tumorigenicity of 2N cancer cells (34, 35). Nevertheless, WGD is often observed in human tumors (1–5) and even subclonal WGD can be associated with cancer progression (12), suggesting an important role for WGD during tumorigenesis. To investigate the contribution of newly formed 4N cells to tumorigenesis, we induced cytokinesis failure in DLD1 colorectal cancer cells to produce 4N cells and injected these newly formed 4N cells and the 2N parental cells into opposing flanks of nude mice (Fig. 1 A and B). We found that tumors induced by injecting 2N or newly formed 4N cells (referred to hereafter as 2N tumors and 4N tumors, respectively) grew at similar rates (Fig. 1 B and C). Chromosome spreads (Fig. 1D) prepared from postxenograft samples showed that most cells from the 2N tumors remained 2N (Fig. 1E), but surprisingly, the majority of cells from the 4N tumors were also 2N (Fig. 1F), although the fraction of 4N cells in the 4N tumors was still larger than the fraction of 4N cells in the 2N tumors. This suggests that the small residual fraction (~10%) of 2N cells present at the time of 4N cell injection may have a fitness advantage over the newly formed 4N cells within the in vivo environment.

Fig. 1.

4N cells are outcompeted by 2N cells and harbor extra centrioles in vivo. (A) Diagrammatic representation of experimental methods for subcutaneous injections of 2N DLD1 cells and newly formed 4N cells into opposing flanks of nude mice. After excision, tumor tissue samples and cell suspension were prepared for downstream analyses. (B) Representative images of xenograft tumors from 2N DLD1 cells (Left flank of each mouse) and newly formed 4N DLD1 cells (Right flank of each mouse). (C) Quantification of xenograft tumor growth in nude mice. Tumor size was averaged from 28 mice and reported as mean ± SEM every 3 d. (D) Representative images of near 2N (Left) and near 4N (Right) chromosome spreads. (E) Cell ploidy in the injected 2N cells and in 2N tumors. (F) Cell ploidy in the injected newly formed 4N cells, at day 12 following evolution in vitro (middle), and in postxenograft samples from 4N tumors. (G) Representative images of 4N cells centriole and kinetochore staining in mitotic cell in vitro. Top row depicts a cell with normal centrioles and Bottom row depicts a cell with extra centrioles. (Scale bars, 5 µm.) (H) Quantification of extra centrioles in the injected cells and 2N postxenograft samples. Individual dots correspond to experimental replicates or individual tumors. At least 50 mitotic cells were analyzed per experiment/tumor. (I) Quantification of extra centrioles in the injected cells and in cells (both 2N and 4N) from postxenograft samples of 4N tumors. Individual dots correspond to experimental replicates or individual tumors. At least 50 mitotic cells (between 2N and 4N) were analyzed per experiment/tumor. Error bars in all graphs report SEM. ****P < 0.0001.

The observation that the 2N and 4N tumors grew at similar rates (Fig. 1C) despite a proliferative disadvantage of the 4N cells raises the possibility that the newly formed 4N cells may contribute to creating a microenvironment that enhances the proliferation or survival of the small fraction of 2N cells at the time of injection. Additionally, stromal cell recruitment could be higher in the 4N tumors compared to the 2N tumors, which would result in overall growth of the tumor mass despite limited proliferation of the 4N cancer cells. To explore these possibilities, we performed immunohistochemistry on xenograft tumor tissue sections for the proliferation marker Ki67, cleaved-caspase-3 (CC3, a marker of apoptosis), α- smooth muscle actin (α-SMA, a marker of cancer-associated fibroblasts), and F4/80 (a marker of murine macrophages). We found that proliferation was higher (Fig. 2 A and B) and apoptosis lower (Fig. 2 C and D) in the 4N compared to the 2N tumors, suggesting that the 4N cells may activate prosurvival signaling. We also found that the area occupied by cancer associated fibroblasts (CAFs) was larger in the 4N tumors compared to the 2N tumors (Fig. 2 E and F). This suggests the 4N cells within the tumor may promote the recruitment and/or activation of CAFs. Accumulation of CAFs could contribute to growth of the tumor mass both by directly increasing the cell mass in the tumor and by promoting proliferation of 2N cells, possibly through secretion of growth factors (36–38). Although the total content of murine tumor-associated macrophages (TAMs) was not different between 2N and 4N tumors (Fig. 2 G and H), the 4N tumors had a higher number of TAMs that were intraepithelial (i.e., within tumor nests) (Fig. 2 I and J). These are areas that are dense and cancer-cell rich, with no stromal characteristics. The presence of TAMs within tumor nests can lead to protumorigenic signaling by enhancing crosstalk between tumor cells, macrophages, and other stromal cells (39, 40). Therefore, the 4N cells may promote TAM invasion into the tumor nests, aiding tumor growth and signaling.

Fig. 2.

4N tumors display lower incidence of apoptosis and higher numbers of protumorigenic stromal cells compared to 2N tumors. (A) Representative images of Ki67 immunohistochemistry staining in 2N (Top) and 4N (Bottom) xenograft tumor sections. (B) Quantification of Ki67-positive area in 2N and 4N xenograft tumors. (C) Representative images of cleaved-caspase3 (CC3) immunohistochemistry staining in 2N (Top) and 4N (Bottom) xenograft tumor sections. (D) Quantification of CC3-positive area in 2N and 4N xenograft tumors. (E) Representative images of α-smooth muscle actin (α-SMA) immunohistochemistry staining in 2N (Top) and 4N (Bottom) xenograft tumor sections. (F) Quantification of α-SMA-positive area in 2N and 4N xenograft tumors. (G) Representative images of F4/80 immunohistochemistry staining in 2N (Top) and 4N (Bottom) xenograft tumors. (H) Quantification of F4/80-positive area in 2N and 4N xenograft tumors. (I) Representative images of F4/80 immunohistochemistry staining within tumor nests in 2N (Top) and 4N (Bottom) xenograft tumors. (J) Quantification F4/80-positive area in the tumor nest regions of 2N and 4N xenograft tumors. All error bars report mean ± SEM, with individual points representing one tumor section. Each tumor (N = 10) was stained and imaged twice (n = 20) on two different sections to account for diversity in the xenograft mass. All scale bars = 1,000 µm. **P < 0.01, ***P < 0.001, ****P < 0.0001.

To assess whether 4N cells may play similar roles in human tumors, we examined TCGA samples from Taylor et al. (3, 41) and found that the stromal fraction was higher in several WGD-positive (WGD+) cancers compared to WGD-negative (WGD−) from the same anatomical site (SI Appendix, Fig. S1A). Moreover, we analyzed H&E whole slide images (WSIs) from TCGA diagnostic slides (as in ref. 35) and found other WGD+ cancers that showed lower numbers of dead cells compared to WGD- cancers from the same anatomical site (SI Appendix, Fig. S1B).

Extra Centrosomes Acquired during Whole Genome Doubling Are Lost during Tumor Evolution In Vivo, But Not to the Same Extent as In Vitro.

Extra centrosomes have been suggested to contribute to the tumorigenic potential of 4N cells by inducing chromosomal instability and aneuploidy (25, 27, 42). In support of this idea, supernumerary centrosomes are often found in solid and hematological tumors (23, 43, 44). However, we previously found that DLD1 cells lose their extra centrosomes very quickly after WGD in standard culture conditions (20). This raises the possibility that factors within the TME may promote the formation or preservation of supernumerary centrosomes. To determine how tumor formation in mice affects centrosome numbers, we quantified centriole numbers in mitotic cells (Fig. 1G) from postxenograft samples. In the 2N tumors, there was a slight, nonsignificant increase in the fraction of cells with extra centrioles in the postxenograft samples compared to the injected population (Fig. 1H). In the postxenograft samples from the 4N tumors, the 2N subpopulation displayed centriole numbers similar to those observed in the postxenograft samples from the 2N tumors (compare Fig. 1 I and H). Moreover, a large portion of 4N cells lost their extra centrosomes during tumor growth (compare injected and 4N in Fig. 1I). However, the fraction of 4N cells with extra centrioles was substantially higher than that observed when newly formed 4N cells were evolved in culture (~15%) (20). We also performed immunohistochemical staining of xenograft tumor tissue (SI Appendix, Fig. S2) and found that the fractions of mitotic cells with extra centrioles were consistent with those found in the postxenograft samples, indicating that briefly placing cells in culture for analysis after recovery from xenograft tumors does not affect centrosome numbers. Because our analysis was performed exclusively in mitotic cells, the presence of extra centrosomes in 4N cells in the postxenograft samples or in the tumor tissue cannot be explained by persistence of nonproliferative 4N cells. Instead, our data suggest that the TME affects centriole number in 4N cells by causing them to retain and/or acquire extra centrosomes.

Mathematical Modeling Shows That the TME Affects 4N Cell Evolution by Promoting Centrosome Overduplication and Altering Proliferation Rates.

As described above, newly formed 4N DLD1 cells are outcompeted by 2N cells during tumorigenesis in mice, and a substantial fraction of 4N cells have extra centrosomes compared to 2N cells. Conversely, when newly formed 4N cells are cultured in vitro, the fraction of 4N cells remains high and the number of cells with extra centrosomes decreases to levels similar to those found in the parental population (20). Due to the inherent challenges of in vivo experiments, it is difficult to dissect the factors that shape the cell population during its evolution. To address this issue, we developed a mathematical model (Fig. 3A and SI Appendix, Modeling Methods), expanding on our previous work (20), that tracks the coevolution of ploidy and centrosome numbers in different cell populations. Additionally, the model splits 4N cells into subpopulations based on their size, as our recent work shows that small 4N cells are more proliferative and better at centrosome clustering than large 4N cells (20, 35, 45). To determine the baseline “control” parameter values of the model, we used a global optimization algorithm to fit the model to previous centrosome number evolution data from standard in vitro culture conditions (20) (Fig. 3B and SI Appendix, Fig. S3 and Table S1).

Fig. 3.

Model for co-evolution of ploidy and centrosome number suggests that spontaneous centrosome amplification and changes in relative growth rates in vivo drive differential cell fates. (A) Model diagram demonstrating the pathways of interconversion between cell subpopulations. Tetraploid cells are separated by size, with large cells (4N4Cl, 4N2Cl) above, and small cells (4N4Cs, 4N2Cs) below. In all abbreviations, the number preceding “N” refers to the ploidy and the number preceding “C” refers to centrosome numbers. The letters “l” and “s” denote large and small cells, respectively. “NP” stands for nonproliferative. See Materials and Methods for detailed model assumptions and equations. (B) ODE simulation results for the model (solid lines) fitted to our previous in vitro data (stars) (20), and time trajectories of each subpopulation over the first 12 d (dashed lines). (C) Posterior distributions (red with shading), prior distributions (gray), and control value for each parameter (blue dashed lines) in the 4N tumors. (Top Left) Probability of cytokinesis failure. (Top Right) Probability of small 4N symmetric division. (Bottom Left) Probability of bipolar symmetric division in 2N and large 4N cells. (Bottom Right) Probability of centrosome overduplication. Posterior distributions were generated from inference results after burn-in and filtering (Materials and Methods). (D) Pairwise plots of effective birth rates (i.e., $b_i-d_i$) in 4N tumors. 2,000 randomly chosen postfiltering parameter sets were plotted. Each circle represents the pair of parameter values in one parameter set from posterior distributions obtained through Bayesian inference. The diagonal lines mark where the rates on the x- and y-axes are equal. Pearson correlation coefficients are reported in each figure’s title.

Our in vivo data (Fig. 1) suggest that the TME alters the fate of 4N cells compared to our observations in vitro (20). To identify cellular process/events that change in the TME, we performed Bayesian Inference parameter estimation (46) and determined which parameters were altered in the in vivo environment (SI Appendix, Fig. S4). Instead of predicting only the best-fitting parameter set for the in vivo data, Bayesian inference predicts the posterior probability distributions for each parameter in the model (Fig. 3C and SI Appendix, Fig. S4), allowing for comprehensive insights into parameter changes. In our model, cells can acquire and/or retain extra centrosomes through cytokinesis failure, symmetric bipolar divisions, or centrosome overduplication (23–26). From our Bayesian analysis, we found that the peak of the posterior distribution for the in vivo probability of cytokinesis failure ($p_{\mathit{Cyto}}$) was located close to the in vitro control value (Fig. 3 C, Top Left). This result suggests that the rate of cytokinesis failure was likely unchanged in 4N tumors when compared to in vitro and hence cannot explain the elevated appearance of extra centrosomes in vivo. Additionally, the probability of retaining 2N and large 4N cells with extra centrosomes through symmetric, bipolar divisions ($p_{\mathit{Bi}}{p}_{\mathit{Sym}}$) also achieved peak density close to its in vitro control (Fig. 3 C, Top Right), whereas the corresponding probability for small 4N cells ($p_{\mathit{SCsym}}$) was concentrated below its in vitro control value (Fig. 3 C, Bottom Left). Such parameter changes would reduce extra centrosomes and fail to explain our in vivo observations. Finally, we found that the probability of centrosome overduplication ($p_{\mathit{Over}}$), established at 0.01% in the control parameter set (SI Appendix, Table S1), must increase dramatically in the 4N tumors (Fig. 3 C, Bottom Right). Thus, our model predicts centriole overduplication as the main contributing factor leading to a higher fraction of cells with extra centrosome in vivo (Fig. 1I) compared to our in vitro data (20), rather than higher rates of symmetric bipolar divisions or generation of new 4N cells through cytokinesis failure.

Since the 2N population outcompetes the new 4N population in vivo (Fig. 1F), we also wanted to use our model to predict how cell fitness (i.e., cell growth and death rates) is altered. However, our experimental data were limited to late time points, when the populations are assumed to have approached their steady states. This prevented us from making accurate inference on individual populations’ growth rates (SI Appendix, Fig. S4). As an alternative, we chose to compare the growth and death rates of each population and determine any relationships between them. Among cells with normal centrosome numbers, we found strong correlations between the effective growth rates of the 2N and 4N cells, with the large 4N cells growing slower than the 2N cells, and small 4N cells growing at a comparable rate to the 2N cells (Fig. 3D). These results corroborate our experimental data, showing that the 2N cells have a fitness advantage in vivo and outgrow the 4N cells.

Serum Starvation and Oxidative Stress Influence the Evolution of Newly Formed 4N Cells.

The modeling results described in the previous section suggest that our in vivo findings are due to conditions in the TME that promote the proliferation of 2N over 4N cells and centrosome overduplication. To assess how conditions found in the TME may affect the evolution of newly formed 4N cells, we performed a number of in vitro experiments in which cells were cultured under nonstandard conditions for 12 d (same duration as in ref. 20).

First, because cells in the tumor grow in a three-dimensional (3-D) space, we evolved cells in collagen hydrogels for 12 d (SI Appendix, Fig. S5A) and then assessed ploidy (SI Appendix, Fig. S5B) and centriole numbers (SI Appendix, Fig. S5C) in the evolved population. We found that the majority of the population consisted of 4N cells (SI Appendix, Fig. S5B) and that ~15% of the cells possessed extra centrioles at day 12 (SI Appendix, Fig. S5C), consistent with previous observations for cells evolved in two-dimensional (2-D) cultures (20). Varying stiffness in the collagen hydrogels (different collagen concentrations) did not have major effects on the fraction of 4N cells or centriole numbers in the evolved population (SI Appendix, Fig. S5 B and C). These results indicate that growth in 3-D and stiffness of the substrate do not, per se, affect the evolution of ploidy or centriole numbers in populations of newly formed 4N cells. These results also suggest that 2-D cultures are appropriate for studying the evolution of 4N cells, as suggested in previous work (15).

Next, we used 2-D cultures to assess the effects of several nonstandard culture conditions meant to mimic stresses found in the TME, including hypoxia, oxidative stress, and serum deprivation (Fig. 4A). We found that in hypoxic conditions (1% O₂), newly formed 4N cells remained 4N and lost their extra centrosomes similarly to what we observe in standard culture conditions [Fig. 4 B and C and (20)]. This was not surprising, given that hypoxia was previously shown to not affect proliferation in DLD1 cells (47). Conversely, when newly formed 4N cells were kept in culture without growth factors (i.e., no serum), the ploidy of the 12-day population was shifted toward 2N and only a small fraction of cells was 4N (Fig. 4B), similar to what we observed in the postxenograft populations (Fig. 1F). To further assess how serum deprivation affected the early evolution of 4N cells, we performed live cell imaging of 4N cells shortly after their formation and found that 4N cells kept in serum-free media displayed higher rates of cell death compared to cells kept in standard conditions (SI Appendix, Fig. S6A and Movies S1 and S2). Moreover, we analyzed the proliferation marker Ki67 (SI Appendix, Fig. S6C) at early time points of the 12-day period and found that proliferation was lower in serum deprivation compared to standard culture conditions (SI Appendix, Fig. S6B). The lack of growth factors had no effect on the evolution of centriole numbers (Fig. 4C). Conversely, we found that, while inducing oxidative stress in newly formed 4N cells did not affect the fraction of 2N and 4N cells (Fig. 4B), the fraction of cells with extra centrioles at day 12 was higher compared to standard conditions (Fig. 4C). Similar results were obtained when these nonstandard culture conditions were tested on 3-D cultures (SI Appendix, Fig. S5 D and E), again indicating that 2-D cell cultures are a good model to study ploidy and centrosome number evolution in 4N cells. In all conditions tested, the cells with extra centrioles in the evolved population are predominantly 4N cells (Fig. 4D and SI Appendix, Fig. S7 A–C), but the number of 4N cells with extra centrioles in cells subjected to oxidative stress was substantially higher than in any other culture condition (compare Fig. 4D and SI Appendix, Fig. S7 A–C). This indicates that the increase in extra centrioles induced by oxidative stress primarily affects 4N cells over 2N cells (Fig. 4D). Importantly, the fraction of cells with extra centrioles after growth in oxidative stress conditions was similar to what we observed in the postxenograft samples (Fig. 1I), suggesting that the centrosome overduplication seen in vivo could be due to the high levels of reactive oxygen species commonly reported in the TME (48–50). Our mathematical modeling specifically predicted that the increase in centrosome numbers seen in vivo may arise from centrosome overduplication rather than retention of the extra centrosomes acquired during WGD. To experimentally test this prediction, we investigated how centrosome numbers at day 12 compared to an earlier time point (day 8) and found that centrosome numbers were similar in cells evolved in standard vs. oxidative stress conditions at day 8 (SI Appendix, Fig. S7D). However, while centrosome numbers further decreased through day 12 in standard culture conditions (SI Appendix, Fig. S7D), they increased in cells subjected to oxidative stress from day 8 to day 12 (SI Appendix, Fig. S7D). This indicates that oxidative stress does not cause cells to retain the extra centrosome acquired during WGD; instead, these observations support the conclusion that oxidative stress causes centrosome overduplication. Taken together, our data identify oxidative stress and growth factor scarcity as the key TME conditions that may affect the evolution of 4N cells following WGD.

Fig. 4.

Serum starvation and oxidative stress influence the evolution of newly formed 4N cells. (A) Diagrammatic representation of experimental design used to induce tetraploidy and culture cells under conditions that mimic the tumor microenvironment in vitro. (B) Ploidy in cells immediately after DCB treatment (Day 0) and after 12 d of culture in various conditions. (C) Quantification of extra centrioles in cells cultured under various conditions for 12 d after DCB treatment. (D) Quantification of extra centrioles in 2N and 4N cells after 12 d of culture under oxidative stress. All error bars report SEM. ***P < 0.001, ****P < 0.0001.

Discussion

Many studies have identified WGD as a driver of tumor progression that can contribute to poor patient prognosis and drug resistance (1–5, 13). Additionally, a recent study reported that even subclonal WGD is associated with poor clinical outcomes (12). Our data unveil a role of tetraploidy in promoting tumorigenesis, which may explain how even small numbers of 4N cells may drive cancer progression. We found that when populations of predominantly 4N cells (~90% 4N cells + ~10% 2N cells, Fig. 1F-injected) were injected into nude mice, the 4N cells did not appear to proliferate very efficiently, yet the tumors (4N tumors) grew at similar rates as those (2N tumors) induced by injecting predominantly 2N cells (~10% 4N cells + ~90% 2N cells, Fig. 1E-injected). We found that this could be explained by higher proliferation, lower levels of apoptosis, and higher fractions of stromal components in the 4N compared to the 2N tumors. These data suggest that at the early stages of tumor formation in mice, when the fraction of 4N cells is still high, these cells may play a role in recruiting stromal cells, which may then promote cancer cell proliferation. Indeed, CAFs have been shown to both secrete growth factors (51) and recruit macrophages (52, 53). The growth factors may promote proliferation of epithelial cancer cells, with the 2N cells displaying a stronger response than their 4N counterparts. We should also note that, at the time of injection, the 4N cells carry extra centrosomes, which can lead to catastrophic mitosis and death of the progeny (20). Therefore, the 2N cells within the population may outgrow the 4N cells in the early stages of tumor formation due to the growth-promoting signals released by CAFs and pervasive cell death in the progeny of 4N cells arising through multipolar cell divisions. The recruitment of macrophages by CAFs may also explain some of our observations. CAF signaling can promote a protumoral transition of recruited macrophages to produce what are known as TAMs (52–54). TAMs, in turn, can promote survival and proliferation of resident cancer cells (55). This is consistent with our observation that 4N tumors had higher proliferation and lower levels of apoptosis compared to the 2N tumors. Consistent with our mouse studies, we also found that certain WGD+ human cancers display higher fractions of stromal cells or lower numbers of dead cells. Overall, our findings suggest that the oncogenic effects of WGD could be due, in part, to stromal cell recruitment, which leads to a protumorigenic TME that favors cancer cell survival and proliferation.

Previous studies have provided evidence that WGD can promote tumorigenesis through the concomitant acquisition of extra centrosomes, which can lead to chromosome missegregation and aneuploidy (21). This is consistent with the widespread aneuploidy in WGD+ tumors (1, 3, 5) and the many reports of extra centrosome in tumor tissues (23, 24, 27, 42). In fact, extra centrosomes were also shown to promote tumorigenesis independent of WGD (25). However, 4N cells evolved in culture rapidly lose their extra centrosomes while maintaining a near-4N chromosome number (20). In our xenograft model, we found that although the population of 4N cells decreases, the number of 4N cells with extra centrioles is higher than what is observed after in vitro evolution. Our modeling and experimental data suggest that certain features of the TME may be the basis for the discrepancies between in vitro and in vivo observations. In particular, we found that oxidative stress can explain the increase in centrosome numbers. Importantly, our modeling work points to centrosome overduplication as the mechanism underlying the high numbers of extra centrioles in the postxenograft cell and tissue samples. Interestingly, centrosome overduplication is also believed to be a major mechanism underlying the supernumerary centrosome phenotype in human cancers (56, 57). Our experimental data from cells cultured in oxidative stress conditions also support centrosome overduplication, rather than retention of extra centrosomes acquired during WGD, as the mechanism causing higher rates of supernumerary centrioles in 4N cells. Our experimental data also showed that extra centrioles were more common in 4N cells than in 2N cells and that the increase in extra centrioles resulting from oxidative stress specifically affected the 4N subpopulation. This could be due to oxidative stress having a stronger effect on 4N cells than 2N cells or to a selection mechanism. Indeed, it is possible that 2N cells are more susceptible to the presence of extra centrioles/centrosomes than 4N cells due to the chance of multipolar division, which would be more deleterious to 2N than 4N cells. Oxidative stress results in the formation of reactive oxygen species (ROS), and high levels of ROS have been reported in the TME of many tumors (58, 59). Thus, our work suggests that anatomical sites or tumors with high levels of oxidative stress and ROS may be more prone to experience centrosome overduplication. Because our work also suggests that centrosome overduplication may occur preferentially in 4N cells, the same anatomical sites and tumors may display a high incidence of tetraploidy.

Overall, our work shows that oxidative stress plays a major role in centrosome overduplication, particularly in 4N cells, suggesting that supernumerary centrosomes and WGD may coexist in tumors with high levels of ROS. We also found that, compared to tumors generated by 2N cells, tumors in which a larger fraction of 4N cells was present displayed differences in stromal cell numbers or localization in vivo. Because stromal cells are known to drive tumor growth and evolution (51, 52, 55), our data suggest that one mechanism by which WGD may drive tumorigenesis is recruiting and/or activating fibroblasts and macrophages.

Materials and Methods

DLD1 (ATCC CCL-221) cells were used for all experiments. Newly formed tetraploid cells were generated by dihydrocytochalsin B-induced cytokinesis failure. Mouse xenograft tumors were induced by subcutaneous injection of 2 × 10⁶ cells in nude mice. After tumor formation, tumors were resected and processed for downstream analyses, which included immunostaining of tumor tissue or resuspended cells and preparation of chromosome spreads for analysis of chromosome numbers. Theoretical work included a mathematical model that tracks the coevolution of ploidy and centrosome numbers in different cell populations combined with Bayesian inference. Detailed information on 2D and 3D cell culture, chromosome and centrosome staining and analysis, mouse xenograft assays, xenograft immunohistochemistry, and mathematical modeling can be found in the SI Appendix.

Supplementary Material

Appendix 01 (PDF)

Movie S1.

Representative time-lapse phase-contrast microscopy video of DLD1 cells after experimental induction of WGD and imaged in standard culture conditions. Time shown as hh:mm.

Movie S2.

Representative time-lapse phase-contrast microscopy video of DLD1 cells after experimental induction of WGD and imaged in serum-deprivation conditions. Time shown as hh:mm.

Data, Materials, and Software Availability

The code for the mathematical model, parameter fitting, global and local parameter sensitivity analyses, and Bayesian inference used in this study is available on the GitHub repository at https://github.com/jingchen-biomodel/cent-ploidy-evo (60), under the MIT license. The data in SI Appendix, Fig. S1B are based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga. Other data are included in the article and/or supporting information.