Comparison of cancer cells in 2D vs 3D culture reveals differences in AKT–mTOR–S6K signaling and drug responses

Established cell lines have been essential in understanding the basic molecular principles of cancer (Bodnar et al., 1998; Hahn et al., 1999; Iyer et al., 1999). However, cancer cells have been continuously cultivated in vitro for a long time and have adapted to in vitro culture conditions. These concerns are often raised to question the relevance of cell lines as proper preclinical cancer models. However, for colorectal cancer (CRC), the analysis of the mutation and gene expression status of about 150 available human CRC cell lines has revealed that the entire variety of molecular and transcriptional subtypes, as previously described in CRCs in vivo (Marisa et al., 2013), is represented in this cell collection (Medico et al., 2015). In addition, these cells display the full range of drug responses to EGFR inhibition from sensitivity to resistance. Thus, CRC cell lines seem to fully cover the genetic, transcriptional and phenotypic make-up of cancer cells in vivo.

However, cells are often cultured as monolayers on flat surfaces, and these conditions do not faithfully reflect the situation in vivo given that proper tissue architecture and cell–cell contacts are lost in such two-dimensional (2D) systems. 3D cell cultures are well documented to regain intrinsic properties and to better mimic the in vivo situation than cells cultured as monolayers on plastic (Dolznig et al., 2011; Fischbach et al., 2009; Okawa et al., 2007; Pickl and Ries, 2009). Studies have shown that the gene expression profiles (Takagi et al., 2007) as well as the responses to treatment (Desoize and Jardillier, 2000) in the multicellular spheroid 3D models resemble more closely the in vivo situation. Similar to human tumors, proliferating, quiescent and dying cells coexist in normoxic, hypoxic or necrotic zones within spheroids (Hirschhaeuser et al., 2010). Therefore, 3D cancer models are increasingly recognized and biologically relevant for drug development and preclinical drug testing. In vivo, carcinoma cells not only form 3D structures but also interact with the surrounding extracellular matrix (ECM) and cells of the tumor stroma. These interactions have been shown to be essential for tumor development and progression, which prominently alter signaling pathways in both the tumor cells and the stromal cells.

Thus far, only a few comprehensive and systematic studies have been found that compare distinct signaling pathways in 2D vs 3D culture, in the presence or absence of ECM and/or stromal cells, or upon treatment with inhibitors (Ekert et al., 2014; Luca et al., 2013; Pickl and Ries, 2009). Changes in signaling cascades in cancer cells or crosstalk with other cell types in 3D cultures are still poorly understood. How these pathways change in response to targeted therapy and whether there is a different response in 2D vs 3D culture remain rather undocumented so far.

In this study, we systematically analyze the signaling changes in 2D vs 3D culture without adding further levels of complexity, such as the interaction of these cells with the ECM or stromal cells. We focus on the AKT and mammalian target of rapamycin (mTOR), and mitogen-activated protein kinase (MAPK) pathways because these two pathways are the most commonly mutated signaling routes in CRC and many other cancers. The PI3K–AKT–mTOR pathway is a central regulator of cell growth, proliferation, survival, metabolism and aging because it integrates different environmental cues. Dysregulation of the PI3K–AKT–mTOR pathway has been implicated in various pathological conditions, including cancer. Based on these findings, different efforts to modulate the PI3K–AKT–mTOR activity by pharmacologically targeting different molecules in the pathway are ongoing (Laplante and Sabatini, 2012).

We provide here some evidence of significant differences in PI3K–AKT–mTOR signaling in a 3D vs a 2D system, including spatial alterations in the signaling strength and responses upon treatment with inhibitors of the AKT, mTOR and S6K (also known as RPS6KB1) axis or of the MAPK (ERK) axis.

We used six colon cancer cell lines in this study. Caco-2 cells display enterocyte-like differentiation and do not harbor known mutations in the PI3K–AKT–mTOR or RAS–RAF–MAPK pathways (Hidalgo et al., 1989; Pinto et al., 1983). LS174T, DLD-1 and HCT116 cells contain mutations in both PI3K and KRAS, HT29 cells have mutations in BRAF and PI3K, and SW620 cells are driven by mutant KRAS (Table S1).

All cell lines formed spheroids within 24–48 h after seeding. Spheroid morphology varied from a compact appearance (DLD-1, HT29) and less-condensed spheroids with smooth (HCT116) or irregular surfaces (LS174T) to loose aggregates (SW620) and adenomatous cell clusters (Caco-2) (Fig. 1A). EdU-incorporation was used to precisely determine the cell cycle distribution of cells grown in 2D or 3D culture (Fig. 1B). A significant reduction (−50%) of cells in S-phase in 3D culture compared to those in 2D culture was observed in HT29, HCT116, Caco-2 and DLD-1 cells. A small reduction (−20%, Fig. 1C) in the number of cells in S-phase was observed in LS174T and SW620 cells. Cell death, as determined by calculating the percentage of subG1 cells by flow cytometry, was generally low in all cell lines when cultured in three or two dimensions (Fig. 1D). Nevertheless, there was a twofold increase in the apoptotic rate in 3D culture compared to the rate in 2D culture of DLD-1 cells (8% and 4% in 3D and 2D culture, respectively). Low levels of apoptosis were confirmed by cleaved-caspase-3 staining in 2D culture as well as in spheroid sections (Fig. 1E). No spatial variations were detected in the spheroids – apoptotic cells appeared randomly without clustering in specific areas.

Next, we determined the molecular differences in signaling pathways in the CRC cells in 2D and 3D culture by using western blot analysis (Fig. 2A). The amount of phosphorylated (phospho) proteins was quantified by using densitometry evaluation (Fig. 2B) of three biological replicates. Phosphorylation of pan-AKT at residue S473 (referring to the position of serine in AKT1; phospho-AKT) was weak in SW620 cells compared to that in the other cell lines. Diminished phospho-AKT corresponded to low AKT activity, as observed by the reduced phosphorylation of the direct AKT target PRAS40 (also known as AKT1S1). Interestingly, in five cell lines (LS174T, HT29, HCT116, Caco-2, DLD-1), AKT phosphorylation was selectively reduced in 3D but not in 2D culture. This was associated with decreased phospho-PRAS40 in four of the five cell lines (i.e. cell lines LS174T, HT29, HCT116, DLD-1); in Caco-2 cells, PRAS40 was equally phosphorylated during 2D and 3D culture (Fig. 2B upper panels). In all six lines tested, S6K and RPS6 phosphorylation was consistently and significantly reduced in spheroids, which indicated reduced mammalian target of rapamycin complex 1 (mTORC1) activity. Accordingly, another direct target of mTORC1, eukaryotic initiation factor 4 binding protein 1 (4E-BP1), was less phosphorylated under 3D culture. The level of hyperphosphorylated 4E-BP1 (i.e. 4E-BP1γ) was reduced to about 50% in LS174T, HT29, Caco-2 and DLD-1 cells cultured in 3D as compared to that under 2D culture, or by 20% in SW620; no change in HCT116 cells was found (Fig. 2B, middle panels). In contrast, phosphorylation of MAPK3 and MAPK1 (phospho-ERK1/2) as an indicator of active RAS–RAF–MAPK signaling was variably influenced by 3D culture conditions. In DLD-1, LS174T and HT29 cells, ERK1/2 phosphorylation remained unchanged in 3D compared to 2D culture. Phospho-ERK1/2 was significantly induced in SW620 and Caco-2 cells in 3D culture but decreased in HCT116 cells. PKCβII (an isoform encoded by PRKCB) phosphorylation was slightly (LS174T, HCT116, DLD-1, HT29) or strongly induced (SW620, Caco-2) in 3D culture, indicating that there was not a general decrease in phosphorylated proteins, thereby ruling out a handling, harvesting or cell lysis artifact that was only present in 3D cultures (Fig. 2B, lower panel).

In summary, these experiments reveal a general profound and highly specific decrease in AKT–mTOR–S6K signaling in cancer cells grown as spheroids as compared to those cultured in the respective conventional 2D system.

The global decrease in the activity of the AKT–mTOR–S6K axis in 3D cultures prompted further investigation of specific inhibitors that target major hubs in this pathway and comparison of the phenotypic changes, as well as molecular responses in signaling, in 2D vs 3D cultures. A MAPK pathway inhibitor targeting MEK1 (also known as MAP2K1) was also included. The inhibitors used in this study were rapamycin, Torin1, PF4708671, MK2206 and AZD6244 to target mTORC1, mTOR, S6K, AKT and MEK1 kinase activities, respectively. DLD-1, Caco-2 and HCT116 were selected for this in depth analysis because these cell lines belong to the same transcriptional subtype of CRC cells (stem-cell like; Marisa et al., 2013; Medico et al., 2015). DLD-1 and HCT116 harbor the same mutations in the PI3K–AKT and MAPK pathways, whereas Caco-2 cells are wild type for those pathways.

Cells were grown in 2D or 3D culture for 1 day and were treated with compounds or solvent control (DMSO) for another 24 h. Proper drug action was verified by western blot analysis of key phospho-proteins in the disrupted signaling pathway. Upon treatment with rapamycin, PF4708671, MK2206 or AZD6244, cell morphology of DLD-1 cells (Fig. 3A) in 2D culture was unchanged. In contrast, DLD-1 spheroids that had been treated with Torin1, MK2206 and AZD6244 appeared darker in phase contrast analyses (Fig. 3A). In comparison to DMSO controls, spheroid size was reduced to about 50% upon treatment with Torin1, and the rapamycin-, MK2206- or AZD6244-treated spheroids displayed intermediate volume reduction (75% of controls), whereas PF4708671 had a small but significant impact on spheroid size (Fig. 3A,B). The volume of HCT116 spheroids was reduced significantly under all treatment conditions. In Caco-2 spheroids, treatment with Torin1 and MK2206 resulted in spheroids that were 80% of the size of controls (Fig. S1A,B).

Cell cycle analysis revealed essential differences in treatment responses between DLD-1 cells that had been cultured in two vs three dimensions (Fig. 3C; HCT116 and Caco-2, Fig. S1C). In the 2D system, rapamycin, PF4708671 and AZD6244 had no effect on cell cycle progression, whereas Torin1 reduced the proportion of cells in S-phase to less than 15% as compared to 38% in DMSO controls. Inhibition of AKT had a small but significant effect on proliferation. In contrast, cell proliferation was affected by all compounds (except for the S6K inhibitor) in the 3D model. MK2206 displayed a more pronounced inhibitory effect by reducing the number the cells in S-phase by 50% (8% vs 16%). Unexpectedly, AZD6244 was even more effective, resulting in only 4% of cells replicating; however, it was ineffective in 2D cultures. Interestingly, total inhibition of mTOR kinase activity led to a complete block in cell proliferation in 3D culture (0.3% of cells were in S-phase), whereas blocking mTORC1 activity with rapamycin had a minor effect (10% vs 16% in controls). HCT116 cells were most effectively slowed down in their proliferation through inhibition of S6K and MEK1 in 2D culture. However, in 3D cultures, AZD6244 and Torin1 treatment resulted in the most pronounced effects on proliferation (Fig. S1C), as also seen in DLD-1 and Caco-2 spheroids (Fig. S1C).

To further corroborate these findings, cellular metabolic capacity was determined (Fig. 3D), which faithfully recapitulated the pattern of the cell cycle analysis in DLD-1 and HCT116 cultures. Interestingly, there was a general decrease in metabolic activity of cells in 3D compared to 2D cultures – a 40% decrease was observed for DLD-1 cells in 3D culture (Fig. 3D) and a 60% decrease in HCT116 and Caco-2 cells (Fig. S1D). Subtle but significant differences were detectable, similar to results observed in the cell cycle analysis. All treatments that appeared to have no effect on proliferation in 2D and in 3D culture (see Fig. 3C; Fig. S1C) nevertheless resulted in small decreases in metabolic activity. Surprisingly, this was not the case in Caco-2 cells in which the level of intracellular ATP was rather inversely correlated with the S-phase data.

As inhibition of the mTOR pathway results in reduced cellular volume (Fingar et al., 2002), we next determined whether a gross decrease of cell size contributes to the effects of the inhibitors on spheroid size. Cell size was reduced by Torin1 in DLD-1 (Fig. 3G), HCT116 (Fig. S1F) and Caco-2 spheroids (Fig. S1F), or in the 2D system only marginally.

Apoptosis levels remained low under treatment with compounds, as determined by staining of cleaved caspase-3 (DLD-1, Fig. 3E) and sub-G1 DNA content (DLD-1, Fig. 3F; HCT116, Caco-2, Fig. S1E). Single inhibitor treatment had no gross apoptotic effect on the CRC cells either in 2D or in 3D culture, which is indicated by the low percentage of cells in sub-G1. Interestingly, only inhibition of MEK1 (by AZD6244) induced low rates of apoptosis in 3D culture compared to the controls. Thus, the small differences in cell size and apoptosis led to the conclusion that reduced spheroid volume, which might serve as an approximation for impaired tumor growth, is predominantly due to a reduction of cellular proliferation capacity under effective treatment conditions. Taken together, there are sharp differences in drug responses in 2D vs 3D culture. Inhibition of mTOR(C1)–AKT resulted in more severe anti-proliferative activity in 3D spheroids as compared to that in the 2D system.

In addition, treated cells in 2D and 3D culture were analyzed for total and phosphorylated proteins involved in AKT–mTOR and MAPK signaling by western blotting. Three biological replicates were analyzed (DLD-1, Fig. 4A), and the mean response was calculated by performing densitometry evaluation (Fig. 4B; Fig. S2). Firstly, this was conducted to demonstrate proper drug action, and secondly, to identify differences in signaling upon inhibitor treatment between 2D and 3D cultures. In all three cell lines, treatment with rapamycin and Torin1 resulted in complete abrogation of phospho-S6K in 2D and 3D cultures, whereas Torin1 further inhibited AKT phosphorylation as a result of mTORC2 blockade. There was a consistent increase in phospho-AKT (residue S473) in the presence of rapamycin or PF4708671, which is indicative of the discontinuation of the negative feedback loop to IRS1 as a result of these two inhibitors (Tremblay et al., 2007). Treatment with PF4708671 led to an accumulation of phospho-S6K, which has been shown to lack kinase activity (Pearce et al., 2010), and further downstream, to a decreased level of phospho-RPS6. Treatment with MK2206 abolished AKT phosphorylation at S473 and T308, and the phosphorylation of the downstream target PRAS40 was eliminated.

Of note, upon MK2206 and AZD6244 treatment, the levels of phospho-S6K and phospho-RPS6 were decreased in spheroids, but not in the 2D cultures, indicating a response of mTORC1 activity to AKT or MEK1 inhibition only in the 3D model. Furthermore, abrogating the AKT–mTOR–S6K axis with any inhibitor increased ERK1/2 phosphorylation effectively in the 2D, but not in the 3D, model.

In conclusion, these data reveal that mTOR activity and the crosstalk between AKT–mTOR–S6K signaling and the MAPK pathway is altered in 3D compared to 2D cultures.

The prominent decrease of phospho-RPS6 in spheroids in western blots could be due to two reasons. Either phospho-RPS6 is reduced in every cell to a similar extent or there are cells with variable expression levels from high to low or absent, which in sum give rise to the diminished phospho-RPS6 levels detected by immunoblotting.

Immunofluorescence analysis (Fig. 5A) confirmed the signaling response to treatment with inhibitors observed by western blotting analysis. In 2D culture, treatment with rapamycin, Torin1 and, to a lesser extent, PF4708671 led to a reduction of phospho-RPS6. Cross-sections of cells grown in two dimensions showed the same result (supporting figure 7). DLD-1 cells that had been treated with MK2206 or AZD6244 in 2D cultures showed no decrease in phospho-RPS6. However, the spheroid sections that had been stained for phospho-RPS6 revealed uneven intensities. A clear gradient of phospho-RPS6 staining intensity from the spheroid surface to the inner core was evident in the controls. This was also seen at reduced levels in PF4708671-, MK2206- and AZD6244- treated spheres. However, the staining inside displayed more variation than that in DMSO controls (Fig. 5A), whereas Torin1 and rapamycin diminished phospho-RPS6 almost completely. Total RPS6 protein was not affected and was expressed at high levels in every cell (Fig. S3A,B).

As a more objective evaluation, image analysis was used (Schlederer et al., 2014). Automated spheroid recognition and zonation (Fig. 5B, Fig. S3C) allowed division of equatorial sphere sections into three zones – an outer zone comprising the surface two-cell layers followed by a middle zone and the inner core (Fig. 5C). Nuclear recognition and creation of a cytoplasmic mask was used to quantitatively determine phospho-RPS6 levels in every single cell in the three distinct zones. In DMSO controls, the outer rim displayed a mean phospho-RPS6 level of 100 (Fig. 5D). In the inner core, phospho-RPS6 was reduced to more than half, and in the middle zone, a mean value of 63 was determined, which indicated a non-linear rapid decrease of S6K phosphorylation from the surface to the center (Fig. 5D). In rapamycin- or Torin1-treated spheroids, there was an almost complete loss of S6K phosphorylation in all zones. S6K, AKT and MEK1 inhibition resulted in a gradient similar to that in the control but with reduced mean intensities. Interestingly, the range of the signals varied from 200 to 25 under MK2206 and AZD6244 treatment in the outer zone, indicating that some cells still displayed maximal phospho-RPS6 levels in this area, comparable to DMSO controls, whereas the minimum was reduced to half. Inhibition of S6K reduced the range of phospho-RPS6 levels from 150 to 25, and the mean was smaller than that with the AKT or MEK1 inhibitors, suggesting that RPS6 phosphorylation was reduced in all cells (Fig. 5D).

Importantly, phospho-RPS6 gradients in in vivo DLD-1 xenografted tumors were present as detected in the DMSO-treated control spheroids (Fig. 5E), underscoring the ability of our model system to faithfully recapitulate the in vivo situation for AKT–mTOR–S6K signaling. In vivo, the highest levels of RPS6 phosphorylation were evident in the direct vicinity of cross-sectioned perpendicular tumor blood vessels. The levels of phospho-RPS6 radially diminished with increasing distance to the endothelium (Fig. 5E). Total RPS6 signals were invariably present in all cells (Fig. S4A). In the DLD-1 xenograft, cellular phospho-RPS6 signals were quantified in vessel-proximal, intermediate and vessel-distal zones (Fig. 5F). This analysis revealed an astonishingly similar pattern of intensity distributions, with highest levels of phospho-RPS6 in a vessel-proximal area that was three-cell-layers thick, and a prominent drop in the intermediate layer and low-to-absent vessel-distal signals (Fig. 5G). On the one hand, this qualifies DLD-1 spheroids as a reliable model for physiological mTOR–S6K signaling in tumors. On the other hand, it reveals that growth factor, nutrient and/or oxygen gradients are a probable explanation for the formation of the gradient from the surface to the center of the spheroid.

However, this recapitulation of the in vivo data by the in vitro assay might be specific for DLD-1 cells and thus be of limited relevance. Hence, the experiments were repeated with HCT116 cells, which differ in their mutational status from DLD-1 cells (Table S1). Indeed, the phospho-RPS6 gradient around blood vessels was confirmed in HCT116 xenografts (Fig. 6A,B), and the gradual decrease of phospho-RPS6 from outside to inside was corroborated in HCT116 spheroid sections (Fig. 6C), whereas total RPS6 levels were unchanged throughout the spheroid (Fig. S4B). Moreover, phospho-RPS6 in HCT116 cells showed the same zonal pattern as in DLD-1 spheroids under the different treatments in 3D (Fig. 6C; Fig. S3B). Phospho-RPS6 levels were quantified by performing image analysis (Fig. 6D). Notably, the mean of all zones (Fig. 6D, red dotted lines) closely recapitulated the pattern found in western blots (Fig. S2A). Ki67 and DAPI signal intensities in the three zones were also determined. The nuclei of the cells in the spheres were evenly stained with DAPI (Fig. 6E). The proliferation marker Ki67 exhibited a well-described decrease of Ki67⁺ cells from the outside to the core of the spheroid (Grimes et al., 2014; Laurent et al., 2013). Ki67⁺ cells were significantly reduced in the treated samples as compared to the control (Fig. 6F; Table S3). Under treatment, Ki67⁺ cells did not completely overlap with the strict pattern of phospho-RPS6 expression, indicating that the mTOR–S6K pathway is not solely or directly involved in the cell cycle changes of HCT116 cells grown as spheres. Taken together, our data show that the spatial differences in the activities of mTOR–S6K in human CRCs in vivo are faithfully reproduced in the spheroid model.

In order to normalize the results of this extensive study for simultaneous comparative evaluation, fold changes relative to the respective DMSO controls in the 2D and 3D systems were calculated and presented in heatmaps (Fig. 7). First, 2D vs 3D cultures were compared under normal growth conditions. The obtained data were arranged according to the mean change calculated for all cell lines (Fig. 7A). This allowed assessment and identification of responsive phenotypes or signaling molecules, and permitted general statements to be made. At first, a general significant down modulation of AKT signaling was detectable in the 3D cultures (loss of S473 and T308 phosphorylation, as well as reduced levels of phospho-PRAS40, one downstream target of AKT). Secondly, phosphorylation of S6K and RPS6 was substantially impaired in 3D cultures of all cell lines. Together with the decrease of hyperphosphorylated 4E-BP1γ, these data clearly indicate a general decrease in mTORC1 activity in the 3D system. Interestingly, phospho-ERK1/2 levels were highly variable, whereas phospho-PKC and E-cadherin expression was induced in the 3D spheroid culture. All the different responses could not be correlated to the mutational status (Fig. 7A, right).

The common and differential responses to treatment with inhibitors in 2D and 3D culture are summarized in Fig. 7B. Most obviously, S6K, RPS6 and 4EBP1 phosphorylation was efficiently blocked in 2D and 3D culture with all tested inhibitors, except for PF4708671 treatment, which resulted in S6K hyperphosphorylation both in 2D and 3D systems. It is well documented that binding of PF4708671 to S6K promotes its phosphorylation by mTORC1 but retains S6K in an inactive state (Pearce et al., 2010). All inhibitors resulted in a considerably decreased spheroid size. However, the proportion of cells in S-phase was unaltered by PF4708671 in DLD-1 cells in 3D culture, but the proportion of S-phase cells in Caco-2 both in the 2D and the 3D models was increased. Intracellular ATP levels were decreased in DLD-1 and HCT116 cells in 2D and 3D cultures, and in Caco-2 cells in 2D culture. However, treatment with rapamycin, Torin1 and AZD6244 slightly induced ATP content in Caco-2 cells in 3D culture for unknown reasons. Most interestingly, there was an increase of phospho-ERK1/2 in 2D cultures when treated with AKT, mTOR or S6K inhibitors, whereas the opposite was true for HCT116 and DLD-1 3D spheroids (Fig. 7B, bottom panel).

We identified spatial differences in RPS6 phosphorylation in spheroids under normal growth conditions. Phospho-RPS6 levels were high at the outer rim of DLD-1 and HCT116 spheroids and gradually decreased towards the inner core of the spheres. Interestingly, a similar phospho-RPS6 gradient was also detected in vivo in the same CRC cells grown as xenografted tumors in mice. Here, phospho-RPS6 levels were highest in close vicinity to blood vessels and diminished with increasing distance from the vessel (Fig. 7C). The phospho-RPS6 signal was almost completely abolished in rapamycin- or Torin1-treated samples, which fully blocked phospho-RPS6 in all three zones. On the contrary, phospho-RPS6 levels were reduced in each zone upon treatment with PF4708671, MK2206 or AZD6244, but the outer zone still displayed higher signals as compared to the respective middle and inner zones (Fig. 7D).

We used phospho-receptor tyrosine kinase (phospho-RTK) arrays as a first attempt to identify potentially altered signaling pathways upstream of AKT or mTOR–S6K that might also be responsible for the decreased AKT and/or mTOR–S6K activities in 3D culture. 28 different RTKs and 11 intracellular phospho-protein signaling molecules were evaluated in DLD-1 and HCT116 cells grown in 2D or 3D culture. The global phospho-RTK profile is shown in a heatmap (Fig. 8A). As a general downregulation of AKT and mTOR–S6K activities was identified, we focused on consistently downregulated phospho-RTKs in 3D compared to 2D culture in DLD-1 and HCT116 cells. Surprisingly, we detected a consistent loss of Eph receptor (Eph; EphA2, EphA3 and EphB4) phosphorylation as well as a decrease in Src phosphorylation (Fig. 8B,C). Interestingly, ErbB2 (also known as HER2) and ErbB3 signaling was non-uniformly changed in 2D vs 3D cultures. In DLD-1 spheroids, ErbB2 and ErbB3 phosphorylation was significantly increased, whereas there was a sharp decrease of ErbB2 and ErbB3 phosphorylation in HCT116 spheroids compared to in the 2D cultures. Importantly, AKT as well as RPS6 phosphorylation was significantly reduced in the 3D system (Fig. 8C), as expected from our previous results.

The overall survival rate of individuals with cancer undergoing systemic therapy has been extended only marginally over the past 30 years. Cancer therapy for advanced cancers is far from being efficient to date (Scannell et al., 2012). Despite some key therapeutic success in targeted therapy, the current fail rate of anti-cancer compounds in clinic trials is 96% (Bhattacharjee, 2012). Clearly, there is an urgent need for more innovative pre-clinical in vitro models to thoroughly test promising compounds before proceeding into clinical trials.

3D models might provide better predictive power by more faithfully recapitulating the response of human tumors to anti-cancer compounds in vivo. This study is not a complete and comprehensive assessment of every organotypic and/or 3D model currently available (Stadler et al., 2015; Unger et al., 2014), but intentionally focuses on the comparison of 2D monolayers with free-floating 3D spheroids. This relatively simple model lacks the interaction of cancer cells with their environment, such as ECM and/or stromal cells, and has been used for long time to better mimic chemotherapeutic and radiation therapy response in vitro.

With regard to metastasis, the predominance and importance of single circulating tumor cells in vivo is well accepted. However, there is recent evidence that primary tumor cell spheroids might be also of in vivo relevance. Cancer cell clusters have been identified to pass through lymphatic endothelial cell layers in human mammary carcinomas (Kerjaschki et al., 2011), whereas colon cancer clusters have been found in blood vessels in human CRC. In CRC, two types of clusters have been identified, which differed in their polarity. On the one hand, they interacted with their environment, and on the other hand, cancer cell spheres with inverse polarity lacked contact to host structures, closely resembling in vitro generated spheroids. Of note, these cancer-tissue-derived spheres have also been shown to be critically involved in liver metastasis in mice (Okuyama et al., 2016). Thus, these results might qualify spheroids as proper models to recapitulate specific subsets of cancer metastasis. Nevertheless, comprehensive systematic investigations on phenotypic changes, potentially altered signaling pathways and the response to drug treatment in 2D vs 3D systems have only occasionally been performed (Ekert et al., 2014; Luca et al., 2013; Pickl and Ries, 2009). We focused on colon carcinoma, being the third most deadly cancer in the world, to thoroughly evaluate the main signaling pathways in 2D vs 3D spheroid cultures under normal and treated conditions.

Our findings that AKT and mTOR signaling is drastically reduced in our 3D system lead to questioning of the reasons and the underlying mechanisms. In fact, it has been shown that RTK signaling upstream of AKT–mTOR is specifically induced in 3D compared to 2D cultures. For EGFR/ErbB signaling in HER2-overexpressing breast and ovarian cancer cells, it has been shown that 3D culture results in HER2 activation in contrast to 2D culture (Pickl and Ries, 2009; Weigelt et al., 2010). Another study using lung cancer cultures has demonstrated elevated basal levels of EGFR and Met phosphorylation levels upon 3D culture compared to the levels in 2D culture (Ekert et al., 2014). Counterintuitively, activation of HER2 signaling is associated with decreased phospho-AKT levels in the breast and ovarian cancer cells. In line with this discrepancy, RTK profiling in HCT116 and DLD-1 cells revealed striking differences in EGFR signaling, showing a significant increase in ErbB2 and ErbB3 phosphorylation in DLD-1 spheroids, whereas HCT116 spheres displayed reduced phospho-ErbB2 and -ErbB3 levels as compared to the levels in 2D cultures. However, consistent with the breast cancer study, AKT signaling was decreased in both cell lines in 3D culture. In fact, all six CRC cell lines used in our study showed decreased AKT activity. Of note, reduction of AKT phosphorylation in 3D compared to 2D systems has been described in another study with CRC (Luca et al., 2013) and breast cancer cells (Weigelt et al., 2010), thus pointing to a general mechanism in 3D cultures. Interestingly, we observed loss of Eph and Src phosphorylation in 3D compared to 2D cultures of HCT116 and DLD-1 cells, as demonstrated by RTK profiling. However, phosphorylation levels of Eph were low. Importantly, these data are only correlative, and further functional experiments are needed to decipher the impact of Ephrin–Eph signaling on AKT–mTOR activities in spheroids. One alternative explanation for the loss of AKT activation in 3D cultures is that signaling is switching from the AKT–mTOR pathway in 2D monolayers to the RAS–RAF–MAPK pathway in the 3D system. Indeed, phospho-MEK1 levels were increased in 3D cultures of breast and ovarian cancers; however, phospho-ERK1/2 levels are not substantially increased (Pickl and Ries, 2009) or have not been assessed (Weigelt et al., 2010). In colon cancers, expression profiling has revealed activation of MAPK signaling and a slight increase of phospho-ERK1/2 levels in Lovo cells in 3D cultures (Luca et al., 2013), whereas six other CRC lines displayed the opposite regulation. Indeed, in our study of ERK1/2 phosphorylation in 3D vs 2D cultures was also highly variable in the six cell lines. Some of the above-described 3D models engage laminin rich ECM (lrECM) to induce three-dimensionality, whereas others do not. Thus, the ECM adds another level of complexity (e.g. through integrin signaling) on top of the 2D vs 3D differences. Hence, further experiments are necessary to interpret decreased AKT signaling in the 3D system. Another possibility for the loss of AKT activity in 3D cultures might be simply the loss of integrin signaling, which is activated in 2D cultures by attachment to the plastic surface (or an ECM substrate) and thus induces AKT activation (Persad and Dedhar, 2003). Unfortunately, phospho-AKT levels in CRC spheroids were so low that it was not possible to detect spatial differences by performing immunofluorescence analysis. However, AKT activity might be induced through integrin signaling by embedding the spheres into ECM (basement membrane extract or collagen I gel), which is readily testable for changes in signaling in future studies by using our comprehensive approach. Alternatively, diminished AKT activity in 3D culture might be mediated through a specific increase in PTEN activity, which was not tested in this study. Notably, all six cell lines tested harbor wild-type PTEN. Nevertheless, AKT inhibitor treatment, which abolished AKT kinase activity, as demonstrated by the loss of downstream target phosphorylation, did not have such a profound effect on S6K activity in 2D as compared to 3D cultures.

To the best of our knowledge, we are the first to demonstrate that spheroid culture is associated with profoundly reduced mTORC1 signaling activity. This was revealed through diminished levels of active S6K and RPS6 phosphorylation and by the decrease of hyperphosphorylated 4E-BP1 (4E-BPγ), another target of mTORC1. Whether the reduced AKT activity is the reason for the downstream impairment of mTORC1 activity is not known. However, the decrease in phospho-RPS6 levels in 3D culture demonstrated in western blot analyses might be due to a gradual decrease of RPS6 phosphorylation in the inner regions of the spheroids, whereas in the surface area (two-cell layers), RPS6 phosphorylation remained high. This opens an alternative possibility for the diminishing mTORC1 activity inside the sphere. The inner regions of spheroids experience decreased nutrient availability (such as amino acids), which translates into diminished nutrient signaling and is mandatory for mTOR activation (Jewell et al., 2013). Indeed, intracellular amino acid concentrations rapidly drop along a gradient to the insides of spheroids (Kasinskas et al., 2014). This might explain the drastic decrease of RPS6 phosphorylation – from high levels in the outer two-cell-layer-thick zone to low levels in the deeper areas. Of note, cell proliferation also decreased gradually in the inner zones of the spheroids, as demonstrated previously (Grimes et al., 2014; Laurent et al., 2013). However, when mTOR activity was inhibited, RPS6 phosphorylation was lost but Ki67 remained, indicating that there is no direct correlation of RPS6 phosphorylation with cell proliferation. Importantly, the drop of RPS6 phosphorylation in the DLD-1 and HCT116 spheroids was also seen in vivo in viable cell clusters around blood vessels in tumor xenografts in mice. These results clearly qualify our spheroids as faithful models of the in vivo situation, at least for RPS6 phosphorylation and most probably for the upstream AKT–mTOR signaling pathway.

It is well known that the RAS–MAPK and the PI3K–AKT pathways can either positively or negatively regulate each other (Mendoza et al., 2011). Inhibition of MEK proteins can induce RTK-mediated activation of AKT through ERK-mediated inhibitory phosphorylation of GAB1 (Yu et al., 2002). On the contrary, AKT can negatively regulate ERK protein activation through inhibitory phosphorylation of RAF (Guan et al., 2000). This might explain why ERK1/2 phosphorylation was induced in all our cell lines in 2D culture upon inhibition of AKT. However, inhibiting mTOR or S6K also resulted in profound MAPK pathway activation in the 2D system. To the best of our knowledge, there are no reports on this obvious cross inhibition, which is released by AKT–mTOR pathway inhibitors. Even more puzzling is the fact that induction of ERK protein phosphorylation only occurred in 2D cultures, whereas in 3D cultures, the ERK phosphorylation was diminished upon inhibition of AKT–mTOR–S6K, suggesting a substantial rewiring of the two pathways in 3D cultures. Intriguingly, differential responses to inhibitor treatment of the three cell lines in 2D systems were abolished in 3D models, and all cell lines behaved in a similar manner, indicating a more consistent response in the 3D system.

Taken together, our results demonstrate a general theme in spheroid culture of colon cancer cells, namely the reduction in the activity of the AKT–mTOR–S6K pathway and a substantial rewiring of signaling in 3D compared to 2D culture. Our comprehensive study on colon cancer cells identifies common mechanisms that are active in 3D cultures and that are considerably different to those in 2D cultures. Despite being a simple and well-known model for many years, many unknown mechanisms are yet to be discovered in spheroids as a 3D model. Notably, the identification of sphere-like cancer cell clusters (Kerjaschki et al., 2011; Okuyama et al., 2016) entering, traveling along and escaping from vessel structures with or without interacting with the surrounding environment could result in novel perspectives for metastasis intervention. Such findings also strengthen the importance of spheroids as a valid model for mimicking certain aspects of human cancer biology. Moreover, here, we demonstrate that mTOR–S6K signaling gradients present in vivo are reliably reproduced in spheroid models. Thus, better understanding of 3D cancer biology in vitro – such as AKT–mTOR and MAPK signaling in spheroids – might aid better comprehension of cancer biology in vivo, prediction of the response of cancer cells to targeted therapy as well as the development of novel therapeutic concepts.

Human colon cancer cell lines LS174T [American Type Culture Collection (ATCC^®)#CL-188™], HT29 (ATCC^®#HTB-38™), SW620 (ATCC^®#CCL-227™), HCT116 (ATCC^®#CCL-247™), Caco-2 (ATCC^®#HTB-37™) and DLD-1 (ATCC^®#CCL-221™) were obtained from the ATCC and used from early passages derived from our master cell bank. Short tandem repeat (STR) profiling was performed for cell authentication. The cells were cultivated in Dulbecco's modified Eagle's medium (DMEM) high glucose (4.5 g/l) supplemented with 10% fetal calf serum, 2 mM L-glutamine and antibiotics [60 mg/l penicillin, 100 mg/l streptomycin sulfate (PenStrep)] at 37°C under 5% CO₂.

Spheroid formation (3D culture) with 3000 cells per spheroid was induced as described (Korff and Augustin, 1998; Walzl et al., 2014). In brief, cells were detached, counted and seeded in 100 µl of DMEM with 5% FCS, glutamine and PenStrep containing 0.3% methylcellulose into 96-well plates (round bottomed, untreated) for 24 h. In parallel, cells were seeded on cell culture plates (2D culture) and cultivated in the same medium. Cells were cultured at 80% humidity, under 5% CO₂ and 20% O₂ (Walzl et al., 2012). Cultures were treated with different inhibitors for an additional 24 h. Spheroid formation time and/or the presence of methylcellulose did not alter drug response (supporting figures 8 and 9, respectively). Cellular and spheroid morphology was evaluated microscopically and was analyzed (Cell^F, Olympus, Tokyo, Japan) to determine spheroid volume by measuring projected areas, followed by radius and volume (µm³) determination.

For inhibitor treatment in 2D and 3D cultures, 100 nM of rapamycin (Grabiner et al., 2014; Kang et al., 2013; Sarbassov et al., 2006), 250 nM of Torin1 (Cheng et al., 2016; Grabiner et al., 2014; Kang et al., 2013; Thoreen et al., 2012), 10 µM of PF47068671 (Pearce et al., 2010; Schipany et al., 2015; Zhang et al., 2015), 1 µM of MK2206 (Devery et al., 2015; Shen et al., 2015; Sung et al., 2016) and 1 µM of AZD6244 (Devery et al., 2015; Ewald et al., 2015) were employed as these concentrations have been most frequently used in cell-based assays and have a complete effect on their respective targets [phospho-protein analysis in western blots (see Fig. 4 and western blots in the supporting figures)].

Cell cycle analysis was performed with the Click-iT^® EdU Alexa Fluor^® 488 Kit (Invitrogen, Thermo Scientific). Cells and spheroids were exposed to EdU (5-ethynyl-2′-deoxyuridine, 10 µM) at 37°C for 20 min. Thereafter, spheroids were trypsinized and cells were analyzed. 7-aminoactinomycin D (7AAD) was used to stain total DNA and set the forward scatter to determine cell size. The metabolic capacity of cells was determined with CellTiter-Glo^® (Promega) according to the manufacturer's protocol and luminescence measurement (Synergy HT, Biotek^®).

Whole-cell lysates were prepared from 2D or 3D cultures. Cells and spheroids were washed in PBS and extracted in RIPA lysis buffer (50 mM Tris-HCl pH 7.6, 150 mM NaCl, 1% TritonX-100, 0.1% SDS, 0.5% sodium deoxycholate, 1 mM PMSF, 4 µg/ml aprotinin, 4 µg/ml leupeptin, 0.6 µg/ml benzamidinchloride, 20 µg/ml trypsin inhibitor). Supernatants were collected after centrifugation (18,800 g at 4°C for 20 min). Equal amounts of protein (10–15 µg) were mixed with loading dye (200 mM Tris-HCl pH 6.8, 400 mM DTT, 8% SDS, 0.4% bromophenol blue, 40% glycerol), denatured, subjected to SDS-PAGE and transferred to nitrocellulose membranes. Blocked membranes were probed with primary antibodies at 4°C overnight and with horseradish-peroxidase (HRP)-conjugated secondary antibodies (anti-mouse-IgG heavy and light chain, and anti-rabbit-IgG heavy and light chain, Bethyl Laboratories) at room temperature for 1 h. Signals were detected using chemiluminescence and X-ray films. The mean integrated density of each band was determined with ImageJ (National Institutes of Health). For antibodies, see Table S2.

Cells and spheroids were fixed in 4% paraformaldehyde (PFA) at room temperature for 30 min. Fixed spheroids were molded into agarose gels (1%), embedded into paraffin, sectioned (5 µm) and subjected for antigen retrieval in citrate buffer pH 6 (DAKO) at 120°C for 10 min.

For immunohistochemical staining, endogenous peroxidase activity was blocked in 3% H₂O₂ for 10 min. Blocking was performed in horse serum (2.5% in PBS, Vector Laboratories) for 20 min at room temperature. Primary antibody incubation was performed at 4°C overnight and secondary antibody (biotin labeled, Vector Laboratories) was performed at room temperature for 1 h, followed by streptavidin–HRP (Leica Microsystems, Germany) incubation for 30 min at 4°C and chromogenic development (AEC, DAKO). Nuclei were counterstained with hematoxylin, and slides were coverslipped (Aquatex^®, Merck-Millipore). 20× (3D) and 40× (2D) objectives (NA=1.3) were used.

For immunofluorescence, spheroids were handled as described above. In 2D culture, cells fixed in PFA (4%) were permeabilized in methanol at −20°C for 10 min. 2D and 3D culture samples were blocked in PBS with 1% BSA at room temperature for 1 h. Primary antibody incubation was performed at 4°C overnight. Alexa-Fluor^®488- and Alexa-Fluor^®546-conjugated secondary antibodies (Thermo Scientific) were used at room temperature for 1 h. Nuclei were counterstained with 2 µg/ml DAPI (Sigma). Slides were mounted with Vectashield^® (Vector Laboratories). Confocal fluorescence images were recorded on a Leica-SP8 with a 20× immersion objective (NA=1.3).

Image analysis of spheroid sections was performed with StrataQuest software (TissueGnostics, Vienna, Austria). For automated spheroid detection, a virtual channel was created followed by threshold segmentation, generation of a background mask followed by distance transformation. Three intervals were set on this distance-transformed image to correspond to three zones of interest in the spheroid. The outer zone was set to be two-cell-layer thick; the remaining zones were equally divided. Nuclear detection was performed in the DAPI channel, and cellular masks were computed. For each cell, the intensities for the DAPI, phospho-RPS6 and Ki67 channels were calculated and assigned to a certain zone. For quantification of phospho-RPS6 in the areas around cross-sectioned blood vessels in HCT116 and DLD-1 xenografted tumors, the same image analysis was performed except that the zonation was performed by hand. The vessel-proximal zone was defined by a three-cell-layer thickness immediately around the vessel, followed by an intermediate and a vessel-proximal area of similar size to the analyzed spheroids.

Phosphorylation levels of RTKs and other signaling nodes were determined (Cell Signaling Technology). DLD-1 and HCT116 colon cancer cells cultivated in 2D and 3D cultures in normal growth medium (DMEM, 5% FCS) were harvested 48 h after seeding. The assay was performed as described in the manufacturer′s protocol. The arrays were developed using X-ray films; spot intensities were quantified and the background was corrected using ImageJ.

Bar graphs are presented as mean±s.d. or s.e.m. For statistical analysis, Student's t-test (unpaired, two-tailed) was performed. P-values are indicated as *P≤0.05, **P≤0.01, ***P≤0.001 or ****P≤0.0001. For exact P-values of all experiments, see Table S3.

We thank Dr Julia Schüler (Oncotest, Freiburg, Germany) for xenograft tumors and Eduard Clim (TissueGnostics Vienna) for help with StrataQuest. Some of this work is also presented in the thesis of Angelika Riedl (2017, Evaluating PI3K/AKT/mTOR and MAPK signaling in 3D colon cancer models to identify effective combination therapy approaches, Medical University of Vienna).