Matrix composition and glucose availability cooperatively determine cancer spheroid bioenergetics in 3D hydrogels.

Introduction
The tumor microenvironment (TME) presents a dynamic and complex landscape where cancer cells constantly integrate diverse biochemical and biophysical signals that drive them toward metabolic reprogramming, a hallmark of cancer [1, 2]. The main metabolic alterations described are the use of opportunistic modes of nutrient acquisition, such as glucose [3] and glutamine [4, 5], fueling biosynthesis and production of many crucial cell biomolecules used for cell growth. Critical interconnected pathways of glycolysis include the pentose phosphate pathway, producing necessary nicotinamide adenine dinucleotide phosphate (NADPH) used in lipid production, as well as an increased glutamine uptake supporting the increased nitrogen demand of a fast-growing cell. Alterations in metabolite-driven genes, dysregulated glucose and amino acid uptake, all depend on the metabolic interactions with the microenvironment [2]. Within this complex system, the extracellular matrix (ECM) is increasingly recognized not just as a structural scaffold but as an active participant, dictating cellular behavior through its distinct physical and chemical properties [6–8]. One of the most well-documented alterations in the tumor microenvironment is an aberrant increase in ECM stiffness, largely driven by excessive collagen deposition, crosslinking, and remodeling [1, 9]. This stiffening is known to activate mechanotransduction pathways that regulate oncogenic signaling and can significantly rewire cellular metabolism, often boosting glucose uptake and glycolytic flux [1, 10–13].
While the impact of ECM stiffness on cancer metabolism is well-documented, the native ECM is a heterogeneous structure characterized by diverse compositions (e.g., collagen-rich stroma versus fibrin-rich provisional matrices) and intricate microarchitectures. These compositional variations offer unique biophysical cues beyond bulk stiffness, including ligand presentations, degradability, porosity, nutrient diffusion, and viscoelasticity. However, when cells encounter fluctuating nutrient availability, the impact of these compositional variations on metabolic programming remains less understood [14, 15]. Understanding how cancer cells differentially interpret and respond to combined signals from distinct ECM components and variable nutrient landscapes, including glucose gradients, is crucial. Moreover, this knowledge is essential for developing more physiologically relevant in vitro models that can accurately mimic and predict in vivo behavior and for identifying context-dependent metabolic vulnerabilities.
Previous work has underscored the profound metabolic differences between traditional (two-dimensional) 2D and (three-dimensional) 3D models that better recapitulate TME architecture and cell-ECM interactions [16–21]. Mechanically tunable hydrogels, especially those derived from natural ECM components like collagen and fibrin, serve as powerful 3D platforms to systematically study these complex interactions [22–25]. Collagen type I, a primary structural protein in many dense tumors, and fibrin, a key component of provisional matrices formed during wound healing, inflammation, and early tumor development, have inherently different mechanical properties and engage distinct cellular pathways [26–29]. A central, yet unexplored, question is how these ECM biomaterials distinctly modulate the metabolic hierarchy when cancer cells face varying glucose levels simultaneously. The degree to which nutrient availability overrides matrix-derived cues—and how this balance differs across cancer types from tissues with divergent mechanical properties—remains an important gap in our understanding of tumor metabolism.
In this study, we systematically investigate how ECM composition (collagen type I vs. fibrin), concentration-dependent biophysical properties (stiffness and microarchitecture), and glucose availability (high, low, or absent) collectively dictate the metabolic phenotype of lung adenocarcinoma (A549) and pancreatic cancer (Panc1) spheroids. These cell lines were chosen due to their distinct tumor origins – lung tissue being relatively soft and elastic [30], while pancreatic tissue is often characterized by a stiffer and denser stroma [31] – potentially priming them for different baseline responses to mechanical and metabolic cues. By integrating comprehensive mechanical characterization of the hydrogels with detailed metabolic profiling (assays using Seahorse flux analysis and metabolite quantification) of spheroids formed within the different matrices, we aim to identify distinct, material-dependent adaptation. Our findings reveal that the hierarchy of influence between ECM properties and glucose availability is critically dependent on the ECM material, highlighting the importance of considering both biomechanical and biochemical ECM factors when studying cancer metabolic reprogramming, offering new perspectives on the role of diffusion and nutrient accessibility in these processes.

Results

Collagen and fibrin hydrogels stiffness depends on polymer concentration
The mechanical properties of the different hydrogels were determined using rheology measurements of the varying concentrations of collagen and fibrin. For both matrices, a “high”, “medium” and “low” concentration was chosen, representing 6, 4 and 2.5 mg/mL for collagen and 5.6, 3.9 and 2.2 mg/mL for fibrinogen, respectively. Initially, a Temperature Sweep (Fig. 1a) was performed to analyze the effect of temperature on hydrogel polymerization. In this way, we considered the gelation point the intersection between Storage (G’) and Loss (G’’) modulus, where the G’ remained higher than G’’. In collagen hydrogels (Fig. 1a, upper row), gelation occurred after reaching 30 °C, with a slight increase in temperature directly related to the concentration. The slightly higher polymerization temperature correlates to a lower amount of collagen in the mixture. However, as fibrin polymerizes based on an enzymatic reaction with thrombin and fibrinogen, this crossing point of G’ and G’’ was not present in the temperature sweeps (Fig. 1a, lower row) [32].
After polymerization, hydrogels were subjected to constant strain for one hour to determine when the polymerization finished and mechanical properties stabilized. In collagen hydrogels, there was an increase in the storage modulus (G’) during the first minutes of measurement compared to the other matrix formulations (Fig. 1b, left). We found that the storage modulus of the polymerized gels depended on the collagen concentration within the studied range. Although this concentration dependence was also noticeable in fibrin hydrogels (Fig. 1b, right), where high fibrin had the highest storage modulus, the stiffness achieved was lower than that of high and medium collagen samples (Table 1).
Finally, we measured both storage and loss modulus during an oscillatory stress sweep assay. The initial storage modulus value was similar to the one recorded during the final stage of the previous step at small strains. From a strain of around 20% all the tested collagen hydrogels began to show strain-stiffening (Fig. 1c), while fibrin hydrogels exhibited this effect earlier, at 7% of strain. As the strain increased, the storage modulus stabilized, peaking for a brief period before decreasing due to hydrogel rupture. Interestingly, although fibrin hydrogels presented a lower stiffness, they exhibited a markedly higher capacity for stiffening, reaching exceptionally high values for both storage and loss modulus (Table 1).

Using these rheology data, we calculated the mesh size of the different hydrogels. This parameter quantifies theoretically the distance between crosslinks in the polymer network and the increase of mesh size is typically inversely proportional to the polymer density or the crosslinking concentration [14, 23]. We calculated the average mesh sizes for collagen and fibrin hydrogels based on shear rheology data and found values between 95 and 34 nm for gels prepared with collagen and between 134 and 74 nm for fibrinogen hydrogels, as shown in Table 1.

Tumor cells alter collagen hydrogel stiffness more than fibrin hydrogels
To study how cells modify the mechanical properties of the hydrogels, the elastic modulus of all collagen and fibrin hydrogels with and without cells was measured after days 2 and 10 of the assay. In addition, at day 10, a comparison between the effects of cells grown at either high or no glucose in the cell media, together with the matrices, was established. The elastic modulus (kPa) of each hydrogel under the corresponding condition is presented in Fig. 2, and shows that overall, collagen hydrogels exhibited a higher elastic modulus compared to fibrin hydrogels across all conditions. On Day 2, the high collagen hydrogels showed the highest stiffness in the Control condition (40.12 ± 0.00 kPa), whereas reducing the collagen concentration to 2.5 mg/mL significantly decreased the modulus (24.22 ± 0.50 kPa) (Table 2). The presence of A549 or Panc1 cells further reduced hydrogel stiffness at day 2, particularly in the lowest collagen concentration, where gel breakage occurred, hindering measurements. Remarkably, by day 10, collagen hydrogels containing spheroids showed a reduction in modulus with respect to the control in both glucose conditions (Fig. 2a). In the medium collagen matrix, Panc1 exhibited a higher tendency to remodulation, decreasing the elastic modulus. Interestingly, this trend was reversed in low collagen hydrogel, where A549 exhibited greater remodeling compared to the acellular control. Notably, the 2.5 mg/mL collagen acellular hydrogels partially recovered their stiffness in both glucose conditions (26.65 ± 1.90 and 28.46 ± 0.00 kPa).
For fibrin hydrogels (Fig. 2b), the elastic modulus values were generally lower than those of collagen, ranging from 30.79 ± 0.73 kPa to 25.40 ± 1.95 kPa (Table 2), depending on concentration and cell presence. Compared to collagen hydrogels, fibrin ones exhibited less variation in stiffness between Day 2 and Day 10 with and without cancer cells, exhibiting lower degradation or remodeling in the presence of tumor cells. However, a key point to highlight is that glucose affected the elastic modulus, as acellular fibrin hydrogels were softer without glucose compared to high glucose at all fibrinogen concentrations. Notably, spheroids caused a slight stiffening effect, which was more evident in Panc1 cells.

Matrix composition affects cell proliferation and spheroid formation capacity
After characterizing the biomechanical properties of the matrices, our aim was to understand how these different environments, together with variations in glucose availability, can influence cellular behavior. We studied the proliferation of two cancer cell lines, derived from tissues with differing biomechanical properties, using three distinct glucose conditions and six different matrix composition. A549 and Panc1 cells were maintained for 10 days in culture before proliferation measurement. Figure 3a shows brightfield images of A549 at day 10 in the different glucose and matrix conditions, while Figure S1 displays the corresponding day 5 images to allow comparison of spheroid evolution over time. Qualitative assessments show that cells in collagen hydrogels did not appear different across the different glucose conditions, however there seemed to be a slight decrease in growth as the gel stiffness decreased, confirmed by the quantitative data (Fig. 3b). Furthermore, cell organization seemed more compact at higher stiffnesses and more disorganized at lower stiffnesses. Interestingly, quantitative data (Fig. 3b) showed that the A549 cells seem to prefer growth in medium collagen hydrogels in both low and no glucose conditions, exhibiting increased cell proliferation. Panc1 showed similar proliferation patterns depending on matrix stiffness (Fig. 4a), with smaller and more disorganized spheroids in softer collagen hydrogels. Additionally, there seemed to be a slight decrease in proliferation related to glucose availability, with significant differences between collagen hydrogels in the absence of glucose (Fig. 4b). However, the major differences in cell growth were seen in fibrin hydrogels, regardless of glucose conditions. While fibrinogen concentration was increased, a small number of A549 spheroids could be detected, although their occurrence was markedly lower compared to collagen hydrogels. At lower fibrinogen concentrations, cells exhibited a strong tendency toward 2D spreading, and no spheroid formation was detectable, only occasional 3D structures were observed (Figure S2). Interestingly, we found that reduced glucose levels slightly favored the appearance of some spheroids, but A549 spheroid formation in fibrin hydrogels remained overall very limited. In contrast, Panc1 cells formed clearer and bigger spheroids, without 2D cell dissemination in fibrin hydrogels (Fig. 4a). It should be noted that softer fibrin hydrogel led to larger spheroids and a clear preference for the low glucose condition was detectable, as reflected in the quantitative data (Fig. 4b). Further characterization of spheroids confirmed our main findings: A549 cells formed more but smaller spheroids in collagen than Panc1, whereas Panc1 formed more in fibrin, although both cell lines produced a higher total number of spheroids in collagen. A549 showed the highest spheroid numbers under high glucose regardless of collagen concentration, and under low collagen regardless of glucose availability. As expected, fibrin supported very few and very small spheroids. Panc1 formed the fewest spheroids in low collagen, with optimal formation at 1 g/L glucose, consistent with previous observations (Figure S3). Regarding spheroid area, A549 followed the same pattern observed for metabolically active cells measured by Alamar Blue, but in fibrin the 2D growth artifact was eliminated, revealing a clear and significant reduction in spheroid size with decreasing fibrin concentration. Panc1 displayed the opposite trend, with the smallest spheroids showing the highest metabolic activity (Figure S4).

Glucose availability shapes the metabolic phenotype of spheroids formed in collagen hydrogels more than matrix stiffness
After matrix characterization and analysis of corresponding cell proliferation, both cell lines were analyzed for real-time metabolic flux using the Seahorse XFe96 system. Oxygen consumption rates (OCR) and extracellular acidification rates (ECAR) data from the Cell Mito Stress Test were analyzed to get key metabolic metrics, providing further insight into the phenotypic characteristics of the cell lines. In Fig. 5a, A549 phenotypic energy maps (upper row) show OCR over ECAR from initial basal respiration measurements to examine how collagen hydrogel stiffness and glucose growth conditions influence mitochondrial responses in the cancer spheroids formed. Here, A549 spheroids adapt their metabolic phenotype in response to glucose availability, displaying a more glycolytic profile when formed in high and low glucose, whereas in the absence of glucose, the cells reverted to a quiescent metabolic state. Although the different collagen concentrations had minimal impact on A549 metabolic phenotype, a subtle stiffness-dependent effect was observed in high and low glucose settings, promoting a shift toward a more glycolytic state in the highest collagen concentration. The different OXPHOS parameters presented in the lower bar charts (Fig. 5a), show a collagen-dependent shift in metabolic activity. No significant changes in ATP-linked respiration or total respiratory capacity (TRC) were detected in either glucose condition. However, although a modest decrease in TRC and spare capacity was observed when formed in medium collagen at high glucose levels, this reduction was only significant in the high glucose condition. This trend changed in no glucose conditions, where the decreasing collagen concentrations caused a reduction in TRC and spare respiratory capacity, with spare capacity significantly affected. Expanded OCR and ECAR traces of A549 cell line for all conditions are shown in Figures S5 and S6, respectively.
Examining the Panc1 energy map (Fig. 5b, upper row), a similar pattern as A549 was observed, with stiffer collagen hydrogels promoting a more aggressive phenotype, except in no glucose conditions. However, as reflected in the top metabolic phenotype maps, pancreatic spheroids were more glycolytic than the A549 lung adenocarcinoma spheroids. It is worth noting that Panc1 grown in low glucose conditions presented the largest difference, where all collagen conditions resulted in a highly energetic phenotype. The difference could be attributed to near-complete oxygen depletion during measurements due to high oxygen consumption by the cells during the basal measurement cycle. As oxygen consumption rates were high, it caused chamber oxygen levels to drop rendering cells unable to increase respiration upon FCCP addition. This was further supported by stable ECAR values throughout the assay (Figure S8). Consequently, in the lower bar charts (Fig. 5b), all Panc1 cells in low glucose collagen conditions resulted in low and non-differentiable ATP-linked oxygen consumption, TRC and spare capacity, explaining why spare capacity was negative. Interestingly, OCR (Figure S7) and respiratory parameters remained unchanged in Panc1 grown in medium collagen without glucose, with significant differences in comparison to the other two collagen conditions, except in the ATP-linked oxygen consumption. However, ECAR (Figure S8) for this condition differed from low glucose, being overall lower indicating a more aerobic metabolic phenotype. In high glucose conditions a significant reduction in TRC and spare capacity for medium collagen matrices were found, similar to what was observed in no glucose conditions.

Fibrin hydrogel concentration and glucose availability equally shape the metabolic phenotype of cancer spheroids
Building on the findings observed in collagen hydrogels, we next analyzed the metabolic behavior of spheroids formed in fibrin matrices using Seahorse assays to assess how both fibrin concentration and glucose availability influence their oxidative and glycolytic activity. In Fig. 6a (upper row), the A549 energy map show initial basal measurements to see the direct effect of fibrin hydrogel stiffness and glucose growth conditions on cancer spheroids formed. Here, a clear difference was observed with respect to collagen hydrogels, where cells displayed a quiescent phenotype in fibrin hydrogels, similar to what was found in collagen when glucose was absent from the media. Interestingly, as opposed to collagen, the cells grown in the low fibrin condition always had a slightly more glycolytic phenotype than the other fibrin matrices. Analyzing OXPHOS capacity, spheroids formed in fibrin matrices showed more pronounced metabolic changes than in collagen (Fig. 6a, lower row) with spheroids formed in medium fibrin hydrogels with high glucose exhibiting higher ATP-linked respiration levels, despite the TRC and spare capacity being the lowest. This effect was opposite in A549 spheroids formed in low glucose, where ATP-linked oxygen consumption was increased with decreasing fibrin concentration, but medium fibrin showed clearly higher levels in TCR and spare respiratory capacity. Without glucose, overall spheroid oxygen consumption rates decreased, although a slight increase was measured in association with lower fibrin concentration hydrogels. OCR and ECAR traces of the A549 cell line under all conditions are shown in Figures S5 and S6.
The Panc1 energy map (Fig. 6b, upper row) displayed a similar pattern as the A549, with fibrin hydrogel leading to less aggressive metabolic phenotypes than collagen ones. Examining the effect of fibrin hydrogels on mitochondrial capacity (Fig. 6b, lower row), no significant differences were observed across fibrin concentrations at high glucose levels. ATP-linked respiration, TRC, and spare capacity remain relatively stable, indicating that fibrin does not strongly influence respiration in this condition, although medium fibrin hydrogels produced a non-significant increase in all the respiratory parameters. Similar patterns appeared in low glucose, with mitochondrial respiration favored in low fibrin hydrogels. In contrast, significant differences emerged under glucose deprivation, particularly in TRC and spare capacity, where low fibrin concentrations showed a marked increase compared to medium and high fibrin conditions. ATP-linked respiration was also higher in low fibrin conditions, although the effect was less pronounced.

Metabolite analysis reveals that matrix stiffness promotes oxidative metabolism in pancreatic cancer spheroids under glucose deprivation
To provide further insight into the metabolic phenotypes observed in the Seahorse assays, the consumption of glucose and glutamine, along with lactate production, was measured in the culture medium. These measurements help characterize the metabolic preferences of each condition and establish a link between extracellular metabolite utilization and mitochondrial function. To this aim, medium was collected at the endpoint (day 10) after 48 h in contact with the cell culture, and metabolites were quantified. Lactate production was assessed explicitly due to its role as the main byproduct of anaerobic glycolysis, while glutamine was analyzed as a crucial alternative fuel for the Krebs cycle in cancer cells [33]. Attending to A549 glucose consumption (Fig. 7a, left row), glucose consumption per cell (cells were counted as shown in Figure S10) in high-glucose media was matrix-dependent, with the highest uptake observed in the stiffest hydrogel. In a low glucose condition, no significant differences were observed. On the contrary, in fibrin hydrogels, the glucose consumption increased as the fibrinogen concentration decreased. Panc1 expressed the same pattern when grown in low glucose and a similar one at high glucose, with a substantial change in high fibrin hydrogels, probably due to a low number of metabolically active cells, as seen in the previous section.
Glutamine measurements indicated production instead of consumption, resulting in negative values in almost all the measured conditions for both cell lines (Fig. 7b). This correlates with available glucose levels, as an increase in glutamine production was observed under glucose-restriction. A549 spheroids exhibited high glutamine production, whereas Panc1 spheroids had lower levels. Notably, glutamine production increased in conditions where cells consumed the least glucose. In Panc1, almost no differences in matrix stiffness were observed, although a slight increase in glutamine production was measured in the glucose-deprived conditions.
The matrix stiffness effect was more pronounced in lactate production (Fig. 7c), with A549 spheroids having higher levels than Panc1. Spheroids cultured in high glucose exhibited a strong dependence on matrix stiffness, with markedly high lactate production in the stiffest collagen hydrogels, while in the softest conditions, lactate appeared to be consumed for both cell lines. This trend was similar in low glucose conditions for A549 but not for Panc1, which rather showed a significant lactate peak in low fibrin hydrogel. In glucose deprivation, A549 and Panc1 exhibited completely different behaviors, whereA549 produced up to 4 ng lactate/cell in low collagen hydrogels, whereas the maximum production of Panc1 was 1 ng lactate/cell in both high collagen and fibrin matrices.

Discussion
This study suggests that the physical properties of the extracellular matrix (ECM) are a dominant regulator of cancer cell metabolic programming, capable of overriding other specific cell intrinsic tendencies [25, 34, 35]. By decoupling matrix stiffness from polymer composition, we demonstrate that while lung adenocarcinoma cells (A549) are acutely sensitive to mechanical cues, pancreatic (Panc1) cancer spheroids deploy a rigid metabolic program that is affected under extreme nutrient deprivation, providing a new framework for understanding metabolic heterogeneity in solid tumors, particularly under nutrient-limiting conditions.
The biomechanical properties of the ECM are critically influenced by both material type and polymer concentration, shaping how cancer cells interact with their environment. In our study, collagen and fibrin hydrogels displayed distinct mechanical characteristics, with increasing concentrations resulting in higher stiffness and reduced mesh size, consistent with prior reports [22, 32, 36, 37]. Despite fibrin’s lower stiffness compared to collagen, its high resistance to breakage and more elastic behavior make it an attractive alternative for 3D culture systems. Importantly, mesh size emerged as a key factor, regulating the diffusion of nutrients and oxygen and thereby impacting cell proliferation and drug accessibility [14, 23, 28]. Although absolute mesh size values varied from those in literature (ranging from nanometers to micrometers depending on measurement techniques), the observed inverse relationship with stiffness held consistently [14, 23, 28, 32, 38, 39]. While in our study mesh size was estimated indirectly from rheological data, more direct methods such as SEM imaging or DNA diffusion assays are commonly used to assess this parameter at different scales. Notably, Xia et al.., successfully decoupled mesh size from hydrogel stiffness and composition, confirming its critical role in regulating molecular diffusion and cellular behavior, and underscoring its importance in cell-substrate interactions [40]. Our data support the idea that larger mesh sizes, as observed in fibrin hydrogels, enhance permeability and facilitate diffusion [24, 41, 42], emphasizing the importance of ECM composition in determining the biochemical landscape encountered by embedded cells. Beyond defining passive properties such as stiffness and mesh size, ECM composition also determines how cells actively remodel their microenvironment. Panc1 cells altered high collagen concentration hydrogels more substantially, while A549 cells had a greater impact on low collagen concentration ones, suggesting cell-specific differences in the balance between matrix degradation and contraction [43]. In contrast, fibrin matrices were minimally affected, reinforcing their structural resilience [44]. Key elements influencing ECM remodeling include the activity of matrix metalloproteinases (MMPs) and their tissue inhibitors (TIMPs). MMP-mediated proteolysis could contribute to local matrix softening, whereas TIMPs may modulate this process, and their expression and activity may be influenced by the metabolic phenotype of the tumor cells [45]. The lack of direct assessment of these pathways represents a limitation of the present study; however, previous reports have shown that MMP/TIMP balance can drive matrix softening or stiffening, suggesting that these mechanisms could underlie the cell-specific remodeling patterns observed here [46, 47]. Together, these findings highlight that matrix remodeling is the result of a dynamic interplay between ECM properties, cell behavior, and environmental cues like glucose availability.
Proliferation results showed that A549 proliferation, under glucose-limiting conditions, was favored in medium-collagen matrices, as high-collagen limited nutrient diffusion. In contrast, low-collagen or fibrin hydrogels promoted cell migration [48] or even 2D growth via durotaxis, a phenomenon whereby cells preferentially proliferate on stiffer substrates [49, 50], making quantitative comparisons with 3D spheroids less straightforward. On the contrary, Panc1 cells formed better spheroids in low stiffness due to better nutrient diffusion and resistance to degradation, but in collagen hydrogels, they appeared more dispersed as degradation of the matrix results in better nutrient access [51–53].
Seahorse analysis revealed distinct responses depending on matrix composition and stiffness, highlighting the interplay between mechanical properties and cellular metabolism. Notably, fibrin hydrogels led to a less aggressive metabolic phenotype compared to collagen, which was particularly evident under glucose deprivation. This suggests that matrix biomechanical properties not only affect cellular proliferation but also metabolic plasticity. Several studies have used the same technique to explore the role of the ECM [34, 54]. Our study extends these previous findings by being the first to investigate the metabolic effects of pure fibrin hydrogels. Interestingly, rather than a straightforward correlation between ECM stiffness and increased oxidative phosphorylation (OXPHOS), as reported in previous studies [34, 55, 56], our results suggest a more complex relationship. In fibrin, the softest matrix, there was a tendency toward a more aerobic metabolic profile. However, this trend was not consistently observed within each ECM material, as glucose availability played a dominant role in shaping metabolic behavior. Under metabolically stressful conditions, such as glucose restriction, ECM properties exerted a more pronounced influence, promoting greater oxidative capacity in medium collagen and low-concentration fibrin hydrogels. This highlights that while matrix composition influences metabolism, the interplay between ECM properties and nutrient availability is crucial in determining the metabolic phenotype of cancer spheroids.
Glucose availability emerged as a key determinant of metabolic behavior, with spheroids formed under low glucose conditions exhibiting the most aggressive phenotype in both cell lines. This is likely due to metabolic reprogramming triggered by nutrient restriction, which forces cancer cells to maximize glycolytic flux and alternative energy pathways to sustain proliferation and survival, driving a more invasive and treatment-resistant phenotype [57–59]. Metabolite analysis supports this hypothesis with previous studies reporting that increased ECM stiffness is associated with higher glucose consumption, enhanced glycolysis, and metabolic adaptations such as increased lactate production and glutamine anaplerosis [60, 61]. However, we also observed that in extremely soft environments, such as fibrin hydrogels, glucose consumption was high, probably due to improved diffusion and permeability, but, unlike in stiff conditions, this increased glucose uptake did not correspond with a shift towards a glycolytic phenotype, as no parallel increase in lactate production or glutamine anaplerosis was detected.
Panc1 spheroids, known for their aggressive and glycolytic phenotype [20, 42, 62–64], showed to be more aggressive and metabolically active across all matrix stiffness conditions when formed in low glucose. This favored growth condition further resulted in high initial oxygen consumption rates during the Seahorse assay, thereby limiting the available oxygen levels to test for their maximal respiratory capacity. Our results suggest that Panc1 cells are less dependent on ECM stiffness for metabolic adaptation, unlike A549 cells, which showed more pronounced changes. This aligns with the research of Tilghman et al., who classified A549 cells as “rigidity-dependent”, unlike pancreatic cancer cells, which appear metabolically rigid [15, 65]. In addition, Panc1 cells exhibited the highest degree of dispersion in all collagen matrices at low glucose, suggesting increased migratory behavior. Previous studies have shown that invasive Panc1 cells adopt distinct metabolic profiles with enhanced oxidative stress resistance [66] and promote aerobic glycolysis by YAP activation [67]. Our findings further support this adaptation, as Panc1 cells produced the highest levels of glutamine and lactate, consistent with metabolic plasticity. However, under glucose deprivation, they shifted to a more oxidative phenotype, as indicated by reduced lactate production and OCR/ECAR profiles. Interestingly, peaks of lactate production were observed in the softest matrices, particularly in Panc1 spheroids cultured in the highest fibrin concentration hydrogels. These peaks correlated with an increase in glutamine production, suggesting that in soft environments and under metabolic stress, cancer cells may activate alternative anaplerotic pathways and potentially use lactate produced through glycolysis to refuel the TCA cycle—even more than glucose—in well-oxygenated tumor regions by lactate import through the monocarboxylate transporter-1 (MCT1) [68–70]. This underscores the context-dependent nature of metabolic adaptation, where not only matrix stiffness but also nutrient availability and cell type define the resulting phenotype. In addition to metabolic changes, matrix–cell interactions and mechanotransduction pathways likely play an important role in shaping the observed responses. Mechanical cues transmitted through integrins and the cytoskeleton can activate YAP/TAZ signaling, leading to transcriptional reprogramming that affects proliferation, survival, and metabolism [13, 43, 71, 72]. As reported by Sorzabal-Bellido et al., some important genes affected by alterations in matrix biomechanics are E-cadherin (CDH1), N-cadherin (CDH2), vimentin (VIM) and ZEB1 [73]. Future work integrating transcriptional and epigenetic analyses could clarify how these mechanoresponsive pathways contribute to the metabolic phenotypes reported in our study.

Conclusion
Our study highlights the intricate relationship between extracellular matrix composition, concentration and stiffness, glucose availability, and cancer cell metabolism. We found that higher collagen concentrations promote anaerobic metabolism, particularly in Panc1 cells, while increased fibrin concentration suppresses mitochondrial respiration, especially under glucose deprivation. Notably, Panc1 cells exhibit metabolic rigidity, especially in low glucose conditions, maintaining their invasive potential regardless of mechanical constraints, whereas A549 cells display a more dynamic metabolic response to ECM stiffness. Nevertheless, metabolic patterns expressed by both cell lines are similar. These findings further support the role of mechanical stress in shaping cancer cell metabolism and emphasize the importance of integrating mechanical and metabolic factors in 3D tumor models, suggesting that targeting ECM properties could offer new therapeutic opportunities for highly glycolytic cancers. In conclusion, the biophysical properties of the ECM are a pivotal factor influencing the metabolism of cancer spheroids, although the metabolic plasticity of cancer cells is also strongly influenced by glucose availability.

Methods

Cell culture
Human lung adenocarcinoma cell line (A549) was obtained from the American Type Culture Collection (ATCC, USA) and human pancreatic cancer cell line (Panc-1) was generously provided by collaborators at the Stavanger University Hospital Molecular Biology Lab. Both cell lines were maintained at 37 °C in a 5% CO2 humidified incubator, in Dulbecco’s modified Eagle’s medium high glucose (DMEM) (Gibco) supplemented with 10% heat-inactivated fetal bovine serum (FBS) (Gibco) and antibiotics (penicillin, streptomycin) (Biowest). Cells were trypsinized/passaged every 2–3 days using Trypsin 0.25% in PBS (Biowest). When cells reached 80% confluence, they were counted using Muse© Cell Analyzer and diluted to the desired experimental density.

Hydrogels fabrication and cell seeding
The collagen hydrogels were made following the protocol established by [74]. In accordance with this method, the hydrogel was created by combining 10X DPBS (Thermo Fisher), DMEM, collagen (Corning) to reach the final desired concentration and 0.5 M NaOH (Sigma Aldrich) for pH adjustment to 7.5. Cells were added individually to the mixture to reach a final concentration of 100,000 cell/mL. This mixture was introduced in drops of 5 µL into 48 well-plates and allowed to polymerize at 37 °C in a humid environment (Figure S9). The hydrogels used for this study were at a concentration of 6, 4 and 2.5 mg/mL (high, medium and low, respectively) of collagen.
For fibrin hydrogel studies, fibrinogen (Merck Life Science) was dissolved in 0.9% NaCl (Sigma) and combined with an individual cell suspension at a density of 55,556 cells/mL. Aprotinin (25 µg/mL) (MP Biomedicals) was added to prevent degradation, along with thrombin (1 U/mL) (Merck Millipore Calbiochem) and PBS. The final fibrinogen concentrations used were 5.6, 3.9, and 2.2 mg/mL (high, medium and low, respectively). The mixture was dispensed in 9 µL droplets into 48-well plates and allowed to polymerize at 37 °C in a humidified environment (Figure S9).

Rheology
The rheological experiments were carried out in collaboration with the Norwegian Institute of Food, Fisheries and Aquaculture Research (NOFIMA). For these measurements a hybrid rheometer (Discovery HR-2, TA Instruments, Newcastle, UK) fitted with 40nn parallel plate was used. The samples were measured without cells, replacing that volume with DMEM media. A total volume of 500 µl was used per sample. Each condition was measured in triplicate. The mixture was prepared on ice and loaded on the Peltier plate with set temperature of 4 °C at the gap of 500 μm. To avoid sample evaporation, a solvent trap filled with distilled water was placed on top of the geometry. Temperature Sweep measurement was performed at the temperature range from 4 °C to 37 °C with constant strain (0.1%) and angular frequency (10 rad/s). Following, a Time Sweep of 60 min was performed at constant strain at 0.5%. The last step was a Strain Amplitude Sweep where the strain was increased from 0.05 to 1000% at constant frequency of 0.1 Hz. Due to the low viscosity of the fibrin hydrogels, a Torque control of 0.1 µN·m was added in the Temperature Sweep step.

Mesh size calculations
We determined the mesh size, ξ, of both hydrogels by the Eq. 1 [14], which relate the shear modulus (G) to the mesh size following the classical theory of rubber elasticity:
Where R is the gas constant, T is the absolute temperature, and NAv is Avogadro’s number. Experimentally, this relation allowed us to estimate the mesh size based on the measurement of the shear modulus of a hydrogel. For our analysis, we considered the shear modulus (G) to be equivalent to the storage modulus (G′) within the linear viscoelastic region.

Texture analyzer
These assays were carried out in collaboration with the Norwegian Institute of Food, Fisheries and Aquaculture Research (NOFIMA). As control, acellular collagen and fibrin hydrogels were measured 48 h after their polymerization. In addition, hydrogels with both cell lines 48 h after the seeding were measured. Hydrogels with and without cells were incubated with 4.5 and 0 g/L glucose for 10 days, when they were measured again in order to study the effect of time, glucose and cells over hydrogel’s Young’s Modulus. Samples were drops of 340 µL in 6 well-plates. A predefined project in the TA.XT PlusC Texture Analyzer (Stable Micro Systems Ltd, Godalming, UK), based on the Kobe method, was used to determine the Elastic modulus (E) of the different hydrogels via confined compression-stress testing. A 5 Kg load cell and a 5 mm diameter stainless steel spherical probe (P/05S) were used, with the probe height calibrated to the bottom of a 96-well plate before each experiment. Force calibration was performed using a 2 kg weight. The test was conducted at a speed of 2 mm/sec, with a predefined start position ensuring consistent compression distances across samples. Test parameters, including stress area, strain height, and data acquisition rate, were defined in the software. Each sample was placed under the probe, and force-distance data were recorded. Macro settings were applied to calculate the gradient of the linear elastic region, obtaining Elastic modulus (Pa) from stress-strain data through the existing project and macro settings in the Texture Analyzer.

Metabolically active cells measurements
For these assays, cells were maintained in the 3D culture for 10 days, with media change every 2 days using different medium according to every condition: DMEM no glucose, DMEM low glucose (1 g/L) or DMEM high glucose (4.5 g/L). DMEM no glucose was complemented with pyruvate at same concentration than the low and high glucose media (1 mM) before treatment. To characterize cell proliferation in 3D models under every glucose condition, spheroids growth formation was evaluated using a Leica TCS SP8 FALCON Lifetime Confocal Microscope. Endpoint snapshots of the hydrogels were captured under brightfield illumination at a magnification of 4X. For quantitative measurements, Alamar Blue reagent (Thermo Fisher Scientific) was used mixed with normal DMEM medium (1:9). After 4 h of incubation at 37 °C, the mixture was collected from the wells and its fluorescence was measured using a plate reader with a fluorescence excitation wavelength of 540–570 nm and emission at 580–610 nm. To establish the relationship between Alamar Blue fluorescence and cell number, a calibration curve was constructed across a range from 30,000 to 0 cells for each cell line, as depicted in Figure S10.

Metabolic flux assays
Mitochondrial respiration and extracellular acidification rates were measured through a Cell Mito Stress Test using the Seahorse XF96e and XFp flux analyzers (Agilent). Before performing the assay, FCCP titration was conducted over a range of concentrations, in both 2D and 3D cultures, following the Seahorse cell characterization procedure. The FCCP concentration that yielded the maximum OCR value was selected for each cell line (Figure S11). A549 and Panc1 were cultured in the same conditions as described in the previous section. They were maintained for 10 days in different matrices of collagen and fibrin with high, low or no glucose in complete DMEM media in 48 well-plates. For Seahorse measurements, specific assay media was made composed of unbuffered, serum-free DMEM (Sigma-Aldrich) with 2 mM pyruvate, 2 mM l-glutamine, and 25, 5 or 0 mM glucose additions, and pH was adjusted to 7.4 before running the experiment.
The ECM domes, containing the spheroids formed over 10 days, were transferred and placed in the centers of the wells in a 96 well Seahorse assay plate pretreated with poly-D-lysine (PDL), before 180 µL assay media was added. Plates were kept in a CO₂-free incubator at 37 °C for 1 h before running the assay. OCR and ECAR were measured over 150 min, consisting of 15 mix-and-measure cycles programmed to be used with spheroid measurements. During the Cell Mito Stress Test, the following compounds were injected sequentially (final concentrations in the wells): oligomycin after cycle 3 (5 µM), FCCP after cycle 6 (2 µM), rotenone and antimycin A after cycle 12 (2 µM). To use the data for comparison of metabolic phenotypes across cell models regardless of absolute metabolic activity, OCR and ECAR were normalized to their own individual basal levels measurements.

Metabolites quantification
Culture medium was collected at endpoint after 48 h in contact with the cells, taking into account that the experiment finished at day 10 and last media change was at day 8. Glucose remaining was measured with a glucometer GlucCell (Cesco Bioengineering). Each measure was repeated twice and any effect of evaporation on concentration was corrected by taking into account the variation of total volume. Lactate was assayed using the Glycolysis Cell-Based Assay Kit (Cayman Chemical) according to the manufacturer’s instructions. Glutamine was analyzed using the Glutamine Assay Kit (Abcam). Glucose and glutamine consumption, as well as lactate production, were determined through a mass balance analysis at the measured timepoint, relative to a control well without cells maintained in the same condition as the experimental ones. Metabolite production and consumption rates were normalized by dividing by the number of metabolically active cells, assessed using the Alamar Blue assay.

Image acquisition and analysis
The spheroid growth was monitored daily with a Leica DMLS Binocular microscope at 4X magnification in brightfield. Endpoint pictures were taken with a Leica TCS SP8 FALCON Lifetime Confocal Microscope. Later, these images were processed and the spheroids area was segmented with the semiautomatic Segmentation3D App developed by C. Borau using MATLAB (Mathworks, Natick, CA, US) as described by Alamán-Díez et al.24. The data obtained were processed and represented using GraphPad Prism 8.

Statistical analysis
All experimental conditions were tested in triplicate, except for the Texture Analyzer assay and metabolite measurements, which were performed in duplicate due to limitations in reagent availability. Statistical analysis was conducted using GraphPad Prism 8 and expressed as the mean ± standard deviation (SD) where possible. Normality of the data was assessed using the Shapiro-Wilk test. As more than two variables were analyzed, analysis of variance (ANOVA) was performed, followed by post hoc Tukey tests to ascertain statistical significance across the continuous variables under different conditions. Statistical analysis for Texture Analyzer data was performed using a one-way ANOVA followed by Dunnett’s post-hoc test to compare each condition with the control group. This statistical analysis was conducted in R (version 4.3.2). All statistical tests performed are two-tailed, and p-values are p < 0.05 (*); p < 0.01 (**); p < 0.001 (***).

Supplementary Information
Below is the link to the electronic supplementary material.