Acidity Generated by the Tumor Microenvironment Drives Local Invasion

The propensity of cancers to invade adjacent normal tissues contributes significantly to local tumor growth and formation of metastases, which are largely responsible for tumor-associated morbidity and mortality (1). The mechanisms by which tumor cells invade are complex and can be modified in response to environmental conditions (2). Because of increased glucose metabolism, H⁺ production and excretion are generally increased in cancers (3). This, combined with poor perfusion, results in an acidic extracellular pH (pHe) in malignant tumors (pH 6.5–6.9) compared with normal tissue under physiologic conditions (pHe 7.2–7.4; refs. 4–6). Acidic pHe can induce release of (cysteine or aspartyl) cathepsin proteinase activity in vitro (7–9), which is generally believed to be involved in local invasion and tissue remodeling (10–12). Furthermore, cells exposed to in vitro low pH show increased invasion both in vitro and in vivo (9, 13, 14). These observations are synthesized in the acid-mediated invasion hypothesis, wherein H⁺ ions flow along concentration gradients from tumor into adjacent normal tissue, promoting tissue remodeling at the tumor–stroma interface (15). The resulting acidic environment is toxic to normal cells, promotes a degradation of the extracellular matrix by proteinases (16), increases angiogenesis through the release of VEGF, and inhibits the immune response to tumor antigens (17–19). Cancer cells, because of their enhanced evolutionary capacity, develop adaptive mechanisms that allow them to survive and even proliferate in acidic environments (20). These adaptations can involve, inter alia, upregulation of the sodium-hydrogen exchanger-1 (NHE-1) or carbonic anhydrase (CA-IX; refs. 21–23). As normal cells die and the extracellular matrix is degraded, cancer cells continue to proliferate and invade this open space. Thus, we propose that the acidic pH of the tumor microenvironment represents a niche engineering strategy that promotes local invasion and subsequent in vivo growth of malignant tumors. Support for this model has come from recent observations that neutralization of the tumor-derived acid with systemic buffers (e.g., bicarbonate, imidazole, lysine, etc.), can inhibit spontaneous and experimental metastases (8, 24, 25). Here, we explicitly examine the acid-mediated invasion model by correlating regional variations in peritumoral acidity with subsequent patterns of tumor invasion.

An important tool to investigate acid-mediated invasion is the dorsal window chamber, first developed in 1987 (26). We have previously used this system to test aspects of the acid mediated invasion hypothesis and have quantitatively measured the export of tumor-derived acid into surrounding stroma (16). In addition, dorsal window chambers were used to observe a measureable decrease in tumor–stroma pH gradient following oral NaHCO₃ treatment, which has been shown to reduce formation of spontaneous or experimental metastases (8). In the current work, we observed that the acidic pHe of peritumoral tissues was coincident with the location of subsequent tumor invasion, which is a specific prediction of the acid-mediated invasion hypothesis. Furthermore, bicarbonate treatment reduced the pH gradient and prevented invasion completely. Thus, the current findings are directly supportive of the acid-mediated invasion hypothesis.

All animals were maintained under Institutional Animal Care and Use Committee at H. Lee Moffitt Cancer Center (Tampa, FL). Eight- to 10-week-old severe combined immunodeficient (SCID) mice (22–25g; Charles River, Inc.) were used as host for MDA-MB-231/GFP, and HCT116/GFP tumors.

In vitro and in vivo experiments were carried out using 3 cells lines. All cell lines were passaged weekly in standard incubation conditions 37°C and 5% CO₂. Normal human mammary epithelial cells (HMEC; Invitrogen Life Technologies Corporation) were maintained as adherent cultures in HuMEC Ready Medium (Invitrogen Life Technologies Corporation). MDA-MB-231 cells [American Type Culture Collection (ATCC)] were derived from a human breast cancer and HCT116 cells (ATCC), a human colon cancer cell line, were stably transfected with a pcDNA3 GFP vector following polyclonal selection. Green fluorescence was used to clearly distinguish tumor edge, to differentiate tumor from surrounding normal tissue, and to accurately measure tumor growth. All 3 cell lines were used for in vitro studies, however, only the cancer cell lines were used in vivo for tumor development.

Basal rates of extracellular acidification (ECAR) for HMEC, HCT116, and MDA-MB-231 cells were determined using the Seahorse Extracellular Flux (XF-96) analyzer (Seahorse Bioscience). The XF-96 measures the rate of extracellular acidification in the medium above a monolayer of cells in real time. This rate can be converted to a concentration of free protons using a measured buffering capacity. Cells (1.5 × 10⁴/well) were seeded in a XF-96 microplate (V3-PET cat# 101104-004) in normal growth media overnight. One hour before measuring basal ECAR, the growth media was replaced with Seahorse assay media (cat# 100965-000) supplemented with 12.5 mmol/L d-glucose, 0.5 mmol/L sodium pyruvate, and 2 mmol/L l-glutamine. Following flux measurements, protein concentrations were determined in situ for each well using a standard BCA protein assay (Thermo Scientific Pierce). Briefly, the wells were rinsed with PBS and stored at −80°C for 24 hours. The XF-96 microplate was then thawed at room temperature upon which the BCA protein assay was conducted directly in each well. Absorbance was determined at a wavelength of 560 nm and mg protein was calculated using a set of standard bovine serum albumin controls. ECAR values were normalized to mg/protein and were plotted as the mean ± SD.

HCT116 (GFP) or MDA-MB-231(GFP) cells were grown as subcutaneous tumors. Once tumors reached a volume of more than 800 mm³, the pHe was measured by microelectrode, as described previously (24). Briefly, pH measurements were obtained using an FE20 Five Easy pH meter (Mettler-Toledo). Animals were sedated with isoflurane (3.5%) before beginning the experiment and remained under anesthesia (1.5%–3.5% isoflurane) for the duration. A reference and pH electrode (MI-401F and MI-408B, respectively, Microelectrodes Inc.) were used to measure the pH by first inserting the reference electrode (outer diameter, 1 mm) under the skin of the mouse near the tumor. The pH electrode (outer diameter, 0.8 mm) was then inserted up to 1.3 cm into the center of each subcutaneous tumor. Electrodes were calibrated before and following each set of measurements using standard pH 7 and 10 buffers (Sigma). Two measurements were taken at each position and 3 positions were interrogated at each time point and averaged.

Tumor constructs were engineered using the tumor droplet method (8). HCT116 (GFP) or MDA-MB-231(GFP) cells were suspended in 0.8 mg/mL of type I collagen (BD Biosciences #354249) and 1× Dulbecco's Modified Eagle's Medium (DMEM) at a final concentration of 2.5 × 10⁶ cells/mL. Using a 48-well nontissue culture plate, a 15-μL drop of the tumor suspension was polymerized in the center of the well. Polymerization occurred after 20 to 30 minutes of incubation at 37°C. Following polymerization, the droplet was surrounded by a layer of type I collagen at a concentration of 1.25 mg/mL. This procedure allowed the tumors to maintain a circular shape with well-defined borders. Following polymerization of the collagen (20–30 minutes), the construct was incubated with 200 μL of growth medium (DMEM with 10% FBS) at 37°C. In parallel, a dorsal window chamber was implanted into recipient mice. After 2 days in culture, the in vitro constructs were aseptically inoculated into the window chamber. Following 2 to 3 days of recovery, intravital images were acquired to assess tumor integrity and subsequent images were captured every 4 to 5 days to assess tumor growth and to develop a pH profile of the tumor microenvironment. To determine tumor volume, the GFP-expressing tumors were excited with an Argon laser tuned to 488 nm and emission was collected with a 498 to 538 nm bandpass filter. Images were captured using an Olympus FV1000 MPE (multiphoton) microscope and analyzed using Image-Pro Plus v6.2 (Media Cybernetics). At time of tumor regression, typically after 22 to 25 days (tumor dependent), intravital imaging was discontinued.

pHe was measured using SNARF-1 Free Acid (Invitrogen #C-1270), a fluorescent pH indicator that exhibits a spectral shift in fluorescence emissions as a result of a change in pHe. pHe experiments were carried out in window chambers before inoculation of tumor constructs to obtain accurate background measurements. Once a tumor was established, pHe measurements were again acquired. For SNARF-1 measurements, mice were sedated with 1.5% isofluorane, covered with a warm pad to maintain appropriate body temperature, and breathing was monitored during the duration of the imaging session. A tail vein injection of 200 μL of 1 mmol/L of SNARF-1 (free acid) solution was administered and diffusion into the tumor was observed by capturing images continuously for the first 15 minutes using the confocal modality of an Olympus FV1000 MPE. Further images were captured at 30, 40, 50, and 60 minutes after SNARF-1 injection. Spatial distribution of pHe in the tumor and adjacent normal tissue was obtained by exciting the dye with a He/Ne laser at 543 nm and emissions were collected with a 570 to 620 nm bandpass filter and with a 640-nm long pass filter. Confocal images from each channel were converted to tif format using Imaris software, followed by the subtraction of background from each fluorescence image and smoothing with a 7 × 7 kernel with Image-Pro Plus v6.2. Ratiometric images were produced by applying algorithms to the smoothed images using the Definiens Developer XD 1.5. Every ratiometric pixel was used to convert to a pH value by applying in vitro calibration data. The pH was calculated using the equation:

where R is the ratio of emissions from the 2 channels, R_A is the maximal ratio (observed at acidic endpoint); R_B is the minimal ratio (observed at the alkaline endpoint), and [(F B(λ2))/(F A(λ2))] is a correction factor for the ratio of fluorescence values (F) between alkaline (B) and acidic (A) endpoints at the higher wavelength (27). At 1.25×, R_A = 1.863; R_B = 0.954 and 4× R_A = 1.725; R_B = 0.966. To determine the pH in different tumor regions and the area adjacent to the tumor, radial lines emanating from the tumor centroid were drawn and average pH distribution along the lines was determined. “0” was assigned as the tumor centroid, and pH images were aligned so that they coincided at the tumor margin using the GFP image to determine the tumor edge.

Collagen degradation was measured by the increase in fluorescence of dye-quenched collagens I and IV (Life Technologies Corp.) according to methods described in ref. 28. For dye-quenched collagen I, 6-well tissue culture plates were coated with 600 μL of 1.25 mg/mL of type I rat tail collagen (BD Biosciences) containing 25 mg/mL of dye-quenched collagen I and placed in a 37°C incubator for 30 minutes to allow for polymerization. For dye-quenched collagen IV, glass coverslips in 35-mm dishes were coated with 45 μL of Cultrex (Trevigen) containing 25 mg/mL of dye-quenched collagen IV and placed in a 37°C incubator for 10 minutes to allow for polymerization. Cells were labeled with 10 μmol/L Cell Tracker Orange (Life Technologies Corp.) and approximately 40,000 or 2,500 cells were seeded on top of the collagen or Cultrex, respectively, and incubated at 37°C for 30 to 60 minutes until adherent, at which point either a neutral or acidic culture medium was applied. To visualize cell nuclei, cells were incubated with 5 μg/mL of Hoechst (Molecular Probes) for 10 minutes at 37°C. All images were obtained with a Leica TCS SP5 AOBS laser scanning confocal microscope through a ×20/0.7NA Plan Apochromat objective lens (Leica Microsystems). A total of 405 diode, argon 488, and He/Ne 543 laser lines were applied to excite the samples and tunable emissions were used to minimize cross-talk between fluorochromes. Image z-sections for each sample were captured with photomultiplier detectors and prepared with the LAS AF software version 2.6 (Leica Microsystems). A ×2 zoom was applied to increase the total magnification to ×400. Analysis of dye-quenched collagen in the z-stack images was conducted using Definiens Developer v1.5 (Definiens) software suite. First the intracellular area was determined with automatic threshold segmentation on the combination of Hoechst stain and Cell Tracker Red. The intracellular segmentation was conducted on all sections of the z-stack and connected in 3 dimensions. Total pixel area was extracted from this intracellular segmentation. Next contrast split segmentation with a minimum threshold of 50 grayscale values was used to segment dye-quenched collagen stain from background. The dye-quenched collagen segmentation was also conducted on all sections and connected in 3 dimensions. Finally, the localization (intracellular vs. extracellular) of dye-quenched collagen was determined on the basis of the initial intracellular segmentation, mentioned earlier, and the fluorescence signal from the dye-quenched collagen was normalized by the intracellular total pixel area.

Images were acquired with 488-nm excitation with a ×1.25 lens with GFP (tumor) and transmitted light emissions. The GFP-expressing tumor images were subtracted from the transmitted light images leaving a 0 intensity (black) value in place of the tumor. The center of mass for each tumor at early (day 4) and late (day 13) time points were identified (white dot). Four quadrants were drawn radiating 3 mm from the tumor centroid. Each image was brightfield flattened (20-pixel width). The darker vessels were identified by an open object threshold between 1 and 75 dynamic range units for this 8 bit image (0 is excluded to avoid the subtracted tumor). Region of interests (ROI) were placed in each quadrant and the area of the vessels was quantified in pixels. The area of the tumor (pixels) was also quantified in each quadrant. Tumor growth was considered the area of tumor on day 13 subtracted from the area of tumor on day 4. These data were graphically plotted against the early vessel density (area in pixels) from the day 4 time point.

Before initiating the NaHCO₃ treatment, animals were randomly divided into control (n = 4) and experimental (n = 8) groups. Six days before the inoculation of the tumor constructs into the dorsal window chamber, the control group drank tap water and the experimental group was provided with 200 mmol/L of NaHCO₃ad libitum. Three days before the inoculation of the tumor constructs, the dorsal window chambers were implanted and a day before the inoculation, pHe images were acquired to use as background measurements. NaHCO₃ treatment continued throughout the course of the experiment. Once the experiment was terminated, xenografted tissues were collected and analyzed ex vivo by histology and immunohistochemistry.

Once imaging sessions were completed, the xenografted tumors were harvested, fixed in 10% neutral buffered formalin (Thermo Scientific) for 24 hours, processed and embedded in paraffin. Routine hematoxylin and eosin stains were conducted on 4-μm sections of tissue. NHE-1, glucose transporter-1 (GLUT-1), and CD31 were detected by immunohistochemistry using the Ventana Discovery XT automated system (Ventana Medical Systems). Rabbit polyclonal NHE-1 antibody (#sc-28758, Santa Cruz Biotechnology) was incubated at a dilution of 1:200; rabbit polyclonal Glut-1 antibody and rabbit polyclonal CD31 antibody (#ab15309 and #ab28364 Abcam) were incubated at a dilution of 1:400. All antibodies were incubated for 32 minutes in Dako antibody diluent, followed by a 20-minute incubation in Ventana OmniMap anti-rabbit secondary antibody and detected with the Ventana ChromoMap Kit system.

Cancer cells have enhanced rates of glucose metabolism, which produces an acidic environment that may promote degradation of the stromal extracellular matrix and local invasion. The basal rate of proton production is a proximal component in this acidity. In the process of generating ATP, cells produce H⁺ through both oxidative phosphorylation and glycolysis. Rates of proton production were determined in normal HMEC cells and established colon (HCT116) and breast (MDA-mb-231) cancer cell lines, as shown in Fig. 1A. These data show that HCT116 had significantly higher rates of extracellular acidification (P < 0.001), compared with HMEC and MDA-mb-231. MDA-mb-231 also produced H⁺ at a significantly higher rate as compared with HMEC. Thus, the acid production rates of both cancer lines were higher as compared with normal epithelial cells, which is a common observation (5, 6, 29, 30). Notably, these data were normalized to protein, and HCT116 are much smaller than either of the breast lines so that H⁺ production rates of HCT116 was comparable with that of MDA-mb-231, on a per-cell basis (Supplementary Figs. S1 and S2).

The higher rates of acid production by HCT116 would predict that tumors of these cells would have a lower steady state pHe, compared with MDA-mb-231 tumors. To investigate this, tumor cells were injected subcutaneously into SCID mice and allowed to reach a tumor volume of 1,000 mm³. Once the desired volume was reached, a pH-sensitive microelectrode system was used to measure the pHe of these tumors in multiple locations (n = 9 for HCT-116 and n = 7 for MDA-mb-231). These measurements indicated a significantly (P < 0.01) more acidic pHe within the HCT116 tumors as compared with MDA-MB-231 tumors (Fig. 1B). Thus, these in vivo tumor data are consistent with the in vitro proton production measurements. Both in vitro and in vivo evaluations suggest that the HCT116 cells created a more acidic extracellular environment, which may be an important factor in the development of a more aggressive tumor phenotype.

The next step in evaluating the extracellular tumor microenvironment was to monitor tumor growth within the dorsal window chamber. Before in vivo testing, we optimized collagen matrix concentrations for in vitro growth and observed that a concentration of 0.8 mg/mL supported maximal in vitro growth for both cell lines (data not shown). Following in vivo implantation in optimized matrixes, we observed that HCT116 xenografts grew rapidly and reproducibly with a significant (P < 0.0001) 95% increase in tumor-associated pixels by day 16 (Fig. 1C). In contrast, the MDA-mb-231 tumors grew more slowly, for example, a 37% increase in tumor area by day 16 (Supplementary Fig. S3). These basal growth experiments also allowed us to identify the optimum time to observe tumor growth and invasion. Intravital microscopy images of the tumor and its microenvironment can normally be captured for up to a month before skin retraction of the dorsal window chamber is observed. Thus, we limited our observation window to 3 weeks, wherein we considered tumor growth would be unimpeded by failure of the dorsal flap. Notably, most observations were made before 15 days of inoculation (Figs. 1 and 2). Figure 1D shows 4 different HCT116/GFP tumors early (day 2–4) and late (day 10–13) during tumor growth, illustrating the increase in the physical size of the tumor within the chamber and that the growth was directionally heterogeneous. In the initial days of imaging, the HCT116/GFP tumors maintained their original circular morphology. By day 16, the tumors had approximately doubled their original size (Fig. 1C). The early and late images were pseudo-colored to superimpose the images and illustrate that tumors grew anisotropically in the dorsal window chamber.

We have previously proposed that a decrease in the pH of the microenvironment might pose a growth advantage to tumors because of their ability to resist acid-induced cellular toxicity. Figure 1D illustrates the growth that took place within the window chamber within the first 2 weeks of inoculation. To relate growth to the pH of the extracellular peritumoral environment, SNARF-1–free acid was used to map the pH of the tumor and its surrounding microenvironment. Representative ratiometric results are shown in Fig. 2A and show that anisotropically distributed decreases in pHe in both the tumor and neighboring environment were observable by day 7. By day 14, acidic regions were distinct and easily identifiable, specifically in the regions into which the tumor had grown (Fig. 2B). Superimposing the green and red images showed that the tumor invaded the normal tissue that was most acidic. To investigate the relationship between the direction of acidity and path of invasion, the peritumoral pH and direction of tumor growth were more deeply analyzed, as shown in Fig. 2C. First, the center-of-gravities (COG), for both the days 4 and 14 tumors were determined, and they were then coregistered using the rim of the window chamber as a fiducial marker. The ratiometric images were also coregistered and the peritumoral pH was measured in an area that spanned 100 μm from the tumor edge. Both tumor growth and pH were measured at every 22.5° of arc from the COG. Tumor growth was quantified as the number of pixels between edges of the tumor on days 4 and 14 (Fig. 2C and Supplementary Fig. S4), allowing us to compare growth and pHe. As a result, we were able to identify that the direction toward which the tumor grew by day 14 was also the predominant area of acidification. In this particular tumor, taking into consideration that 0° was “true west” relative to the ring clip of the dorsal window chamber, maximal tumor growth and the greatest acidity both occurred at 45° (arrows in Fig. 2B). The entire radial relationship between peritumoral pH and tumor growth is shown in Fig. 2D. The vertical dotted line in the graph represents maximal growth and minimal pH occurring coincidentally at 45°. In regions where the normal tissue had a more alkaline pH, little or no invasion was observed. This pattern was observed across all 5 tumors in this study, wherein growth was highly correlated (P < 0.02) with acidic pH below a threshold that varied from pH 6.8 to 7.1 in individual mice (Supplementary Fig. S5). This variability in threshold is expected because of inaccuracies in assigning scalar pH values using ratiometric methods (27). To address the question: “Is growth greatest in areas with lowest pH?,” we compared the growth in the volumes with the highest and lowest terciles of pH values for each mouse. Across individual mice, the t test P values were 0.003, 0.07, 0.006, 0.12, and 0.008. Across all mice, the difference in growth between low and high pH volumes had a P value of 4e-9.

Following the final image acquisition, the mice were sacrificed and the tumors were removed from the chamber and immediately fixed for histology. To better understand the underlying biology of this regional acidosis and tumor growth, ex vivo biomarker analyses were conducted to determine the expression and spatial distribution of metabolic and angiogenic tumor markers such as the GLUT-1, NHE-1, and the endothelial marker, CD31. It was reasoned that GLUT-1 and NHE-1 work coordinately to acidify the extracellular environment. As shown in Fig. 3A, there was a higher level of GLUT-1 expression at the tumor edge compared with the core. Because GLUT-1 levels are correlated to high glycolysis (31) and high glycolysis is correlated to acid production (3), this regional heterogeneity of GLUT-1 expression was consistent with the lower pHe at the tumor edge (cf. Fig. 2A). Cancer cells adapt to low pHe by upregulating membrane transporters, which help maintain intracellular pH. One of these transporters is NHE-1, which showed a similar regional pattern of overexpression compared with GLUT-1. Image analysis software was used to more accurately characterize these expression profiles. For these analyses, individual cells were identified by nuclei and segmented and the expression level of each immunohistochemical marker was then quantified and expressed on a per-cell basis. These expression levels were then normalized to radial position, from the edge to the center of the tumor. Figure 3B shows that NHE-1–normalized expression levels were highest at the tumor edge, whereas maximal expression of GLUT-1 was approximately 100 μm from the edge. This pattern was statistically significant and was observed consistently across all 4 preparations in this study, as shown by the trend graph (Fig. 3C). These results are consistent with the acidification shown in the peritumoral region in Fig. 2A and B.

Figure 4A shows high-resolution immunohistochemical images of localized microinvasion at the tumor–host interface. The large arrows in the figure denote the region of invasion and smaller arrows designate the tumor–host interface. As earlier, high expression of GLUT-1 and NHE-1 were localized to tumor cells in this invasive region, in contrast to the vessel marker, CD31, which was elevated in the adjacent (reactive) stroma. To more deeply quantify the expression of these markers, staining levels were expressed on a per-cell basis (Fig. 4B), wherein red, orange, and yellow were used to demark strong, moderate, and weak uptake, respectively. Deeper analysis of CD31 expression showed that vessel density was significantly (P < 0.05) higher in reactive stroma, compared with the tumor or distant stroma (Supplementary Fig. S6). These findings suggested that both GLUT-1 and NHE-1 were localized at the tumor front and at regions of localized invasion, consistent with the decrease in pH observed in the in vivo chamber studies. The presence of CD31-positive vessels in the reactive stroma may act as a sink for the tumor-derived acids, and thus would limit the penetration of the acidification deeper into the stroma. Although at lower resolution compared with ex vivo data, spatial analyses of vessel density in vivo showed significant radial heterogeneity in peritumoral regions (Supplementary Fig. S7). When these data were compared with spatially explicit invasive growth, it seemed that there was an inverse relationship between vessel density and tumor growth (Supplementary Fig. S8). These observations were also consistent with the invasion into areas of lower pH, as the more poorly perfused regions can be expected to be more acidic.

Prior work has suggested that acidic pH conditions can stimulate the release of cysteine cathepsins (8) and increase levels of activated collagenases (32). To investigate this, in vitro studies measured the effect of culture pH on proteolytic activity of HCT116 cells grown in matrixes containing dye-quenched collagens I or IV, using methods previously described (28). Collagens I and IV were interrogated because collagen I was used in the dorsal window chamber system and collagen IV is found in basement membrane. Proteolytic activity was assessed using confocal microscopy and analyzed using Definiens Developer v1.5. These data showed collagen degradation was significantly higher at pH 6.8, compared with 7.4 in HCT116 cultures (Supplementary Fig. S9). This is identical to results observed for MDA-mb-231 cells using the same system (33). While these data do not identify the exact mechanism by which acidity promotes invasion, they do indicate that the mechanism involves the production, release, and/or activation of matrix-degrading proteases.

Although the earlier data are consistent with acid-mediated invasion, it remains a possibility that the acid pH is a consequence, and not a cause, of invasion. To test this, the tumor pH was neutralized using a systemic pH buffer, sodium bicarbonate. We have previously shown that a variety of buffers including bicarbonate, Tris, lysine, and imidazoles, can neutralize tumor pHe and inhibit tumor invasion and metastasis (8, 24). In this current study, window chamber-bearing mice were randomized into control (tap water) and bicarbonate (200 mmol/L ad libitum) groups, which continued for the duration of the experiment. After 6 days of acclimation, the window chambers were inoculated with HCT116/GFP droplet constructs and the tumors monitored as earlier. Intravital images were captured 2 days after inoculation to examine integrity of the tumor and to capture initial pHe measurements. After 2 days, there were no differences in the size or density of inoculated tumors between the 2 groups. Monitoring of tumor growth and pHe levels was continued throughout the experiment every 4 to 5 days. As shown in Fig. 5A, tumors treated with bicarbonate displayed a different growth pattern than the controls (cf. Fig. 1D). Figure 5B shows that the control tumors almost doubled in volume by day 13, whereas the volumes of NaHCO₃ treated tumors were not significantly changed between days 1 and 13. Inhibition of growth by bicarbonate was highly significant, as shown in Supplementary Fig. S10 (P < 0.02). Figure 5C shows the pH values mapped within the chamber on both days 8 and 19 postinoculation and showed that pHe of the tumor and its microenvironment were higher in the bicarbonate group as compared with controls (cf. Fig. 2A). Using calibration data, the ratiometric data were converted into scalar pH values and the intratumoral and peritumoral pH values of the bicarbonate and control tumors were obtained. Measurements were calculated using all intravital images per group along 4 radial directions (Fig. 6A and B). The mean values were plotted to show a pH gradient from the center of the tumor to the tumor edge and from the tumor edge into and across the stromal extracellular matrix adjacent to each tumor (Fig. 6C and D). The pH values for all control and bicarbonate tumors were collected and mean measurements were plotted. By day 15 postinoculation, there were no significant differences in the pH values at the centers of the tumors for control (pH 6.78 ± 0.25) and bicarbonate-treated (pH 6.97 ± 0.15) tumors. In contrast, the pH at the edge of control tumors were significantly (P < 0.05) more acidic (pH 6.75 ± 0.19 SD) compared with that of bicarbonate treated tumors (pH 7.14 ± 0.05). These results show that, in a window chamber model, systemic buffering with NaHCO₃ increased the peritumoral pH and reduced local invasion.

The morbidity and mortality associated with cancer is largely related to tumor invasion and formation of metastases. Extensive application of 2[18F]fluoro-2-deoxy-d-glucose–positron emission tomography (FDG-PET) imaging to clinical cancers has clearly shown the vast majority of malignant tumors metabolize glucose at high rates. (34, 35). We propose there is a direct, causative link between increased glucose metabolism and the ability of cancer cells to invade and metastasize.

Elevated glucose metabolism is the proximate cause of increased acidity in the tumor microenvironment. Furthermore, most tumors develop an aberrant vasculature network that tends to be poorly organized and leaky, disrupting blood flow, and hampering the delivery of oxygen (36). This has a 2-fold effect on tumor acidity. First, it subjects tumor regions to poor perfusion, and hence poor oxygenation (37). Low oxygenation increases glycolytic flux via the Pasteur effect (38). Notably, even in tumor regions with adequate oxygen supply, glycolysis and acid production are upregulated via the Warburg effect (39). Second, poor perfusion hampers the ability of the microenvironment to remove tumor derived acid through diffusion. Consequently, the pHe of tumors is typically highly acidic, and this will inevitably result in acid diffusion into the surrounding stroma.

We have previously proposed that the invasive phenotype and increased glucose metabolism are, in fact, closely linked (23). Much of this work stems from viewing cancer biology as an ecologic and evolutionary process. Within this context, all commonly observed phenotypes in cancers must be conferring an evolutionary advantage. In this case, we ask “how does increased glucose metabolism and consequent interstitial acidosis confer an evolutionary advantage that promotes tumor cell proliferation?” This has led to the proposal that regional acidosis represents a strategy commonly observed in nature as niche engineering in which plants and animal alter their environment in ways that promote their own growth and survival and/or diminishes that of their competitors. The conceptual model that tumors use increased glucose metabolism, even in the presence of adequate oxygen, as a form of niche engineering is the basis of the acid-mediated invasion hypothesis (15, 16). The model assumes that through evolutionary events during carcinogenesis, as cancer cells proliferate on epithelial surfaces (40), they must adapt to increased acid production by expression of proton export systems, such as NHE-1 or CA-IX, which transport H⁺ from intra- to extracellular space, resulting in extracellular acidification. However, normal mammalian tissue, lacking the evolutionary capacity of cancer cells, typically remains intolerant of prolonged exposure to acidic pH. This model posits H⁺ flows along concentration gradients, from the tumor into the peritumoral normal tissue, causing disruption that favors subsequent tumor growth. A variety of studies have shown that an acidic peritumoral pH is associated with a degradation of the extracellular matrix, possibly through the release and activation of proteolytic enzymes (7, 16). The low pH also leads normal cells to undergo apoptosis and necrosis, whereas cancer cells survive due to acquired resistance mechanisms (41, 42). In summary, regional acidosis causes substantial niche engineering through normal cell death, breakdown of extracellular matrix, promotion of new vessel formation, and suppression of the immune response (10, 17, 18).

The results of this study suggest that tumor cells do, indeed, perform niche engineering by creating an acidic environment that is nontoxic to the malignant cells, but through its negative effects on normal cells and tissue, promotes local invasion. Targeting this evolutionary strategy through systemic buffers and other mechanisms to reduce peritumoral pH will likely provide a valuable adjunct or alternative to traditional therapies focused entirely on killing the tumor cells.

No potential conflicts of interest were disclosed.

Conception and design: V. Estrella, T. Chen, M. Lloyd, J. Wojtkowiak, J.M. Rothberg, R.A. Gatenby, R.J. Gillies

Development of methodology: V. Estrella, T. Chen, M. Lloyd, K. Bailey, J.M. Rothberg, B.F. Sloane, R.A. Gatenby, R.J. Gillies

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): V. Estrella, T. Chen, M. Lloyd, J. Wojtkowiak, H.H. Cornnell, A. Ibrahim-Hashim, J. Johnson

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): V. Estrella, T. Chen, M. Lloyd, H.H. Cornnell, Y. Balagurunathan, J. Johnson, R.J. Gillies

Writing, review, and/or revision of the manuscript: V. Estrella, T. Chen, M. Lloyd, H.H. Cornnell, A. Ibrahim-Hashim, K. Bailey, Y. Balagurunathan, J.M. Rothberg, B.F. Sloane, R.A. Gatenby, R.J. Gillies

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V. Estrella, R.A. Gatenby

Study supervision: V. Estrella, R.A. Gatenby, R.J. Gillies

This study was supported by grants U54 CA143970 (R.A. Gatenby and R.J. Gillies); R01 CA 077575 (R.J. Gillies and R.A. Gatenby); and R01 CA 131990S (B.F. Sloane and J.M. Rothberg).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.