Metabolic Alterations in Lung Cancer–Associated Fibroblasts Correlated with Increased Glycolytic Metabolism of the Tumor

This article is featured in Highlights of This Issue, p. 555

Tumors are composed of malignant cancer cells and stromal cells that constitute the tumor microenvironment. Research efforts have predominantly focused on understanding the biology of the epithelial component of tumors, leading to the identification of genes and signaling pathways dysregulated in cancer cells. The nonepithelial cellular elements of the tumor mass, although not by themselves malignant, have essential roles in carcinogenesis by supporting the cancer cells (1, 2). Cancer-associated fibroblasts (CAF) of the tumor microenvironment are protumorigenic, although the exact functions of CAFs have yet to be fully delineated (3, 4). Proposed functions of CAFs include secretion of growth factors, of proangiogenic VEGF, and/or of proteases that remodel the extracellular matrix (5–9). The distinct protumorigenic activities of CAFs identified in the different studies might reflect tumor type–specific roles for CAFs.

Proliferation requires growth-promoting signaling and an appropriate nutrient milieu. Cancer cells frequently have a predominantly glycolytic metabolism, the “Warburg effect” (10, 11), that is required for the growth of some tumors (e.g., ref. 12). The emerging view is that reliance on a predominantly glycolytic metabolism supports the anabolic needs of cancer cells at the expense of efficient production of ATP from glucose (11). Little is known about possible metabolic alterations in the stroma cells of the tumor microenvironment. Alterations in stromal cell metabolism might be advantageous to the cancer cells, as would be the case if stroma cell metabolism was coupled to that of the cancer cells (e.g., ref. 13).

Here, we use mass spectrometry–based profiling of the abundances of 203 biochemicals of 46 metabolic pathways/groups to compare primary human lung tumor CAFs to “normal” fibroblasts (NF) isolated from nonneoplastic lung tissue located at least 5 cm away from the tumor. We found differences in several metabolic pathways that distinguish CAFs from NFs, suggesting that alterations in cellular metabolism are not limited to epithelial cancer cells of the tumor. Moreover, we found differences between CAFs and NFs that correlate with the glycolytic capacity of the tumor, one of which was a relative increase in CAFs of 17 unrelated dipeptides. The relative difference in dipeptides was most significant between CAFs derived from highly glycolytic tumors and their paired NFs, although dipeptides in CAFs from less glycolytic tumors were also increased compared with paired NFs. Unexpectedly, dipeptides were also increased in NFs isolated from lung tissue of individuals with highly glycolytic tumors compared with those from less glycolytic tumors, suggesting either a tumor-specific field effect extending beyond 5 cm or metabolic alteration in NFs induced by a widespread inflammatory or other protumorigenic parenchymal milieu. Dipeptides arise from the breakdown of proteins. In CAFs, we find evidence for increased autophagy, a lysosome-related catabolic mechanism, which likely explains the increase in dipeptides. We also find that reducing the amount of glucose in the medium from 25 to 5 mmol/L causes an increase of dipeptides and autophagy in NFs but not CAFs, suggesting that CAFs and NFs have different set points for glucose induction of autophagy. There is mounting evidence that increased degradative pathways (e.g., autophagy) in cancer cells, potentially as stress response mechanisms, have roles in tumor maintenance and possibly in resistance to chemotherapy (14–16). Our studies provide evidence that increased autophagy and lysosomal capacity are characteristics of human lung CAFs, alterations that might contribute to the protumorigenic activity of CAFs.

The clinical study (Fig. 3) was an Institutional Review Board–approved review of a prospectively assembled thoracic surgery database of patients with non–small cell lung cancer (from February 2000 to December 2009). Patient consent was waived. Patients with surgically resected tumors (stage I–III) and documented SUV_max values by positron emission tomography (PET) were considered eligible for this review (n = 693). Patients were segregated into 2 cohorts those with SUV_max ≥8 (n = 231, 33%) and those with SUV_max ≤5 (n = 350, 50%). Disease-free survival (DFS) for the cohort (from date of surgery to recurrent or death) was evaluated by Kaplan–Meier survival analysis. Log-rank test was used for comparison of DFS between the 2 groups. Median follow-up was 23 months.

Cells were derived from resected tumors and adjacent nonneoplastic tissue in an Institutional Review Board–approved study. Patients consented to use of cells derived from their tissue for this study. Tissue, which was processed within 2 hours of resection, was minced with a razor and incubated in 1 mg/mL collagenase D for 60 minutes with agitation at 37°C. The material was passed through a 70-μm sieve, red blood cells lysed, and extensively washed in HEPES-buffered Dulbeccos' Modified Eagle's Medium (DMEM; 25 mmol/L glucose) before plating in DMEM supplemented with 10% FBS. When the cultures reached about 80% confluence, cells were passaged and expanded further by growth. Cell pellets (≥2.5 × 10⁶ cells) were washed with PBS, flash frozen in a dry ice ethanol bath, and stored at −80°C. In some cases, cells were expanded in 5 mmol/L glucose–DMEM–10% FBS.

Each sample received was accessioned into the Metabolon LIMS system and was assigned by the LIMS a unique identifier that was associated with the original source identifier only. This identifier was used to track all sample handling, tasks, results, etc. The samples (and all derived aliquots) were tracked by the LIMS system. All portions of any sample were automatically assigned their own unique identifiers by the LIMS when a new task is created; the relationship of these samples is also tracked. All samples were maintained at −80ºC until processed.

Samples were prepared using the automated MicroLab STAR system from Hamilton Company. A recovery standard was added before the first step in the extraction process for quality control (QC) purposes. Sample preparation was conducted using aqueous methanol extraction process to remove the protein fraction while allowing maximum recovery of small molecules. The resulting extract was divided into 4 fractions: one for analysis by ultrahigh performance liquid chromatography/mass spectroscopy (UPLC/MS/MS; positive mode), one for UPLC/MS/MS (negative mode), one for gas chromatography/mass spectroscopy (GC/MS), and one for backup. Samples were placed briefly on a TurboVap (Zymark) to remove the organic solvent. Each sample was then frozen and dried under vacuum. Samples were then prepared for the appropriate instrument either UPLC/MS/MS or GC/MS.

The liquid chromatography/mass spectrometry (LC/MS) was based on a Waters ACQUITY UPLC and a Thermo-Finnigan linear trap quadrupole (LTQ) mass spectrometer, which consisted of an electrospray ionization (ESI) source and linear ion-trap (LIT) mass analyzer. The sample extract was dried then reconstituted in acidic or basic LC-compatible solvents, each of which contained 8 or more injection standards at fixed concentrations to ensure injection and chromatographic consistency. One aliquot was analyzed using acidic positive ion–optimized conditions and the other using basic negative ion–optimized conditions in 2 independent injections using separate dedicated columns. Extracts reconstituted in acidic conditions were gradient-eluted using water and methanol containing 0.1% formic acid, whereas the basic extracts, which also used water/methanol, contained 6.5 mmol/L ammonium bicarbonate. The MS analysis alternated between MS and data-dependent MS/MS scans using dynamic exclusion. Raw data files were archived and extracted.

The samples destined for gas chromatography/mass spectroscopy (GC/MS) analysis were redried under vacuum desiccation for a minimum of 24 hours before being derivatized under dried nitrogen using bistrimethyl-silyl-triflouroacetamide. The GC column was 5% phenyl and the temperature ramp was from 40° to 300°C in a 16-minute period. Samples were analyzed on a Thermo-Finnigan Trace DSQ fast-scanning single-quadrupole mass spectrometer using electron impact ionization. The instrument was tuned and calibrated for mass resolution and mass accuracy on a daily basis. The information output from the raw data files was automatically extracted as discussed below.

For quality assurance (QA)/QC purposes, additional samples were included with the analysis of each day. These samples included extracts of a pool of well-characterized human plasma, extracts of a pool created from a small aliquot of the experimental samples, and process blanks. QC samples were spaced evenly among the injections, and all experimental samples were randomly distributed throughout the run. A selection of QC compounds was added to every sample for chromatographic alignment, including those under test. These compounds were carefully chosen so as not to interfere with the measurement of the endogenous compounds.

Raw data were extracted, peak-identified, and QC processed using Metabolon's hardware and software. These systems are built on a web service platform using Microsoft's.NET technologies, which run on high-performance application servers and fiber channel storage arrays in clusters to provide active failover and load balancing. Compounds were identified by comparison to library entries of purified standards or recurrent unknown entities. More than 2,400 commercially available purified standard compounds have been acquired and registered into LIMS for distribution to both the LC and GC platforms for determination of their analytic characteristics.

Missing values (if any) were assumed to be below the level of detection. However, biochemicals that were detected in all samples from one or more groups but not in samples from other groups were assumed to be near the lower limit of detection in the groups in which they were not detected. In these cases, the missing values were replaced with the minimum value obtained for the biochemical within all groups. Each biochemical value was then normalized to cell number and scaled to their median value. The data were also analyzed after normalizing the individual metabolites in each sample to the median of all metabolites in that sample. This scaling accounts for any difference between samples due to a change in the abundance of all metabolites, as would be the case if there were differences in cell size among groups. Similar results were obtained in all comparisons regardless of whether the data for each cell line were scaled to the median value of that cell line. The data sets without and with scaling to the median of all metabolites in that sample are presented in Supplementary Table SI. The results presented are the significant differences common to the analyses of the data without and with the scaling.

Heatmap data were scaled up for zero mean and unit variance by first subtracting the mean of each row for the matrix needs to be visualized and then dividing each row by its SD. This provides a matrix where each metabolite concentrations have mean 0 and SD 1.

An unpaired t test for differential abundance with a standard 2-tailed and 2-sample t test was done using mattes function in MATLAB on each metabolite to calculate the Wilcoxon t statistic for the 2 groups being compared. An unpaired t test was then used to conduct t test over the t scores and fold change (log₂) of metabolites grouped as required for the analysis. Resultant P values were then adjusted for false discovery rate as described by Benjamini–Hochberg correction for multiple testing. The same analysis was repeated with random shuffling (10⁴ times) of the group labels (permutation testing). Results that had a probability of less than 0.05 of appearing in this shuffling exercise were selected as significant. Complete analysis was done in MATLAB R2011b with Bioinformatics (ver 4.0) and Statistics (ver 7.6) toolboxes.

To analyze biochemicals grouped into 7 main groups, we compared the distributions of t scores of the metabolites of that group (obtained from t-test as described above) to the distribution of the t scores of the rest of the metabolites. The P values for this comparison for each group were determined after correcting for multiple testing. To analyze biochemicals grouped into 46 subgroups, we compared the distributions of t scores (obtained from t test as described above) and the mean log₂ fold change of the metabolites in each group with those of rest of the metabolites combined. Significance was calculated by conducting an unpaired t test assuming normal distributions of unequal variances. P values corrected for multiple testing were obtained for each subgroup as described above. Subgroups having an adjusted P < 0.05 from t scores comparisons (which captures the variance of data) and adjusted P < 0.05 from mean fold change comparison were selected. The data were analyzed in 2 ways, with and without scaling the metabolites of an individual sample to the median value of all metabolites in that sample. There were only minor differences in the results with and without the scaling. The results presented are the significant differences between groups that were common to the 2 different methods of analyses.

Lysosomal pH was measured as described previously (17). Cells were incubated for 18 hours in growth medium supplemented with 2.5 mg/mL fluorescein rhodamine–labeled dextran (Invitrogen, Inc.). Following this incubation, cells were washed and incubated for an additional 3 hours in growth medium to chase the dextran into late endosomes and lysosomes. The cells were transferred to a HEPES-balanced salt solution with 10 mmol/L glucose and imaged live by epifluorescence in both fluorescein and rhodamine channels. To construct a pH calibration curve a set of cells similarly labeled were fixed and incubated in monensin containing buffers of pH 3, 4, 5, 6, 7, and 8. These cells were imaged with the same settings as the live cells. Metamorph software was used to quantify the rhodamine and fluorescein fluorescence of regions of interest (ROI) identified in the rhodamine channel. The rhodamine–fluorescein ratios of the ROIs of the fixed cells were used to create the pH calibration curve that was used to assign pH values to the ROIs of the live cells.

Paired CAFs and NFs were grown in culture to 100% confluence in 6-well plates (DMEM-Hams F12 1:1 blend, 10% FBS). Cells were either treated with 30 μmol/L chloroquine for 3 or 5 hours or kept in normal growth media. Protein extracts were prepared in 150 μL of 2× Laemmli buffer, boiled, sonicated, and run on a 15% SDS-PAGE gel. Membranes were incubated overnight with primary antibodies: rabbit anti-LC3B 1:1,000 (Invitrogen, Inc); rabbit anti-actin 1:5,000 (Cytoskeleton Inc.); rabbit anti-Caveolin-1 (BD Biosciences); and anti-Lamp-2A (Abcam, Inc) and for 1 hour with secondary antibodies (goat anti-rabbit 680 nm, 1:5,000). Membranes were read on a LiCor imager and quantified using Odyssey software (version 3.0). Alternatively, LC3 autophagosomes were detected using indirect immunofluorescence in epifluorescence microscopy. Caveolin and LAMP-2A blotting were conducted on the non-chloroquine–treated extracts only.

In an Institutional Review Board–approved study, we generated by long-term culture 21 CAF lines from surgically resected primary lung cancers and matched NFs from nonneoplastic lung tissue obtained at least 5 cm from the tumor, and we used these samples for mass spectrometry metabolomic profiling of CAFs (Fig. 1A). Tumor and nonneoplastic tissues were processed by mechanical and collagenase disruption, and the resultant heterogeneous populations of dispersed cells were plated in growth medium supplemented with 10% FBS. One day after plating, keratin-expressing and vimentin-expressing cells were detected by immunofluorescence (Fig. 1B). Overtime in culture without passage, both keratin- and vimentin-expressing cells proliferated, with the keratin-expressing cells growing in clusters characteristic of epithelial cells. However, upon expansion of the cultures by passage, only the vimentin-expressing cells proliferated in samples from both tumor and nonneoplastic tissue (Fig. 1C). These vimentin-expressing cells were further expanded and prepared for metabolomic analysis. Numerous previous studies have isolated CAF and NF lines by culture without specific enrichment steps. To be consistent with the terminology in the literature, we refer to the expanded vimentin-expressing cultures as CAFs cell lines and NFs cell lines, although they are not clonal lines. There were no consistent differences in growth rates of the expanded CAF and NF cultures. Pertinent clinical data associated with the 21 CAFs/NFs paired lines are presented in Table 1.

Increased expression of α-smooth muscle actin (αSMA) is one commonly reported characteristic of CAFs, presumably reflecting that CAFs are in an “activated state” (3, 4). Consistent with those previous observations, expression of αSMA was generally higher in the CAFs than in matched NFs; however, there was a considerable variation in αSMA expression among the different lines (Fig. 1D). This heterogeneity was observed in the percentage of cells expressing αSMA within any given CAF and NF pair as well as in the amount of αSMA expression per cell. No correlation between αSMA expression and clinical data was noted. Heterogeneity in αSMA expression in CAFs has been previously reported for CAFs (e.g., ref. 9).

One group has proposed that reduced caveolin-1 expression drives CAF activation and is responsible for a metabolic reprogramming of CAFs (18). We did not find differences in total caveolin-1 expression between CAFs and paired NFs detected by Western blotting (Fig. 1E). In addition, caveolin-1 was homogeneously expressed at the individual cell level in both CAFs and NFs, labeling both the plasma membranes and internal membranes (Fig. 1F). Thus, reduced caveolin-1 expression is not a feature of lung tumor–derived CAFs.

Mass spectrometry was used to profile 203 metabolites in the 21 pairs of cultured primary lung CAFs and their matched NFs as previously described (19, 20). Metabolite levels were corrected for cell numbers, and each metabolite was scaled to its median across all samples (Fig. 2A and Supplementary Table SI). There was a slight, albeit significant, increase in all 203 metabolites considered as a group in NFs compared with the CAFs (Fig. 2B). The data are corrected for cell number, and we do not know the reason for this difference. To account for this change, we also analyzed the data after normalizing the metabolites of each individual sample to the median value of all metabolites of each individual sample (Supplementary Table SI). This normalization did not change the results of the CAF to NF comparison. No single metabolite was significantly different between CAFs and NFs in a comparison of the means of the individual metabolites (P ≤ 0.05, two-tailed, unpaired Student t test, Benjamini–Hochberg corrected for multiple testing).

To minimize subject-to-subject variations that are independent of possible CAF to NF differences, we examined the ratio of the individual metabolites in each CAF to the amount of that metabolite in the NF derived from the same individual (Supplementary Fig. S1B). In this analysis, only the abundance of glucose was significantly different between CAFs and NFs at P ≤ 0.05 (2-tailed unpaired Student t test, Benjamini–Hochberg–corrected). The biologic significance of this change in glucose without corresponding changes in glucose metabolites is not clear.

Cellular metabolism is an interconnected network; therefore, steady-state analyses can be confounded because changes in biochemicals of a metabolic pathway will be distributed across individual intermediates of that pathway as well as potentially distributed more broadly over connected pathways. To capture differences in abundance of metabolites of metabolic pathways, and thereby potentially increase the sensitivity of our analysis to detect differences between CAFs and NFs, we assigned the 203 metabolites to 7 groups based on Kyoto Encyclopedia of Genes and Genomes (KEGG; ref. 21) and compared CAFs with NFs based on these groups (Supplementary Table SI). To accomplish this, we used the t test to compare the t scores of the metabolites of each of the 7 groups to the t scores of all the other metabolites excluding those of the pathway being examined. In this analysis, we determined whether differences in any of the 7 groupings of metabolites between CAFs and NFs were more pronounced than the difference in the abundance of all metabolites between CAFs and NFs. Two of the 7 groups were significantly enriched in CAFs: amino acids and peptides (P ≤ 0.05, unpaired 2-tailed Student t test, Benjamini–Hochberg–corrected; Fig. 2C). These data indicate differences in the cellular metabolisms of CAFs and NFs that were not apparent at an individual metabolite/biochemical level.

To refine the comparison, we analyzed the data by further categorizing the metabolites into 46 subgroups based on KEGG designation (ref. 21; Supplementary Table SI). We compared both the t scores and the fold changes (log₂) in abundance of the metabolites for each subgroup to the respective t scores or fold changes of all the other metabolites excluding those of the subgroup being examined (Supplementary Fig. S2). Thus, in this analysis, we determined whether differences in any of the 46 subgroupings of metabolites between CAFs and NFs were more pronounced than the difference in the abundance of all metabolites between CAFs and NFs. To consider a subgroup significantly different between CAFs and NFs, we imposed the criteria that the distribution of t scores of the metabolites of a subgroup and the mean log₂ fold differences in abundance of the metabolites of a subgroup should both be significantly different (P ≤ 0.05, unpaired 2-tailed Student t test, Benjamini–Hochberg–corrected).

This analysis revealed that 3 subgroups were significantly different between CAFs and NFs (Fig. 2D and E). Metabolites of the glycine/serine/threonine subgroup include these amino acids as well as n-acetlylserine and betaine (Supplementary Table SI). Creatinine subgroup consists of creatine and creatinine. The dipeptides subgroup consists of 17 unrelated dipeptides (Supplementary Table SI). The differences between CAFs and NFs in the glycine/serine/threonine and the creatinine subgroups, which are both part of the KEGG designation amino acid group, likely account for this group distinguishing the CAFs from the NFs (Fig. 2C), whereas the change in the dipeptide subgroup accounted for the peptide group distinguishing CAFs from NFs (Fig. 2C). The metabolites of the 3 subgroups that were different between CAFs and NFs were all enriched in CAFs.

These metabolite profiling data support the hypothesis that there are steady-state metabolic differences between CAFs and NFs.

To further interrogate the data, we analyzed the metabolites after subdividing the CAFs and NFs based on the glycolytic activity of the parent primary tumor measured by standardized uptake value (SUV_max) on fluorodeoxyglucose PET (Table 1). SUV_max, a measure of intracellular accumulation of phosphorylated deoxyglucose, is a surrogate measure of the glycolytic metabolism of tumors (22). To determine whether there were differences in the metabolite profiles of CAFs and NFs related to the SUV_max of the tumor, the CAF/NF pairs were segregated into “low” SUV_max (≤5) and “high” SUV_max (≥8) groups. Data collected from our clinic show significant differences in DFS for patients based on these SUV_max groupings, in accord with the potential clinical relevance of SUV_max stratification (Fig. 3; refs. 23, 24). On the basis of these criteria, 7 CAF/NF pairs were categorized as low SUV_max and 11 as high SUV_max (Table 1). The 3 CAF/NF pairs that fell between SUV_max of 5 and 7 were excluded from this analysis.

No individual metabolites were significantly different between CAFs derived from high SUV_max tumors and their corresponding NFs. Three of the 46 KEGG subgroups were significantly different between high SUV_max CAFs and their matched NFs (Fig. 4A). Both the dipeptides and the glycine/serine/threonine subgroups were significantly enriched in CAFs. These subgroups were also identified in the comparison of CAFs to NFs independent of SUV_max categorization (Fig. 2D). In addition, the mean abundance of metabolites of the lysolipids subgroup (16 different lysolipid species) was reduced in high SUV_max CAFs relative to their paired NFs (Fig. 4A).

No individual metabolites were significantly different between CAFs derived from low SUV_max tumors and their paired NFs. Two of the 46 KEGG subgroups were significantly different between low SUV_max CAFs and their matched NFs (Fig. 4B). As was the case for the comparison of high SUV_max CAFs to NFs, the dipeptide subgroup was significantly enriched in the low SUV_max CAFs relative to their paired NFs, although the relative enrichment in the in low SUV_max CAFs was smaller than in the high SUV_max comparison. There was also a significant decrease in the glycolysis/pyruvate metabolites subgroup in the low SUV_max CAFs relative to their paired NFs (Fig. 4B). The metabolite subgroups that distinguish low SUV_max CAFs from their matched NFs were not the same as those that distinguish high SUV_max CAFs from their paired NFs, suggesting a link between the metabolic alterations in CAFs and the glycolytic activity of the tumor (i.e., SUV_max) from which the CAFs were derived.

To further explore the possibility of a relationship between CAF metabolism and tumor SUV_max, we compared CAFs from low SUV_max tumors to CAFs from high SUV_max tumors. Two subgroups of metabolites were different between CAFs from low and high SUV_max tumors: lysolipid and dipeptides (Fig. 4C). The abundance of lysolipids was reduced in the CAFs from high SUV_max tumors, whereas the dipeptides were increased in CAFs from high SUV_max tumors. These 2 subgroups also distinguished high SUV_max CAFs from their paired NFs (Fig. 4A). Interestingly, in both the comparisons, the lysolipids were reduced and dipeptides were increased in the high SUV_max CAFs, suggesting these changes are characteristics of CAFs derived from high SUV_max tumors, reinforcing the hypothesis that metabolic changes in CAFs are in some way linked to the glycolytic capacity of the tumor.

The above data provide direct support for the hypothesis that there are differences between CAFs from high and low SUV_max tumors. We next compared the metabolite profiles of NFs assigned to either high or low SUV_max groups based on the SUV_max of the tumor from which their corresponding CAFs were derived. Two of the KEGG designated subgroups of metabolites were different between NFs associated with low and high SUV_max tumors (Fig. 4D). Dipeptides were enriched in NFs associated with high SUV_max tumors, whereas the carnitine subgroup (acetylcarnitine, carnitine, deoxycarnitine, and hexanoylcarnitine) was enriched in NFs associated with low SUV_max tumors. These data suggest an unexpected link between the glycolytic activity of the tumor (i.e., SUV_max) and the metabolic profile of fibroblasts derived from nonneoplastic tissue of the same lobe as the tumor.

Dipeptides were increased in each comparison of CAFs to NFs, suggesting an increase in dipeptides is a characteristic of CAFs (Figs. 2 and 4). In addition, dipeptides were also increased in CAFs from high SUV_max tumors relative to those from low SUV_max tumors, and they were increased in NFs associated with high SUV_max tumors relative to those associated with low SUV_max tumors (Fig. 4D). Thus, an increase in dipeptides is also correlated with the higher SUV_max tumors.

Proteolysis is a source of dipeptides and therefore the differences in the dipeptides could reflect differences in proteolysis among the cell groupings. Alterations in autophagy, a lysosome-related proteolytic-degradative mechanism, have been noted in cancer cells, a change that might reflect a response to energy balance alterations in the cancer cells (25). Two autophagic mechanisms, macroautophagy and chaperone-mediated autophagy, have been linked to cancer (14–16, 26).

We first determined whether there were differences in macroautophagy between CAFs and NFs. Upon induction of macroautophagy (from here on out referred to as autophagy), microtubule-associated protein 1 light chain 3 (LC3I) is processed by lipidation to form LC3II that associates with membranes during the elongation phase of autophagosome formation. An increase in LC3II, evident by altered electrophoretic migration, is a measure of increased autophagy (Fig. 5A; refs. 27–29). Although there was variability in the total amount of LC3 (LC3I + II) among CAF/NF pairs (reflecting individual-to-individual variation), there were no consistent differences in total LC3 between CAFs and NFs, nor was there a correlation in the total LC3 with the tumor SUV_max (Fig. 5B). There was, however, a significant increase in the basal steady-state LC3II as a percent of total LC3 in CAFs, indicating increased autophagy in CAFs (Fig. 5C). To establish that the basal steady-state increase in LC3II was due to increased production of LC3II, rather than a decrease in LC3II turnover, we used chloroquine to inhibit degradation of LC3II (29). Chloroquine induced an increase of LC3II in CAFs and NFs lines; however, the relative increase in LC3II as a percentage of total LC3 in CAFs was maintained, consistent with the higher basal LC3II in CAFs being due to increased autophagy (Fig. 5C). When we segregated the data for the tumor SUV_max, we found that the increase in LC3II in CAFs relative to their paired NFs was significant for pairs associated with high SUV_max tumors but not for those associated with low SUV_max tumors (Fig. 5C). These results are consistent with the more robust difference in dipeptides between high SUV_max CAF/NF pairs (Fig. 4), and the steady-state LC3II measurement may not be of sufficient sensitivity to detect the smaller differences between low SUV_max CAF/NF pairs.

We also investigated whether there were differences between CAFs and NFs in chaperone-mediated autophagy (CMA), a process in which cytosolic proteins are transported into lysosomes for degradation that has been shown to be increased in some cancer cells (14–16, (25, 26). Upregulation of LAMP-2A is a marker for increased CMA. Although there was individual-to-individual variation in LAMP-2A expression, there were no consistent differences between CAFs and NFs (Supplemental Fig. 3). These data establish that CMA was not upregulated in CAFs.

The above results established an increase in basal autophagy in CAFs. Nutrient limitation is a primary inducer of autophagy, prompting us to investigate whether CAFs and NFs differ in the activation of autophagy in response to nutrient reductions. We first determined the effect of acute, stringent nutrient deprivation by incubating cells in medium lacking glucose and glutamine for 3 hours. Induction of autophagy was investigated by measuring the LC3II in Western blotting and by fluorescent microscopy measurement of LC3-stained puncta. Increased basal autophagy in CAFs was evident by an increase in the number and intensity of LC3-positive juxtanuclear structures relative to NFs (Fig. 5D). The LC3 labeling and the concentration in the juxtanuclear region of cells are characteristics of autophagosomes. Acute nutrient deprivation of cells in medium lacking glucose and glutamine resulted in an increase in autophagosomes in CAFs and NFs, although there was some cell-to-cell heterogeneity in both CAFs and NFs (Fig. 5D). Acute nutrient deprivation induced an increase of LC3II, measured by Western blotting, to the same level in CAFs and NFs (Fig. 5E). Thus, despite increased basal autophagy in CAFs, there was no difference between CAFs and NFs in the induction of autophagy in response to acute nutrient deprivation, showing that CAFs further induce autophagy in response to nutrient stress.

We next examined differences in autophagy between CAFs and NFs in response to chronic reduction in glucose availability rather than acute challenge with no glucose. Because of the high-glucose demands of cancer cells, glucose available for metabolism by stroma cells within the tumor microenvironment might be chronically limited. Cells are typically grown in culture in media supplemented with 25 mmol/L glucose, and our metabolomic profiling was conducted in cells expanded in 25 mmol/L glucose. To determine the effect of reduced glucose, cells were grown in media with 25 or 5 mmol/L glucose. We used quantitative fluorescence microscopy of LC3 to detect autophagosomes. The amount of LC3 (increased staining intensity) associated with juxtanuclear puncta, characteristic of lysosomes/autophagosome, was increased in NFs grown in 5 versus 25 mmol/L glucose, although the increase was not as great as when the cells were incubated in medium lacking glucose and glutamine (Fig. 6A). These data establish that reduced glucose in the growth medium–induced autophagy in NFs. However, there was no increase in autophagosomes of CAFs grown in 5 mmol/L glucose relative to growth in 25 mmol/L glucose (Fig. 6A). These data suggest that the set point for induction of autophagy by chronically reduced glucose is different between CAFs and NFs.

To determine whether reduced glucose affects dipeptide production, the abundance of the 17 dipeptides measured in the metabolomic profiling were determined for 7 CAFs/NFs grown in 5 mmol/L glucose (Fig. 6B). As was the case for the 21 CAF/NFs pairs (Fig. 2), the abundance of dipeptides was increased in CAFs of this subset of 7 pairs grown in 25 mmol/L glucose. However, there was no difference in the abundance of dipeptides in CAFs compared to NFs when the cells were grown in 5 mmol/L glucose (Fig. 6B). The abundance of dipeptides was significantly increased in NFs grown in 5 versus 25 mmol/L glucose, whereas the abundance of dipeptides in CAFs was not significantly altered by growth of cells in 5 mmol/L glucose. Thus, the increase in autophagosome formation induced in NFs by growth in reduced glucose was paralleled by an increase of dipeptides.

Growth of CAFs and NFs in media with varying amounts of glucose (25, 5, or 2 mmol/L) did not reveal a proliferative advantage of CAFs in response to reduced glucose (Supplementary Fig. S4), although the ability of NFs to upregulate autophagy upon chronic reduction of glucose might in this in vitro assay mitigate any advantage of the increased basal autophagy in CAFs.

Autophagic degradation ultimately depends on an active lysosomal system. Acidification is essential for the hydrolytic function of lysosomes and regulation of lysosomal pH is one mechanism for controlling lysosomal activity (30–32). We investigated whether there were changes in lysosomal acidification between CAFs and NFs. Cells were incubated with dual labeled fluorescein rhodamine dextran for 18 hours, and the endocytosed dextran was chased into lysosomes with a 3-hour incubation in medium without dextran. In both CAFs and NFs, the label accumulated in puncta distributed throughout the cell but with some degree of concentration in the perinuclear region characteristic of lysosomes (Fig. 6C). The rhodamine–fluorescein fluorescence ratio of the puncta was measured and converted into a pH values using a pH standard curve (Supplementary Figs. S5 and S6). There were no consistent differences in the average lysosomal pH between CAFs and NFs when grown in 25 or 5 mmol/L glucose (Fig. 6D). In all conditions, lysosomes had a pH value of 4.5, in line with previous measurements of lysosomal pH.

Many tumors are metabolically distinct from nonneoplastic tissue (e.g., ¹⁸F-deoxyglucose PET activity), reflecting the pronounced glycolytic metabolism of cancer cells. It is not known, however, whether the metabolisms of the nonepithelial cells of the tumor are also altered. To begin to explore that question, we used quantitative mass spectrometry profiling of the steady-state abundances of 203 metabolites in primary, nontransformed human lung CAFs and their paired NFs. Although we found no differences in individual metabolites distinguishing CAFs from NFs, we report herein the novel observation of significant differences between CAFs and NFs in the steady-state abundances of metabolites of select metabolic pathways.

In this initial study, because of some practical limitations of working with primary nontransformed cells, we conducted a steady-state metabolite profiling. We analyzed the data using a systems pathway approach by identifying changes in KEGG designation groups of metabolites that differed significantly from the overall differences in the abundance of metabolites between comparison samples. In addition to comparing CAFs with NFs, we also analyzed the data after segregating the samples based on the SUV_max of the tumor: “high” SUV_max (≥8) and “low” SUV_max (≤5). The impetus for this classification was that SUV_max is a measure of the glycolytic activity of the tumor and therefore it was of interest to know whether this is related to the metabolite profile of CAFs. Furthermore, individuals with high SUV_max tumors have a worse outcome on 5-year disease-free survival, so this classification is of potential clinical significance. We detected changes in groups of metabolites that indicate possible changes in metabolic pathways. We followed up one of these changes, increased dipeptides in CAFs, with additional experiments. Additional future studies are required to investigate other metabolic changes identified in this study. Nonetheless, our results establish for the first time metabolic differences between CAFs and their paired NFs that persist in in vitro culture and that are correlated with the glycolytic activity of the parent tumor.

The relative abundance of the 17 unrelated dipeptides was significantly different in the comparisons of the different fibroblast groups. There was an enrichment of dipeptides, relative to other metabolites, in CAFs versus NFs, and in CAFs from high SUV_max tumors compared with CAFs from low SUV_max tumors. Thus, increase in dipeptide abundance is a general characteristic of CAFs that is correlated with the glycolytic activity of the tumor.

The primary source of cellular dipeptides is protein turnover, suggesting differences in protein degradation among the different fibroblast groups. In parallel with the difference in dipepitde abundance, we found an increase in autophagy in CAFs relative to their paired NFs based on increases of LC3-labeled phagosomes as well as an increase in LC3II/I ratio. Autophagy was originally described as a response to nutrient deprivation in which cellular components are degraded to provide substrates for oxidative energy production. However, autophagy is now recognized to have roles in normal basal cellular physiology by mediating the turnover of long-lived proteins, the clearance of damaged mitochondria and other organelles, and the clearance of potentially toxic protein aggregates as a protection against oxidative damage (14, 25). In addition to these prosurvival activities, autophagy can, under certain conditions, result in cell death (14).

Studies of autophagy in tumors have focused on cancer cells (14). Our data support an increase in autophagy in fibroblasts of the tumor microenvironment, suggesting that alterations in autophagy within tumors are not restricted to the cancer cells. We have also discovered a positive correlation between increased dipeptide abundance/increased basal autophagy in CAFs and the tumor SUV_max, revealing a possible link between the glycolytic activity of the tumor and autophagy.

Increased autophagy is believed to promote survival of cancer cells by providing nutrients (amino acids, lipids, and nucleic acids) and/or as a protection against oxidative damage. Increased CAF autophagy might also provide a survival benefit to CAFs in the tumor. Thus, elevated autophagy, by promoting survival of CAFs, could indirectly contribute to the protumorigenic effects of CAFs, as CAFs thriving in the tumor would be able to support cancer cells and tumor growth by one or more of the previously described mechanisms (e.g., paracrine support). An alternative and not mutually exclusive view is that increased autophagy is not limited to cell-autonomous benefits for CAF survival but that increased autophagy in CAFs might be of direct benefit to cancer cells. Some possibilities include increased CAF autophagy as a means of sparing nutrients in the tumor microenvironment for use by the cancer cells or the secretion of nutrients from CAFs generated by autophagy to enrich the nutrient milieu of the tumor microenvironment. The latter is of particular interest with regard to dipeptides, as supplementation of media with certain dipeptides enhances in vitro cell viability (33).

CAFs and NFs similarly induced autophagy when acutely challenged with the severe nutrient restriction of removing glucose and glutamine from the medium. However, only the NFs induced autophagy when challenged with a chronic reduction of glucose in the medium, which indicates that NFs have a lower set point for induction of autophagy controlled by alterations in the nutrient milieu. Because of the high glucose consumption of cancer cells, glucose availability for other cells within the tumor is likely limited. Thus, it is tempting to speculate that the basal upregulation of autophagy in CAFs reflects this difference in set point for induction of autophagy and that it is advantageous for CAF growth/survival in the tumor environment. Although we did not observe consistent differences between growth of CAFs and NFs in normal growth medium (25 mmol/L glucose) or in conditions of “nutrient stress” induced by reduced glucose (5 or 2 mmol/L glucose), the in vitro growth assay does not recapitulate the conditions experienced by cells within the tumor microenvironment. Thus, future studies are required to address the functional consequences of increased CAF autophagy. Regardless, our results establish that autophagy changes in cells of the tumor microenvironment must be taken into account when considering the role of autophagy in solid tumors.

In addition to an increase in the abundance of dipeptides in CAFs from high SUV_max tumors relative to those from low SUV_max tumors, there was an increase, albeit less robust, in the abundance of dipeptides in NFs isolated from individuals with high SUV_max tumors relative to those from individuals with low SUV_max tumors. This link between the SUV_max of the tumor and an increased abundance of dipeptides in fibroblasts outside the tumor microenvironment was unexpected. An explanation for these results is that the possible tumor effects on CAFs are far reaching and that the less robust difference in the fibroblasts from nonneoplastic lung tissue reflects a diminution of the effect due to distance from the tumor. An alternative possibility is that the increased dipeptide abundance of NFs paired with high SUV_max tumors reflects a pre-existing difference independent of the tumor, and that the more robust differences between CAFs from high and low SUV_max tumors is an effect of the tumor re-enforcing this pre-existing difference. Another alternative possibility is that the differences in metabolites between NF derived from high and low SUV_max tumors may reflect differences in host epigenetic alterations in response to exposure to tobacco-related carcinogens.

Here, we have focused on the changes in dipeptides, which led us to consider differences in autophagy/proteolysis between CAFs and NFs. However, other groups of metabolites were significantly different in some of the fibroblast comparisons. In a number of comparisons, there were changes in metabolites of amino acid pathways, and in each case, the change in abundance was in the same direction as that of the dipeptides. These data implicate changes in protein/amino acid metabolism as important differences between CAFs and NFs.

It has been previously suggested that CAFs might undergo alterations in carbohydrate metabolism to support the “Warburg” metabolism of cancer cells. We found a reduction in the steady-state abundance of glycolytic intermediates in CAFs in the comparison of CAFs to NFs from low SUV_max tumors, consistent with this hypothesis. However, this difference was not observed in the comparison between high SUV_max CAFs and NFs. A possible explanation for these results is that the cancer cells of the low SUV_max tumors require alterations in CAF carbohydrate metabolism, whereas glycolytic metabolism of the cancer cells of the high SUV_max tumors is sufficiently robust independent of accommodation in the metabolism of CAFs within the tumor. Regardless, our study reveals the importance of considering the glycolytic nature of the tumor in studies of CAFs.

The profiling was conducted on cells grown in culture. An advantage of this approach over profiling whole tumor tissue is that we were able to specifically examine the nonepithelial cells of the tumor mass. We focused our analysis on vimentin-positive, keratin-negative cells that grew out of dispersed cells from tumor and adjacent nonneoplastic tissues, which is the standard method for isolating CAFs (e.g., ref. 9). Consistent with the literature, the CAF lines as a group expressed more α-smooth muscle actin (α-SMA), a marker for activated fibroblasts, than did paired NFs, but it is important to note that there was considerable heterogeneity among the different lines. However, despite this heterogeneity, our analyses revealed differences between CAFs and NFs that were maintained during in vitro culture. Previous studies have shown that CAFs retain differences from NFs after in vitro culture, including differences in gene expression as well as protumorigenic activities measured in vitro and in vivo (e.g., ref. 9). Thus, our findings of metabolic difference between CAFs and NFs maintained after in vitro culture are in agreement with previous studies.

No potential conflicts of interest were disclosed.

Conception and design: G. Salzler, N.K. Altorki, T.E. McGraw

Development of methodology: S. Dick, O. Elemento, T.E. McGraw

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Salzler, M. Buckman, R. Mohney, B. Stiles

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Salzler, S. Dick, M. Buckman, R. Sordella, E.D. Karoly, R. Mohney, O. Elemento, N.K. Altorki, T.E. McGraw

Writing, review, and/or revision of the manuscript: G. Salzler, S. Dick, M. Buckman, B. Stiles, O. Elemento, N.K. Altorki, T.E. McGraw

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Buckman, N.K. Altorki, T.E. McGraw

Study supervision: O. Elemento, N.K. Altorki, T.E. McGraw

The authors thank Fred Maxfield, Vivek Mittal, and the members of the McGraw and Altorki group for comments on the manuscript.

The study was supported by grants from Columbia University DIRC P30 DK063608-09 (principal investigator: D. Accili; sub-award to N.K. Altorki and T.E. McGraw), Neuberger Berman Lung Cancer Research Center (N.K. Altorki), Seth Sprague Education and Charitable Foundation Award (N.K. Altorki), WCMC Discovery award (T.E. McGraw), STARR Cancer Foundation (R. Sordella), NSF career award 1054964 (O. Elemento), Starr Cancer Foundation (O. Elemento), and Clinical and Translational Science Center at Weill Cornell Medical College TL1RR02499 (G. Salzler).

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.