MEK activation modulates glycolysis and supports suppressive myeloid cells in TNBC

Triple-negative breast cancers (TNBCs) are heterogeneous and aggressive, with high mortality rates. TNBCs frequently respond to chemotherapy, yet many patients develop chemoresistance. The molecular basis and roles for tumor cell–stromal crosstalk in establishing chemoresistance are complex and largely unclear. Here we report molecular studies of paired TNBC patient–derived xenografts (PDXs) established before and after the development of chemoresistance. Interestingly, the chemoresistant model acquired a distinct KRASQ61R mutation that activates K-Ras. The chemoresistant KRAS-mutant model showed gene expression and proteomic changes indicative of altered tumor cell metabolism. Specifically, KRAS-mutant PDXs exhibited increased redox ratios and decreased activation of AMPK, a protein involved in responding to metabolic homeostasis. Additionally, the chemoresistant model exhibited increased immunosuppression, including expression of CXCL1 and CXCL2, cytokines responsible for recruiting immunosuppressive leukocytes to tumors. Notably, chemoresistant KRAS-mutant tumors harbored increased numbers of granulocytic myeloid-derived suppressor cells (gMDSCs). Interestingly, previously established Ras/MAPK-associated gene expression signatures correlated with myeloid/neutrophil-recruiting CXCL1/2 expression and negatively with T cell–recruiting chemokines (CXCL9/10/11) across patients with TNBC, even in the absence of KRAS mutations. MEK inhibition induced tumor suppression in mice while reversing metabolic and immunosuppressive phenotypes, including chemokine production and gMDSC tumor recruitment in the chemoresistant KRAS-mutant tumors. These results suggest that Ras/MAPK pathway inhibitors may be effective in some breast cancer patients to reverse Ras/MAPK-driven tumor metabolism and immunosuppression, particularly in the setting of chemoresistance.


Introduction
Triple-negative breast cancer (TNBC) is a unique clinical subtype of breast cancer characterized by uniform lack of estrogen receptor and progesterone receptor expression and the absence of human epidermal growth factor receptor amplification. TNBC is characterized by diverse chromosomal aberrations and frequent loss of tumor suppressor p53 functions but infrequent druggable mutations. This genomic instability contributes to its heterogeneity and the perception that TNBC is a diverse collection of malignancies rather than a unique molecular entity (1). TNBC also demonstrates high rates of mortality, primarily due to its propensity to metastasize to visceral organs early in the clinical course.
Triple-negative breast cancers (TNBCs) are heterogeneous and aggressive, with high mortality rates. TNBCs frequently respond to chemotherapy, yet many patients develop chemoresistance. The molecular basis and roles for tumor cell-stromal crosstalk in establishing chemoresistance are complex and largely unclear. Here we report molecular studies of paired TNBC patientderived xenografts (PDXs) established before and after the development of chemoresistance. Interestingly, the chemoresistant model acquired a distinct KRAS Q61R mutation that activates K-Ras. The chemoresistant KRAS-mutant model showed gene expression and proteomic changes indicative of altered tumor cell metabolism. Specifically, KRAS-mutant PDXs exhibited increased redox ratios and decreased activation of AMPK, a protein involved in responding to metabolic homeostasis. Additionally, the chemoresistant model exhibited increased immunosuppression, including expression of CXCL1 and CXCL2, cytokines responsible for recruiting immunosuppressive leukocytes to tumors. Notably, chemoresistant KRAS-mutant tumors harbored increased numbers of granulocytic myeloid-derived suppressor cells (gMDSCs). Interestingly, previously established Ras/MAPK-associated gene expression signatures correlated with myeloid/neutrophil-recruiting CXCL1/2 expression and negatively with T cell-recruiting chemokines (CXCL9/10/11) across patients with TNBC, even in the absence of KRAS mutations. MEK inhibition induced tumor suppression in mice while reversing metabolic and immunosuppressive phenotypes, including chemokine production and gMDSC tumor recruitment in the chemoresistant KRAS-mutant tumors. These results suggest that Ras/MAPK pathway inhibitors may be effective in some breast cancer patients to reverse Ras/MAPK-driven tumor metabolism and immunosuppression, particularly in the setting of chemoresistance.

Results
Sensitivity of TNBC PDX models to doxorubicin/cyclophosphamide → taxane ± MEK inhibition. We sought to determine whether addition of a MEKi improved responses to a standard chemotherapy combination regimen in the setting of chemoresistance and to further explore the effects of MEKi treatment on the mammary tumor microenvironment. We began by exploring 5 TNBC PDX models by Western blot analysis to determine the activation status of the Ras/MAPK pathway ( Figure 1A). PDX models BCM-2277 and HBCx1 exhibited the highest activation of phosphorylated ERK1/2 (p-ERK) and thus were selected for further analysis, along with BCM-4013 and BCM-2147 as controls. Of note, BCM-2147 and BCM-2277 were derived from the same patient. Specifically, BCM-2147 was isolated before chemotherapy treatment and progression, whereas BCM-2277 was derived after progression.
Using an accelerated therapeutic regimen mirroring an approved therapy for TNBC (doxorubicin/ cyclophosphamide → taxane; AC→T) with or without the addition of the MEKi trametinib in the taxane (docetaxel) phase (AC→TM), we tracked tumor growth over time. After 4 weeks of therapy (2 weeks of weekly doxorubicin and cyclophosphamide and 2 weeks of docetaxel ± trametinib), tumors were harvested for molecular analysis ( Figure 1B). Interestingly, the only model that exhibited statistically significant reductions in tumor growth with MEKi over the 2-week treatment period were the postchemotherapy/ progression BCM-2277 tumors (P < 0.0001) (Figure 1, C and D). HBCx1, which also showed MEK activation by Western blot, and BCM-2147 showed marginal effects of MEKi that were not statistically significant during the treatment period. HBCx1 tumors also appeared to be more sensitive to taxane treatment than the other models tested. Western blot analysis of treated tumor replicates demonstrated that MEKi treatment decreased both p-ERK and p-S6 ribosomal protein, a marker of increased translation and a downstream target of MAPK signaling, only in BCM-2277, whereas diminished p-ERK but not p-S6 was observed in HBCx1 ( Figure 1E). Previous studies have shown that MEKi treatment activates prosurvival feedback loops in various contexts, which may partially explain the lack of MEKi-specific responses within HBCx1 tumors (23,24). As mentioned previously, HBCx1 tumors appear to be sensitive to taxane therapy, which could also make observing MEKi-specific responses more difficult.
Molecular metabolic imaging of tumor organoids reveals a link between MEK activation and glycolysis. To molecularly characterize the residual tumors from PDX models treated with AC→T or AC→TM, we collected tumor samples at the end of the 4-week treatment. We then used NanoString 3D Biology technology to simultaneously analyze common DNA alterations (104 single nucleotide variants across 25 genes), RNA expression (192 cancer-targeted mRNAs), and protein expression/activation (26 proteins and phosphoproteins) from a single formalin-fixed, paraffin-embedded (FFPE) sample (Supplemental Figure 1; supplemental material available online with this article; https://doi.org/10.1172/jci. insight.134290DS1). Replicate samples from the same tumor demonstrated high reproducibility (r > 0.99) with lower correlations observed between tumors from the same model and even lower correlations between tumors from different models, as expected (Supplemental Figure 2). We initially focused on analyzing the gene expression data generated from the 3D Biology analysis by performing unbiased clustering across the 192 transcripts analyzed. Importantly, each model clustered separately, suggesting gene expression differences between each PDX, as expected (Supplemental Figure 3A). However, only BCM-2277 clustered by MEKi treatment, suggesting that MEKi treatment did not affect the other PDXs as uniformly. HBCx1 did exhibit a partial response to MEKi treatment but had a single MEKi-treated tumor that clustered with the untreated tumor samples. These data suggest that (a) the models are transcriptionally distinct from one another and (b) a substantial effect of MEKi was observed only in BCM-2277, and to a lesser extent in HBCx1, replicating our tumor growth and protein expression data. A correlation matrix across samples also demonstrated that transcription in BCM-2277 bore general resemblance to its parental model, BCM-2147. Interestingly, MEKi treatment in BCM-2277 resulted in gene expression profiles that exhibited increased similarity with the parental BCM-2147 model (Supplemental Figure 3A). These data suggest that MEK activation was a driving force in the molecular differences observed between the 2 models in response to combination chemotherapy treatment and supported an association of Ras/MAPK activation with chemotherapeutic resistance.
DNA mutation analysis was completed by using PCR to amplify genomic loci within 25 genes frequently altered in solid tumors prior to hybridization with specially designed probes for detection of short nucleotide variants. This analysis did not identify any alterations in BCM-2147, HBCx1, or BCM-4013 samples. Yet, a single KRAS Q61R mutation was present in 8/8 BCM-2277 model samples (Supplemental Figure 3B). This mutation is known to be oncogenic (25) but is generally rare in primary human tumors (Supplemental Figure 4) (26). Interestingly, a recent report identified KRAS codon 61 mutations as arising specifically under therapeutic selection, rather than as direct tumor initiators (27). Analysis of thousands of genetic profiles of human breast cancers revealed an enrichment of KRAS mutations in metastatic breast cancers versus primary breast cancer; however, these mutations were all codon 12 and codon 13 mutations (Supplemental Figure 4). Yet, limited data exist on chemotherapy-resistant breast cancers profiled at the end of therapy, representing a knowledge gap of KRAS mutational rates in that population. Thus, the prevalence of KRAS codon 61 mutations in breast cancers that have undergone therapeutic selection remains to be determined.
To gain additional insights into the molecular phenotypes altered by MEK inhibition in the BCM-2277 KRAS Q61R model, we performed differential gene set analysis using the RNA data generated by 3D Biology analysis. Significantly altered gene sets in tumors treated with MEKi suggested suppression of proliferation and suppression of MEK and Ras/MAPK pathway activation, as expected based on our in vivo study (Figure 2A). Interestingly, MEKi treatment also reduced expression of gene sets representing inflammation and growth/metabolism in BCM-2277, suggesting Ras/MAPK signaling drives multiple oncogenic pathways in response to KRAS mutation ( Figure 2A). Further analysis of (phospho)-proteins altered with MEKi also demonstrated suppression of p-AMPKα, a known regulator of metabolic homeostasis (Table 1).
To evaluate the possibility of a MEK-dependent metabolic phenotype, we performed optical metabolic imaging (OMI) (28,29) of tumor organoids derived from BCM-2147 and BCM-2277 PDXs, grown in the presence or absence of MEKi for 72 hours. OMI is a multiphoton fluorescence microscopy technique that quantifies the endogenous fluorescence intensity of cellular NAD(P)H and flavin adenine dinucleotide (FAD) in single cells. The optical redox ratio, defined as the ratio of NAD(P)H fluorescence intensity to that of FAD, is sensitive to shifts in metabolic pathways that oxidize/reduce these coenzymes and reflects the overall redox state of the cell (28,30,31). Thus, inhibition of glycolysis by MEKi would be reflected in a decrease in the redox ratio. Single-cell analysis of tumor organoids across hundreds of cells in each organoid revealed a shift in the redox ratio (corresponding to enhancement of glycolysis versus oxidative phosphorylation) in the BCM-2277 KRAS Q61R model that was completely reversed by MEK inhibition (Figure 2, B and C). OMI studies in TNBC cell lines (MDA-231 [KRAS G13D ], SUM159PT, and BT549) revealed that MEKi had little effect on glycolytic phenotypes in 2D culture (Supplemental Figure 5A); however, in 3D organoids of the same cell lines, MEKi resulted in a marked effect on the redox ratio, particularly in the KRAS-mutant MDA-231 cells (Supplemental Figure 5, B-D). Furthermore, a previously published signature of Ras/ MAPK activation exhibited a positive correlation with the glucose transporter SLC2A3 (GLUT3) and a nonsignificant trend toward positive correlation with the glucose transporter SLC2A1 (GLUT1), both of which are necessary for glycolysis, in publicly available TNBC The Cancer Genome Atlas (TCGA) data (Supplemental Figure 6) (32). Thus, the metabolic phenotypes (i.e., Warburg effect) induced by mutant KRAS (12) appear to be primarily a result of MEK activation, and specific to 3D growth conditions, suggesting a role for hypoxia and/or matrix constituency. These glycolytic phenotypes have been reproducibly linked to chemoresistance (33)(34)(35), resistance to targeted inhibitors (36)(37)(38), and general immunosuppression via reduced T cell activation (33,39). Furthermore, glycolytic function promotes tumor cell survival in hypoxic conditions, further contributing to an immunosuppressive tumor niche (40,41).

Ras/MAPK activation increases cytokine expression in TNBC PDXs and cell lines.
Given the similarity in gene expression and mutational profile between the matched BCM-2147 and BCM-2277 PDXs, we asked what genes were differentially expressed between the 2 models in response to chemotherapy. We identified a distinct gene expression signature of primarily upregulated genes in the KRAS Q61R BCM-2277 model in response to chemotherapy treatment ( Figure 3A). A substantial portion (9/32 altered at least 2-fold with P < 0.05) were also downregulated with the addition of MEKi to the combination chemotherapy regimen by comparing AC→T-and AC→TM-treated samples from the BCM-2277 model. These genes included myeloid-recruiting chemokines CXCL1 and CXCL2 ( Figure 3B and Supplemental Figure 7A). A signature derived from these genes was also downregulated, albeit heterogeneously, with 4 or 24 hours of MEKi treatment (7)   Myeloid recruitment to TNBC is mediated by Ras/MEK-dependent CXCL1/2 expression. CXCL1/2/8, all chemokines that were coordinately regulated by Ras/MAPK activity in PDX models and TNBC cell lines, bind CXCR2 to recruit myeloid cells, including neutrophils and MDSCs, to sites of inflammation. To determine the effects of MAPK signaling on the recruitment of immunosuppressive myeloid cells, we performed immunohistochemistry (IHC) for Gr1 (Ly6C/Ly6G) in tumors from mice treated according to the schema in Figure 1B. Given the marked differences between BCM-2147 and BCM-2277 tumors and the propensity for larger tumors to have increased levels of necrosis, necrotic regions were excluded from IHC scoring. We observed a striking enrichment of Gr1 + myeloid cells in BCM-2277 (KRAS Q61R ), which was markedly reduced with MEKi ( Figure 4, A  tumors demonstrated the majority of CD11b + GR1 + cells to be Ly6G hi Ly6C lo , likely representing a granulocytic MDSC (gMDSC) population, which was high in programed cell death ligand 1 (PD-L1) expression and low in MHC-II expression (Figure 4, E-G). To verify that these were suppressive myeloid cells, Gr1 + cells were isolated by affinity column before coculture with CD3/CD28-stimulated and fluorescently labeled T cells at various dilutions for 72 hours before flow cytometry analysis for cell proliferation (Supplemental Figure 8). An increased proportion of slowly dividing T cells were observed in the 1:1 and 1:2 (T cell/Gr1 + ) cocultures, demonstrating functional T cell suppression by Gr1 + tumor-derived MDSCs (Figure 4, H and I). Furthermore, these Gr1 + cells exhibited marked enrichment for suppressive myeloid genes, including Arg1, INOS, NOX2, and S100A8, compared with initial tumor dissociates or Gr1-depleted dissociates (Figure 4, J-M). Taken together, these findings suggest that MAPK activation in BCM-2277 correlates with increased MDSC tumor recruitment.
Systemic MEKi or CXCR2i treatment reduces suppressive gMDSC accumulation within BCM-2277 tumors. To further explore whether the reduced Gr1 + and Arg1 + cells visualized by IHC were a result of MEKi treatment and not induced by synergistic or chemotherapy-dependent effects, we treated BCM-2277 tumor-bearing mice daily with or without MEKi for 7 days, without any chemotherapy, before flow cytometry analysis of CD45 + CD11b + cells for Ly6G or Ly6C expression. Importantly, CD45 + CD-11b + Ly6G hi Ly6C lo cells, representing the gMDSC population, were decreased from approximately 6% to 1% of live cells within tumors treated with MEKi, but no changes were observed in CD45 + CD11b + populations within the spleen, suggesting that myeloid cell recruitment rather than differentiation was affected by MEK activity (Figure 5, A-C). Given this relatively high degree of myeloid cell infiltration, it is possible that a small portion of the previously described MAPK signaling and glycolytic activation within PDX tumors was due to myeloid cell population changes. To gain further insight on the molecular mechanism mediating gMDSC recruitment to BCM-2277 tumors, we then treated a cohort of tumor-bearing mice with an inhibitor of CXCR2, the receptor through which CXCL1 and CXCL2 signal. Systemic CXCR2i treatment for 7 days decreased both gMDSC and monocytic MDSC accumulation within tumors ( Figure 5D). Importantly, in the absence of an adaptive immune response, the inhibition of CXCR2 + cell recruitment to the tumor exhibited a nonsignificant trend toward reduced tumor growth, suggesting that the loss of CXCR2 + cell recruitment was not sufficient to rescue MEKi-induced tumor suppression (data not shown). Moreover, these data suggest that MEKi and CXCR2i both affect myeloid cell recruitment to tumors and that CXCL1/CXCL2 mediate MEKi-induced effects on myeloid cell populations within BCM-2277 tumors.

CXCR1/2 ligands are associated with Ras/MAPK transcriptional activity in breast cancer cell lines and tumors.
To evaluate the relevance of the association between Ras/MAPK activation and MDSC-recruiting chemokines to human disease, we first explored the Cancer Cell Line Encyclopedia (CCLE database). Across cell lines from all tumor types, we observed enrichment of CXCL1 expression in KRAS MUT cell lines versus KRAS WT cell lines (Supplemental Figure 9). Interestingly, the same effect was not observed in NRAS MUT cell lines, reflecting possible differences in RAS isoform biology. Refining the data to all breast cancer cell lines and using a previously published signature of Ras/MAPK activation (43) because of the low preponderance of activating KRAS mutations in breast cancer, we found strong positive associations of transcriptional activation of the Ras/MAPK pathway with MDSC-recruiting CXCR2 ligands (CXCL1/2/8) but not T cell-recruiting CXCR3 ligands (CXCL9/10/11) ( Figure 6A).
To examine the relevance of these findings to human breast cancer, we analyzed breast cancers in TCGA data set (44) using cBioPortal (45). The Ras/MAPK transcriptional signature was highly associated with expression of CXCL1/2/8 and CSF1/2/3 across TNBC/basal-like breast tumors, while T cell-recruiting CXCR3 chemokines were negatively associated with Ras/MAPK activity ( Figure 6B). Moreover, these correlations between myeloid-associated or T cell-recruiting chemokines were also statistically significant when measured across more than 1000 breast cancer samples from TCGA independent of clinical subtypes, suggesting that Ras/MAPK activity may also regulate chemokine production in other forms of breast cancer (Supplemental Figure 10). In a microarray data series of 201 genetically engineered mouse models of breast cancer, similar associations for Cxcl1 were also identified (Supplemental Figure 11). Thus, activation of the Ras/MAPK pathway, either through oncogenic KRAS activation, or other mechanisms, may drive MDSC recruitment in breast cancers in both mice and humans. While Ras/MAPK score was not associated with poor survival in TCGA TNBC/basal-like breast cancer subset, previous work by our group demonstrated that increased Ras/MAPK score correlated with poor prognosis in residual disease after neoadjuvant chemotherapy in a TNBC cohort (21). To confirm the relationship between MDSC recruitment and Ras/ MAPK activity at the protein level in human TNBC, we performed multiplexed immunofluorescence for the MHC-II protein human leukocyte antigen DR isotype (HLA-DR), pan-cytokeratin, and CD11b in a tissue microarray comprising 61 cases of TNBC after neoadjuvant chemotherapy (all residual disease) (6,21) where mRNA expression data for the Ras/MAPK transcriptional signature (10, 43) were available (Figure 6C). Using HLA-DR -(MHC-II -) CD11b + cells as a marker of immunosuppressive myeloid cells, we confirmed a positive correlation (P = 0.02; r = 0.28) between Ras/MAPK signaling and suppressive myeloid cell presence in TNBC ( Figure 6D). Given that MEKi elicit immunologic effects as well as direct antitumor effects (10,46) and are actively being combined with immune checkpoint inhibitors in breast cancer, suppression of MDSC recruitment via transcriptional inhibition of CXCL1/2/8 may be a novel mechanism of combinatorial activity.

Discussion
Our work has focused on the study of paired breast cancer PDXs before and after the development of chemotherapy resistance to identify targetable, KRAS mutant-associated signaling mechanisms that promote tumor progression and alter the tumor immune microenvironment. Specifically, 3D Biology analysis of PDX tumors treated with a combination of chemotherapy and MEKi identified immune and metabolic pathways that were inhibited by MEKi treatment. Using OMI (28, 29), we observed glycolytic reduction/oxidation changes in organoids derived from PDXs that corresponded to the acquisition of KRAS Q61R and that were reversed completely by MEKi. Furthermore, we identify recruitment of immunosuppressive Gr1 + myeloid cells via tumor cell expression of CXCL1, CXCL2, and CXCL8 (CXCR2 ligands) expression, which were substantially reduced with MEKi or CXCR2 blockade. These findings were validated across 115 cases of human TNBC, where we observed that Ras/MAPK transcriptional patterns demonstrate correlations with expression of myeloid-recruiting chemokines and inverse correlations with T cell-recruiting chemokines.
The activation of the Ras/MAPK pathway has previously been implicated in breast cancer and other tumor types as a source of immunosuppressive signals (11). Our recent work has identified suppression of antigen presentation as an additional mechanism whereby Ras/MAPK suppresses antitumor immunity, 72 hours of coculture with Gr1 + cells and CD3/CD28 bead stimulation measured by CellTrace Far Red fluorescence. (I) Distribution of T cell proliferation in 72-hour cocultures with Gr1 + cells across 3 independent experiments. (J-M) RNA isolated from tumor dissociates, Gr1 + cells, and Gr1-depleted dissociates was probed for Arg1, INOS, NOX2, and S100A8 by qRTPCR (n = 3). despite its proliferative effects on naive T cells (10,47). Recent studies of Kras-driven murine models have found associations between tumors induced by activated Kras and secretion of CXCR2-binding cytokines (13,48,49), although these studies have focused on the role of Kras-induced NF-κB pathway activity, rather than MEK. Allegrezza et al. demonstrated that MEKi treatment could reduce tumor-infiltrating MDSCs in Kras-mutant models; however, the mechanism was not defined within that study (46). Recent work in Kras-mutant colorectal cancer models suggests that Kras regulates MDSC recruitment via IRF2 and tumor cell secretion of CXCL3 (50). While the mechanism seems to vary between tissues, the connection between increased Ras/MAPK activity and MDSC recruitment appears to be conserved, because our data suggest that MEK activation downstream of oncogenic KRAS drives MDSC recruitment to breast tumors.
MEK activation and MDSC recruitment may be targetable for some breast cancer patients; however, MEK inhibition did not completely reverse suppressive myeloid cell recruitment in vivo, suggesting additional pathways may contribute to this phenotype. Furthermore, a high degree of Gr1 + cells was not observed in the other 3 models tested, despite HBCx1 demonstrating moderate ERK activation and marginal sensitivity to MEKi. This finding could be explained by the need for additional activation of NF-κB resulting from the consistent signaling induced by activated KRAS (46,48,49) in the BCM-2277 model. Several studies have attributed effects of oncogenic KRAS to this pathway, and chemokine/secretory phenotypes such as expression of CXCL1/2/8 have been shown to rely on NF-κB pathway activation.
Nonetheless, our findings directly implicate activation of MEK in the recruitment of suppressive myeloid cells and in the generation of glycolytic phenotypes that may further feed into immunosuppressive tumor microenvironments. Because early-phase trials show promising activity in TNBC with combinations of MEKi and taxanes (18), the effects on MDSC recruitment and metabolic phenotypes should be explored as potential biomarkers associated with outcome. Furthermore, new clinical trials are now being initiated using MEKi, taxanes, and anti-PD-L1 antibodies, which unleash suppressed T cells in the microenvironment (51). Given the role of MDSCs in T cell suppression (52), the blockade of MDSC recruitment coupled with therapies that reinvigorate T cell responses may be more successful therapeutically in these trials. To reduce the likelihood of genetic drift, only PDXs in the first 5 passages were used for study. Mice bearing tumor sizes at least 150 mm 3 were randomized to treatment with doxorubicin (2 mg/kg/wk i.p.) and cyclophosphamide (100 mg/ kg/wk i.p.) for 2 weeks, followed by docetaxel (20 mg/kg/wk i.p.) and oral gavage vehicle or docetaxel + trametinib (1 mg/kg/d by mouth). Trametinib dosing was completed on a per weight basis, building on a previous pharmacokinetic study in rodents (54). At the end of 4 total weeks of treatment (2 weeks of doxorubicin + cyclophosphamide and 2 weeks of docetaxel ± trametinib), mice were euthanized, and residual tumors were resected for analysis. During the study, tumor diameters were measured using calipers 3 times per week, and volume in mm 3 was calculated with the following formula: volume = width 2 × length/2.  (6,21). Data were assessed in tissue microarray format, using the average cell number across 3 independent cores per patient sample.
NanoString 3D analysis. Two 5 μm sections from FFPE blocks were cut onto slides for simultaneous DNA, RNA, and protein analysis. The expression of 26 proteins and 192 RNAs as well as the analysis of 104 clinically actionable single nucleotide variants (SNVs) were simultaneously measured using the nCounter Vantage 3D Solid Tumor Assay for FFPE (NanoString Technologies) reagent using NanoString protocols. Briefly, for protein analysis, sections were subjected to deparaffinization and rehydration followed by epitope retrieval with pH 6.0 citrate buffer and overnight incubation with the DNA-labeled antibody mix at 4°C. Following washes to remove nonspecific antibodies, the slides were placed on an ultraviolet transilluminator for 3 minutes to release the photo-cleavable DNA tags. The DNA tags were denatured and then hybridized to target specific fluorescent barcodes. For RNA expression and SNV analysis, RNA and DNA were purified from a single 5 μm section using the QIAGEN AllPrep kit. DNA mutational analysis was conducted using a commercial multilocus targeting assay kit, the nCounter Vantage 3D DNA SNV Solid Tumor Panel, on an nCounter MAX Analysis System. The assay uses multiplex PCR to amplify 40 human genomic loci from 25 genes that are frequently mutated in solid tumors. Following the PCR step, the DNA amplicons are interrogated by specialized DNA probes designed to specifically hybridize to short nucleotide variants (single-and dinucleotide substitutions and insertions or deletions of up to 18 nucleotides) and additional probes that specifically hybridize to the GRCh37/hg19 reference sequence that corresponds to the position of each assayed variant. After hybridization, stable complexes are immobilized and counted on the nCounter system. For RNA expression analysis, 100 ng of RNA was hybridized to target specific fluorescent barcodes. The hybridized samples for DNA, RNA, and protein analysis were simultaneously analyzed on the NanoString nCounter MAX Analysis system followed by data processing and primary analysis with the NanoString nSolver data analysis software.
Tumor cell dissociation and in vitro T cell assay. Splenocytes were isolated from 8-week-old BALB/cAnNCr mice (Envigo, formerly Harlan) and labeled with CellTrace Far Red dye for 30 minutes in serum-free PBS (Thermo Fisher Scientific, 1:1000). BCM-2277 tumors were harvested from mice after reaching more than 500 mm 3 for tumor cell dissociation (in Serum-Free RPMI from Gibco, Thermo Fisher Scientific, 2.5 mg/mL, and Collagenase 3, 62.5 μg/mL, from Worthington) using the gentleMACs Octo dissociator (Miltenyi Biotec) default tumor protocol for 30 minutes at 37°C under constant agitation. The tumor dissociate was then passed through a 40 μm filter and washed with 20-30 mL of PBS. MDSCs were isolated using the Myeloid-Derived Suppressor Cell Isolation Kit (Miltenyi Biotec, mouse) according to the manufacturer's protocol. The Gr1 + enriched cells from this isolation were then counted using trypan blue and a LUNA-II automated cell counter (Logos Biosystems) and plated with negatively selected labeled T cells (Pan T Cell Isolation Kit, Miltenyi Biotec) isolated from WT BALB/c mice (Envigo, formerly Harlan), for a total of 30,000 cells per well in RPMI medium (1% HEPES, 50 μM BME, 10 ng/μL mouse IL-2, BD Biosciences), in a round-bottom 96-well cell culture plate (Corning). After 72 hours of CD3/CD28 Dynabead (Thermo Fisher Scientific) stimulation, the cells were processed for flow cytometry analysis using an Attune NxT flow cytometer (Life Technologies, Thermo Fisher Scientific).
Organoid OMI. Autofluorescence images were acquired using an inverted custom-built multiphoton system (Bruker Fluorescence Microscopy), using either a ×40 oil immersion objective (1.3 NA, Nikon) or a ×40 water immersion objective (1.15 NA, Nikon). A titanium/sapphire laser (Chameleon Ultra II, Coherent or InSight DS+, Spectra-Physics) was used for 2-photon fluorescence excitation. NAD(P)H was excited using 750 nm light, and a 440/80 nm bandpass filter was used to collect its emission. FAD was excited at 890 nm and a 550/100 nm filter was used to collect its emission. A pixel dwell time of 4.8 μs and a total integration time of 60 seconds was used to collect 256 × 256. A GaAsP photomultiplier tube (H7422P-40, Hamamatsu Photonics) detected emitted photons.
After 72 hours of treatment, cell lines and organoids were imaged at 3-4 locations per dish for a total of 100-1000 cells imaged per treatment group for cell lines and 60-280 cells per group for organoids. NAD(P) H images were first acquired, followed immediately by an FAD image of the same field of view. All OMI experiments were repeated in triplicate.
CellProfiler was used to automatically identify individual cells and isolate average fluorescence intensity values for each (minus background and nuclear signals) (58). Optical redox ratio values were calculated for each cell by dividing the average intensity of NAD(P)H by the average intensity of FAD. Redox ratios for all cells in a treatment group were averaged together and normalized to control values within an experiment.
Statistics. Statistics were performed in GraphPad Prism or R (www.r-project.org). In data with 2 groups, 2-sample 1-or 2-tailed t tests were used. For analyses with more than 2 groups, significant differences were determined by 1-way ANOVA with a Tukey's post hoc adjustment for multiple comparisons. For all multiple comparisons, statistical significance is noted by *P < 0.05, **P < 0.01, ***P < 0.001, and ***P < 0.0001. A P value of less than 0.05 was considered statistically significant. Bar graphs show mean ± SEM, unless otherwise stated in the figure legend. For correlations, the Pearson correlation coefficient was used to test significance of association.
Study approval. Athymic mouse experiments were approved by Vanderbilt University's (VUMC) comprehensive Animal Care and Use Program (ACUP). The VUMC ACUP is registered with the United States Department of Agriculture (USDA registration 63-R-0129) and operates under a Public Health Service Animal Welfare Assurance Statement (PHS Assurance A3227-01). The VUMC ACUP has been accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care, International, since 1967 (AAALAC file 000020) and most recently received continued full accreditation on June 21, 2017.