Reagents and kits

Microbiome collection kits were obtained from TransnetYX (Cordova, TN). McCoy’s 5a Medium (30–2007) and fetal bovine serum (30–2020) were purchased from American Type Culture Collection (ATCC, Manassas, Virginia). Puromycin (ant-pr) was purchased from InvivoGen (San Diego, CA). XenoLight D-luciferin potassium salt (122799) was purchased from PerkinElmer (Waltham, MA). ThermoScientific (Waltham, MA) was the source of glutathione mouse monoclonal antibody (D8; MA1-7620) and the Pierce™ reagent HENS buffer (90106). PerkinElmer (Waltham, MA) was the source of Bioware® Brite Cell Line HCT116 Red-FLuc (BW124318). 10X Genomics (Pleasanton, CA) was the source of all Visium reagent kits for whole transcriptome profiling of intact formalin fixed paraffin embedded (FFPE) tissue sections including test slide (PN-1000347), slide kit (PN-1000188), reagent kit (PN-1000361), human transcriptome probe kit (PN-1000361), accessory kit (PN-1000194) and dual index kit TS Set A (PN-1000251). KAPA SYBR FAST qPCR master mix (KK4600) was purchased from Roche Diagnostic Corporation (Indianapolis, IN). SPRIselect (B23317) was purchased from Beckman Coulter Life Sciences (Indianapolis, IN). ProLong Gold antifade reagent with DAPI (P36935) and Streptavidin-FITC (19538–050) were from Invitrogen by ThermoFisher Scientific. 3-(N-Maleimidylpropionyl) biocytin (MPB) was from Santa Cruz Biotechnology (sc-216373). The following were purchased from Millipore Sigma: NADPH, tetrasodium salt (481973), Glutaredoxin-S2, E. coli (GRX1, 354406), Glutathione reductase (GSSG, G3664), and glutathione (GSH, Y0000517). Primary and secondary antibodies used with FFPE tissue sections in immunohistochemistry (IHC-P) and confocal immunocytochemistry (ICC/IF) are listed in Table 1. All protocols are accessible in protocols.io (doi.org/10.17504/protocols.io.ewov1oe37lr2/v1).

Gut microbiome

Individual mice were allowed to defecate normally in autoclaved cages with no bedding. Using a sterile 26Gx1/2 needle (one per mouse), the first two fecal pellets per mouse were submerged in stabilization buffer in a barcoded sample collection tube. Collection tubes were shipped to TransnetYX for DNA extraction (Qiagen DNeasy 96 PowerSoil Pro QIAcube HT kit/protocol), library preparation (KAPA HyperPlus protocol), and sequencing. High molecular weight genomic DNA captured the true microbial diversity of stool samples from two cohorts: six- to eight-week-old Hltf^+/+ (n = 9) and Hltf^-/- (n = 5) immune deficient (ID) male mice. Libraries were sequenced using Illumina NovaSeq at a dept of 2 million read pairs (2x150 configuration) sufficient for species and strain level taxonomic resolution. FASTQ files were uploaded onto One Codex analysis software [11] and analyzed against the One Codex database of ~127K whole microbial reference genomes. Classification results were filtered through several statistical post-processing steps to eliminate false positive results caused by contamination or sequencing artifacts. Mouse 12K metagenome-assembled genomes were included to screen out host reads. One Codex annotated metagenomes and generated a taxa plot of the top 10 species for all samples, and calculated alpha (within-sample) family diversity for comparison of the two cohorts.

HLTF-deleted isogenic HCT116-Red-FLuc cells

Bioware® Brite Cell Line HCT116-Red-FLuc is a luciferase-expressing derivative of a highly metastatic cell line (HCT116 parental line, CCL-247). PerkinElmer’s recommendation that the Bioware® Brite Cell Line HCT116-Red-FLuc only be passaged 10 times, precluded the use of CRISPER/Cas9 to generate a stable knockout of HLTF. Thus, System Biosciences (Palo Alto, CA) produced a custom stable cell line with shRNA sequence targeting of HLTF [12]. Briefly, synthesized HLTF shRNA sequence was cloned into pGreenPuro shRNA expression lenti vector (SI505A-1). HCT116-Red-Fluc cells infected with 10 MOI of HLTF-shRNA-SI505 virus (lot#171009–008) yielded puromycin-resistant polyclonal HCT116-Red-FLuc HLTF^-/- cells. For each CDX experiment, cell stocks in liquid nitrogen at passage 2 were thawed using T25 flasks, expanded in T150 flasks for 2 days, passaged overnight and harvested at 70–75% confluency. We used this protocol as a unified framework for comparative gene expression analysis.

Hltf-deleted mice

Immune competent (IC) global Hltf knockout (KO) mice were developed in collaboration with genOway (Lyon, France) as previously described [13]. Speed congenics accelerated the generation of IC Hltf KO C57BL/6J mice [14] that were bred (IACUC# 02007) into the recombinase activating gene 2 (Rag2)/common gamma (IL2rg) double knockout background [15] to generate ID Hltf KO mice >99% C57BL/6 background genome. ID Hltf KO mice were housed in a barrier facility with a 12:12 light/dark cycle, access to food and water ad libitum and bedding changes 2–3 times/week. Routine testing of sentinel mice ensured the colony was disease free. All studies and the anticipated mortality were conducted in accord with the NIH Guidelines for the Care and Use of Laboratory Animals, as reviewed and approved by the Institutional Animal Care and Use Committee at Texas Tech University Health Sciences Center (NIH Assurance of Compliance A3056-01; USDA Certification 74-R-0050, Customer 1481, S1 Checklist). TTUHSC’s IACUC (# 02009) specifically approved this study.

For the orthotopic HLTF^-/-HCT116 xenograft model, six- to eight-week old ID Hltf KO mice (n = 7) received direct OCMI of HLTF^-/-HCT116-Red-Fluc cells (2x10⁶ cells/10 μl) between the mucosa and the muscularis layers of the cecal wall. Each surgery was performed with isoflurane (isothesia) and the SomnoSuite® Low-Flow anesthesia system (Kent Scientific) with far infrared warming pads during surgery and recovery. Efforts to minimize suffering included an IP injection of Buprenorphine (Buprenex, 0.1 mg/kg) prior to surgery to manage incisional pain followed by a second dose 4–8 hours later, if needed. The cecum was exteriorized via small midline incision on the linea alba between the rectus abdominus muscles to eliminate bleeding. Non-invasive bioluminescence imaging (BLI) with an IVIS Spectrum In Vivo Imaging System was used to validate the quality and accuracy of the injection, and to track and quantify tumor growth and metastasis. Histopathology at necropsy confirmed placement of the inoculum. Mouse behavior and well-being were monitored daily. Tumor growth/metastasis was monitored weekly with BLI.

Mice were placed under continuous isothesia for imaging and killed immediately thereafter (<15 seconds) by cervical dislocation at the humane endpoint of 28 days post CDX establishment. Primary HLTF^-/-CDX (hereafter HLTF^-/-CDX) tumors were quickly removed, rinsed in physiological saline and either flash frozen for biochemical evaluation (RNAseq, Western blotting, MALDI-TOF/TOF MS, nanoLC-MS/MS) or formalin fixed paraffin embedded (FFPE) for routine histopathology, IHC-P, laser scanning confocal ICC-IF and spatial transcriptomics.

Microscopy

FFPE HLTF^-/-CDX were serially sectioned (3–4 μm) for histopathology, IHC-IP and ICC-IF. Two sections were placed on each slide and deparaffinized prior to staining. Beginning with the first slide, sections on every fifth slide were stained with hematoxylin and eosin (H&E) for evaluation by light microscopy. Sections on alternate slides were processed for IHC-IP with heat-induced epitope retrieval (HIER). Two tissue sections per slide facilitated the use of one section for positive immunostaining, and the companion section for negative (minus primary antibody) control staining. Pairing of primary antibodies with secondary antibodies is shown in Table 1. Nuclei were counterstained (blue) with hematoxylin. ICC-IF (Table 1) was performed with serial sections from the above described FFPE tumors with HIER, aldehyde quench (50 mM NH₄Cl in PBS), and ProLong Gold DAPI (DNA-specific blue-violet fluorescent dye).

Protein S-glutathionylation (Pr-SSG) was detected in situ in FFPE HLTF^-/-CDX sections (3–4 μm) with a biotin switch assay [16]. Free thiol groups were blocked in HENS buffer (100mM HEPES, pH 7.8, 1 mM EDTA, 0.1 mM Neocuproine, 1% SDS) containing 20 mM MMTS (methyl methanoethiosulfonate) and 1% triton (vol/vol) for 10 minutes. After three rinses in PBS, S-glutathionylated cysteine groups were reduced by incubation with 27 μg/ml E coli GRX1, 4 U/ml GSSX, 1 mM GSH, 1 mM NADPH in 1 mm EDTA/10mM Tris pH 8 for 20 minutes. After three rinses in PBS, newly reduced cysteine residues were labeled with 1 mM MPB for 60 minutes. Excess MPB was removed with six rinses in PBS and tissue samples were incubated with 0.5 μg/ml streptavidin-FITC for 60 minutes. Excess streptavidin-FITC was removed with 6 rinses in PSB and cover slipped with ProLong Gold antifade reagent with DAPI to stain nuclei. Slides were allowed to cure overnight at room temperature and then stored at -20C until confocal imaging. Controls were treated the same except E. coli GRX1 was omitted from the deglutathionylation step.

Tumor transcriptome (RNAseq)

HLTF^-/-CDX (1 per ID Hltf KO mouse x 3 biological replicates) were flash frozen and sent to Genewiz, the next generation sequencing (NGS) division of Azenta Life Sciences (South Plainfield, NJ) RNA-sequencing. Briefly, total RNA was isolated, and evaluated for its integrity and purity with an Agilent Bioanalyzer (Table 2). RNA was polyA selected prior to library preparation (Illumina Stranded mRNA prep) followed by Illumina HiSeq, 2x150 configuration, single index per lane with 350 million raw paired-end (PE) reads per lane, i.e. 40–60 million PE reads per sample (>80% of bases >Q30). Sequence reads were trimmed to remove possible adapter sequences and nucleotides with poor quality using Trimmomatic v.0.36. The trimmed reads were mapped to either the Homo sapiens GRCh38 reference genome or the Mus musculus GRCm38 reference genome on ENSEMBL using the STAR aligner (v.2.5.2b) that detects splice junctions and incorporates them to help align the entire read sequences. BAM files were generated as a result of this step.

Unique gene hit counts were calculated by using featureCounts from the Subread package v.1.5.2. The hit counts were summarized and reported using the gene_id feature in the annotation file. Only unique reads that fell within exon regions were counted. After extraction, gene hit counts were used for downstream differential gene expression analysis (DESeq2) of HLTF^-/- and HLTF^+/+CDX from ID Hltf KO mice. Log2 fold changes (L2 FC) were calculated. P-values were obtained with the Wald test, and adjusted p-values (padj) were calculated with the Benjamini-Hochberg method. Genes with padj < 0.05 and absolute L2 FC >1 were called as differentially expressed genes for each comparison. New RNAseq data for Hltf^-/-CDX in ID Hltf KO are accessible through NCBI’s Gene Expression Omnibus (GEO) SuperSeries accession number GSE234728, SubSeries GSE234725 (RNA-Seq). RNAseq data for Hltf^+/+CDX in ID Hltf KO are accessible through GSE161961.

Spatial transcriptomics

Bioinformatics analysis

Visium Spatial Gene Expression Software Suite includes Space Ranger and Loupe Browser. Spaceranger mkfastq demultiplexed the Illumina sequencer’s base call files (BCLs) for each flow cell directory into FASTQ files. Spaceranger count combined a brightfield microscope slide image and FASTQ files from spaceranger mkfastq and performed alignment, tissue detection, fiducial detection, barcode/UMI counting, and prepared a full resolution slide image for visualization in Loupe Browser. The pipeline used the Visium spatial barcodes to generate feature-spot matrices, determine clusters, and perform gene expression analyses. The pipeline used a probe aligner algorithm for FFPE tissues. Spaceranger aggr used the output of multiple runs of spaceranger count from related samples and aggregated their input, normalizing those runs to the same sequencing depth, and then recomputed the feature-barcode matrices and the analysis on the combined data. The aggr pipeline combined data from multiple samples into an experiment-wide feature-barcode matrix and analysis. Loupe Browser was used to interrogate significant genes, characterize and refine gene clusters, and perform differential expression analyses. Spatial transcriptomics data for HLTF^-/- and HLTF^+/+CDX tumors in ID Hltf KO mice are accessible through NCBI’s Gene Expression Omnibus (GEO) SuperSeries accession number GSE234728, SubSeries GSE234517 (Spatial).

Protein S-Glutathionylation (Pr-SSG)

Parallel analysis of Pr-SSG in HLTF^-/-HCT116-Red-Fluc cells used for OCMI and resultant HLTF^-/-CDX in ID Hltf KO mice was performed by Applied Biomics, Inc (Hayward, CA). Briefly: cell pellet (control) and CDX (n = 2 mice) were homogenized/sonicated (Polytron) in 4 volumes HENS buffer. Protein concentrations were determined (OD₂₈₀) with NanoDrop One (ThermoFisher), adjusted to a final concentration of 2 μg/μl with HENS buffer and frozen for analysis.

Differential Pr-SSG between cell and tumor proteins was evaluated by large-format two-dimensional gel electrophoresis (Western blot) and two-dimensional difference gel electrophoresis (2D DIGE) in which cell (gel 1) and tumor proteins (gel 2) labeled with Cy2 (internal standard) were interrogated with mouse monoclonal anti-glutathione (0.5 μg/ml) and Cy5-labeled donkey anti-mouse IgG (1:2000 dilution). These two gels provided information on the difference in glutathionylation between cells and tumor (glutathionylation ratio). Difference in protein levels between the two samples (ratio) was determined by comparing Cy2 labeled cell proteins with Cy3 labeled tumor proteins by 2D DIGE (gel 3). The final glutathionylation ratio for each spot was obtained by adjusting the glutathionylation ratio by the protein ratio. 2D-DIGE guided the excision of 61 spots. Protein identification was based on peptide fingerprint mass mapping (using MS data) and peptide fragmentation mapping (using MS/MS data). MASCOT search engine was used to identify human and mouse proteins from primary sequence databases (SWISSProt). Posttranslational modification (glutathionylation) authentication was based on matrix-assisted laser-desorption ionization time-of-flight (MALDI-TOF/TOF).

HLTF-deleted isogenic HCT116-Red-FLuc cells

The goal of the second experiment was to characterize the HCT116-Red-FLuc isogenic cell line. The shRNA-silenced stable isogenic cell line was expanded and monitored for GFP expression (Fig 2). When grown in McCoy’s 5a Modified Medium supplemented with 10% fetal bovine serum and puromycin (2 μg/mL) HLTF^-/-HCT116-Red-FLuc cells averaged the same doubling time (16 hours), shared the same morphology in culture, and had the same kinetic curve for their bioluminescence signal post-Luciferin injection (peak time 10–20 minutes) as parental HLTF^+/+HCT116-Red-FLuc cells. Cells were further characterized as controls in the 2D-DIGE MALDI-TOF/TOF mass spectrometry experiments.

Experimental timeline ensured mouse survival

The goal of the third experiment was to establish primary xenografts with HLTF^-/-HCT116-Red-FLuc cells in ID Hltf KO mice. When Cespedes et al [19] developed the OCMI model, HCT116 cells in Swiss Nu/Nu mice produced highly aggressive, disseminating stage IV tumors that caused death in 39+4 days. We implemented the model to confirm the negative correlation between HLTF expression in tumor cells and survival in mice previously identified with a non-metastatic AOM/DSS model [20]. ID Hltf KO mice are associated with a worse prognosis and are 5-times more likely to die during the treatment protocol compared to ID Hltf^+/+ mice [10]. Establishing a survival timeline at 28 days for ID Hltf KO mice inoculated with HLTF^-/-HCT116-Red-FLuc cells (Fig 2) ensured all mice survived to the end of this study.

HLTF expression is negatively associated with survival in human CRC

The negative correlation between HLTF expression and the progression of CRC is well documented in humans [5, 21–25]. With early-stage CRC, HLTF is detectable in tumor cells but negligible in the TME. With progression, epigenetic silencing of HLTF in tumor cells coincides with negligible HLTF expression in the TME. The HLTF^-/-CDX/TME model in this study recapitulates end stage disease progression. As shown in Fig 2, the absence of HLTF protein expression is indistinguishable from the moderately differentiated adenocarcinoma of cecum from a 54-year-old male (T4aN2bM1). The patient’s primary tumor invaded through the muscularis propria and penetrated to the surface of the visceral peritoneum (T4a). Metastasis was found in a total of 19/27 regional lymph nodes (N2b), with spread to liver, lung, and omentum (M1). All CDX attached to the parietal peritoneum (T4b) as with end stage human CRC (peritoneal carcinomatosis).

Comparative CDX transcriptomics

In the fourth experiment, DESeq2 was used for comparison gene expression analysis between HLTF^-/- and HLTF^+/+CDX from ID Hltf KO mice. Genes with padj < 0.05 and absolute L2FC > 1 were called as differentially expressed (DE) genes. There was a total of 1,599 upregulated genes and 3,493 downregulated genes (Fig 3). Following species-specific transcriptome alignment, DE genes were clustered by their gene ontology (GO). Enrichment of GO terms for HLTF^-/-CDX vs HLTF^+/+CDX in ID Hltf KO mice was tested using Fisher exact test (GeneSCFv1.1-p2).

GO analysis identified transcriptional activation (padj 0.04649) of 39 genes involved in the cellular response to a DNA damage stimulus (GO:0006974; Fig 3) coincident with γH2AX detection in HLTF^-/-CDX ID Hltf KO mice (Fig 4). Human HLTF is a multidomain protein that safeguards genomic integrity by exerting independent functions in nucleotide excision repair (NER) and post replication repair (PRR) including translesion synthesis [25, 26]. While there is a paucity of information regarding HLTF loss of function in NER and PRR during in vivo tumorigenesis, DNA damage from double stranded breaks was quickly followed by phosphorylation of Ser-139 of the histone variant H2AX forming γH2AX [27]. Detection of nuclear foci of newly phosphorylated protein—a reliable marker of DNA damage initiation and resolution [28]—was accompanied by transcriptional activation (padj 0.00588) of 43 genes involved in DNA repair (GO:0006281; Fig 3).

Oxidative stress and HIF-1α

Excessive production of reactive oxygen/nitrogen species (ROS/RNS) causes oxidative stress that has been implicated in the development and progression of CRC [29]. Mitochondrial dysfunction results from oxidative stress [30]. Transcription factor hypoxia inducible factor-1α (HIF-1α) mediates the mitochondrial response to oxidative stress [31] via two mechanisms [32, 33]. One is transcriptional activation in response to hypoxia, and the second is by a GSH adduct on human Cys520 (mouse Cys533) that stabilizes HIF-1α protein. Hltf is a transcriptional regulator of murine hypoxia inducible factor-1α (Hif-1α) in heart [14]. We previously reported [10] HIF-1α gene transcription in HLTF^+/+CDX ID Hltf KO mice was unaffected by the presence or absence of Hltf in the TME. In the fifth experiment, we show deletion of HLTF from the tumor and the TME had no effect on total HIF-1α transcription or on its downstream targets. Neither the hypoxia-responsive genes involved in angiogenesis (VEGF-A, PIGA, ANG2, PDGF-B) nor the glycolytic enzymes (ALDA, LDHA, PGK1, PKM) with hypoxia response elements were transcriptionally activated. Cluster experiments from spatial transcriptomics allowed us to visualize HIF-1a throughout the tumors (Fig 5). The data set contained 88.7% of the reads in spots under tissue, encompassing 17,998 unique genes, and approximately 5,909 unique genes were identified per spot. Additional information about the spatially resolved dataset is provided in Table 4. The t-distributed neighbor embedding (t-SNE) projection plots—a statistical method for understanding high-dimensional datasets—identified UMI counts and 16 distinct clusters that strongly recapitulated CDX histology (Fig 5).

OXPHOS and Pr-SSG

In the absence of HIF-1α activation, a total of 23 human genes in mitochondrial electron transport, NADH to ubiquinone (GO:0006120), were transcriptionally activated (padj 3.73E-05) in conjunction with 25 human genes (padj 9.83E-05) in the mitochondrial respiratory chain complex 1 assembly (GO:0032981). Human MT-CYB (also CYTB)—a mitochondrially encoded component of the ubiquinol-cytochrome c reductase complex (complex III or cytochrome b-C1 complex), a respiratory chain that generates an electrochemical potential coupled to ATP synthesis—was one of the genes transcriptionally activated (Fig 3) in HLTF^-/-CDX/TME (L2FC 3.03; padj 1.7E-44). Immunolabeling of MT-CYB protein—a marker for OXPHOS—identified mitochondria in tumor cells in the Hltf^-/-CDX and in the host lymphovasculature (Fig 6). These findings in our preclinical model are consistent with elevated MT-CYB expression as a prognostic marker for metastatic and advanced clinical stage disease [34].

Mitochondria are key generators of cellular reactive ROS [35] and they have multiple antioxidant strategies to neutralize ROS including glutathione (GSH; γ-glutamyl-cysteinyl-glycine), a low molecular weight antioxidant exclusively synthesized in the cytosol in two ATP-dependent steps [36]. Solute carrier family 7 member 11 (SLC7A11) also known as xCT mediates the uptake of extracellular cystine in exchange for glutamate [37]. Cystine is reduced to cysteine. In the first step, γ-glutamylcysteine is synthesized from L-glutamate and cysteine by glutamate cysteine synthetase (GCL). In the second step, glycine is added to the C-terminal of γ-glutamylcysteine by glutathione synthetase (GSS), which is also a potent antioxidant. In the CDX tumors, transcriptional availability of the SCL7A11 gene is the same for HLTF^+/+ and HLTF^-/-CDX ID Hltf KO mice. SLC7A11-AS1, an overlapping cis-natural antisense transcript at the xCT gene locus that inhibits xCT expression was transcriptionally downregulated (L2FC -6.87; padj 3.86E-09) in HLTF^-/-CDX.

Spatial transcriptomics provided an unprecedented view of region-specific human GSH gene expression and compensated for the loss of structure that resulted from tissue dissociation for RNAseq experiments. It enabled us to simultaneously document cancer cell location in the CDX histology sections, and view the combinatorial expression of genes in GSH synthesis within that morphological context (Fig 5). GSH diffuses across the outer mitochondrial membrane, is actively transported across the inner mitochondrial membrane, and concentrated in the matrix. In the human HLTF^-/-CDX, SLC25A10, the mitochondrial glutathione transporter [38] was upregulated (L2FC = 1.22; padj 3.12E-06) commensurate with down regulation (L2FC = -1.2; padj 9.04E-07) of its transcriptional repressor, BTB and CNC Homology 1 (BACH1) [39]. Region-specific genes in GSH synthesis and the expression of genes for mitochondrial GSH transport overlap in HLTF^-/-CDX (Fig 6).

In mouse TME a total of 22 genes in the glutathione metabolic process (GO:0006749) were upregulated (padj 0.025974). GSH plays a critical role in endothelial function especially vasodilation and paracellular permeability [40]. Glutathione also has a role in protein post-translational modification [PTM, 41]. S-glutathionylation is a reversible PTM on thiol groups in protein cysteine (Cys) residues. As shown in Fig 6, the biotin-switch assay with FFPE HLTF^-/-CDX/TME showed protein S-glutathionylation in human tumor cells and endothelium of mouse lymphovasculature. These data support the conclusion that increased availability of glutathione buffers oxidative stress, and drives tumor/endothelial cell proliferation and migration in the HLTF^-/-CDX/TME model.

Pr-SSG diverts glycolysis

Hypoxia induced S-glutathionylation of HIF-1α on Cys⁵²⁰ (mouse Cys⁵³³) stabilizes the protein [33]. Thus, we directly tested the hypothesis that protein S-glutathionylation in HLTF^-/-CDX/TME targeted Hif-1α protein in the sixth experiment. A total of 61 differentially expressed proteins were identified by 2D-DIGE comparison of glutathione-labeled proteins from Hltf^-/-HCT116-Red-Fluc cells at the time of OCMI and resultant HLTF^-/-CDX from ID Hltf KO mice (Fig 7 and S1 Raw_images). Final adjusted WB ratios for tumor proteins/cell proteins were calculated (S2). Protein authentication by MALDI-TOF/TOF led to the authentication of site-specific Pr-SSG of 6 proteins (Table 4). Fragments were omitted, and only protein and total ion C.I.% scores equal to 100% were included.

Neither human (Cys⁵²⁰) nor mouse (Cys⁵³³) HIF-1α is a target of S-glutathionylation in this CDX model. Further use of MALDI-TOF/TOF followed by species-specific data analysis identified highly conserved cysteines in identical peptide fragment ion spectra (S2) from human and mouse, leading us to conclude humans and mice share the same Pr-SSG protein targets in this HLTF^-/-CDX/TME model. Protein conservation is shown in Table 4.

Importantly, three of five Pr-SSG protein targets are glycolytic enzymes that control reversible steps 7, 8 and 9 in the 10-step glycolysis pathway (Fig 8). Glycolytic intermediates including glucose-6-phosphate (G6P) can be diverted into the oxidative pentose phosphate pathway to produce the antioxidant NADPH. Glyceraldehyde 3-phosphate dehydrogenase (GAPDH), the enzyme that catalyzes step 6, was not targeted for site-specific Pr-SSG thus avoiding programmed cell death. S-glutathionylation of GAPDH Cys²⁴⁷ shuttles GAPDH to the nucleus where it transfers the glutathione to Sirtuin-1 (SirT1) during SirT1-p53-mediated apoptosis [42].

Phosphoglycerate kinase (PGK1) and phosphoglyceromutase (PGAM1) are redox sensitive and exposure to glutathione stabilizes them (prevents inactivation) presumably via the newly identified S-glutathionylation sites reported here. PGK1 is a highly conserved so-called moonlighting protein because of its involvement in varied cellular processes as a protein kinase [43]. As the first ATP generating enzyme in glycolysis, PGK1 catalyzes the conversion of 1,3-diphosphoglycerate (1,3-DPG) to 3-phosphoglycerate (3PG), and PGAM1 catalyzes the interconversion of 3PG to 2PG.

PGAM1 protein was the strongest candidate for further experimentation because its expression in HLTF-deleted CDX/TME had the highest (>5.0) final WB ratio indicating Pr-SSG occurred in cells of the CDX rather than the cultured cells from which it was derived. PGAM1 was also transcriptionally activated (L2FC 1.81; padj 1.59E-08) in HLTF^-/-CDX compared to HLTF^+/+CDX from ID Hltf KO mice. Upregulation of PGAM1 is consistent with its upregulation in several human cancers [44]. PGAM1 immunolocalized to tumor and endothelial cells in the HLTF^-/-CDX/TME (Fig 9).

3PG and 2PG are allosteric regulators of glycolytic branching pathways [45]. 3PG inhibits 6-phosphogluconate dehydrogenase (G6PD) in the pentose phosphate pathway. 2PG activates phosphoglycerate dehydrogenase (PHGDH) to provide feedback control of 3PG. PHGDH is the rate-limiting step in the de novo serine biosynthesis cascade that provides precursors for GSH synthesis and production of NADPH [46]. PHGDH was not targeted for site-specific Pr-SSG in our HLTF^-/-CDX/TME model. This suggests PHGDH is functional because covalent modification of PHGDH Cys²⁹⁵ inactivates this enzyme [47].

The penultimate step in glycolysis—the conversion of 2PG to phosphoenolpyruvic acid (PEP)—is catalyzed by the rate-limiting enzyme, alpha enolase (ENOA1), another moonlighting enzyme [48]. Because ENOA1 is overexpressed in numerous different types of cancer, it is a target for therapeutic intervention. Glycolysis is significantly reduced (OXPHOS is rescued) when Pr-SSG of ENO1A Cys¹¹⁹ causes enzyme inactivation [49, 50]. Resultant 2PG accumulation activates PHGDH to deplete excess 3PG, thereby promoting serine synthesis and preventing 3PG-induced inhibition of NADPH production. ENOA1 immunolocalizes to tumor and endothelial cells in the HLTF^-/-CDX/TME (Fig 9).

The Pr-SSG target, Annexin A1 (ANXA1, Cys²⁶³), was neither transcriptionally activated nor did it have a high final WB ratio (Table 4). Nevertheless, mouse ANXA1 is a known target for Pr-SSG at Cys³²⁴ with a putative role in redox signaling [52]. Pr-SSG at Cys²⁶³ in this HLTF^-/-CDX/TME model suggests functional importance as does immunolocalization of ANXA1 protein in tumor and endothelial cells (Fig 9).

Independent of its metabolic activity, PGAM1 interacts with α-smooth muscle actin (ACTA2), and ANXA1 is linked to the actin cytoskeleton. Importantly, glutathionylated actin (GS-actin) on Cys³⁷⁴, a site conserved in all actin isoforms [53], was not a target in this HLTF^-/-CDX/TME model. GS-actin has altered mechanical properties in vitro due to decreased polymerization G- to F-actin. Otherwise stated, deglutathionylation on Cys³⁷⁴ increases the rate of actin polymerization and cell motility [53]. Co-localization of PGAM1 and ANXA1 in invading lymphatic intravascular cancer cells and extravascular amoeboid cancer cells (Fig 9) was confirmed by laser scanning confocal microscopy. Front-rear asymmetry is characteristic of migrating lymphocytes [54]. However, the genetic background of the CDX model (Rag2-/IL2-double mutant) is completely alymphoid, which precludes confusion of cellular identify. Extravascular amoeboid cancer cells displayed aggressive metastatic directional movement consistent with an epithelial-to-mesenchymal transition [55].

PGAM1/ANXA1-positive tumor cells in intravascular niches corresponded to membership in spatial transcriptional Cluster 1 (Fig 5) concomitant with locoregional enrichment of the thymosin beta 4 X-linked (TMSB4X) gene (L2FC 1.86; padj 8.60e-32) whose secreted G-actin sequestering protein regulates actin dynamics. Transcriptional Cluster 1 was subepithelial and extended into the tumors resulting from circumferential cellular migration. Extravasculature PGAM1/ANXA1-positive tumor cells corresponded to membership in spatial transcriptional Cluster 14 (Fig 5) enriched for the epithelial membrane protein 1 (EMP1) gene (L2FC 2.99; padj 4.31e-27), a pro-metastatic factor responsible for metastatic recurrence in CRC. Transcriptional Cluster 14 subtends the epithelium independent of the tumor.

HSP60 is evolutionarily conserved. HSP60 typically resides in mitochondria where it maintains protein homeostasis via a protein folding apparatus with its co-chaperone HSP10 [56]. HSP60 monomers are composed of three structural domains designated apical, intermediate and equatorial, that are subjected to extensive modifications by TPMs [57]. Human and mouse HSP60 proteins are 97.56% identical, with three cysteine resides (Cys²³⁷, Cys⁴⁴², and Cys⁴⁴⁷). Previously identified PTMs [57] related to nitration include S-nitrosylation at HSP60 Cys²³⁷ and S-guanylation at HSP60 Cys⁴⁴⁰. This is the first report of S-glutathionylation at HSP60 Cys⁴⁴⁷.

Results from this HLTF^-/-CDX/TME model—show for the first time—that tumor cell recovery from DNA damage via robust transcriptional activation of DNA repair and OXPHOS genes coincided with a highly metastatic phenotype. Loss of HLTF function throughout the CDX/TME co-sponsored an amoeboid-associated TME at the tumor border. Cancer and endothelial cells of the TME share site-specific Pr-SSG signatures for 3 glycolytic enzymes that divert glycolysis in support of glutathione biosynthesis that prevents OXPHOS inhibition and promotes redox homeostasis.

Experimental advantages of this CDX model of CRC are validated by results from this study. First, by manipulating the genotype of the mice and HCT116 cells, the CDX model recapitulates the beginning and end of disease progression. Second, when human CRC is propagated in immune-deficient mice, not only does murine stroma and vasculature begin to replace the human equivalents immediately [58], but the human tumor induces reorganization of the quiescent, murine stroma into a pro-tumorigenic phenotype supporting tumor growth and metastasis [59]. We exploited this synergy in our preclinical CDX model. Third, the CDX model is predicated on direct orthotopic microinjection (OCMI) of HCT116 cells between the mucosa and the muscularis layers of the cecal wall known to replicate metastatic spread observed in advanced human CRC [19]. Fourth, RNAseq combined with species-specific mapping unambiguously differentiated gene expression in human tumor cells from cells of the mouse TME [10]. Fifth, a CDX histology section is entirely accommodated on the spatial transcriptomics array. As a result, human tumor cells were transcriptionally profiled as they interacted with the mouse TME, especially in lymphatic intravascular niches at the tumor margin. Sixth, protein posttranslational modifications were authenticated by using MS/MS data and MALDI-TOF/TOF through comparative analysis of tumor cells in culture with resultant CDXs.

One potential disadvantage of the CDX model is its development in an immune-deficient mouse. In defense of that approach, ~80–85% of CRC patients have so-called “cold tumors” meaning they do not have a strong immune response (do not express T-cells). Consequently, cold tumors are resistant to immunotherapy [60]. We posit our CDX model is relevant to immune-cold CRC. More importantly, spatial transcriptomics can be used to identify patients with active vs suppressed immune systems as a screening tool for therapy and to predict survival. Coupling mass spectrometry-based proteomics for FFPE [61] with spatial transcriptomics and immune profiling of FFPE could revolutionize our diagnostic/treatment approach to CRC.