Reshaping the Gut Microbiome in An Apc-Mutant Genetic Background: Mechanistic Insights From Integrated Multi-Omics

Ying-Shiuan Chen The University of Texas MD Anderson Cancer Center Jia Li Texas A&M University System Health Science Center College of Medicine: Texas A&M University College of Medicine Rani Menon Texas A&M University College Station Arul Jayaraman Texas A&M University College Station Kyongbum Lee Tufts University Yun Huang Texas A&M University System Health Science Center College of Medicine: Texas A&M University College of Medicine Wan Mohaiza Dashwood Texas A&M University System Health Science Center College of Medicine: Texas A&M University College of Medicine Ke Zhang Texas A&M University System Health Science Center College of Medicine: Texas A&M University College of Medicine Deqiang Sun Texas A&M University System Health Science Center College of Medicine: Texas A&M University College of Medicine Roderick Dashwood (  rdashwood@tamu.edu ) Texas A&M University Health Sciences Center https://orcid.org/0000-0003-0351-4034


Background
The mammalian gut microbiome is strongly implicated in host physiology and pathophysiology [1][2][3]. For example, studies in germ-free models of colorectal cancer (CRC) revealed decreased bowel in ammation and tumor outcomes as compared with the corresponding animals under conventional housing conditions [4,5]. Fecal microbiota transplantation was used successfully to treat recurrent Clostridium di cile infection [6], and provided bene t to patients with in ammatory bowel diseases, functional gastrointestinal disorders, and obesity [7].
There is increasing interest in de ning interventions that alter the gut microbiota for disease prevention and treatment. Epidemiological studies indicate that CRC is associated with low consumption of green vegetables and ber, whereas intake of dark leafy vegetables is linked to decreased risk [8]. However, little is known about how these dietary intakes in uence the crosstalk between gut microbiota, host transcriptomics, and pathogenesis in the gastrointestinal tract.
Spinach is a dark leafy vegetable with a high chlorophyll content and bioactive phytochemicals such as betaine, carotenoids, avonoids, and polyunsaturated fatty acids [9][10][11]. We employed an adenomatous polyposis coli (Apc)-mutant rat model [12][13][14][15] to examine anticancer outcomes from dietary spinach consumption. The polyposis in rat colon (Pirc) model mimics disease progression in human familial adenomatous polyposis patients, involving spontaneous tumor development both in the colon and in the small intestine [15,16]. The genetic model circumvents the need for carcinogen treatment, used previously with dietary spinach [17], and the burden of adenomatous polyps facilitates temporal tracking of tumor suppression via colonoscopy [12,14,18].
We observed signi cant antitumor e cacy from dietary spinach consumption and, despite the Apcmutant genetic background, b-catenin protein levels remained highly overexpressed in colon polyps. Subsequently, mechanisms were pursued linking gut microbiome to host multi-omic changes in fatty acid metabolism, the tricarboxylic acid (TCA) cycle, and pathways in cancer.

Methods Preclinical
Prior approval was obtained from the Institutional Animal Care and Use Committee. After weaning, Pirc (F344/NTac-Apc am1137 , Taconic Farms, Inc. USA) and wild type (WT) F344 male rats were assigned randomly to basal AIN93 control diet (Ctrl) or AIN93 diet containing 10% w/w freeze-dried spinach by weight (SPI). Diets were replenished every 2 to 3 days, and animal body weights were monitored weekly. Monthly endoscopy was employed for temporal tracking of polyp development in the rat colon, as described before [12,14]. Each polyp was assigned a unique 'address' in the colon, based on a reported methodology [18]. Prior to termination, 3-4 rats in each group were injected with BrdU (100 mg/kg body weight), and animals were euthanized 1 h later by CO 2 inhalation. A thorough necropsy examination was performed, and tissue samples were taken for histopathology and molecular analysis, as reported [12,14].

Histology
Tissue sections (5 mm) were stained with hematoxylin and eosin (H&E), or immunostained with antibodies for b-catenin (BD Biosciences # 610153), cleaved caspase-3 (Cell signaling #9661), and BrdU (BD Biosciences # 347580), at the Research Histology, Pathology and Imaging Core, The University of Texas MD Anderson Cancer Center. A BrdU labeling index was determined, as described [61], and cleaved caspase-3 was quanti ed as percent positive-stained crypts in a given eld. Two to three biological replicates were employed for WT-Ctrl, WT-SPI, Pirc-Ctrl, and Pirc-SPI groups. For each tissue section, at least 15 independent elds were quanti ed in the proximal, middle, and distal regions of the colon. Other proteins were immunoblotted as reported [12][13][14][62][63][64][65], using Cell Signaling primary antibodies for PARP and small RNA sequencing was conducted by single-end sequencing with 50 bp read length. Fastq les were generated on BaseSpace for further bioinformatics processing and analysis.

Metabolome
Pre-weighed samples of rat colon tumor and normal colonic mucosa (n = 6-7 biological replicates), collected at the time of necropsy, were homogenized in 0.5 ml cold methanol and 0.2 ml chloroform in pre-cooled Garnet bead tubes using a Precellys®24 beadbeater. Samples were centrifuged at 3000 rpm for 10 mins at 4°C and 0.7 ml cold water was added to the supernatant. The aqueous phase was collected by centrifugation at 3000 rpm for 1 min, and the extraction procedure was repeated. The pooled aqueous (upper) phase was passed through a sterile nylon cell strainer and lyophilized (Labconco TM ). Lyophilized samples were reconstituted in 50 ml methanol/water (1:1, v/v) and stored at −80°C until analysis. Untargeted liquid chromatography high-resolution accurate-mass spectrometry (LC-HRAM TM ) pro ling was conducted on a Q Exactive™ Plus Hybrid Quadrupole-Orbitrap™ Mass Spectrometer coupled to a Dionex UltiMate 3000 high-performance liquid chromatography system (Thermo Scienti c TM ). A Synergi Fusion-RP C-18 column (Phenomenex) was used with a methanol/acetonitrile solvent gradient, and mass scanning in the positive mode was in the range 50 to 750. The MS1 and MS1-dependent MS2 spectra were collected at an m/z resolution of 70,000 and 17,500, respectively, with the autosampler maintained at 4°C. Methanol/water (1:1 v/v) blanks were injected between each run to prevent sample carryover. Parallel studies also were conducted on the freeze-dried spinach.
Raw metabolomic data were imported into Progenesis QI (Waters) for alignment, peak picking, and compound identi cation. Among the 17243 features detected, candidates were identi ed by reference to the Human Metabolome Database (HMDB) and KEGG. Raw abundance data were normalized to initial sample weights, incorporating Partial Least Squares Discriminant Analysis (PLSDA). Features were further ltered by their appearance in three independent metabolomic databases, with at least three biological replicates and a signi cant ANOVA test. This resulted in 5946 differential features for further analysis. Signi cant features were subjected to clustering and correlation by MetaboAnalyst 4.0 [69][70][71][72].
The p-values (two-tailed t-test) and t-scores (standardized test statistic) were generated for multiple group comparisons of metabolic networks and functional metabolite prediction via Mummichog version 2 in R [73]. Primary prediction of 883 compound names was mapped to the KEGG COMPOUND Database, and pathway analyses by Mummichog were ranked according to the p-value, using p=0.05 as the cutoff.
Bioinformatics RNA-seq data were processed and analyzed as reported [12]. DEGs were called using DESeq2, with adjusted p-value <0.05. MiRNA-seq raw fastq data were trimmed using cutadaptor, and reads with the same sequence were collapsed and counted. Mature and hairpin miRNA sequence data for rat were downloaded from miRbase (http://www.mirbase.org/ftp.shtml). Collapsed reads were mapped to rat mature miRNA sequences using blastn. The counts table was input to DESeq2 to call DEmiRs with adjusted p-value <0.05. PCA was performed using DESeq2 for both RNA-seq and miRNA-seq datasets.

Quantitative PCR
Basic methodologies for RNA extraction and puri cation were as reported, using three biological replicates per group [12][13][14][79][80][81]. After SuperScript III (Thermo Fisher Scienti c) or miScript RT II (Qiagen) kits, quantitative PCR (qPCR) reactions were performed by LightCycler® FastStart DNA Master SYBR Green (Roche Applied Science) on a LightCycler96 instrument. Primers for mRNA qPCR were designed by NCBI-BLAST (see Table 11 in Additional le 3), whereas primers for miRNA qPCR were purchased custom-made from Qiagen. Internal controls were Gapdh and U6B small nuclear RNA for mRNA and miRNA analyses, respectively. RNAs and miRNAs were selected based on prior validation and ranking in GSEA data. Original gene lists for tight junction and anti-microbial function were obtained from KEGG and Gene Ontology (GO) resources, respectively, and further sorted by the sequencing data to generate genes of interest. We focused on six miRNAs consistently altered in Pirc colon tumors, and ltered miRNA-mRNA pairs by conserved UTR target site in human and rat, with a linear correlation <-0.7 in sequencing data. Veri cation of miRNAs and mRNAs was by qPCR (Pearson's test, with r <-0.5).

Statistics
Statistical analysis of two-group comparisons was performed using an unpaired two-tailed t-test. Correlation analysis was performed by linear and Pearson's correlation, for miRNA-RNA target and microbiome-host correlations, in tumor and diet intervention groups. For matched tumor outcomemicrobiome correlation analysis, seven biological replicates of Pirc/Ctrl or Pirc/SPI were undertaken, whereas three biological replicates were used for SPI responsive gene-microbiome correlations. Unbiased metabolomic analyses used n=6-7 biological replicates per group. In the gures, each datapoint designates a single colon tumor or normal colonic mucosa sample from individual rats in the corresponding groups. One-way ANOVA was used to compare the mean of each column with the mean of every other column, with Tukey correction for multiple comparisons (GraphPad Prism 9.0). The level of signi cance was designated in the gures as follows: *p<0.05; **p<0.01; ***p<0.001; ****p<0.0001, or with the exact p value [80][81][82][83][84].

Results
Antitumor e cacy of dietary spinach in an Apc-mutant rat model Pirc and wild-type (WT) rats were fed AIN93 control (Ctrl) diet, or AIN93 diet containing 10% w/w freezedried baby spinach (SPI), starting at 4 weeks of age (Fig. 1a). No signi cant treatment-related effects were observed with respect to food consumption and body weight throughout the study (Fig. 1b). By colonoscopy [12,14,18], SPI suppressed adenomatous polyps as early as 12 weeks into the experiment ( Fig. 1c), i.e., after 8 weeks of SPI treatment. During this period and at later times, colonoscopy data revealed consistent inhibition of small colon polyps, and signi cant suppression of large colon polyps after week 20 (Fig. 1d).
When the experiment was terminated, after the rats had reached 30 weeks of age, tumor multiplicity was decreased signi cantly both in the colon and in the small intestine, and tumor volume also was reduced signi cantly in the colon by SPI treatment (Fig. 1e). No marked changes were observed histologically, but bromodeoxyuridine (BrdU) labeling hinted at reduced cell proliferation rates by SPI in some regions of the colonic crypt (Additional le 1). Immunohistochemistry and immunoblotting experiments indicated that bcatenin overexpression in colon tumors was unaffected by SPI treatment (Fig. 1f and Additional le 2). Thus, despite the oncogenic driver of the Apc-mutant genetic background, antitumor mechanisms other than b-catenin downregulation were pursued.
The gut microbiota is altered by dietary spinach We performed 16S rRNA sequencing of the gut microbial community in Pirc and WT rats. For a complete view of the taxonomic and other data, refer to Additional le 3. The observed Operational Taxonomic Units (OTUs, Table 1 in Additional le 3) and Shannon index revealed that a-diversity was unaffected by host genotype, but was increased signi cantly by SPI treatment in Pirc and WT rats (Fig. 2a, black vs. green symbols). There was no segregation between Pirc and WT rats for weighted UniFrac principal coordinates analysis (PCoA) (Fig. 2b), but a signi cant separation was observed in unweighted UniFrac PCoA (Fig. 2c). The gut microbiome in both genotypes clustered separately in weighted UniFrac PCoA between Ctrl and SPI groups (Fig. 2b), with a marked shift in unweighted UniFrac PCoA (Fig. 2c). These data are consistent with previous ndings indicating that diet plays a dominant role over genetic background with respect to shaping interindividual variations in host-associated microbial communities [19,20].
The Pirc model had a higher abundance of Bacteroidetes and Proteobacteria than WT rats, while Firmicutes, Actinobacteria, and Tenericutes were lower (Fig. 2d), as observed in mouse and human microbiomes [21][22][23][24]. These abundances were reversed by SPI intake, independent of host genotype, as evidenced by the increased relative abundance of Firmicutes and decreased Bacteroidetes (Fig. 2e). Similar ndings were noted at the Family (Fig. 2f) and genus level (Fig. 2g). For example, in Pirc and WT rats, SPI treatment increased the relative abundance of Lachnospiraceae and decreased Ruminococcaceae (Fig 2f, green bars), and at the genus level SPI ingestion reduced the relative abundance of Bacteroides and Desulfovibrio (Fig. 2g, green bars). These results suggested that SPI consumption reshapes the microbiome composition, reversing the effects of the Apc-mutant background and host genetic predisposition.
Linear discriminant effect size (LEfSe) was used to further analyze the OTU microbiome data (Tables 2-4 in Additional le 3). From the corresponding cladograms ( Fig. 3a-c), host genotype and dietary SPI intake both in uenced Ruminococcaceae and Lachnospiraceae family members. In response to SPI treatment, LEfSe analyses revealed that Pirc and WT rats shared ~50% commonality among changes at the genus level (Fig. 3d). Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt) categorized 328 terms following Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis ( Table 5 in Additional le 3). Linoleate and ether lipid metabolism were altered signi cantly in Pirc vs. WT rats, and PICRUSt revealed a marked effect of SPI intake. Thus, after SPI consumption, 85 terms (57 decreased and 28 increased) and 112 terms (71 decreased and 41 increased) were changed in Pirc and WT rats, respectively, and seventy-ve terms overlapped between the two genotypes, i.e., 54 decreased and 21 increased (Fig. 3e). Increases in membrane transporters, cell motility, signal transduction, transcription, carbohydrate metabolism, and kinases were among the top 10 terms prioritized by KEGG analysis in Pirc and WT rats given SPI, along with porphyrin and chlorophyll metabolism (Fig. 3f). Decreased terms were related to protein translation, replication/repair, and energy/nucleotide metabolism.
Pathway changes included an increase in linoleate and butanoate metabolism, and a decrease in the TCA cycle and pathways in cancer (Fig. 3g).
Spinach consumption impacts key genes associated with pathogenesis RNA-sequencing (RNA-seq) mapped 17,378 transcripts in colonic tissues from Pirc and WT rats, and principal component analysis (PCA) completely segregated tumor tissues from normal tissues (Fig. 4a).
Tumor development had a more marked effect on overall transcriptome levels than host genetics (Pirc vs. WT rats) and SPI consumption. There were 261 differentially expressed genes (DEGs) identi ed between Pirc and WT normal-looking tissues, half of which (138 genes) overlapped with 2180 DEGs associated with tumor development (Fig. 4b). Heatmaps of all DEGs showed a distinct tumor feature when compared to Pirc and WT normal-looking tissues (Fig. 4c, PCT vs. PCN and PCT vs. WCN).
Geneset Enrichment Analysis (GSEA) combined with HALLMARK identi ed nine pathways altered signi cantly in Pirc normal-looking colon compared to WT normal colon (Fig. 4d, upper panel), indicating differences at the level of host genetics. Five of these pathways were further altered in colon tumors (Fig.   4d, lower panel, underlined), i.e., In ammatory Response, Allograft Rejection, TNFa signaling through NF-kB, Adipogenesis, and Myogenesis. As expected in the Apc-mutant background, Wnt/b-catenin signaling was upregulated in Pirc colon tumors (Fig. 4d, red font). Immunoblotting corroborated that b-catenin overexpression was associated with poly(ADP-ribose)polymerase (PARP) cleavage, increased cyclin D1, and decreased p53 in Pirc tumors compared to adjacent normal and WT normal colonic tissues (Additional le 2).
Other pathways of note were related to cell cycle changes, immune response, oxidative stress, and metabolism (Fig. 4d). Using RT-qPCR for validation, genes upregulated signi cantly in Pirc colon tumors vs. Pirc adjacent normal-looking colon included Cxcl6, Serpine1, and Il-1b, whereas Hspb8, Tpm2 and Fhl1 were downregulated signi cantly (Fig. 4e). Compared to adjacent normal colon, upregulation of Defa6 and Bcl3 and downregulation of App, Myh11, and Myl9 indicated changes in tight junctions and antimicrobial activity in Pirc colon tumors (Fig. 4f).
We also mapped 559 microRNAs (miRNAs) via small RNA-seq, which segregated tumor vs. normallooking colon (Fig. 4g). Similar to the mRNA pro les (Fig. 4c), miRNAs had a distinct tumor feature as compared to Pirc and WT normal-looking tissues (Fig. 4h, PCT vs. PCN and PCT vs. WCN). There were 115 differentially expressed miRNAs (DEmiRs) associated with tumor formation (Fig. 4h and Table 6 in Additional le 3). We combined TargetScan with RNA-seq and small RNA-seq datasets to identify miRNA-RNA pairs most altered in the Apc-mutant background (Fig. 4i). Validation by qPCR corroborated signi cant downregulation in Pirc colon tumors of miR-215, miR-143, and mir-145 compared with adjacent normal-looking colonic mucosa (Fig. 4j). Other candidates, such as mir-146b, mir-34a, and mir-21, did not reach statistical signi cance.
Attention shifted next to SPI effects on predicted targets ( Fig. 5 and Additional le 4). Compared to the AIN basal diet control group, SPI consumption altered 4 genes in common among the 101 DEGs in WT rats and 80 DEGs in Pirc normal colon (Fig. 5a). GSEA indicated signi cant downregulation of cell cyclerelated pathways and upregulation of immune-related pathways in Pirc and WT rats fed SPI, with ve pathways in common among the genotypes (Fig. 5b). The latter pathways included TNFa Signaling through NFkB, Hypoxia, Epithelial Mesenchymal Transition, Apoptosis, and KRAS Signaling. After SPI consumption, 2945 DEGs were identi ed in Pirc colon tumors compared to adjacent normal colon, and 1754 of the DEGs also were detected in tumors vs. adjacent normal colon from rats given control diet (Fig. 5c). Among the pathways most strongly implicated were IFN-a and IFN-g for tumors from SPI-fed rats compared to rats given control diet.
We also considered two scenarios for the antitumor e cacy: (1) genes up-or downregulated in colon tumors relative to adjacent normal colon that were reversed by SPI in Pirc normal colon, and (2) genes that were normalized in colon tumors from rats given SPI compared with colon tumors from Pirc rats given control diet. The rst scenario would implicate primary prevention of colonic aberrant crypt foci or microadenomas, before they advanced to later stages. These genes included Serpine1, Itga6,Duoxa2, Tcf7l1, Plcd1 and Slc30a10 (Fig. 5d). Comparing tumor to tumor in scenario 2, Ccl21 and Klf7 were normalized by SPI ingestion, relative to basal diet (Fig. 5e).
In terms of miRNAs, among 66 DEmiRs in colon tumors from SPI-fed rats, 41 DEmiRs similarly were detected in colon tumors from animals on Ctrl diet (Fig. 5f). After investigating RNA-miRNA pairs and validating as before (Fig 4j), colon tumors had loss of miR-145 with increased Serpine1 and gain of mir-34a with reduced Klf4 (Fig. 5g and Table 7 in Additional le 3). A negative correlation for mir-145/Serpine1 was maintained after SPI consumption, whereas the mir-34a/Klf4 trend was reversed by SPI treatment (Fig. 5g).

Crosstalk between microbiome and host transcriptome responses
Integrating antitumor outcomes (Fig. 1) with a-diversity (Fig. 2), we observed a signi cant inverse association for tumor multiplicity (Fig. 6a, left panel) but not tumor volume (Fig. 6a, right panel). Tumor multiplicity was inversely correlated with three unclassi ed Bacteroidales families ( Fig. 6b and Table 8 in Additional le 3). At the genus level, one unclassi ed Lachnospiraceae and one unclassi ed Ruminococcaceae genus were negatively correlated with tumor multiplicity, whereas one other unclassi ed Ruminococcaceae genus was positively correlated (Fig. 6c). Metagenome prediction in relation to tumor multiplicity outcomes found signi cant inverse correlations for butanoate metabolism and calcium signaling, and positive associations for peptidases and pathways in cancer (Fig. 6d).
We also compared microbiome and host gene expression changes based on the transcriptomic data (Fig.  6e). Signi cant positive correlations were noted for Lachnospiraceae (Unc0396i) and the e ux transporter Slc30a10, Bacteroidales (Unc00krl) and Ruminococcaceae (Unc01k4o) and the phospholipase C family member Plcd1, and Ruminococcaceae (Unide781 and Unc00vst) and the serine protease inhibitor Serpine1. Negative correlations were detected for Ruminococcaceae (Unide781 and Unc00vst) and the transcription factor Tcf7l1.

Metabolomic corroboration of mechanistic leads
To validate correlations from the microbiome and transcriptome studies, metabolomics was performed on adenomatous colon polyps and normal colon tissues obtained from Pirc rats at 30 weeks (Fig. 1a). As predicted from the microbiome data (Fig. 3g), among the fty-one metabolites identi ed ( Table 9 in Additional le 3) several were associated with fatty acid metabolism, the TCA cycle, and pathways in cancer (Fig. 7). Linoleate and its downstream metabolites from the 15-lipoxygenase-1 (15-LOX-1) pathway exert proapoptotic antitumor mechanisms in CRC [25][26][27]; notably, lower levels of these metabolites in Pirc colon tumors tended to be normalized in adenomatous polyps following SPI treatment, comparable to the levels detected in normal-looking Pirc colon ±SPI (Fig. 7a). Similar trends were observed for 2-aceto-2-hydroxybutanoate, which was increased signi cantly in colon tumors after SPI treatment (Fig. 7b). On the other hand, L-glutamate and N-acetylneuraminate were detected at higher levels in colon tumors, and SPI treatment reduced these metabolites in adenomatous polyps, comparable to the levels observed in normal-looking Pirc colon ±SPI (Figs. 7c and 7d). Key intermediates are discussed further below.

Discussion
We examined the interrelationships between host genetics, gut microbial composition, dietary exposure, and disease outcome in an Apc-mutant preclinical model that mimics hereditary human CRC [12][13][14][15]. A lower diversity in the gut microbiome was circumvented by feeding dietary SPI under conditions in which signi cant suppression of adenomatous polyps occurred in the colon and small intestine. This work extends prior observations on the decreased microbial diversity and loss of Firmicutes, Clostridia, and Lachnospiraceae in Apc Min/+ mouse [21] and human colorectal cancers [28,29]. In a short-term clinical study, Firmicutes was decreased by consumption of an animal-based diet, whereas Bacteroidetes was lowered by a plant-based diet [20]. Ruminococcaceae phylotypes were increased by resistant starch in obese men, whereas Lachnospiraceae phylotypes were increased by non-starch polysaccharides [30].
These trends were recapitulated in the current investigation, with decreasing Bacteroidetes and increasing Firmicutes and Lachnospiraceae after intervention with dietary SPI. Spinach consumption also diminished Desulfovibrio sulfur-reducing bacteria, implicating a role for anti-in ammatory responses and enhanced gut barrier function [31,32].
In contrast to the marked effect on the gut microbiome, long-term SPI intake exerted a relatively mild impact on host transcriptomics, based on mRNA and miRNA sequencing. We prioritized genes associated with adenomatous polyp suppression by SPI that were correlated with microbiome abundance. Serpine1 encodes plasminogen activator inhibitor type 1 (PAI-1), which is elevated in sporadic and hereditary CRC, and has an essential role in extracellular matrix proteolysis and matrix metalloproteinase activity [33][34][35]. Mucosal gene expression pro ling of SERPINE1, plus in ammatory regulators such as CXCL1, STAT3, and IL family members, revealed associations with the decreased abundance of Firmicutes and Bacteroidetes subsets, as in human CRC [36]. DUOXA2 is a maturation factor for the epithelial antimicrobial dual oxidase DUOX2, which is among several NADPH oxidase/dual-oxidase family members deregulated in CRC and Crohn's disease, acting via NFkB [37,38]. DUOX2 expression was negatively correlated with Bacteroides, Lachnospiraceae, and Blautia, but positively correlated with Pasteurellaceae, Enterobacteriaceae, and Gammaproteobacteria [38].
We also extended our prior work on carcinogen-induced rat colon tumors that examined miRNAs and the mRNA targets [17]. Mir-145 was among the most highly downregulated miRNAs in Pirc colon tumors, consistent with its proposed tumor suppressor role in human CRC, although mir-145 was unexpectedly upregulated in carcinogen-induced colon tumors, for reasons that remain unclear. Previous studies prioritized SERPINE1 as a target of mir-143/145 in bladder cancer [39], as well as mir-34a in liver cancer, regulating KLF4 [40]. KLF4 is targeted by multiple miRNAs [40,41], including mir-34a, and in the Pirc model the mir-34a/Klf4 axis was altered by SPI treatment. KLF4 is an important zinc-nger transcription factor involved in cell cycle regulation, somatic cell reprogramming, and tumorigenesis. Reduction of KLF4 is well documented in rat [41,42] and human colon tumors [43,44], providing a potential avenue for next generation precision nutrition via SPI bioactives in the clinical setting [45].
Untargeted metabolomics corroborated several key ndings from the microbiome and transcriptome studies, providing valuable insights and new mechanistic leads into the anticancer effects of SPI. Tumorassociated linoleate and its 15-LOX-1-dependent intermediates were lower in Pirc controls fed basal AIN diet, and they were normalized in adenomatous polyps after SPI treatment (Fig. 7a), consistent with the proposed anticancer mechanisms of these metabolites [25][26][27]. Several other intermediates associated with fatty acid metabolism also were detected in metabolomic analyses, such as 6-keto-prostaglandin E 1 (6-keto-PGE 1 ) and PGE 2 ethanolamine; changes after SPI consumption were consistent with an overall shift towards anti-in ammatory, proapoptotic and tumor suppression pathways (Additional le 5). Thus, in SPI-fed rats, a decreased ux through pro-in ammatory leukotrienes and prostaglandins was paralleled by increased levels of 9-LOX, 15-LOX-1, prostacyclin, and cytochrome P450 (CYP) metabolites that are linked to anticancer outcomes [25][26][27].
One noteworthy observation was that enzymes associated with fatty acid metabolism (Additional le 5, blue font) were not among the main candidates prioritized by RNA-seq analyses (Additional le 4). We speculated that certain intermediates detected in rat tissues at 30 weeks (Fig. 1a) might derive directly from the SPI incorporated into the AIN basal diet, and this was con rmed via unbiased metabolomic analyses of the freeze-dried spinach. Thus, among the 700+ SPI analytes (Table 10 in Additional le 3), several corresponded to key intermediates detected in Pirc colon tumors and normal-looking tissues (Tables 9 and 10, Additional le 3, green font). These included linoleate, 13(S)-hydroperoxy-9Z,11Eoctadecadienoic acid , and (9Z,11E)-13-oxooctadeca-9,11-dienoic acid , as well as L-glutamate and N-acetylneuraminate. Interestingly, the 2-aceto-2-hydroxybutanoate that was increased in colon tumors from SPI-fed rats (Fig. 7b) was not detected in freeze-dried spinach, implicating bene cial butyrate-producing gut bacteria linked to increased a-diversity [46][47][48].
Although L-glutamate and N-acetylneuraminate were detected via metabolomics in freeze-dried spinach, colon tumors at 30 weeks had reduced rather than increased levels of these intermediates following dietary SPI administration (Figs. 7c and 7d). Diminished L-glutamate levels in adenomatous polyps from SPI-fed rats would be synonymous with synthetic lethality [49], circumventing TCA cycle functions that are dependent on glutamine metabolism as a means of sustaining mitochondrial energetics. Lower levels of N-acetylneuraminate in tumors from SPI-fed rats would implicate altered cell surface glycans that are critical for pathways in cancer, including immune evasion, resistance to apoptosis, and enhanced proliferation, metastasis and angiogenesis [50][51][52]. Notably, altered sialyation has been linked to activation of the in ammasome mediator eIF2 [50], which in an APC-de cient background attenuates MYC-dependent apoptosis [53], unless circumvented by mechanisms that downregulate EIF2 -as observed for Eif2b2 in colon tumors from SPI-fed rats (Additional le 4, panel b, green font).
This investigation might be advanced in the future by further corroboration of mechanistic targets at the mRNA and protein level, by veri cation and quanti cation of key tumor tissue metabolites using refence standards and NMR-based methodologies, and by expanded metabolomic analyses that go beyond the hydrophilic analytes prioritized here, recognizing that chlorophylls and other lipophilic phytochemicals can exert anticancer effects in the colon [9][10][11][54][55][56][57][58][59][60].

Conclusions
We demonstrated, for the rst time, the marked anticancer e cacy of dietary spinach in the Apc-mutant Pirc model. After eliminating deregulated b-catenin as the primary mechanistic target, subsequent work identi ed signi cant reshaping of the gut microbiome in SPI-fed rats, along with changes in host transcriptomics and RNA-miRNA networks. Metabolomic analyses corroborated the predicted changes in linoleate and butanoate metabolism, TCA cycle, and pathways in cancer. Whereas butanoate metabolism was probably associated with increased a-diversity of the gut microbiome, multiple SPI-derived linoleate intermediates with reported anti-in ammatory and proapoptotic mechanisms were detected at increased levels in the colon tumors from rats treated with dietary SPI. Despite the presence of L-glutamate and Nacetylneuraminate in freeze-dried spinach, these intermediates were reduced markedly in colon tumors from SPI-fed rats, consistent with anticancer outcomes of the associated metabolites. The latter ndings warrant further investigation with respect to the potential for deregulated mitochondrial energetics leading to synthetic lethality [49], and altered cell surface glycans involved in immune evasion, oncogenic signaling networks and other pathways in cancer [50][51][52][53].  catenin in Pirc rats given Ctrl or SPI diets. Scale bar, 100 μm for 20X and 500 μm for 4X magni cation.

Figure 7
Metabolomic data pertaining to SPI consumption in the Pirc model. represents one colon tumor or one normal colonic mucosa sample from rats in the corresponding groups.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.