Multiomic analysis of FAP samples

We collected and analyzed n = 93 samples from the colons of six persons with FAP whose characteristics are shown in Table 1. The samples were isolated from multiple regions of the colon: ascending, transverse, descending (including sigmoid) and rectum (Fig. 1a). Samples spanned a range of sizes and degrees of dysplasia determined using histological staining: 26 polyp-adjacent ‘normal’ mucosa samples (hereafter referred to in the text as ‘mucosa’), 16 benign polyps (usually small) and 51 dysplastic polyps (small to large). The tumor percentage and degree of dysplasia in the dysplastic polyps were scored by a trained pathologist.

Table 1 Demographic and clinical characteristics of participants in the cohortFig. 1: Overview of the multiomic FAP study.

a, Schematic of tissue and blood collections, assays and downstream single-omic and multiomic analyses performed in this study. b, Summary of the clinical features of each of 93 colorectal samples derived from six persons with FAP, with up to four assays were performed on each sample. Portions of this figure were created with BioRender.com.

Source data

Samples were subjected to a battery of multiomic assays including bulk RNA sequencing (RNA-seq), proteomics (tandem mass tag (TMT) labeling), untargeted metabolomics (reverse-phase liquid chromatography (RPLC) and hydrophilic interaction LC (HILIC) combined with mass spectrometry (MS)8) and targeted lipidomics (Fig. 1b). High-resolution single-cell (sc) assays, including single-nucleus (sn) assays for RNA-seq and assay for transposase-accessible chromatin with sequencing (ATAC-seq) were performed (described in a recent manuscript9), along with methylation analyses10.

Extensive molecular changes in polyp development

To understand the molecular pathways activated during the early stages of polyp formation and cancer progression, we performed deep multiomic profiling of histologically normal mucosa, benign polyps and dysplastic polyps (Fig. 1a). RNA-seq, TMT-based proteomics, untargeted metabolomics and targeted lipidomics were performed on 60, 89, 77 and 77 samples, respectively. Because of limited material, particularly for the small polyps, not all assays could be performed for all samples. The relative levels of transcripts (n = 23,512), proteins (n = 12,389), metabolites (n = 1,157) and lipids (n = 514) were obtained after data curation and normalization.

Principal component analysis (PCA) of each -ome revealed a delineation among mucosa, benign polyps and dysplastic polyps, suggesting a progression of malignancy in precancer colonic tissue (transcriptomic PCA was omitted because of technical reasons; Methods); little difference was observed between benign and dysplastic polyps, while mucosa samples were clearly separated from both categories of polyp (Extended Data Fig. 1). Differential analysis was performed for each -ome between each pair of transitions between the three precancerous stages: mucosa versus benign polyp (M–B), mucosa versus dysplastic polyp (M–D) and benign polyp versus dysplastic polyp (B–D). Differential analysis revealed thousands of molecular changes (Fig. 2 and Supplementary Table 1a–d; P < 0.05). The largest numbers of molecular changes were found for transcripts (n = 2,110, 8.97% of 23,512 tests for one transition) and proteins (n = 1,451, 3.9% of 37,167 tests for three transitions) and the fewest were found for metabolites and lipid components (n = 486, 14% of 3,471 tests and n = 900, 52.5% of 1,713 tests across three transitions, respectively; Fig. 2 and Supplementary Table 1a–d). The number of unique differential molecules by -ome was as follows: transcriptome (n = 2,110, 8.97% of 23,512 transcripts), proteome (n = 978, 7.9% of 12,389 proteins), metabolome (n = 364, 31.5% of 1,157 metabolites) and lipidome (n = 466, 90.7% of 514 lipids). More downregulated molecules were identified than upregulated molecules (that is, less abundant in the more advanced cancerous stage). At the gene expression level (Supplementary Table 1a), most changes were evident in the M–D transition, as technical issues did not permit the identification of changes in the other transitions with statistical reliability (Extended Data Fig. 2). Similarly, the most significant differences in the proteome, lipidome and metabolome (Supplementary Table 1b–d) occurred between mucosa and polyp, whether M–B or M–D. Differential analysis was performed across all participants, but participant-specific differential analysis was also performed (Methods; the experimental design and statistics of differential analysis are shown in Supplementary Table 4a–g and the differential transcripts, proteins, metabolites and lipids are reported in Supplementary Table 5a–d); the differential analysis henceforth refers to the differential analysis across all participants.

Fig. 2: Differentially abundant molecules per -omic.

Top, histograms of the number of differentially abundant molecules per -omic analysis and contrast (M–B, M–D and B–D) separated by upregulated (red) and downregulated (blue) directionality (for example, red or upregulated in M–D indicates increasing abundance from mucosa to dysplastic polyp). Differential abundance was defined by an FC of ±15% and FDR of 5% for the transcriptome, an FC of ±15% and FDR of 1% for the proteome and a P value of 5% for the lipidome and metabolome. Bottom, Venn diagrams show the degree of differentially abundant molecules shared between the contrasts regardless of directionality for each single -omic. The absence of a bar (histogram) or circle (Venn diagram) means that no significant analytes were identified. The statistical tests used were two-sided Wald tests for the transcriptomic, lipidomic and metabolomic data and one-sided F-tests for the proteomic data.

Source data

Differentially abundant transcripts and proteins across the three precancerous transitions were analyzed using the Qiagen Ingenuity Pathway Analysis (IPA). Enrichments were performed separately for upregulated and downregulated molecules. Similar analyses were performed for lipid and metabolite species using ConsensusPathDB (CPDB)11 overrepresentation analysis (ORA12) and metabolite set enrichment analysis (MSEA13). Across all -omes and transitions, a total of 824 pathway enrichments were detected at a false discovery rate (FDR) of 5% (Supplementary Table 2), with 431 unique enriched pathways (Figs. 3 and 4a). These pathways cover a diverse set of biological areas (for example, immune, cancer and cell signaling). Because the number of pathways is extremely large only several of the key pathways are highlighted in later sections.

Fig. 3: Selected pathway enrichments run for differentially abundant molecules for each -omic.

Each dot indicates an enrichment of a pathway for a particular -omic; its size indicates the −log10FDR of the enrichment (a larger dot indicates a smaller adjusted P value) and the color indicates the median log2FC. a, Enrichment analyses for the transcriptome and proteome (separately) were performed using IPA and upregulated and downregulated molecules were enriched separately, although only downregulated enrichments were significant. Pathways were selected that reflected strong agreement between the proteome and transcriptome and were distributed across a variety of cellular and molecular functions relevant to CRC. b, Enrichment analysis for the lipidome and metabolome used the CPDB and MSEA tools rather than IPA and, similarly to the transcriptome and proteome, selected pathways spanning a variety of molecular and cellular functions relevant to CRC were selected (for example, arachidonic acid). Enrichments were performed with all differential molecules and pathways were deemed upregulated and downregulated according to the median log2FC of the enriched molecules. E1, estrone; CA, caffeine; ALA, alpha-linolenic acid; LA, linoleic acid; PL, phospholipid; S, sphingosine; HcyH, homocysteine; Met, methionine; [Hypo]Taurine, taurine and hypotaurine.

Source data

Fig. 4: A multiomic diagram describing the relationship between various enriched pathways in the M–D transition in terms of the hundreds of molecules they share in common.

The molecules are colored according to the log2FC and are generally shaped according to their molecular function. Lipids and metabolites are both symbolized as chemicals, while the remaining molecules are genes or proteins. Some molecules have both gene and protein information and are outlined with dots; otherwise, a bold outline indicates that the molecule is a protein. To reduce the complexity, molecules with similar prefixes and belonging to the same family were combined; subscript characters denote the different members of the family and the molecule complex is colored by the median log2FC. In situations where a complex contains both upregulated and downregulated members, half the molecule is colored with the median log2FC for one direction and the other half of the molecule is colored with the median log2FC for the other direction. Large circles are pathways from IPA, MSEA or CPDB ORA and are colored blue to indicate that all are downregulated from M–D. Lines flow from molecules to pathways to indicate membership. Thick lines that intersect with multiple thin lines simplify the flow of information and indicate that all thin lines move along the direction of thick lines. Lines that are dotted obviate lines they so happen to intersect. The asterisk collapses the following molecules with the IGHV (immunoglobulin heavy chain variable region) prefix: 1-18, 1-2, 1-46, 1-69, 2-26, 2-5, 2-70D, 3-13, 3-15, 3-30, 3-35, 3-43, 3-49, 3-7, 3-74, 4-28, 4-34, 4-4, 5-51 and 6-1.

Source data

M–B comparison

Comparing benign polyps to FAP mucosa identified 1,347 altered molecules (402 lipids, 420 metabolites, 525 proteins and 0 transcripts; Extended Data Fig. 2 and Methods) and over 100 altered pathways (175 pathway enrichments consisting of 162 unique pathways at FDR ≤ 5%; Supplementary Table 2). In the proteome (Fig. 3a), all M–B significant pathways were enriched for downregulated proteins; the figure shows 24 of 39 enriched pathways. The most enriched pathway (−log10FDR = 21.70) was acute-phase response signaling, a cytokine signaling-related pathway. It was one of many immune-related (cellular and humoral immune response and cytokine signaling) pathways (n = 15) in the 39. Only one pathway of the 39, glutathione redox reactions I (−log10FDR = 2.04) involved in biosynthesis, was unique in the M–B transition (Supplementary Table 2). Other top proteomic pathways (Fig. 3a) included nuclear receptor signaling pathways LXR and RXR activation (−log10FDR = 20.40) and FXR and RXR activation (−log10FDR = 17.90), a cellular stress or injury GP6 signaling pathway (−log10FDR = 12.50), a humoral immune response complement system (−log10FDR = 11.80) pathway and disease-specific pathways atherosclerosis signaling (−log10FDR = 11.20) and hepatic fibrosis and hepatic stellate cell activation (−log10FDR = 10.70). Key pathways related to CRC growth and treatment included phospholipase C (PLC) signaling14 (represses CRC growth) and interleukin (IL)-12 signaling and production in macrophages15 (increases radiosensitivity in mouse models of colon cancer because of increases in IL-12-dependent T helper cell 1 response).

Metabolites and lipids were also significantly dysregulated. Amino acids, nucleotides and neurotransmitters, notably, S-adenosyl-homocysteine, adenine, deoxyadenosine, xanthosine, nicotinamide adenine dinucleotide and citicoline were generally elevated, although histamine, a histidine product, decreased in this transition. Fatty acids (FAs), lipids and lipid immune mediators were also generally elevated, including docosahexaenoic acid, adrenic acid, linoleic acid, eicosatrienoic acid, hydroxyoctadecadienoic acid, eicosadienoic acid, docosapentaenoic acid, 5-hydroxyeicosatetraenoic acid, 1-(sn-glycero-3-phospho)-1d-myo-inositol, cyclic phosphatidic acids, glycolipids (phosphatidylcholine (PC), lysophosphatidylcholine (LPC), phosphatidylethanolamine (PE) and lysophosphatidylethanolamine (LPE)) and cholesterol esters (CEs). Downregulated molecules notably consisted of ceramides (CERs), sphingolipids (SPs), diglycerols and triglycerols, although exceptions were evident such as elevation of LacCer(d18:1/18:0). The known role of these molecules suggests potential implications for the regulation of inflammatory and immune pathways.

There were 35 pathways or pathway groups from 31 and 105 metabolomic and lipidomic enriched pathways in the M–B transition (Fig. 3b). SP metabolism (−log10FDR = 5.2) and arachidonic acid metabolism (metabolome, −log10FDR = 2.96; lipidome, −log10FDR = 2.0) were the most enriched lipid and metabolic pathways, respectively (five and six pathways had equivalent −log10FDR, respectively). Dysregulation of the arachidonic acid pathway is particularly interesting because it is involved in inflammation and can be suppressed by aspirin or other nonsteroidal anti-inflammatory drugs (NSAIDs), a treatment for persons with FAP to slow polyp progression and reduce CRC formation16,17,18. The pathway was upregulated in both lipidome and metabolome data for the M–B transition. Furthermore, the acetylsalicylic acid (ASA) pathway, which interacts with arachidonic acid, was also enriched for upregulated lipids in our dataset. Thus, our results provide a molecular explanation for the therapeutic strategy for FAP treatment. Altered pathways related to CER and SP metabolism and signaling were also altered in the lipidome; SP changes were previously shown to be evident in colon cancer19. CER and sphingosine are two important lipid molecules that have notable effects on T cell function20.

Strong lipidomic changes were observed in the pathway titled ‘immune system’ (Reactome; −log10FDR = 5.12, driven by CERs, CEs and glycerophospholipids (GPLs) (Supplementary Table 1c)). We also observed lipidomic alterations in Ca-dependent events (Reactome; −log10FDR = 3.30, driven by increases in PC and LPC (Supplementary Table 1c)), with Ca as a vital component of the immune system and intimately involved in the arachidonic acid pathway. PCs and LPCs are major components of cell membranes and also serve as reservoirs of FAs for energy and signaling21. Glycerolipids (GLs) can affect the localization or activity of Ca-binding proteins, Ca channels or other components of the Ca signaling pathway22 and, thus, impact processes such as immune cell migration, cytokine release and other immune functions. Lastly, despite a lack of Kyoto Encyclopedia of Genes and Genomes (KEGG)-identified pathway enrichments related to triacylglycerol (TAG) biology emerging in this analysis, the extensive alteration of TAG (Supplementary Table 1c) species in the M–B transition is noteworthy and relevant when comparing to later transitions.

M–D comparison

Large numbers of molecular (3,707 sourced from 2,110 transcripts; Extended Data Fig. 2 and Methods; 926 proteins, 438 lipids and 233 metabolites) and pathway (485 from 395 unique at FDR ≤ 5%; Supplementary Table 2) changes were also observed during the comparison of dysplastic polyp to mucosa. This transition accounted for the highest number of molecular and pathway changes (Fig. 2) across the three comparisons and was the only comparison that identified differential transcripts detected across more than a single batch or participant (Extended Data Fig. 2). Differentially expressed molecules were often shared between the M–B and M–D transitions for the lipidome, metabolome and proteome (Fig. 2). Key pathways are shown in Fig. 3.

Proteomic and transcriptomic pathway enrichments

Consistent with the M–B results, nearly all transcriptomic and proteomic pathways for M–D were downregulated (except the transcriptomic M–D enriched pathway intrinsic prothrombin activation, which was upregulated with −log10FDR = 1.30, marginally meeting the 5% FDR cutoff). Overall, 70 pathways of a total of 212 transcriptomic and 93 proteomic pathways were enriched across both modalities (Supplementary Table 2). Of the 98 total proteomic pathways, 38 were also seen in the M–B proteomic comparison.

The five most enriched transcriptomic pathways were the disease pathway pathogen-induced cytokine storm signaling (−log10FDR = 10.20), the cellular immune response pathway phagosome formation (−log10FDR = 9.54) and the organismal growth and development pathways axonal guidance signaling (−log10FDR = 9.47), cardiac hypertrophy signaling (enhanced) (−log10FDR = 9.47) and sperm motility (−log10FDR = 9.25); all pathways but axonal guidance signaling were supported by the proteome. The five most enriched proteomic pathways were cytokine signaling pathway acute-phase response signaling (−log10FDR = 33.30), cell growth and proliferation pathway B cell development (−log10FDR = 32.30), nuclear receptor signaling pathway LXR and RXR activation (−log10FDR = 27.70) and humoral immune response pathways complement system (−log10FDR = 26.80) and FcγRIIB signaling in B lymphocytes (−log10FDR = 24.70); all pathways but acute-phase response signaling were supported by the transcriptome for M–D (Supplementary Table 2). Notably, 57 and 29 M–D transcriptomic and proteomic pathways, respectively, were involved in the immune system (cytokine signaling, cellular immune response or humoral immune response), with 23 overlapping between the two. The 39 pathways depicted in Fig. 3a display a selection of enriched pathways with high concordance across transitions in the proteome and between the transcriptome and proteome; these pathways were among the most significant and relevant enrichments spread across a diverse group of cellular and molecular functions, including pathways relevant to CRC such as CRC metastasis signaling (−log10FDR = 6.03), PLC signaling and FXR and RXR activation, which was shown to antagonize Wnt and β-catenin signaling in CRC tumorigenesis23. Thus, the transcriptome and proteome showed a high degree of concordance in enrichments in the M–D transition and even between the M–B and M–D transitions in the proteome and all such enrichments (except for one) consisted of downregulated genes or proteins; these enrichments spanned a variety of cellular and molecular functions and many were related to CRC.

Lipidomic and metabolomic pathway enrichments

The M–D transition in persons with FAP highlighted extensive changes within both the metabolome and the lipidome profiles (Fig. 2); many of these were similar to those observed during the M–B transition but their magnitude in numbers and levels was usually greater. We observed an increase in both nucleotides and their derivatives, such as cytidine, cytosine, cytidine diphosphate and deoxyuridine, alongside amino acids and their derivatives. Notably, higher levels of specific amino acids and their derivatives such as serine, asparagine and deaminohistidine (imidazolepropionic acid) were evident. The M–D transition also exhibited an extensive number and increased level of CEs and GLs, accompanied by a higher number of decreased SPs, CER, monoglycerol, diglycerol and triglycerol. Interestingly, leukotriene A4, which was found to be increased during the M–D transition, along with 5-hydroxyeicosatetraenoic acid, has an important role in the arachidonic acid pathway, contributing to immune cell function.

Examination of the lipidomic and metabolomic pathway enrichments in Fig. 3b and Supplementary Table 2 reveals enrichments of both upregulated and downregulated metabolite and lipid sets for M–D. Between the 144 lipidome and 37 metabolome enrichments at FDR ≤ 5%, only three metabolomic pathways were enriched for the nearest analog in the lipidome in the M–D transition: arachidonic acid (metabolome, −log10FDR = 3.40; lipidome, −log10FDR = 1.94), phospholipid biosynthesis (metabolome, −log10FDR = 3.40; lipidome, −log10FDR = 5.81) and SP metabolism (metabolome, −log10FDR = 4.81; lipidome, −log10FDR = 7.24). In the metabolome, the six most enriched pathways were primarily related to amino acid and nucleotide metabolism: histidine metabolism (−log10FDR = 6.82), β-alanine metabolism (−log10FDR = 6.57), pyrimidine metabolism (−log10FDR = 6.54), pterine biosynthesis (−log10FDR = 6.30), arginine and proline metabolism (−log10FDR = 5.66) and purine metabolism (−log10FDR = 5.49). The six most enriched lipidomic pathways were naturally related to lipid metabolism and biosynthesis: methylenetetrahydrofolate reductase (MTHFR) deficiency (−log10FDR = 7.53), SP metabolism (−log10FDR = 7.24), SP de novo biosynthesis (−log10FDR = 6.90), high-density lipoprotein remodeling (−log10FDR = 6.09), plasma lipoprotein assembly, remodeling and clearance (−log10FDR = 6.09) and phospholipid biosynthesis (−log10FDR = 5.81) (Supplementary Table 2).

These altered metabolomic and lipidomic pathways are expected to have an abundance of molecular and cellular consequences. Metabolic decreases in the histidine pathway are reported to induce a decrease in amino acid oxidation and a decrease in protein turnover24; altered histidine metabolism also attenuates antioxidant defense, redox balance and signaling25,26. Nucleotide metabolic pathways are crucial for cell division27 and are shifted the M–D transition, presumably to support rapid cell rapid cell proliferation28. Alteration of SP metabolism is observed in the M–D transition and has connections to immune cell function20.

Similarly to the M–B transition, dysregulation of lipid metabolism and immune cell response emerged in the M–D transition as the most prominently altered pathways enriched in both the metabolome and the lipidome of persons with FAP. Enrichment of arachidonic acid and linoleic acid metabolism, both of which intersect with immune regulation29,30 (Fig. 3b), was observed for the M–B transition. Matching against the Reactome database revealed an increased number of GLs and enrichment of diacylglycerol (DAG) and Ca+-dependent events in the M–D transition (Fig. 3b and Supplementary Table 2). A higher proportion of TAG species was also observed in the M–D transition compared to the M–B transition. This shift in the abundance of TAG species could indicate a notable metabolic shift that is likely related to changes in energy use, signaling pathways or other cellular processes.

Figure 3b also showcases other important lipidomic and metabolomic pathways that were enriched in M–D and spanned a broad number of molecular and cellular functions. As found for the M–B transition, notable downregulated enriched lipid pathways and pathway groups for M–D were SP metabolism (Wikipathways, −log10FDR = 7.24), SP biosynthesis (Reactome, −log10FDR = 6.90), CER signaling (BioCarta, −log10FDR = 4.61) and glycosphingolipid (GSL) metabolism (Reactome, −log10FDR = 3.67), which was also downregulated enriched for M–B. Among upregulated enriched lipid pathways and pathway groups for M–D were immune-related pathways, Ca-dependent events and immune system, lipids acting as immune mediators, arachidonic acid and ASA (aspirin) pathways and lipid-related pathway groups (GPL metabolism, phospholipid biosynthesis and alpha-linolenic acid, linoleic acid and linoleate metabolism). Some of these pathways are related to CRC. For instance, Ca-dependent events such as Ca2+ remodeling is associated with CRC. Ca2+ also interacts with aspirin, which inhibits CRC by blocking CRC from free intracellular Ca2+ (ref. 31). Ca is also relevant to PLC because it is ejected from intracellular compartments in T cells during PLC activation by DAG, protein kinase C (PKC) and inositol 1,4,5-trisphosphate32. PLC is important for regulating both Ca and phosphoinositide33. Ca-dependent events are connected intimately with other molecules and pathways associated with CRC. In addition, SP biosynthesis is related to CRC; an increase in the ratio of sphingosine-1-phosphate (S1P) to CER is associated with cancer progression19. We found S1P signaling in the transcriptome and CER signaling in the lipidome to be enriched in the M–D transition and, interestingly, both S1P and CER are critical components for the formation of arachidonic acid34. The high connectivity of the molecules in the S1P and CER signaling and PLC signaling pathways are shown in Fig. 4 for the M–D transition. Notable downregulated enriched metabolite pathways and pathway groups for M–D included FA biosynthesis, SP metabolism, inositol metabolism and gluconeogenesis; notable downregulated enriched metabolite pathways and pathway groups for M–D included purine and pyrimidine metabolism, arachidonic acid, alpha-linolenic acid, linoleic acid and linoleate metabolism and assorted amino acid pathways. These pathways are relevant to CRC in a fashion similar to that described for the lipidome.

Multiomic pathways and networks characterizing M–D transition

To illustrate the connections between CRC related pathways in the M–D transition, six such pathways (IL-12 signaling and production in macrophages, PLC signaling, FXR and RXR activation, SP and sphingomyelin (SPH) metabolism, CER signaling and S1P signaling) were selected and a multiomic pathway diagram was constructed (Fig. 4). Numerous proteins and transcripts relevant to CRC emerged as differentially altered including downregulated Serpin family members such as SERPINF2 (ref. 35) and SERPINA1 (refs. 35,36) (although these molecules are upregulated in adenomas and cancer relative to mucosa), upregulated NOTUM (ref. 37) and upregulated S100A8 (refs. 38,39). Several lipids and metabolites whose downregulation is associated with CRC tie SP membrane metabolism and signaling pathways together explicitly, such as CER and SPH.

CER is a proapoptotic anticancer SP synthesized in several pathways, one of which is an alternative salvage pathway that converts SPH to CER40; the biosynthesis of SPs and signaling of CER were both enriched in the M–D transition but CER itself was downregulated. Additionally, even though the CER salvage pathway was not enriched until the B–D transition, SPH was downregulated in the M–D transition, indicating that there may be evidence of compensatory lipidomic and metabolomic mechanisms to halt the progression of CRC in precancerous tissue. Not only do lipids and metabolites connect these pathways but two important receptors detected in the transcriptome also do; S1P receptors S1PR1 and S1PR3 connect S1P signaling with SP and SPH metabolism and CER signaling and were downregulated in the M–D transition. Classically, activation of S1P signaling heralds the progression of cancer in colonic tissue41. The same is generally true of the receptors S1PR1 and S1PR3; however, the literature is not in complete agreement on this41 and it may be that the precancerous dysplastic polyps are not advanced enough to evince upregulation of these receptors at the mRNA level. Other important molecules connecting CRC-associated pathways include PIK3CD, which normally upregulates protein kinase B, glycogen synthase kinase 3β and β‐catenin and, thus, colon cancer growth and proliferation42. However, as shown in Fig. 4, in the M–D transition, both AKT3 and PIK3CD, which connect IL-12 signaling with SP and SPH metabolism and CER signaling, were downregulated, again re-emphasizing that dysplastic tissue does not yet exhibit the full characteristics of CRC. Overall, Fig. 4 shows the numerous connections of key molecules governing important pathways that occur early in cancer formation and that the molecules between these pathways were overwhelmingly downregulated.

The multiomic data were further subject to analysis by inferring a latent axis correlated with dysplastic progression (Extended Data Figs. 3 and 4) along which molecules of all molecular types covaried (Fig. 5). This method identified the top covarying transcripts, proteins, lipids and metabolites associated with dysplasia (Fig. 5). Verification was restricted to the top ten transcripts along the latent factor and half had some connection to CRC. SERPINB5 is involved in CRC progression; it depends on and covaries positively with the cancer marker carcinoembryonic antigen43. SGK is involved in cell survival, proliferation and ion transport and has been implicated in CRC progression44,45. CLCN2 expression is reduced in CRC compared to normal colonic tissue46. NECTIN4 is a cell adhesion molecule involved in cell–cell interaction that is dysregulated in CRC47 and COLEC12 attenuates stem cell-like features in CRC on knockdown48. Lastly, DPYD is related to CRC treatment response46.

Fig. 5: Selected molecules from MOFA with high weight alongside the latent factor (factor 1) correlated with dysplasia.