Multi-omic Analysis Reveals Metabolic Pathways that Characterize Right-Sided Colon Cancer Liver Metastasis

Abstract

There are well demonstrated differences in tumor cell metabolism between right sided (RCC) and left sided (LCC) colon cancer, which could underlie the robust differences observed in their clinical behavior, particularly in metastatic disease. As such, we utilized liquid chromatography-mass spectrometry to perform an untargeted metabolomics analysis comparing frozen liver metastasis (LM) biobank samples derived from patients with RCC (N=32) and LCC (N=58) to further elucidate the unique biology of each. We also performed an untargeted RNA-seq and subsequent network analysis on samples derived from an overlapping subset of patients (RCC: N=10; LCC: N=18). Our biobank redemonstrates the inferior survival of patients with RCC-derived LM (P=0.04), a well-established finding. Our metabolomic results demonstrate the increase of reactive oxygen species associated metabolites and bile acids in RCC. Conversely, carnitines, indicators of fatty acid oxidation, were relatively increased in LCC. The transcriptomic analysis implicated increased MEK-ERK, PI3K-AKT and Transcription Growth Factor Beta signaling in RCC LM. Our multi-omic analysis reveals several key differences in cellular physiology which taken together may be relevant to clinical differences in tumor behavior between RCC and LCC liver metastasis.

1. Introduction

The anatomic laterality of colon cancer is a factor that is growing in salience from both a clinical and biological perspective. Primary tumors can be divided into right-sided colon cancers (RCC) – those that occur in the cecum, ascending colon, and hepatic flexure – and left-sided colon cancers (LCC) – those that occur in the splenic flexure, descending colon, and sigmoid colon. Several studies demonstrate that RCC tends to have worse overall survival, independent of stage, the effect being most profound in metastatic disease [1–4]. Tumor laterality furthermore impacts the efficacy of targeted therapies – notably Epidermal Growth Factor Receptor-inhibitors (EGFRi) – on metastatic colon cancer; while effective in prolonging survival in LCC they have been shown to be less efficacious in RCC [5, 6]. A number of studies have begun to elucidate the differences in tumor biology across laterality underlying these differences in clinical behavior, utilizing a wide array of techniques including genomics [7], transcriptomics [8–10], and most recently metabolomics, which can capture a more direct picture of cellular physiology [11, 12].

While the clinical effects of laterality are most prevalent in metastatic disease, few studies have investigated differences in metastatic tumors themselves [13]. Therefore, we carried out the first untargeted metabolomic study of colon cancer liver metastases (LM) to continue to probe differences in tumor laterality. We chose liver metastases, as the liver is the most common site of metastasis in colon cancer overall [14]. This approach has the added benefit of reducing the impact of the local metabolomic environment (e.g., such the local microbiome and gut physiology), that vary between the left and right colon. To aid in contextualizing our findings, we also carried out a differential gene expression analysis across laterality utilizing an overlapping cohort. We hypothesize that there are metabolomic and transcriptomic differences between colon cancer LMs derived from RCC and LCC reflective of key differences in tumor biology that may underpin their differences in clinical behavior.

2. Materials and Methods

2.1. Study Cohort

Detailed descriptions of all methods can be found in the supplemental methods section. For the metabolomic analysis, patient-matched samples of colon cancer LMs and adjacent normal liver (NL) (RCC-derived: N=32; LCC-derived: N=58) were collected from a previously published biobank [12]. The biobank consists of prospectively collected, perioperatively snap-frozen (liquid nitrogen) tissue of 736 informed and consenting colorectal cancer patients undergoing primary and/or metastasis resection or biopsy at Memorial Sloan Kettering Cancer Center in New York, NY during a ten-year period (1991–2001). Samples from the cecum, ascending colon and hepatic flexure were classified as RCC, and those from the splenic flexure, descending colon and sigmoid colon were classified as LCC. A majority of the patients received 5-fluorouracil (5-FU)-based chemotherapy prior to liver surgery. Transcriptomic analysis was performed on a secondary group of samples derived from an overlapping subset of the biobank. We selected colon cancer LM samples according to the same criteria as above (RCC derived: N=10; LCC derived: N=18) and analyzed the untargeted differential gene expression between the two groups.

2.2. Methods Description

2.2.1. Metabolomics Sample Preparation and LCMS Analysis

Briefly, snap frozen tissue samples from patients in the metabolomics cohort were processed following a previously published protocol [12]. Pooled quality control (QC) samples were generated from combined aliquots of each processed sample. An untargeted liquid chromatography-mass spectrometry (LC-MS) analysis was carried out following an established procedure [15]. Both hydrophilic interaction liquid chromatography (HILIC) and reversed-phase liquid chromatography (RPLC) column chemistries were used in electrospray ionization negative and positive modes respectively, in order to obtain broad coverage of the metabolome. QC samples were injected every 10 samples to monitor consistency of data acquisition and for subsequent abundance normalization.

2.2.2. Transcriptomics Sample Preparation

50 ng of total RNA was extracted from tissue samples with a Qiagen (Germantown, MD) RNeasy Mini Kit according to the manufacturer’s instructions, in a trizol-free manner, and treated with DNase. The purity and quality of RNA samples were validated with a Nanodrop ND-1000 (Thermofisher Scientific; Waltham, MA) and Agilent (Santa Clara, CA) Bioanalyzer 2100, respectively. Purified RNA samples were submitted to the Yale Center for Genomic Analysis (YCGA) according to their specifications, where they underwent mRNA-Seq profiling with a NovaSeq 6000 Illumina Sequencer (San Diego, CA).

2.2.2. Immunohistochemistry Preparation

Tumor blocks were randomly selected from the RCC liver metastasis (N=4) and LCC liver metastasis (N=4) cohorts for immunohistochemistry. Tissues was fixed in 4% paraformaldehyde overnight, embedded in cryo-embedding media, sectioned, and mounted on slides. Primary antibodies, TGF-β (1:500, 21898-1 AP, Proteintech, IL) and CERCAM (1:500, PA5-32118, ThermoFisher Scientific, Waltham MA), were incubated overnight at 4°C. Staining was performed using the VECTASTATIN Elite ABC kit and DAB peroxidase substrate kit (Vector Laboratories, Newark, CA).

2.3. Statistical Analysis

2.3.1. Cohort Demographics and Survival Analyses

All statistical analysis was performed in R unless otherwise specified. Chi-squared tests were used to assess for significant differences in sex and stage between the LCC and RCC groups used in each of the analyses. In cases in which there were <5 samples in a subgroup, Fisher’s exact test was used. Significant differences in age were assessed using a Mann-Whitney U test. A survival analysis was carried out on all stage IV patients in the biobank who presented with liver metastases, stratified by laterality (N=177). A Mantel-Cox test used to assess statistical significance.

2.3.2. LC-MS Data Processing and Metabolite Identification

Raw spectral data were processed following previously published methods, to generate a list of filtered features [12, 15]. To identify metabolites, both the mass-to charge-ratio (m/z) and retention time (rt) of the spectral features were compared with those of an in-house library of metabolites analyzed in both HILIC (808 standards) and RPLC (1021 standards) modes. This yielded 341 matched metabolites in HILIC mode, and 124 in RPLC. Top metabolites were confirmed with MS/MS fragmentation analysis, and annotated with identification confidence levels following standard practice [16].

2.3.3. Metabolomics Data Analysis

MS data acquired using either HILIC or RPLC columns were analyzed separately. The log₂ paired fold change of LM over NL was calculated for each identified metabolite in each patient, and then averaged by laterality group for each metabolite. A paired Mann-Whitney U-test was used to compare these paired metabolite abundance values and adjusted for multiple comparisons using the Benjamini-Hochberg (BH) false discovery rate (FDR) method. This was done in order to normalize the metabolite abundances in LM compared to the patient-specific normal tissue background in which the metastases were growing.

We also employed a statistical method we deemed the Right-Left (RL) Index. The difference between the mean paired log₂ fold change (LM/NL) between RCC and LCC was taken for each metabolite (RCC-LCC), and a Welch’s t-test used to compare the two groups (BH-FDR adjusted). The distribution of these differences identifies the metabolites that are more abundant in RCC- versus LCC-derived LMs. The z-score for each metabolite was calculated to determine which metabolites could be classified as “outliers” (>1.96 standard deviations from the mean), favoring either RCC or LCC. The two modes (HILIC and RPLC) were compared separately given their different sensitivities and biases in feature detection.

2.3.4. Transcriptomic Analysis

Gene-expression data was input into Galaxy RNA-seq analysis software for processing [17]. Filtered FPKM data was generated, and Log₂ fold change (log-fc) was calculated for each gene between RCC and LCC (RCC/LCC) derived LMs as well as a Cohen’s D statistic; significance was assessed with a Welch’s t-test, controlled for multiple comparisons with BH-FDR. An Ingenuity Pathway Analysis (IPA) [18] (Qiagen) was conducted on the top differentially expressed genes between RCC- and LCC-derived LMs to assess differences in molecular interaction networks and predicted upstream regulation. For this, genes with differential expression significant for an α <0.1 were utilized as is common practice [19–21].

2.3.4. Immunohistochemistry Analysis

Image J software was used to obtain the percentage of positively stained cells. Statistical significance between groups was assessed by students t-test (after normality test) on GraphPad Prism (version 9.2.0).

3. Results

3.1. Metabolomics

We performed a metabolomic analysis on samples from all patients from our biobank that had both LM and NL tissue available, which totaled 32 patients with RCC-derived LMs and 58 patients with LCC-derived LMs (Table 1A). Between the two groups no significant differences were observed regarding age (P=0.950) and sex (P=0.738). A limited number of samples were tested for pathogenic KRAS mutation status [22], which also did not statistically differ by laterality (P=0.22).^* Of the samples tested for MSI status, only two (one of each RCC and LCC) were found to be MSI-H, and no samples had BRAF mutations (Table S1). Additionally, a survival analysis was carried out on all patients from the biobank with stage IV liver metastases (from which the metabolomics cohort was primarily drawn) that demonstrated inferior survival of RCC LMs (P=0.047) (Table S2, Figure S1). Nearly 60% of patients received chemotherapy prior to liver surgery in both cohorts, a majority received chemotherapy, in general, for their metastatic colon cancer treatment, and no patients received epidermal growth factor receptor (EGFR) inhibitors (Table S2). There were no significant differences between the RCC and LCC LMs cohorts in regard to number of liver metastasis, size of largest liver metastasis, and synchronous vs. metachronous liver metastasis (Table S2). Chronic liver disease, when assessed, was exceedingly low (N=1) for each cohort.

Table 1. Cohort Demographics.

Panel A depicts the breakdown of the metabolomics cohort. P-values correspond to analyses of the distribution of age and sex between the right sided liver metastasis (RCC LM) and left sided liver metastasis (LCC LM) groups. Panel B depicts the same for the transcriptomics cohort, which is a subset of patients overlapping with the metabolomics cohort from the original biobank.

A
	Total	RCC	LCC	P-value
Sex (n)	M: 57 (63.3%)	M: 21 (65.5%)	M: 36 (62.1%)	0.738
	F: 33 (36.7%)	F: 11 (34.4%)	F: 22 (37.9%)
Median Age (years)	59.5 (±13.3)	59.5 (±15.3)	60.5 (±12.1)	0.950
Total (N)	90	32	58	--
B
	Total	RCC	LCC	P-value
Sex (n)	M: 13	M: 3 (30%)	M: 10 (55.6%)	0.254
	F: 15	F: 7 (70%)	F: 8 (44.4%)
Median Age (years)	64 (±9.5)	67(±8.1)	63(±9.9)	0.191
Total (N)	28	10	18	--

3.1.1. Top Differentially Abundant Metabolites

We compared the normalized metabolomic profiles of LMs between patients with RCC and LCC primary tumors utilizing the Right-Left (RL) Index (see methods section). Across both modes, we identified 17 metabolites of interest relatively more abundant in RCC, and 11 in LCC, the characteristics and statistics of which are summarized in Table 2. When the p-values of the differences were adjusted for multiple comparisons, none were significant. However, unadjusted differences in cystine (RL: 3.52, P=0.046), tetradecanediodic acid (RL: −2.28, P=0.033), reduced glutathione (GSH) (RL: −2.667, P=0.030) and octanoylcarnitine (RL: −3.14, P=0.019) were significant for α<0.05.

Table 2. Top Differentially Abundant Metabolites by RL-Index.

This table lists the top differentially abundant metabolites, as determined by their Right-Left Index (|RL-Index|>1.96). Positive RL-Index values correspond to favoring Right-derived tumors (RCC), and negative to left side derived (LCC). Corresponding mean difference in paired log fold change (RCC-LCC) from normal liver (NL), used to calculate the RL-Index value is displayed. The corresponding p-values, both raw and BH FDR adjusted are likewise displayed (Welch’s T-test). The paired log fold changes (log FC) from NL, with corresponding Benjamini-Hochberg adjusted q-values (BH FDR) are furthermore shown for both RCC and LCC (Mann-Whitney U-test). Other characteristics include the Human Metabolite Database (HMDB) ID of each metabolite, the mode in which they were detected, their true mass-to-charge ratio (mz), the error of the experimentally measured mz from true mz (ppm), the standard retention time (RT) (s), the error in experimentally measured RT from standard RT (s), as well as the level of confidence in the ID identification (see methods).

Name	ID Level	HMDB ID	Mode	MZ	MZ Error (ppm)	RT (sec)	RT Error (sec)	LCC LM v NL FDR	LCC Log FC	RCC LM v NL FDR	RCC Log FC	Log FC Diff.	R-L Index	Diff. p-value	Diff. FDR
L-Cystine	1	HMDB0000192	HILIC (−)	240.02385	3.92122	418.3	11.4	3.20E-06	−1.72	2.81E-02	−0.72	1.00	3.52	4.55E-02	9.97E-01
Cysteine-S-sulfate	1	HMDB0000731	HILIC (−)	200.97656	4.13282	301.2	13.0	2.77E-01	0.54	3.79E-02	1.50	0.95	3.35	1.24E-01	9.97E-01
Asparaginyl-lysine	1	HMDB0004987	HILIC (−)	261.13247	2.58581	450.6	10.7	1.69E-05	−1.88	6.50E-02	−1.12	0.76	2.68	2.45E-01	9.97E-01
Dehydroascorbic acid	1	HMDB0001264	HILIC (−)	174.01644	6.17563	159.0	12.4	9.92E-06	−1.70	7.29E-03	−0.94	0.76	2.67	2.25E-01	9.97E-01
Glycodeoxycholic acid	2	HMDB0000631	RPLC (+)	449.31412	2.51036	329.4	4.6	3.63E-09	−3.24	5.74E-08	−2.45	0.79	2.62	9.49E-02	9.81E-01
Prolyl-Alanine	2	HMDB0029010	HILIC (−)	186.10044	0.88352	339.0	7.1	3.20E-04	−1.27	1.86E-01	−0.56	0.71	2.48	1.39E-01	9.97E-01
Farnesyl acetate	2	HMDB0240268	HILIC (−)	264.20893	2.56433	49.2	7.6	3.09E-03	−0.51	8.11E-01	0.17	0.68	2.39	2.39E-02	9.97E-01
Glycerophosphocholine	2	HMDB0000086	HILIC (−)	257.10282	1.17977	374.4	0.0	3.06E-09	−3.53	1.13E-06	−2.88	0.66	2.30	2.80E-01	9.97E-01
L-Phenylalanyl-L-proline	2	HMDB0011177	HILIC (−)	262.13174	3.17180	232.8	4.9	8.46E-07	−1.66	5.89E-03	−1.09	0.57	1.98	2.97E-01	9.97E-01
Orotic acid	2	HMDB0000226	HILIC (−)	156.01711	3.37162	207.6	3.6	3.38E-01	0.63	4.37E-01	0.08	−0.55	−1.97	4.71E-01	9.97E-01
3-Hydroxybenzyl alcohol	2	HMDB0059712	HILIC (−)	124.05243	2.37908	159.0	12.8	2.05E-04	1.44	2.46E-01	0.81	−0.63	−2.25	1.27E-01	9.97E-01
Tetradecanedioic acid	2	HMDB0000872	HILIC (−)	258.18311	3.38771	205.8	5.6	1.76E-08	−1.53	1.62E-07	−2.17	−0.64	−2.28	3.39E-02	9.97E-01
Dimethylguanosine	2	HMDB0004824	HILIC (−)	311.12297	7.96397	177.2	10.6	1.88E-02	0.58	9.06E-01	−0.12	−0.70	−2.51	5.89E-02	9.97E-01
Adenine	1	HMDB0000034	HILIC (−)	135.05450	2.36509	162.0	1.7	3.10E-02	0.59	3.19E-01	−0.12	−0.71	−2.53	2.39E-01	9.97E-01
CMP sialic acid	2	HMDB0001176	RPLC (+)	614.14727	0.07564	39.6	5.1	1.49E-01	−0.18	3.28E-02	−0.94	−0.76	−2.54	6.34E-02	9.81E-01
L-Glutathione reduced	1	HMDB0000125	HILIC (−)	307.08381	2.56255	475.0	0.0	5.33E-04	−0.45	1.09E-05	−1.19	−0.75	−2.67	3.05E-02	9.97E-01
N-Acetyl-L-tyrosine	1	HMDB0000866	HILIC (−)	223.08446	3.97041	214.8	12.7	2.30E-05	−1.49	2.80E-06	−2.30	−0.81	−2.89	1.13E-01	9.97E-01
Decanoylcarnitine	1	HMDB0000651	RPLC (+)	315.24096	7.16362	286.8	6.7	1.05E-08	3.84	8.35E-07	2.95	−0.88	−2.95	9.89E-02	9.81E-01
Octanoylcarnitine	1	HMDB0000791	RPLC (+)	287.20966	7.05885	250.8	0.9	2.30E-07	1.97	7.20E-03	1.03	−0.94	−3.14	1.85E-02	9.81E-01
UDP-D-glucuronate	1	HMDB0000935	HILIC (−)	580.03429	1.10761	469.1	13.9	5.40E-03	1.37	2.20E-01	0.36	−1.01	−3.58	3.84E-01	9.97E-01

3.1.2. Patterns in Molecular Class

Based on the presence of molecules with shared characteristics and known roles in colon cancer pathology, the RL distribution of two molecular classes were further assessed. Bile acids are a class of molecule with well-known roles in colon cancer pathology [23]. Glycodeoxycholic acid (GDCA) (RL: 2.62, P=0.094) was shown to be more abundant in RCC in our analysis. Therefore, we assessed the RL distribution of all bile acids in present in our data and found a pattern of distribution favoring RCC LMs (Figure 1A): 4/7 bile acids were found to have an RL index value > 1.0 (greater than one standard deviation). Statistical details can be found in Table S3. Carnitines are another class of molecules associated with colon cancer pathology via their specific role in fatty acid oxidation [24]; both octanoylcarnitine (RL: −3.14, P=0.019) and decanoylcarnitine (RL: −2.95, P=0.099) are metabolites of interest increased in LCC over RCC derived LMs. Consequently, we also assessed the RL distribution of all carnitines present in our analysis. Figure 1B (Table S4) demonstrates a pattern of distribution of carnitines favoring LCC LMs, with 7/10 demonstrating an RL Index value < −1.0.

Figure 1. RL Distribution of Bile Acids and Carnitines.

This figure shows the location on each Right-Left (RL) distribution of key metabolites belonging to the class of carnitine (orange) or bile acid (dark green). Panel A represents those metabolites detected in HILIC (−) mode, and panel B those detected in RPLC (+) mode. The RL-Index of each metabolite is a measure of their preference for either the left or right side, based on the distribution of differences in normalized abundances (log fold change difference from normal liver) for each metabolite between the two sides (RCC-LCC).

-CA = carnitine; GCDCA = glycochenodeoxycholic acid; TDCA = taurodeoxycholic acid; TLCA-SO₄ = Taurolithocholic acid sulfate; GDCA = glycodeoxycholic acid; DCA = deoxycholic acid; HILIC = hydrophilic interaction liquid chromatography; RPLC = reverse phase liquid chromatography

3.2. Transcriptomics

We additionally performed a differential gene expression analysis comparing LMs from 10 RCC patients and 18 LCC patients, representing an overlapping subset of patients in the biobank (Table 1B). Within this cohort, there were no significant differences between RCC and LCC LM patients regarding sex (P=0.254) or age (P=0.191).

3.2.1. Differential Gene Expression

Differential gene expression analysis between RCC LMs and LCC LMs was conducted on 12,401 filtered genes, the results of which are plotted in Figure 2. Genes of interest were selected based on a q-value of less than 0.1 resulting in a set of 40; the statistics are summarized in Table S5. Of note the genes all demonstrate |Cohen’s D statistic| >1.5, indicating a high effect size. There are seven genes that meet the threshold of α<0.05, and each have a |log-fc RCC/LCC|>1 and |Cohen’s D|>2: CERCAM, SCARF2, SMARCD3, PALLD, LEPREL2, LOX and ZNF681. All favored RCC except ZNF681. Only CERCAM meets the threshold for α<0.01.

Figure 2. Differential Expression Analysis of RCC versus LCC.

This figure shows a volcano plot of all the genes selected for analysis by differential expression between RCC and LCC. The significance is calculated by taking the negative log₁₀ of the Benjamini-Hochberg (BH FDR)-corrected Welch’s T-test for each comparison.

Differential expression is displayed as the log of the fold change of FPKM (Fragments per kilobase of transcript per million mapped reads) normalized abundance between right (RCC) and left (LCC) side derived tumors. The horizonal line represents a significance cutoff of α<0.1, and those genes meeting the threshold of α<0.05 are labelled.

3.2.2. Pathway and Network Analysis

IPA software was used to analyze all the genes differentially regulated between RCC LMs and LCC LMs significant to a level of α < 0.1, in order to identify relevant differences on the network and pathway level, as well as predict possible upstream regulation. The top predicted molecular network (score 52 – highly significant) from network analysis is displayed in Figure 3, which demonstrates MAPK, ERK, AKT, PI3K and TGF-β as some of the most connected nodes. Key elements of both the PI3K-AKT (represented by both PI3K and AKT), and the MEK-ERK (represented by MAPK, and ERK) pathways demonstrate predicted upregulation. Both of these are well known effector pathways downstream of EGFR [25, 26], however notably EGFR is not implicated in the network. Finally, TGF-β is predicted to play a central role in the upregulation of many of the nodes in the network; indeed, it is also predicted to be highly upregulated in RCC as one of the top upstream regulators in the analysis (P=4.34E-04, z=2.34).

Figure 3. IPA Network Analysis of Top Differentially Expressed (DE) Genes.

An IPA nodal network analysis was carried out with default network settings, utilizing all genes from the DE analysis meeting the significance threshold of α<0.1. The figure shows the top network generated, with a score of 52. The legend in the top left guides interpretation of the relationships. Nodal shapes correspond to molecular functions. Vertical Rhombi represent enzymes, horizontal rhombi represent peptidases, double circles represent complexes and molecular groups, horizontal ovals represent transcription regulators, vertical ovals represent transmembrane receptions and squares represent cytokines. Circles are nonspecific.

3.2.3. Immunohistochemistry

Results from immunohistochemistry (IHC) showed that the expression of TGF-β (P=0.0010) and CERCAM (P=0.0009) in the RCC tissues was higher than that in the LCC tissues, which is consistent with our hypothesis that CECAM and TGF-β play important roles in RCC-derived LM (Figure 4).

Figure 4. Immunohistochemistry of TGF-β and CERCAM of RCC and LCC Liver Metastasis Cohorts.

Left panel shows representative immunohistochemical staining for TGF-β and CERCAM in RCC and LCC-derived metastatic liver tumor samples. Right panel shows the percentage of cells positively stained for TGF-β and CERCAM, with higher expression in the RCC cohort (N=4) than that in the LCC-derived liver metastasis cohort (N=4). All values are presented as mean ± standard deviation. Student’s t-test was used ***p ≤ 0.001

4. Discussion

By integrating untargeted metabolomic and transcriptomic analyses derived from metastases and background normal tissue of patients undergoing liver surgery for colon cancer liver metastasis, our study provides insight into potential mechanisms of the differing clinical behavior of RCC and LCC LMs; this was done in a clinically relevant cohort taken from a biobank that recreated the inferior survival of patients with RCC derived LMs. The metabolomics analysis revealed increased reactive oxygen species (ROS) and bile acids, as well as downregulated fatty acid oxidation (FAO) in RCC versus LCC LMs. Concurrently, the transcriptomics pathway analysis demonstrated increased TGF-β, and evidence of increased MEK-ERK and PI3K-AKT signaling in RCC LM. Neither of the analyses showed significant variation with respect to sex or age between RCC and LCC patients, suggesting laterality was an independent driver of the observed differences between the groups beyond these factors. Taken together, the results provide evidence of putative cellular mechanisms potentially underlying the inferior prognosis of metastatic RCC, as well as its resistance to EGFR inhibition (Figure 5).

Figure 5. Potential Mechanisms of Pathogenicity and EGFRi Resistance in RCC LM.

This summarizes the findings that suggest potential candidate mechanisms for RCC resistance to EGFR inhibitor therapy. Dark colored or glowing elements indicate observed or IPA-predicted findings (i.e., a differentially expressed gene, or metabolite class), with green elements signifying upregulation, and red downregulation. These elements are also labelled by corresponding colored arrows. Differentially expressed genes are indicated with asterisks corresponding to level of significance and predicted differential regulation of molecules (via IPA) are indicated with crosses (*<0.05, **<0.01; ††<0.01, †††<0.001). Finally, both the MEK-ERK and PI3K-AKT pathways were predicted to be significantly upregulated by the IPA Network Analysis, which is indicated by their enlarged activation arrows and green highlights.

4.1. Metabolomics

GSH was among the notable differentially abundant metabolites, and found to be more strongly downregulated in RCC LMs. GSH serves as a major antioxidant in human cells [27], and its dearth in RCC suggests that ROS may be more abundant in liver metastases derived from the right colon.^† This is bolstered by the presence of several metabolites associated with increased ROS among those most relatively increased in RCC. Cysteine-S-sulfate (Cys-SO₃H) is a highly oxidized derivative of cysteine [28], and thus evidence of an oxidized cellular state in RCC. Of note, cystine, the dominant form of cysteine in the cell, was also shown to be more abundant in RCC. Dehydroascorbic acid (DHAA) – the oxidized form of vitamin C – is a known to be a potent oxidizer of colon cancer cells via its depletion of GSH [29]. Finally, GDCA is a secondary bile acid known to generate ROS via mitochondrial activation and is implicated in colon cancer development [30, 31]. While it is acknowledged that the differences in Cys-SO₃H, DHAA and GDCA do not individually reach statistical significance, and GSH and does not meet adjusted significance, the fact that they are all present among the most differentially abundant metabolites increased in RCC LMs, is a highly suggestive pattern of distribution implicating increased ROS in RCC LMs. ROS have well-known associations with colon cancer, shown to stimulate tumor growth, proliferation, survival, and chemoresistance in cancer through multiple mechanisms [32].

GDCA is notable not only due to its ability to generate ROS, but also because it is one of several bile acids in the analysis found to collectively favor RCC, with a trend towards statistical significance. Among the roles played by bile acids in colon cancer proliferation are PKC and NF-kB signaling, pathways implicated in colon cancer proliferation, survival, metastasis, angiogenesis and drug resistance [31, 33, 34]. This is consistent with recent findings [35] from our own group demonstrating increased bile acids among primary RCCs; our data intriguingly suggests that this feature of RCCs may persist despite metastasis and thus be a core aspect of their cellular biology.

Conversely, carnitines were collectively more abundant in LCC LMs. Carnitines are specific intermediates of FAO, the process by which cells utilize fatty acids for energy production via oxidative phosphorylation. The fact that tetradecanedioic acid, a fatty acid, was also found to be increased in LCC over RCC further serves to bolster this notion. Additionally the gene, CPT1A, a key regulator of FAO, was found to be significantly upregulated in LCC (log-fc: −0.47, Cohen’s D = −0.86, P=0.03), though this did not reach adjusted significance. Though the role of FAO in colon cancer has yet to be fully clarified [12, 24], one intriguing finding is that FAO is known to decrease ROS formation via generation of NADPH.[36] It therefore is consistent that RCC would have higher levels of ROS than LCC, given its relative aversion toward FAO.

4.2. Transcriptomics

There were several differentially expressed genes of interest identified between RCC and LCC LMs. CERCAM, coding for a cell-adhesion molecule of the same name, was overexpressed in RCC LMs. Previous studies have revealed its association with invasion, metastasis, chemoresistance, and PI3K signaling in a colon cancer metastasis model [37, 38]. Interestingly, SMARCD3, another differentially expressed gene coding for an actin-associated protein involved in chromatin remodeling, was implicated alongside CERCAM in the same study as part of the handful of genes prognostic of worse survival in colon cancer. The authors hypothesize based on gene set enrichment analysis and protein-protein interaction networks that SMARCD3 potentiates epithelial mesenchymal transition (EMT) through a positive feedback loop with TGF-β, promoting downstream signaling.

Some of the most connected nodes in the top predicted network in the IPA included MAPK, ERK, AKT and PI3K, which are all upregulated. These are key members of both the PI3K-AKT and the MEK-ERK pathways, well-known drivers of colon cancer generally [25, 26, 39], and RCC LMs specifically [7]. As further discussed below, TGF-β, although not directly observed, is predicted be activated and play a central role in the upregulation of many of the elements in the network, including those of both pathways.

4.3. Integrated Analysis: EGFRi Resistance and TGF-β

Beyond spotlighting various aspects of tumor biology that may underly the increased pathogenicity of metastatic RCC, our study points toward more specific models of the resistance to EGFR inhibition (Figure 4). First, several of the findings provide evidence of stimulation of EGFR in RCC LM; these include the inhibition of PTP1B – a direct inhibitor of EGFR – by ROS [40–42], as well as the alteration of cell membrane characteristics by bile acids that promote EGFR dimerization [43]. However, perhaps a more intriguing possibility is that EGFR is bypassed by downstream signaling activation, thus rendering its inhibition inconsequential. One such downstream pathway is PI3K-AKT, which is activated by ROS via inhibition of PTEN and PTP1B [40–42]. Furthermore, recall that CERCAM, the top gene favoring RCC in the transcriptomic analysis, is known to drive PI3K-AKT signaling [37].

Another pathway predicted to be upregulated is MEK-ERK, the canonical downstream effector of EGFR [25, 26]. Of note, while pathologic KRAS mutation mediates this signaling, and is known to be more prevalent in RCC [44], there are many additional routes by which MEK-ERK may be stimulated; indeed, it is also independently stimulated by ROS [45–47]. As an aside, this comports well with clinical evidence showing that both the inferior prognosis and resistance to EGFRi inhibition in RCC persist in the absence of mutant KRAS [3, 44, 48–51].

However, perhaps the most intriguing candidate in EGFRi resistance in RCC LM is TGF-β; predicted to be among the most highly upregulated elements in RCC LMs by IPA upstream analysis, it stands out due to its unifying explanatory power for many of the findings discussed thus far. Results from immunohistochemistry showed that the expression of the TGF-β and CERCAM in the RCC liver metastasis tissue was higher than that in the LCC liver metastasis tissues, further supporting the hypothesis generated from analysis of the metabolomics and transcriptomics results. TGF-β is an activator of both MEK-ERK [52] and PI3K-AKT [53]. TGF-β has further been found to generate ROS, via several mechanisms including induction of mitochondrial production, and inhibition of antioxidant enzymes in multiple model systems [54, 55]. The interplay of cellular energy metabolism and TGF-β represents another area where our metabolomic findings comport. TGF-β has been shown in multiple cancer model systems, including colon cancer, to stimulate glycolysis and decrease FAO and oxidative phosphorylation [56, 57]. With regard to the transcriptomics, recall that there is evidence that TGF-β can partner with SMARCD3, another top gene upregulated in RCC, in a positive feedback loop to promote EMT. Finally, concordant with previous studies, enrichment for TGF-β signaling in metastatic RCC [7], and colon cancer in general, is well described [58, 59]. In metastatic disease it has been shown to play key roles in metastatic immune evasion, angiogenesis, and metastatic proliferation [58, 60, 61]. Therefore, increased TGF-β could serve to not only underpin the resistance of metastatic RCC to EGFRi, but also its inferior prognosis.

4.4. Challenges and Limitations

While a strength of the study is its use of primary human tissues, a subsequent notable limitation is that the sample sizes of the cohorts, particularly the transcriptomics cohort, were relatively small, which limited our ability to match study cohorts for several clinicopathologic features. For example, while data on KRAS status did not indicate a significant difference between the groups either cohort, it was incomplete and therefore not conclusive (though as mentioned above, clinical evidence suggests KRAS status is unlikely to be the primary driver of differences by laterality). MSI-H and BRAF mutation status did not occur in sufficient abundance in our cohort for analysis. In regard to survival, for the entire metabolomics cohort, the median overall survival of RCC LMs (22.8 mo, CI:13.3 mo-32.6 mo) was also significantly shorter compared to LCC LMs patients (38.9 mo, CI: 28.2 mo-49.5 mo) (p=0.0030). In the subsequent stage IV only analysis, this trend was redemonstrated however the difference did not reach statistical significance (p=0.213). The survival data for the transcriptomic cohort was not available, precluding a survival analysis. Additionally, the heterogeneity of chemotherapeutic protocols (both with regard to timing and agent) was too great to allow for stratification, but we acknowledge the disparities observed in the metabolites and transcriptome could be influenced by tumor cells’ differential response to 5-FU, and not directly related to primary differences in tumor biology. Thus, expanding the number of samples in the biobank to allow for an updated analysis controlling for these potential confounding variables will be an important priority. Finally, our study was necessarily correlational, not causal. This points to the most pressing future direction: designing and conducting suitable functional experiments based on the findings presented here. This will include developing both in vitro and in vivo models of RCC and LCC LMs, and independent modulation of the key genes, molecules, and signaling pathways implicated by this study.

4.5. Summary and Conclusions

This study sought to probe metabolomic and transcriptomic differences between RCC and LCC derived liver metastases to generate putative explanations for differences in clinical behavior across tumor laterality. The metabolomics findings point to increased ROS and bile acids, as well as downregulated FAO, in RCC LMs. The transcriptomic network analysis revealed a top molecular network centered around TGF-β, MEK-ERK, and PI3K-AKT signaling. TGF-β stood out in its ability to unify many of the findings into one mechanism. This makes it a leading candidate potentially driving the inferior survival and EGFRi resistance of RCC metastatic disease. While functional analyses confirming relevance and causality remain a necessary next step, this is the first attempt using a multi-omics approach to explain the differences in clinical behavior between RCC and LCC derived liver metastases. The findings represent promising leads in the paramount pursuit of more precise prognostication and targeted therapy in the realm of metastatic colorectal cancer.

Highlights.

• First integrated metabolomics characterization of right colon liver metastases

• Increased ROS, bile acids and TGF-B signaling in right colon liver metastases

• Potential mechanisms of EGFR-inhibitor resistance in right colon liver metastases

Glossary

RCC

right sided colon cancer

LCC

left sided colon cancer

LM

liver metastasis

EGFRi

epidermal growth factor inhibitors

LCMS

liquid chromatography mass spectrometry

mz

mass to charge ratio

rt

retention time

CC

colon cancer

HILIC

hydrophilic interaction liquid chromatography

RPLC

reverse phase liquid chromatography

NL

normal liver

Log-fc

log fold change

IPA

ingenuity pathway analysis

BH-FDR

Benjamini-Hochberg False Discover Rate

RL Index

Right-Left Index

GSH

Reduced Glutathione

GDCA

glycodeoxycholic acid

FAO

fatty acid oxidation

Cys-SO₃H

cysteine-s-sulfate

DHAA

dehydroascorbic acid

EMT

epithelial mesenchymal transition