Evaluation of a 92 multiplex protein panel in detection of colorectal cancer and high-risk adenoma in 784 symptomatic individuals

Abstract

BACKGROUND:

Blood-based protein biomarkers for detection of colorectal cancer (CRC) have been submitted to intense research to improve the full potential in screening for CRC.

OBJECTIVE:

The aim was to explore the diagnostic performance of 92 proteins related to inflammation and carcinogenesis in detection of CRC or precancerous lesions.

METHODS:

Blood-samples were collected from 4,698 individuals undergoing colonoscopy. An explorative unmatched case-control study was designed with 294 cases (individuals with CRC or high-risk colorectal adenoma) and 490 controls (individuals with low-risk colorectal adenoma, non-malignant findings or clean colorectum at colonoscopy). Protein profiling was performed by multiplex proximity extension assay. Statistical analyses were performed as univariate and multivariate logistic regression analyses.

RESULTS:

Univariably, CSF-1, MMP12 and IL8 demonstrated superior performance in discrimination of individuals with CRC. Recurrently, IL8 was included as contributor in majority of multivariate models discriminating individuals with CRC. The multivariate evaluation in discrimination of individuals with CRC demonstrated AUC ${}_{\text{ROC}}=$ 0.82, sensitivity $=$ 0.39 at specificity $=$ 0.80. Discrimination of individuals with late stage CRC from individuals with clean colorectum demonstrated AUC ${}_{\text{ROC}}=$ 0.90, sensitivity $=$ 0.58 at specificity $=$ 0.80.

CONCLUSIONS:

A subset of biomarker candidates, specifically IL8, investigated in the present study suggest a potential as blood-based biomarkers in screening of CRC.

Keywords

Protein biomarker IL8 colorectal neoplasia proximity extension assay early detection of cancer inflammation carcinogenesis

Abbreviations

ANGPT2	Angiopoietin 2
ARG1	Arginase 1
AUC ${}_{\text{ROC}}$	Area under the ROC curve
CAIX	Carbon anhydrase 9

CCL3	C-C motif chemokine 3
CCL20	Chemokine ligand 20
CD28	T-cell surface glycoprotein 28
CRC	Colorectal cancer
CSF-1	Colony stimulation factor 1
CXCL11	C-X-C motif chemokine
FIT	Fecal immunochemical test
FOBT	Fecal occult blood test
Gal-9	Galectin 9
HGF	Hepatocyte growth factor
HO-1	Heme oxygenase 1
HRA	High risk adenoma

ICOSLG	Inducible T-cell co-stimulator
	ligand
IL6	Interleukin 6
IL8	Interleukin 8
IL18	Interleukin 18
LRA	Low risk adenoma
MCP3	C-C motif chemokine 7
MMP12	Matrix metalloproteinase 12
mSEPT9	Test for methylated septin 9
NPX	Normal protein extression
PEA	Proximity extension assay
qPCR	Quantitative real-time PCR
TNF	Tumor necrosis factor
TNFRSF12A	TNF receptor superfamily
	member 12A
TWEAK	TNF ligand superfamily
	member 12

1. Introduction

Early detection of colorectal cancer (CRC) and precancerous bowel lesions (high-risk adenoma (HRA)) by screening has been shown to reduce mortality [37], which has led to the implementation of national screening programs in some countries, while others are still considering to initiate such programs [36]. Bowel examinations including colonoscopy and sigmoidoscopy are well-established modalities for diagnostic approaches. However, bowel endoscopy as a screening concept is challenged by disadvantages, such as bowel preparation, inconvenience for the individual, invasiveness and risk of severe complications [35], in addition to costs and limited healthcare budgets. Current screening programs for CRC are mostly based on non-invasive stool-based test for occult blood (fecal occult blood test (FOBT)) [36]; fecal immunochemical test (FIT) is the preferred choice due to superior performance, cost and convenience compared to other available FOBTs. However, the required self-sampling and handling of stool for testing may challenge the adherence to feces based screening [34]. In Denmark for instance, the compliance to FIT is 58% [28] and the applied FIT test (OC-Sensor, Eiken Pharmaceuticals, Japan) has a sensitivity 76% at 94% specificity [22] and thereby population screening does only identify 44% of the individuals with a CRC (compliance 58% X sensitivity 76% $=$ 44%). Therefore, the need for a minimal invasive, reliable, highly sensitive and specific biomarker test is critical to improve the full potential of population-based screening for CRC.

Consequently, major research efforts have been put into the development of blood-based, cancer-associated biomarkers or panels of combined biomarkers for the detection of colorectal neoplasia. Recent promising results suggest a variety of blood-based biomarkers for early detection of colorectal neoplasia including proteins [41, 6, 2], circulating tumor DNA methylations/ mutations/fragmentations [31, 14, 8], microRNA [33], nucleosomes and histone modifications [32] and meta- bolomes [4]. However, a blood-based biomarker test as an alternative to FIT is still anticipated. Presently, the test for methylated Septin 9 (mSEPT9) (Epi proColon, Epigenomics Inc.) remains the only FDA approved blood-based biomarker test in CRC screening, but reported diagnostic performance is not competitive to FIT [7, 16]. The combination of biomarkers and even combination of different modalities in screening test could potentially provide the required performance.

The development of CRC occurs predominantly from accumulated genetic and epigenetic DNA alterations of the epithelial cells in the colorectum, which eventually undergo malignant transformation into adenocarcinoma [24]. The solid tumor-tissue of CRC is inhomogeneous and consists of a variety of different cells including malignant, immune and inflammatory cells [38, 25]. Solitary or in combination, these cells contribute to the surrounding inflammatory and immune response, growth of tumor cells, metastasis and angiogenesis by cell-to-cell contact and/or the production of inflammatory mediators such as cytokines. Consequently, the expression and secretion of various proteins from tumor cells and the surroundings are altered and possibly released into the blood stream with a potential influence on protein levels in the circulation.

The aim of the present study was to explore the diagnostic performance of 92 proteins related to inflammation and carcinogenesis in detection of early CRC or precancerous lesions (HRA) in a population of individuals with symptoms attributable to CRC. In addition, the diagnostic performance of a combined biomarker panel consisting of the best performing biomarkers was identified and subsequently evaluated.

2. Methods

2.1 Study design and population

The present study was conducted in the context of the Endoscopy II protocol as previously described [41, 32] and was approved by the Ethics Committee of the Capital Region of Denmark and the Danish Data Protection Agency and performed according to the Helsinki II Declaration. The Endoscopy II protocol has founded a biobank comprising blood samples from individuals undergoing colonoscopy due to symptoms attributable to CRC from May 1, 2010 to November 30, 2012 at seven hospitals in Denmark (Aarhus, Bispebjerg, Herning, Hillerød, Horsens, Hvidovre and Randers). In addition to symptoms, the inclusion criteria included age $\geqslant$ 18 years and signed informed consent. The exclusion criteria were member of a hereditary nonpolyposis CRC or familiar adenomatosis polyposis family, surgical intervention within 3 months prior to inclusion as well as previous history of large bowel endoscopy, colorectal adenoma or CRC. The Endoscopy II biobank consists of blood samples from 4,698 individuals including 512 individuals with CRC and 689 individuals with colorectal adenomas counting 399 individuals with HRA. CRC and HRA were classified according to findings at colonoscopy and pathological diagnosis, and CRC were stage classified according to UICC stage-classification (stage I–IV).

The proposed design of the present study was an explorative unmatched case-control study with cases defined as individuals with either CRC or HRA and controls defined as individuals with low-risk adenoma (LRA), non-malignant findings or no findings (clean colorectum) at colonoscopy. Cases and controls were randomly selected from the Endoscopy II population including 196 individuals with CRC (49 individuals with stage I CRC, 49 individuals with stage II CRC, 49 individuals with stage III CRC and 49 individuals with stage IV CRC), 98 individuals with HRA, 98 individuals with LRA, 196 individuals with non-malignant findings and 196 individuals with clean colorectum at colonoscopy. The total number of included individuals was 784. Additional data regarding age, gender, localization of CRC or colorectal adenoma (colon or rectum) were recorded along with the presence of comorbidity (diabetes I/II, cardiovascular, lung or rheumatic disease or previous history of malignant disease). Figure 1 provides a flowchart which demonstrate descriptive data and random selection of the study participants.

Figure 1.

Flowchart illustration the inclusion and descriptive data of the present study. CRC: colorectal cancer, CC: colon cancer, RC: rectal cancer, HRA: high-risk adenoma, LRA: low-risk adenoma.

Three primary Endpoints and five secondary Endpoints were defined based on outcome at colonoscopy:

Primary Endpoints: Endpoint 1: The discrimination of individuals with CRC or HRA from all other individuals. Endpoint 2: The discrimination of individuals with CRC from all other individuals. Endpoint 3: The discrimination of individuals with HRA from all other individuals but excluding individuals with CRC. Secondary Endpoints: Endpoint 4: The discrimination of individuals with CRC stage I or II from all other individuals but excluding individuals with CRC stage III or IV. Endpoint 5: The discrimination of individuals with CRC stage III or IV from all other individuals but excluding individuals with CRC stage I or II. Endpoint 6: The discrimination of individuals with CRC stage I or II from individuals with clean colorectum. Endpoint 7: The discrimination of individuals with CRC stage III or IV from individuals with clean colorectum. Endpoint 8: The discrimination of individuals with HRA from individuals with clean colorectum.

2.2 Sample collection, handling and storage

The blood samples for EDTA plasma were collected by dedicated research nurses on the day of planned colonoscopy prior to the procedure and after bowel preparation. The blood samples were handled according to a predefined, validated standard operating procedure including centrifugation at 3,000 g for 10 minutes at 21 ${}^{\circ}$ C followed by isolation and transfer of plasma, serum and buffy coats to separate freezing tubes. The entire procedure was performed within two hours from the collection of the samples. All samples were stored at $-$ 80 ${}^{\circ}$ C under electronic 24/7 surveillance until analysis.

2.3 Laboratory procedures

The protein profiling was performed by using the Olink AB (Stockholm, Sweden) custom multiplex proximity extension assay (PEA) which allows simultaneous quantification of 92 human immunological and tumor-associated proteins (IMMUNO-ONCOLOGY panel). A full list of the 92 protein biomarkers is specified in the Supplementary (Table S1). The IMMUNO-ONCOLOGY panel was chosen as it reflects various cellular mechanisms in carcinogenesis and inflammation associated with early CRC or its precancerous lesions. The laboratory analyses were performed according to manufactures instructions at BioXpedia A/S (Aarhus, Denmark). In brief, 92 oligonucleotide-labeled antibodies in pairs were allowed binding to their respective target protein in the plasma sample thereby bringing the paired oligonucleotides into proximity to provide proximity-dependent DNA polymerization. The subsequent detection and quantification were performed by quantitative real-time PCR (qPCR).

Standardization and normalization of raw data were performed according to manufactures instructions [29]. For each sample analyzed, internal controls (incubation, extension and detection controls) were added. In addition, external controls (inter-plate controls, negative controls and sample controls) were added to each 96-well plate. For each protein-assay, controls were used to calculate limit of detection and to normalize raw data as Normal Protein Expression (NPX); an arbitrary, relative quantification on the log2 scale, where a higher NPX value represents a higher level of the target protein in plasma with a background level at approximately zero.

2.4 Statistics

Data management and statistical analyses were done using SAS (v9.4, Cary, N.C., USA), and further statistical analyses were performed with R version 3.6.3 (2020-02-29) using the package “rms” (Frank E Harrell Jr (2019). rms: Regression Modeling Strategies. R package version 5.1-4). The significance level was 5%.

The primary as well as secondary Endpoints are described above. The explanatory variables were age, gender and the 92 markers (expressed as NPX). The statistical analysis was done using logistic regression analysis modelling the probability of the respective endpoints. All NPX values were log transformed base 2, with results presented by the p-values. In addition, sensitivities for 70, 80 and 90% specificities were presented for multivariate models. The area under the receiver operating characteristic curves (AUC ${}_{\text{ROC}}$ ) were shown as a measure of discrimination.

Initial analysis for the selection of relevant biomarkers was done using univariate logistic regression for each of the 92 biomarkers, choosing those biomarkers with $p$ -values $<$ 5% as candidates for multivariate analysis. Multivariate analysis was then done including the selected biomarkers as well as age and gender. For each endpoint, a final model was found reducing the multivariate model removing those biomarkers not significant in a 10-fold cross validated model at the 5% level [11].

Comorbidity was assessed by including each comorbidity in the multivariate model with an interaction between the biomarkers and the comorbidity.

Table 1
Results of the multivariate analyses for Endpoint 1–8 are presented with regulation ( $\uparrow$ / $\downarrow$ ), $p$ -value and AUC ${}_{\text{ROC}}$ with sensitivities at specificities 70%, 80% and 90%

	Protein	Regulation ( $\uparrow$ / $\downarrow$ )	AUC	$P$	Sensitivity at 70% specificity	Sensitivity at 80% specificity	Sensitivity at 90% specificity
Endpoint 1	ICOSLG	$\downarrow$	0.73	$<$ 0.001	0.54	0.36	0.18
	IL8	$\uparrow$		$<$ 0.001
Endpoint 2	ANGPT2	$\downarrow$	0.82	$<$ 0.001	0.58	0.39	0.18
	ARG1	$\downarrow$		0.003
	CSF-1	$\uparrow$		$<$ 0.001
	Gal-9	$\downarrow$		0.001
	ICOSLG	$\downarrow$		$<$ 0.001
	IL8	$\uparrow$		$<$ 0.001
Endpoint 3	CD28	$\uparrow$	0.61	0.008	0.43	0.31	0.13
Endpoint 4	IL8	$\uparrow$	0.73	$<$ 0.001	0.58	0.40	0.18
Endpoint 5	ARG1	$\downarrow$	0.83	$<$ 0.001	0.53	0.36	0.19
	CAIX	$\uparrow$		$<$ 0.001
	CD27	$\downarrow$		0.002
	CSF-1	$\uparrow$		$<$ 0.001
	CXCL11	$\uparrow$		0.003
	ICOSLG	$\downarrow$		$<$ 0.001
Endpoint 6	CSF-1	$\uparrow$	0.79	$<$ 0.001	0.71	0.61	0.30
Endpoint 7	ARG1	$\downarrow$	0.90	0.004	0.76	0.58	0.32
	ICOSLG	$\downarrow$		$<$ 0.001
	IL8	$\uparrow$		$<$ 0.001
Endpoint 8	IL18	$\uparrow$	0.71	0.004	0.57	0.35	0.21
	TWEAK	$\downarrow$		0.003

3. Results

3.1 Characteristics of the population

The study design is shown in Fig. 1 including descriptive statistics.

3.2 Individuals protein assessment

Results of the univariate analyses of all 92 proteins for all 8 Endpoints are presented by the $p$ -value and AUC ${}_{\text{ROC}}$ and showing if the marker is up or down regulated (U/D) in the Supplementary Table 1S.

In the overall identification of individuals with CRC (Endpoint 1, 2 and 4–7), three proteins (colony stimulating factor 1 (CSF-1), matrix metalloproteinase 12 (MMP12) and interleukin 8 (IL8)) consistently demonstrated significant superior performance in the univariate evaluation. Restricting the discrimination to individuals with late stage CRC (stage III–IV) demonstrated improved performance, and further improvements were shown by restricting the control group to individuals with clean colorectum. In details, 50 of the 92 proteins demonstrated significance in discrimination of individuals with CRC or HRA (Endpoint 1) with superior performance by CSF-1, MMP12 and IL8 demonstrating AUC ${}_{\text{ROC}}$ of 0.66, 0.67 and 0.69, respectively. Similar results were obtained in discrimination of individuals with CRC (Endpoint 2) with 50 proteins demonstrating significance at the 5% and CSF-1, MMP12 and IL8 demonstrating the highest AUC ${}_{\text{ROC}}$ (0.70, 0.73 and 0.76, respectively). Twenty-five proteins significantly discriminated individuals with CRC stage I and II (Endpoint 4) in the univariate analyses. As for Endpoint 1 and 2, CSF-1, MMP12 and IL8 demonstrated the highest AUC ${}_{\text{ROC}}$ (0.66, 0.65 and 0.69, respectively). Significant differences were shown for 55 proteins in discrimination of individuals with CRC stage III and IV (Endpoint 5). In addition to CSF-1, MMP12 and IL8 (AUC ${}_{\text{ROC}}$ of 0.73, 0.81 and 0.83, respectively), superior univariate performance was demonstrated for C-C motif chemokine 7 (MCP-3) (AUC ${}_{\text{ROC}}$ of 0.73). Discriminating individuals with CRC stage I and II from individuals with clean colon (Endpoint 6), 38 proteins were significantly different. As for the previous endpoints discriminating individuals with CRC, CSF-1, MMP12 and IL8 demonstrated the highest AUC ${}_{\text{ROC}}$ of 0.70, 0.69 and 0.71, respectively. Restricting the analysis to identification of individuals with CRC stage III and IV from individuals with clean colon (Endpoint 7), 53 proteins demonstrated significance. In similarity with the previous univariable results for discrimination of individuals with CRC, CSF-1, MMP12 and IL8 demonstrated superior performance with AUC ${}_{\text{ROC}}$ of 0.77, 0.84 and 0.84, respectively. In addition, C-C motif chemokine 3 (CCL3), interleukin 6 (IL6) and MCP-3 all demonstrated AUC ${}_{\text{ROC}}$ of 0.77.

The overall identification of individuals with HRA (Endpoint 3 and 8) demonstrated superior discrimination by two proteins (hepatocyte growth factor (HGF) and MCP-3). In specific, nine proteins demonstrated significance in discrimination of individuals with HRA from all other individuals excluding individuals with CRC (Endpoint 3). MCP-3, Chemokine ligand 20 (CCL20), heme oxygenase 1 (HO-1) and TNF receptor superfamily member 12A (TNFRSF12A) all demonstrated AUC ${}_{\text{ROC}}$ of 0.58 and HGF demonstrated the highest AUC ${}_{\text{ROC}}$ of 0.60. Identification of individuals with HRA from individuals with clean colon (Endpoint 8) demonstrated significance for 26 proteins. In addition to MCP-3 and HGF (AUC ${}_{\text{ROC}}$ of 0.63 and 0.64, respectively), levels of IL6 showed superior discrimination with AUC ${}_{\text{ROC}}$ of 0.63.

3.3 Assessment of the multivariate models

Results of the cross validated multivariate analyses for Endpoint 1–8 are presented with p-values and AUC ${}_{\text{ROC}}$ with sensitivities at specificities 0.70, 0.80 and 0.90 in Table 1 and indicating whether the markers were up or down regulated.

An overall improved discrimination of individuals with CRC (Endpoint 1,2 and 4–7) was observed in the multivariate evaluation compared to univariate evaluation. A subset of proteins from the 92 protein-panel was included in the multivariate models with IL8 recurrently identified as contributor. In specific, the discrimination of individuals with CRC or HRA (Endpoint 1) included Inducible T-cell co-stimulator ligand (ICOSLG) and IL8 in the model and demonstrated AUC ${}_{\text{ROC}}$ of 0.73 (sensitivity 0.36 at specificity 0.80). The multivariate model for individuals with CRC (Endpoint 2) included 6 proteins (angiopoietin 2 (ANGPT2), arginase 1 (ARG1), CSF-1, galectin-9 (Gal-9), ICOSLG, and IL8) with AUC ${}_{\text{ROC}}$ of 0.82 with sensitivity 0.39 at specificity 0.80. Exclusively, IL8 was included in the multivariate model in discrimination of individuals with early stage CRC (Endpoint 4); AUC ${}_{\text{ROC}}$ of 0.73 with sensitivity 0.40 at specificity 0.80. For individuals with late stage CRC (Endpoint 5), 6 proteins were included in the multivariate model (ARG1, carbonic anhydrase 9 (CAIX), CD27, CSF-1, C-X-C motif chemokine (CXCL11) and ICOSLG); AUC ${}_{\text{ROC}}$ of 0.83, sensitivity 0.36 at specificity 0.80. Restricting the control group to individuals with clean colon, the discrimination of individuals with early stage CRC (Endpoint 6) exclusively included CSF-1 in the multivariate model (AUC ${}_{\text{ROC}}$ of 0.79, sensitivity 0.61 at specificity 0.80), and the discrimination of individuals with late stage CRC included 3 proteins (ARG1, ICOSLG and IL8) in the multivariate model (AUC ${}_{\text{ROC}}$ of 0.90, sensitivity 0.58 at specificity 0.80).

The overall discrimination in the multivariate evaluation of individuals with HRA (Endpoint 3 and 8) was inferior compared to the discrimination of individuals with CRC. A single protein (T-cell surface glycoprotein 28 (CD28)) was included in the model in discrimination of individuals with HRA (Endpoint 3); AUC ${}_{\text{ROC}}$ of 0.61 with sensitivity 0.31 at specificity 0.80. Restricting the control group to individuals with clean colorectum, the discrimination of individuals with HRA (Endpoint 8) included 2 proteins in the multivariate model (interleukin 18 (IL18) and TNF ligand superfamily member 12 (TWEAK)); AUC ${}_{\text{ROC}}$ of 0.71 with sensitivity 0.35 at specificity 0.80.

3.4 Assessment of influence of co-morbidity

The potential influence of co-morbidity (diabetes I/II, cardiovascular, lung or rheumatic disease or previous history of malignant disease) in the multivariate analyses of Endpoints 1–8 was evaluated, and significance was not demonstrated for the interaction between levels of proteins and the presence of co-morbidity for any of the eight Endpoints as demonstrated in Table 2.

Table 2
The influence of comorbidity in the multivariate analyses of Endpoint 1–8 presented with $p$ -value for the comorbidity and the interaction between comorbidity and the protein biomarkers included in the multivariate model

	N	Endpoint
		1	2	3	4	5	6	7	8
Cardiovascular disease	89	0.47	0.08	0.30	1.00	0.87	0.15	0.46	0.69
X ANGPT2		–	0.89	–	–	–	–	–	–
X ARG1		–	0.26	–	–	0.20	–	0.69	–
X CAIX		–	–	–	–	0.06	–	–	–
X CD27		–	–	–	–	0.08	–	–	–
X CD28		–	–	0.65	–	–	–	–	–
X CSF-1		–	0.48	–	–	0.74	0.15	–	–
X CXCL11		–	–	–	–	0.56	–	–	–
X Gal-9		–	0.17	–	–	–	–	–	–
X ICOSLG		0.44	0.94	–	–	0.25	–	0.33	–
X IL8		0.80	0.17	–	0.98	–	–	0.29	–
X IL18		–	–	–	–	–	–	–	0.78
X TWEAK		–	–	–	–	–	–	–	0.89
Diabetes	90	0.88	0.63	0.73	0.18	0.59	0.61	0.89	0.60
X ANGPT2		–	0.78	–	–	–	–	–	–
X ARG1		–	0.30	–	–	0.30	–	0.80	–
X CAIX		–	–	–	–	0.27	–	–	–
X CD27		–	–	–	–	0.13	–	–	–
X CD28		–	–	0.02	–	–	–	–	–
X CSF-1		–	0.92	–	–	0.24	0.61	–	–
X CXCL11		–	–	–	–	0.92	–	–	–
X Gal-9		–	0.95	–	–	–	–	–	–
X ICOSLG		0.89	0.28	–	–	0.96	–	0.87	–
X IL8		0.50	0.49	–	0.16	–	–	0.41	–
X IL18		–	–	–	–	–	–	–	0.54
X TWEAK		–	–	–	–	–	–	–	0.96
Lung disease	86	0.62	0.29	0.31	0.39	0.61	0.43	0.52	0.38
X ANGPT2		–	0.18	–	–	–	–	–	–
X ARG1		–	0.82	–	–	0.19	–	0.50	–
X CAIX		–	–	–	–	0.80	–	–	–
X CD27		–	–	–	–	0.90	–	–	–
X CD28		–	–	0.53	–	–	–	–	–
X CSF-1		–	0.20	–	–	0.67	0.43	–	–
X CXCL11		–	–	–	–	0.61	–	–	–
X Gal-9		–	0.50	–	–	–	–	–	–
X ICOSLG		0.52	0.59	–	–	0.96	–	0.44	–
X IL8		0.94	0.21	–	0.38	–	–	0.97	–
X IL18		–	–	–	–	–	–	–	0.66
X TWEAK		–	–	–	–	–	–	–	0.24
Rheumatic disease	11	0.60	1.00	0.52	0.99	1.00	0.98	1.00	0.21
X ANGPT2		–	0.99	–	–	–	–	–	–
X ARG1		–	0.99	–	–	0.99	–	0.99	–
X CAIX		–	–	–	–	0.99	–	–	–
X CD27		–	–	–	–	1.00	–	–	–
X CD28		–	–	0.23	–	–	–	–	–
X CSF-1		–	0.99	–	–	0.99	0.98	–	–
X CXCL11		–	–	–	–	0.99	–	–	–
X Gal-9		–	1.00	–	–	–	–	–	–
X ICOSLG		0.51	1.00	–	–	1.00	–	1.00	–
X IL8		0.53	1.00	–	1.00	–	–	0.98	–
X IL18		–	–	–	–	–	–	–	0.78
X TWEAK		–	–	–	–	–	–	–	0.36
Previous malignant disease	25	0.29	0.38	0.77	0.33	0.80	0.25	0.48	0.99
X ANGPT2		–	0.63	–	–	–	–	–	–
X ARG1		–	0.27	–	–	0.35	–	0.48	–

Table 2, continued
	N	Endpoint
		1	2	3	4	5	6	7	8
X CAIX		–	–	–	–	0.52	–	–	–
X CD27		–	–	–	–	0.92	–	–	–
X CD28		–	–	0.52	–	–	–	–	–
X CSF-1		–	0.94	–	–	0.73	0.29	–	–
X CXCL11		–	–	–	–	0.45	–	–	–
X Gal-9		–	0.12	–	–	–	–	–	–
X ICOSLG		0.29	0.84	–	–	0.68	–	0.50	–
X IL8		0.96	0.28	–	0.43	–	–	0.48	–
X IL18		–	–	–	–	–	–	–	0.83
X TWEAK		–	–	–	–	–	–	–	0.36

4. Discussion

The aim of the present study was to explore and evaluate the diagnostic potential of 92 proteins associated with inflammation and carcinogenesis, individually or in combination, in detection of individuals with CRC or HRA.

Evaluated individually, majority of the proteins demonstrated significance in discrimination of individuals with CRC (Endpoint 1, 2, 4–7). Three proteins (CSF-1, MMP-12 and IL8) demonstrated consistent superior performance with IL8 achieving the highest performance in discrimination of individuals with CRC. Superior results for CSF-1, MMP-12 and IL8 were obtained when the analysis was restricted to discriminate individuals with late stage CRC (Endpoint 5 and 7 with AUC ${}_{\text{ROC}}$ of 0.73–0.83 and 0.77–0.84, respectively).

In the combined multivariate evaluation in prediction of individuals with CRC (Endpoint 1, 2, 4–7), improved diagnostic performance was observed. Noteworthy, IL8 was recurrently included as contributor in majority of the multivariate models. The overall superior performance was demonstrated in identification of individuals with late stage CRC from individuals with clean colorectum (Endpoint 7) with inclusion of ARG1, ICOSLG and IL8 in the multivariate model (AUC ${}_{\text{ROC}}=$ 0.90, sensitivity $=$ 0.58 at specificity $=$ 0.80).

The results of the present study indicate a potential of CSF-1, MMP12 and, specifically, IL8 as biomarker candidates in CRC screening due to the persistent selection as superior discriminators of CRC in the univariate evaluation and the recurrent selection of IL8 as contributor in the multivariate evaluation. In the literature, these three proteins are involved in diverse biological processes and classified as modulators of the inflammatory and carcinogenic response, and changes in levels in the circulation as well as changes in expression in tumor tissue have been associated with CRC. In specific, CSF-1 is a cytokine primarily involved in the migration, activation and proliferation of macrophages in inflammation and carcinogenesis [13, 18, 10]. In relations to CRC, significantly altered expression in tumor tissue have been described [39]. MMP12 is an extracellular enzyme involved in breakdown and re-modeling of the extracellular matrix and are associated with cell proliferation, differentiation, inflammation, tissue invasion as well as angiogenesis [9]. In CRC, upregulation of MMP12 in tumor-tissue of CRC have been described as well as an indication of increased levels in the circulation of individuals with CRC [20, 43]. IL8 is a chemotactic cytokine inducing migration and activation of neutrophils as well as being associated with tumor growth, angiogenesis and metastasis [40, 1, 27, 5, 17, 19]. Upregulation in CRC tissue as well as in the circulation of individuals with CRC have been described [17, 26], and a potential as prognostic marker for CRC has been previously suggested [42].

An essential requirement to a screening test for CRC is the identification of individuals with early stage CRC prior to dissemination and with a potential of surgical resection. In the present study, discrimination of individuals with early stage CRC from individuals with clean colon (Endpoint 6) demonstrated an AUC ${}_{\text{ROC}}$ of 0.79 with sensitivity of 0.61 and 0.30 at specificity 0.80 and 0.90, respectively. Reported performances of FIT and mSEPT9 as approved and validated screening tests for CRC in an asymptomatic screening population specifies sensitivity of 0.53–0.85 and 0.48–0.75 at specificity 0.90–0.96 and 0.87–0.92, respectively [22, 7, 16, 30, 3]. Precautions must be taken in comparison between the results of the present study and the specified performances of FIT and mSEPT9 due to the design of the present study as an explorative unmatched case-control study. Furthermore, the study cohort of the present study have symptoms of CRC which could indicate underlying, undiagnosed disease or even cancer with locations outside the colorectum. Since the investigated proteins are associated inflammation or carcinogenesis, this may cause increased base-line level in comparison to levels in an asymptomatic population, particularly in individuals of the control group. Consequently, this may lead to an underestimation of the performance of the present investigated proteins. Overall, the diagnostic performances found in the present study are promising and may even be competitive to FIT or MSEPT9 in a future validation study.

In the individual as well as combined evaluation, improved diagnostic performance was observed when discrimination was restricted to individuals with late stage CRC (Endpoint 5 and 7). Progression of CRC with increased tumor burden, increased inflammatory response in the surrounding tissue as well as dissemination may have influence on levels of proteins associated with carcinogenesis and inflammation which provides an explanation for the observed superior discrimination of individuals with late stage CRC compared to early stage CRC.

The discrimination of individuals with HRA (Endpoint 3 and 8) was inferior to that of individuals with CRC. In the individual evaluation, a smaller subset of the investigated proteins reached significance with inferior performance. HGF and MCP-3 demonstrated superior performance for Endpoint 3 and 8; AUC ${}_{\text{ROC}}$ of 0.60–0.64 and 0.58–0.63, respectively. HGF is a cytokine associated with cell growth, regeneration and homeostasis of epithelial tissues [12, 15]. MCP-3 is a chemotactic cytokine associated with recruitment, activation and regulation of leukocytes including macrophages during inflammation and metastasis [21, 23]. None of the two proteins were included in the combined multivariate models for individuals with HRA. In the combined evaluation, discrimination of individuals with HRA from all other individuals (Endpoint 3) demonstrated similar performance compared to the individual evaluation; AUC ${}_{\text{ROC}}$ $=$ 0.61, sensitivity $=$ 0.31 at specificity $=$ 0.80. By restricting the control group to individuals with clean colorectum (Endpoint 8), the discrimination improved; AUC ${}_{\text{ROC}}$ $=$ 0.71, sensitivity $=$ 0.35 at specificity $=$ 0.80.

Superior discrimination of individuals with CRC or HRA was obtained when the control group was restricted to individuals without findings at colonoscopy (clean colorectum). The study design includes a potential risk of underestimation of performance as the study cohort include individuals undergoing colonoscopy due to symptoms attributable to CRC which, consequently, lead to a symptomatic control group. The purpose of restricting the control group to individuals with clean colorectum was to indicate a potential true diagnostic performance. However, the design of the present study as an explorative unmatched case-control study specifies limitations for the interpretation of results and conclusions. A trend and a possible potential of the investigated proteins as biomarkers in identification of individuals with CRC or HRA can be insinuated, but the results are not referable to the study population of Endoscopy II or to the Danish population in general due to a selected and unrepresentative cohort. To refer the results to the background population, the results and conclusion of the present study must be validated in a non-selected, representative and asymptomatic CRC-screening population.

The analysis technique of PEA applied in the present study enables simultaneous quantification of 92 proteins in a single sample with high reproducibility and modest requirements to sample size (1 $\mu$ L per sample). The same technique employing PEA have been applied in comparable studies where alternative panels of blood-based proteins including a few of the proteins of the present investigation were evaluated [6, 2]. The alternative protein panels evaluated by Chen and colleagues [6] and Bhardwaj and colleagues [2] have shown similar promising results regarding identification of individuals with CRC. However, majority of the proteins evaluated in the present study have not previously been evaluated by PEA in individuals with CRC.

The various protein-candidates for biomarkers in CRC identified by PEA in present and previous studies could possibly reflect the diversity of biological pathways and mechanisms involved in the carcinogenic and inflammatory response affecting levels of various proteins and mediators. Furthermore, the proteins of interest are not specific in development and progression of CRC but are associated to inflammation and carcinogenesis in general.

The promising potential of IL8 demonstrated by PEA technique in the present study is recognized in previous research by different techniques such as biochip array technology [5], enzyme-linked immunosorbent assay [17] and multiplex immunoassay technology [19]. Diversity in study designs hinders comparison of the different techniques. To determine the optimal technique for IL8 measurement, future studies comparing the different techniques are needed.

The potential of specific proteins such as IL8 as biomarker for CRC indicated in the present study contribute to the foundation of knowledge to form hypotheses for future validation studies and could lead to improvement of the existing screening possibilities. The individual performance of IL8 indicated by the present study is insufficient to qualify as a biomarker for CRC. Nevertheless, possible combination with other promising or validated blood-based biomarkers could improve the diagnostic potential and provide a screening concept superior or competitive to FIT. It could even provide the opportunity to choose between modalities of screening (blood or fecal sampling) and possibly improve the adherence to a screening program by offering an alternative to fecal-based testing leading to improved efficiency and cost/benefit of existing screening programs.

The advantage of the applied technique of PEA is the simultaneous determination of 92 proteins in a single sample providing the opportunity to perform large scale studies in the search for blood-based protein biomarkers in various cancer types. The design of the protein panel of the present study was predefined as a commercial panel by the manufacturer (Olink AB, Stockholm, Sweden). The panels offered by Olink AB are designed in consideration of various biological processes (panels including proteins associated with cell-regulation, metabolism, organ damage, neurological disease and cardiovascular disease in addition to the IMMUNO-ONCOLOGY panel included in the present study). The future perspective of PEA in the context of cancer screening test could be the design of a protein panel composed of subunits of validated organ-specific cancer-associated protein biomarkers. The panel could identify “at risk” individuals with increased incidence of the various cancer types, and the organ-specific subunit protein profile could then be used as guidance for the choice of further diagnostic tests or examinations.

5. Conclusion

In summary, a subset of the proteins associated with inflammation and carcinogenesis from the panel of 92 proteins investigated in the present study with a systematic, prospectively collected cohort suggest a potential as blood-based biomarkers in screening of CRC. Of specific interest is CSF-1, MMP12 and particularly IL8 with superior performance in discrimination of individuals with CRC. The future perspectives could include the combination of IL8 with other biomarkers or screening modalities for CRC (established as well as experimental); a combination which could lead to further improvement in the screening of CRC.

Author contributions

Conception: LR, HJN, IJC.

Interpretation or analysis of data: LR, IJC.

Preparation of the manuscript: LR.

Revision for important intellectual content: LR, HJN, IJC.

Supervision: HJN, IJC.

Supplementary data

The supplementary files are available to download from http://dx.doi.org/10.3233/CBM-203211.

sj-pdf-1-cbm-10.3233_CBM-203211.pdf - Supplemental material

Supplemental material, sj-pdf-1-cbm-10.3233_CBM-203211.pdf

Footnotes

Acknowledgments

The research nurses, secretaries and technicians at the participating hospital departments and laboratories are thanked for their skillful work.

The study received financial support from The Andersen Isted Fund, The Augustinus Foundation, The Beckett Fund, The Inger Bonnén Fund, The Hans & Nora Buchard Fund, CEO Jens Bærentsen (private donation), The Walter Christensen Family Fund, The P.M. Christiansen Family Fund, The Aase & Ejnar Danielsen Fund, The Erichsen Family Fund, The Knud & Edith Eriksen Fund, The Svend Espersen Fund, The Elna and Jørgen Fagerholt Fund, The Sofus Carl Emil Friis Fund, The Torben & Alice Frimodt Fund, The Eva & Henry Frænkel Fund, The Gangsted Fund, The Thora & Viggo Grove Fund, The H Foundation, The Erna Hamilton Fund, The Sven & Ina Hansen Fund, The Søren & Helene Hempel Fund, The Henrik Henriksen Fund, The Jørgen Holm Family Fund, Foundation Jochum, The KID Fund, The Kornerup Fund, The Linex Fund, The Dagmar Marshall Fund, The “Midtjyske Bladfund”, The Axel Muusfeldt Fund, The Børge Nielsen Family Fund, The Michael Hermann Nielsen Fund, The Arvid Nilsson Fund, The Obel Family Fund, The Krista & Viggo Petersen Fund, The Willy & Ingeborg Reinhard Fund, The Kathrine & Vigo Skovgaard Fund, The Toyota Fund, The Vissing Fund, The Wedell-Wedellsborg Fund and Hvidovre University Hospital (The Capital Region of Denmark).

References

Alfaro

et al., Interleukin-8 in cancer pathogenesis, treatment and follow-up, Cancer Treat Rev 60 (2017), 24–31. doi: 10.1016/j.ctrv.2017.08.004.

Bhardwaj

et al., Multiplex screening of 275 plasma protein biomarkers to identify a signature for early detection of colorectal cancer, Mol Oncol 14 (2020), 8–21. doi: 10.1002/1878-0261.12591.

Brenner

and Tao

, Superior diagnostic performance of faecal immunochemical tests for haemoglobin in a head-to-head comparison with guaiac based faecal occult blood test among 2235 participants of screening colonoscopy, Eur J Cancer 49 (2013), 3049–3054. doi: 10.1016/j.ejca.2013.04.023.

Bro

et al., Data fusion in metabolomic cancer diagnostics, Metabolomics 9 (2013), 3–8. doi: 10.1007/s11306-012-0446-0.

Bünger

et al., A novel multiplex-protein array for serum diagnostics of colon cancer: a case-control study, BMC Cancer 12 (2012), 393. doi: 10.1186/1471-2407-12-393.

Chen

et al., Head-to-head comparison and evaluation of 92 plasma protein biomarkers for early detection of colorectal cancer in a true screening setting, Clin Cancer Res 21 (2015), 3318–3326. doi: 10.1158/1078-0432.CCR-14-3051.

Church

T.R.

et al., Prospective evaluation of methylated SEPT9 in plasma for detection of asymptomatic colorectal cancer, Gut 63 (2014), 317–325. doi: 10.1136/gutjnl-2012-304149.

Cristiano

et al., Genome-wide cell-free DNA fragmentation in patients with cancer, Nature 570 (2019), 385–389. doi: 10.1038/s41586-019-1272-6.

Cui

and Khalil

R.A.

, Biochemical and biological attributes of matrix metalloproteinases, Prog Mol Biol Transl Sci 147 (2017), 1–73. doi: 10.1016/bs.pmbts.2017.02.005.

10.

Dwyer

A.R.

Greenland

E.L.

and Pixley

F.J.

, Promotion of tumor invasion by tumor-associated macrophages: the role of CSF-1-activated phosphatidylinositol 3 kinase and src family kinase motility signaling, Cancers (Basel) 9 (2017). doi: 10.3390/cancers9060068.

11.

Harrell

F.E.

Lee

K.L.

and Mark

D.B.

, Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors, Stat Med 15 (1996), 361–387. doi: 10.1002/(SICI)1097-0258(19960229)15:4<361::AID-SIM168>3.0.CO;2-4.

12.

Huang

et al., Hepatocyte growth factor is a prognostic marker in patients with colorectal cancer: a meta-analysis, Oncotarget 8 (2017), 23459–23469. doi: 10.18632/oncotarget.15589.

13.

Hume

D.A.

and MacDonald

K.P.A.

, Therapeutic applications of macrophage colony-stimulating factor-1 (CSF-1) and antagonists of CSF-1 receptor (CSF-1R) signaling, Blood 119 (2012), 1810–1820. doi: 10.1182/blood-2011-09-379214.

14.

Jensen

S.Ø.

et al., Novel DNA methylation biomarkers show high sensitivity and specificity for blood-based detection of colorectal cancer – a clinical biomarker discovery and validation study, Clin Epigenetics 11 (2019). doi: 10.1186/s13148-019-0757-3.

15.

Jiang

W.G.

Hallett

M.B.

and Puntis

M.C.A.

, Hepatocyte growth factor/scatter factor, liver regeneration and cancer metastasis, Br J Surg 80 (1993), 1368–1373. doi: 10.1002/bjs.1800801104.

16.

Jin

et al., Performance of a second-generation methylated SEPT9 test in detecting colorectal neoplasm, Journal of Gastroenterology and Hepatology 30 (2015), 830–833. doi: 10.1111/jgh.12855.

17.

Jin

W.J.

et al., Diagnostic value of interleukin-8 in colorectal cancer: a case-control study and meta-analysis, World J Gastroenterol 20 (2014), 16334–16342. doi: 10.3748/wjg.v20.i43.16334.

18.

Jones

C.V.

and Ricardo

S.D.

, Macrophages and CSF-1: implications for development and beyond, Organogenesis 9 (2013), 249–260. doi: 10.4161/org.25676.

19.

Kantola

et al., Stage-dependent alterations of the serum cytokine pattern in colorectal carcinoma, Br J Cancer 107 (2012), 1729–1736. doi: 10.1038/bjc.2012.456.

20.

Klupp

et al., Serum MMP7, MMP10 and MMP12 level as negative prognostic markers in colon cancer patients, BMC Cancer 16 (2016), 494. doi: 10.1186/s12885-016-2515-7.

21.

Kurzejamska

et al., Effect of chemokine (C-C Motif) ligand 7 (CCL7) and its receptor (CCR2) expression on colorectal cancer behaviors, Int J Mol Sci 20 (2019). doi: 10.3390/ijms20030686.

22.

Lee

J.K.

et al., Accuracy of fecal immunochemical tests for colorectal cancer: systematic review and meta-analysis, Ann Intern Med 160 (2014), 171. doi: 10.7326/M13-1484.

23.

Lee

Y.S.

and Cho

Y.B.

, CCL7 signaling in the tumor microenvironment, Adv Exp Med Biol 1231 (2020), 33–43. doi: 10.1007/978-3-030-36667-4_4.

24.

Morson

B.C.

, The evolution of colorectal carcinoma, Clinical Radiology 35 (1984), 425–431. doi: 10.1016/S0009-9260(84)80033-1.

25.

Nielsen

H.J.

et al., Independent prognostic value of eosinophil and mast cell infiltration in colorectal cancer tissue, The Journal of Pathology 189 (1999), 487–495. doi: 10.1002/(SICI)1096-9896(199912)189:4<487::AID-PATH484>3.0.CO;2-I.

26.

Nikolaou

et al., Systematic review ofÂ blood diagnostic markers in colorectal cancer, Tech Coloproctol 22 (2018), 481–498. doi: 10.1007/s10151-018-1820-3.

27.

Ning

and Lenz

H.-J.

, Targeting IL-8 in colorectal cancer, Expert Opin Ther Targets 16 (2012), 491–497. doi: 10.1517/14728222.2012.677440.

28.

Njor

S.H.

et al., Three years of colorectal cancer screening in Denmark, Cancer Epidemiol 57 (2018), 39–44. doi: 10.1016/j.canep.2018.09.003.

29.

Olink, Data normalization and standardization. Olink Document download center. https://www.olink.com/resources-support/document-download-center (assessed May 7, 2021).

30.

Park

D.I.

et al., Comparison of guaiac-based and quantitative immunochemical fecal occult blood testing in a population at average risk undergoing colorectal cancer screening, The American Journal of Gastroenterology 105 (2010), 2017–2025. doi: 10.1038/ajg.2010.179.

31.

Phallen

et al., Direct detection of early-stage cancers using circulating tumor DNA, Sci Transl Med 9 (2017). doi: 10.1126/scitranslmed.aan2415.

32.

Rasmussen

et al., Circulating cell-free nucleosomes as biomarkers for early detection of colorectal cancer, Oncotarget 9 (2017), 10247–10258. doi: 10.18632/oncotarget.21908.

33.

Ren

Dong

Tsoi

and Yu

, Detection of miRNA as non-invasive biomarkers of colorectal cancer, Int J Mol Sci 16 (2015), 2810–2823. doi: 10.3390/ijms16022810.

34.

Robertson

D.J.

et al., Recommendations on fecal immunochemical testing to screen for colorectal neoplasia: a consensus statement by the US multi-society task force on colorectal cancer, Gastroenterology 152 (2017), 1217–1237.e3. doi: 10.1053/j.gastro.2016.08.053.

35.

Saraste

et al., Complications after colonoscopy and surgery in a population-based colorectal cancer screening programme, J Med Screen (2016). doi: 10.1177/0969141315625701.

36.

Schreuders

E.H.

et al., Colorectal cancer screening: a global overview of existing programmes, Gut 64 (2015), 1637–1649. doi: 10.1136/gutjnl-2014-309086.

37.

Shaukat

et al., Long-term mortality after screening for colorectal cancer, N Engl J Med 369 (2013), 1106–1114. doi: 10.1056/NEJMoa1300720.

38.

Taube

J.M.

et al., Implications of the tumor immune microenvironment for staging and therapeutics, Mod Pathol 31 (2018), 214–234. doi: 10.1038/modpathol.2017.156.

39.

Tuomisto

A.E.

Mäkinen

M.J.

and Väyrynen

J.P.

, Systemic inflammation in colorectal cancer: underlying factors, effects, and prognostic significance, World J Gastroenterol 25 (2019), 4383–4404. doi: 10.3748/wjg.v25.i31.4383.

40.

Waugh

D.J.J.

and Wilson

, The interleukin-8 pathway in cancer, Clin Cancer Res 14 (2008), 6735–6741. doi: 10.1158/1078-0432.CCR-07-4843.

41.

Wilhelmsen

et al., Detection of colorectal neoplasia: combination of eight blood-based, cancer-associated protein biomarkers, Int J Cancer 140 (2017), 1436–1446. doi: 10.1002/ijc.30558.

42.

Xia

et al., Prognostic value, clinicopathologic features and diagnostic accuracy of interleukin-8 in colorectal cancer: a meta-analysis, PLOS ONE 10 (2015), e0123484. doi: 10.1371/journal.pone.0123484.

43.

Zucker

and Vacirca

, Role of matrix metalloproteinases (MMPs) in colorectal cancer, Cancer Metastasis Rev 23 (2004), 101–117. doi: 10.1023/a:1025867130437.