Development of liquid biopsy for screening colorectal cancer through the combination of an antibody microarray-based metal-enhanced sandwich immunofluorescent assay of cytokines with machine learning

Abstract

The simultaneous determination of the expression levels of multiple inflammation-associated cytokines in blood holds great promise for the early screening of cancer including colorectal cancer (CRC). Herein, an antibody microarray-based sandwich metal-enhanced immunofluorescent assay (AMSMEIFA) is developed for the quantitative measurement of five cytokines simultaneously through the fabrication of an antibody microarray on a slide coated with a poly(glycidyl methacrylate-co-2-hydroxyethyl methacrylate) layer and modified with gold nanorods (GNR@P(GMA-HEMA) slide). Benefiting from the metal-enhanced fluorescence (MEF) property and abundant antibody immobilization sites in the GNR@P(GMA-HEMA) slide, the newly developed AMSMEIFA enables the selective measurement of five pro-inflammatory soluble cytokines (interleukins (IL-1β, IL-2 and IL-6), tumor necrosis factor-α (TNF-α), and interferon-γ (IFN-γ)) with low limits of detection (LODs) at the sub-pg mL⁻¹ level. The practicability of AMSMEIFA is demonstrated by the simultaneous determination of five cytokines in 35 clinical serum samples, which are obtained from 25 CRC patients and 10 healthy donors (HDs). After analyzing the expression levels of five pro-inflammatory soluble cytokines using a machine learning (ML) model based on the least absolute shrinkage and selector operator (LASSO) regression, the area under the receiver operator characteristic curve (AUC) of CRC is as high as 0.92, demonstrating that ML-assisted AMSMEIFA could be used as a liquid biopsy for screening CRC with reasonable accuracy.

---

Introduction

It has been demonstrated that the constant cross-talk between tumor-associated macrophages (TAMs) and tumor cells results in the accelerated release of several cytokines including interleukins (IL), growth factors, and tumor necrosis factor-α (TNF-α).^1–3 These soluble inflammatory factors collectively establish a distinctive inflammatory network within the tumor microenvironment (TME), thereby fulfilling essential functions throughout tumorigenesis, malignant progression, and invasive dissemination. For instance, IL-6, IL-1β, epidermal growth factor (EGF) and transforming growth factor β (TGF-β) released by myeloid cells (TAM, neutrophils, etc.) provide a favorable environment for tumor progression,^4–6 while IL-1β, IL-6, and TNFα secreted by tumor cells promote carcinogenesis.^7–10 Moreover, the cytokine release syndrome (CRS) (i.e., cytokine storm) of cancer immunotherapy results in significantly elevated levels of proinflammatory cytokines (e.g., TNF-α and interferon-γ (IFN-γ)) in the serum, which may ultimately cause dysfunction of multiple organs, and even death.¹¹ Therefore, it is important to develop high-performance strategies/procedures for profiling inflammation-associated cytokines in complex biological matrices.

The enzyme-linked immunosorbent assay (ELISA) remains the well-established gold standard methodology for the quantification and detection of cytokine biomarkers. Although the ELISA shows excellent sensing specificity towards cytokines, it suffers from poor multiplex analysis and relatively low sensitivity. Currently, various sensing platforms have been developed to sensitively detect cytokines, including surface-enhanced Raman spectroscopy (SERS) assay, fluorescent assays, and electrochemical (EC) biosensors.^12–20 Although these optical and EC sensing platforms exhibit high sensitivity and reliability, they are primarily designed to detect individual cytokines. During tumor initiation and progression, several cytokines normally exhibit diverse biological functions by forming complicated cytokine networks.^21–23 Therefore, the development of methods/assays enabling the simultaneous measurement of multiple cytokines in complex biological media with high sensitivity and selectivity is highly desirable.

Antibody microarrays provide an ideal platform for parallelly analyzing cytokines in a high-throughput manner, while minimizing sample consumption, offering valuable insights into disease processes, drug efficacy, and biomarker discovery.^24–26 There is continued interest in developing microarrays with high sensitivity because the concentrations of cytokines in the blood of tumor patients span a wide dynamic range (from the sub-pg mL⁻¹ level to hundred μg mL⁻¹ level).²⁷ This method relies on the immobilization of massive antibodies as an organized grid pattern consisting of individual spots containing specific reagents on a specific solid substrate (e.g., microscope slide) to capture the corresponding cytokines in the mixture, which are then detected by a secondary antibody in combination with a desirable signal readout. Consequently, the detection sensitivity can be improved significantly by increasing the antibody-cytokine binding efficiency and enhancing the signal intensity. For instance, a substantial enhancement in microarray sensitivity can be achieved through metal-enhanced fluorescence (MEF), which originates from the plasmonic coupling occurring between metallic nanostructures and proximal fluorophores.^28,29

Colorectal cancer (CRC) represents a prevalent malignant tumor that contributes substantially to the global disease burden, exhibiting considerable morbidity and mortality rates worldwide.³⁰ Colonoscopy offers direct visualization and therapeutic removal of precancerous adenomatous polyps.³¹ However, because its invasive nature carries risks of intestinal perforation and bleeding, a colonoscopy requires well-trained operators, particularly limiting accessibility in resource-constrained regions.³² Non-invasive examinations including fecal occult blood testing (FOBT) and fecal immunochemical testing (FIT) show insufficient sensitivity for screening CRC in the early stage.^33,34 Several studies demonstrate that cytokines including IFN-γ and IL-6 have an impact on the development of CRC,^35,36 which may not only be therapeutic targets for CRC biotherapy, but also serve as critical biomarkers for the advancement of liquid biopsy technologies to facilitate and support early diagnostic and screening procedures for CRC.

In this study, an antibody microarray-based sandwich metal-enhanced immunofluorescent assay (AMSMEIFA) is established for the simultaneous profiling of five pro-inflammatory soluble cytokines (IL-2, IL-6, TNF-α, IFN-γ, and IL-1β) on a slide coated with a poly(glycidyl methacrylate-co-2-hydroxyethyl methacrylate) layer and modified with gold nanorods (GNR@P(GMA-HEMA) slide). AMSMEIFA exhibits an excellent analytical performance including a low limit of detection (LOD) at the sub-pg mL⁻¹ level, high specificity, and wide dynamic range. AMSMEIFA was successfully employed to simultaneously determine five cytokines in 35 human serum samples from 25 CRC patients and 10 healthy donors (HDs). Furthermore, the area under the receiver operator characteristic curve (AUC) of CRC is as high as 0.92 by analyzing the expression levels of five cytokines through least absolute shrinkage and selector operator (LASSO) regression modeling, demonstrating the clinical validation of AMSMEIFA.

Experimental section

Reagents and instruments

Capture antibodies (cAbs) including mouse anti-human IL-2 monoclonal antibody (IL-2 cAb Cat#5375706), mouse anti-human IL-6 monoclonal antibody (IL-6 cAb Cat#4IL6-L152), mouse anti-human TNF-α monoclonal antibody (TNF-α cAb Cat#4T10-2C8cc), mouse anti-human IFN-γ monoclonal antibody (IFN-γ cAb Cat#5375906) and mouse anti-human IL-1β monoclonal antibody (IL-1β cAb Cat#53762206), and the corresponding detection antibodies (dAbs) including biotinylated mouse anti-human IL-2 monoclonal antibody (IL-2 dAb Cat#5175706), biotinylated mouse anti-human IL-6 monoclonal antibody (IL-6 dAb Cat#4IL6-L137), biotinylated mouse anti-human TNF-α monoclonal antibody (TNF-α dAb Cat#4T10-F6C5cc), biotinylated mouse anti-human IFN-γ monoclonal antibody (IFN-γ dAb Cat#5175906) and biotinylated mouse anti-human IL-1β monoclonal antibody (IL-1β dAb Cat#5176206) were purchased from HyTest Ltd (Turku, Finland). Cytokines IL-2, IL-6, TNF-α, IFN-γ and IL-1β were procured commercially from Linc-Bio Science Co. Ltd (Shanghai, China). The fluorescent-labeled probe strepavidin-Cy5 was obtained from Thermo Fisher Scientific Inc. (Waltham, USA). Milli-Q H₂O (18.2 MW cm⁻¹) was used in all experiments. The details of other reagents, materials and instruments are listed in the SI.

Fabrication of antibody microarray

The GNR@P(GMA-HEMA) slide was produced using a previously reported strategy with slight modification.³⁷ The details of the preparation of the GNR@P(GMA-HEMA) slide and fabrication of the antibody microarray are shown in the SI. Briefly, various concentrations of cAbs in spotting buffer (0.05 mol L⁻¹ sodium citrate, 0.2 mol L⁻¹ ammonium sulfate, 5 mg mL⁻¹ isoleucine, 5 mg mL⁻¹ glycine, and 10 mg mL⁻¹ trehalose, pH 7.2) were spotted utilizing the functionalized GNR@P(GMA-HEMA) slide surface via standard contact printing. After incubation under vacuum at 30 °C for 9 h, the antibody microarray was gently rinsed with 45 mL H₂O to remove excessed cAbs. Subsequently, the antibody microarray was transferred into 500 mL PBST (5 mmol L⁻¹ NaH₂PO₄, 5 mmol L⁻¹ Na₂HPO₄, 150 mmol L⁻¹ NaCl, and 0.05% (v/v) Tween-20, pH 7.5), and incubated for 20 min. Then, the antibody microarray was incubated in 45 mL blocking solution (10 mmol L⁻¹ PBS containing 10 mg mL⁻¹ BSA) for blocking unreacted epoxy groups at 27 °C for 2 h. Finally, the antibody microarray was washed with PBST (45 mL, 4 times) and PBS (5 mmol L⁻¹ Na₂HPO₄, 5 mmol L⁻¹ NaH₂PO₄, 150 mmol L⁻¹ NaCl, pH 7.5, 45 mL, 4 times), respectively.

Detection of cytokines

The cytokines used, including IL-2, IL-6, TNF-α, IFN-γ, and IL-1β, were dissolved in reaction buffer (10 mmol L⁻¹ PBS containing 10 mg mL⁻¹ BSA) or reaction buffer containing 10% clinical serum. 30 μL various concentrations of cytokines solutions were added to each subarray, respectively, incubated at 30 °C under 70% relative humidity (RH) for 2 h, and then washed with PBST (45 mL, 4 times) PBS (45 mL, 4 times), and H₂O (45 mL, 4 times), respectively. After drying by centrifugation (2000 rpm, 2 min), 30 μL dAbs with the desired concentration in reaction buffer was introduced to each subarray, respectively. After incubation at 30 °C under 70% RH for 2 h, the microarray was washed and dried as previously described. Then, 30 μL streptavidin-Cy5 conjugate (0.02 mg mL⁻¹ in reaction buffer) was added to each subarray, respectively. After incubation at 30 °C under 70% RH for 1 h, the polytetrafluoroethylene (PTFE) mask was removed. The microarray was washed and dried as previously described, and subjected to the data acquisition procedure. To assess the method specificity, 30 μL mixture of cytokines (100 pg mL⁻¹) was applied to a subarray, and treated as previously described.

Data analysis and modeling

The relative fluorescence intensity (F_R) was calculated using the formula F_R = (F − F₀)/F₀, where F₀ is defined as the mean fluorescence intensity of 5 replicate spots of blank sample, while F is defined as the mean fluorescence intensity of 5 replicate spots of sample containing a certain amount of cytokine and/or cytokine mixture. The LASSO regression model was set up to distinguish CRC patients and HDs by the as-obtained F_R from clinical serum samples. The training cohort was comprised of 17 CRC patients and 7 HDs, while the testing cohort contained 8 CRC patients and 3 HDs. The model effectiveness was assessed by analyzing the receiver operator characteristic (ROC) curve from the training set, and additional evaluation utilized confusion matrix analysis of the test set.

Results and discussion

Analytical performance of AMSMEIFA

AMSMEIFA was fabricated using our previously reported strategy, except anti-cytokine antibodies were used instead of anti-cardiovascular marker antibodies.³⁷ The detection principle of AMSMEIFA is shown in Scheme 1. Briefly, a microarray containing 5 anti-cytokine antibodies (cAbs) is fabricated on the GNR@P(GMA-HEMA) slide, which is employed to capture cytokines in the different samples. The corresponding 5 biotinylated anti-cytokine antibodies (dAbs) are used to recognize the antibody-cytokine pairs in the microarray, and form sandwich structures, which are subsequently labeled by strepavidin-Cy5 via the interaction of biotin with streptavidin. The synergistic combination of enhanced fluorescence of GNR and high protein loading capacity of P(GMA-HEMA) yielded an exceptional analytical performance for satisfying the clinical requirement for the determination of cytokines.³⁷ In this case, five soluble cytokines including IL-2, IL-6, TNF-α, IFN-γ, and IL-1β were selected as analytical targets. In CRC, IL-2 serves as a pivotal effector cytokine. The comparative analysis revealed that the serum concentrations of IL-2 were notably elevated within the population of individuals diagnosed with CRC compared to demographically matched healthy control subjects. Multivariate analysis identified IL-2 as a key factor in constructing the prognostic model.³⁸ Functioning as a pro-inflammatory signaling molecule, IFN-γ enhances the transcriptional activity of metastasis-linked genes, and thereby facilitates the migratory capacity of CRC cells through various oncogenic processes.³⁹ IL-1β, IL-6, and TNF-α are pro-inflammatory cytokines downstream of MK2 signaling. They drive tumor growth, invasion, and metastasis in CRC. Supplementing these cytokines restores tumor proliferation and invasion, indicating that they maintain cancer progression through a positive feedback loop.⁴⁰

Scheme 1 Schematic of ML-assisted AMSMEIFA for screening CRC through the simultaneous determination of five cytokines in clinical serum samples. (a) Collection of the clinical serum samples from HDs and CRC patients, (b) profiling cytokines in sera by AMSMEIFA, and (c) data analysis by LASSO regression.

GNRs with an average aspect ratio of 2.4 (average length, 56.5 ± 3.9 nm and average width, 23.0 ± 1.7 nm) were used to enhance the fluorescence emission of Cy5 because of the overlapping local surface plasmonic resonance (LSPR) band of GNR at 655 nm with the excitation profile of Cy5 (as shown in Fig. 1a). This phenomenon leads to an enhancement in the quantum yield and improvement in the photostability of Cy5.⁴¹ The successful formation of a GNR self-assembly monolayer on the amino-silanized slide was confirmed by scanning electron microscopy (SEM) (as shown in Fig. 1b). A comprehensive series of sequentially performed X-ray photoelectron spectroscopy (XPS) analyses combined with detailed water contact angle (WCA) measurements demonstrate the successful graft-fabrication of the GNR@P(GMA-HEMA) slide on the GNR self-assembly monolayer (as shown in Fig. S1 and S2). The P(GMA-HEMA) coating performs several critical roles, including serving as a physical barrier to avoid immediate contact between the fluorescent molecules (e.g., Cy5) and the GNR surface, providing an antifouling layer for the minimization of nonspecific protein binding, increasing the antibody surface density for enhancing the antibody-cytokine binding efficiency. In addition, there are small inter-batch and intra-batch differences on AMSMEIFA for the determination of five cytokines in the same serum sample of CRC patients, demonstrating that the as-proposed AMSMEIFA has good reproducibility (as shown in Fig. S3).

Fig. 1 (a) LSPR band of GNR (black line), excitation spectrum of Cy5 (red line), and emission spectrum of Cy5 (blue line), inset of (a) TEM micrograph of the as-prepared GNRs and (b) SEM micrograph of the GNR@P(GMA-HEMA) slide.

To improve the analytical performance of AMSMEIFA, systematic optimization of the concentrations of cAbs and dAbs was performed. F_R increased by increasing the concentration of cAbs in the spotting solution, and gradually reached saturation when the concentration of cAbs was more than 500 μg mL⁻¹ (as shown in Fig. S4). In addition, F_R increased by increasing the concentration of dAbs, and gradually reached saturation when the concentration of dAbs was more than 10 μg mL⁻¹ (as shown in Fig. S5). Considering both the detection cost and analytical performance, the optimized details of AMSMEIFA are shown in Table S1. Under the optimal conditions, various concentrations of cytokines in the reaction buffer and/or reaction buffer containing 10% (v/v) serum from HDs were detected by AMSMEIFA. The values of F_R increased by increasing the concentration of cytokines (as shown in Fig. 2). AMSMEIFA displayed LODs at as low as the sub-pg mL⁻¹ level and wide dynamic ranges with more than 5 orders of magnitude both in the reaction buffer and/or reaction buffer containing 10% serum (as shown in Table 1). The LOD was calculated using the formula LOD = 3 × S × (C/X), where S refers to the standard deviation of signals from three parallel low-level samples, C is the concentration of analyte, and X is the average value of the signal intensity of analyte. The as-obtained LODs and dynamic ranges of AMSMEIFA are better than that of most of the sensing platforms reported including immunoassay, SERS, and microfluidic platforms (as shown in Table S2), demonstrating that AMSMEIFA is suitable for profiling both a low abundance of cytokines (at pg mL⁻¹ level) in healthy people, and high abundance of cytokines (at hundred pg mL⁻¹ level) in tumor patients, respectively.⁴²

Fig. 2 Calibration curves for the quantitative analysis of (a) IL-2, (b) IL-6, (c) TNF-α, (d) IFN-γ and (e) IL-1β in the reaction buffer and reaction buffer plus 10% (v/v) clinical serum. (f) Determination of the individual cytokines and mixed cytokines in reaction buffer plus 10% clinical serum by AMSMEIFA.

Table 1 Summary of the analytical performance of AMSMEIFA for detecting cytokines in different matrices

Targets	Matrix	Linear ranges	Linear regression equations	Limits of detection
IL-2	PBS	0.005–10 ng mL^−1	F R = 2.52 lg C + 6.11	0.42 pg mL^−1
	10% serum	0.005–10 ng mL^−1	F R = 1.88 lg C + 4.52	0.67 pg mL^−1
IL-6	PBS	0.001–100 ng mL^−1	F R = 7.06 lg C + 19.62	0.26 pg mL^−1
	10% serum	0.005–100 ng mL^−1	F R = 7.48 lg C + 17.25	0.53 pg mL^−1
TNF-α	PBS	0.001–100 ng mL^−1	F R = 2.94 lg C + 9.08	0.25 pg mL^−1
	10% serum	0.001–100 ng mL^−1	F R = 2.92 lg C + 8.48	0.25 pg mL^−1
IFN-γ	PBS	0.001–100 ng mL^−1	F R = 3.95 lg C + 12.57	0.19 pg mL^−1
	10% serum	0.001–50 ng mL^−1	F R = 3.53 lg C + 11.36	0.21 pg mL^−1
IL-1β	PBS	0.001–50 ng mL^−1	F R = 3.36 lg C + 9.99	0.38 pg mL^−1
	10% serum	0.001–100 ng mL^−1	F R = 2.17 lg C + 6.54	0.38 pg mL^−1

---

Detection of cytokines in clinical serum samples

To evaluate its clinical practicability, AMSMEIFA was employed for the simultaneous detection of five cytokines in clinical sera, which were collected from 25 CRC patients and 10 HDs. Considering potential matrix interference, the serum samples were first diluted systematically with reaction buffer. The F_R initially increased, and then decreased with an increase in serum concentration due to competing matrix effects. Low concentrations reduce non-specific adsorption, improving the efficiency, while higher levels introduce steric hindrance, site competition, and background interference, which diminish the signal. A progressive increase in F_R was observed as the serum concentration increased from 0.1% (v/v) to 10% (v/v). The opposite trend was observed at higher concentrations, with F_R gradually declining when the serum concentration exceeds 10% (v/v) (as shown in Fig. S6). Therefore, 10% (v/v) diluted serum was selected to quantitatively analyze the abundance of five cytokines. In addition, the five cytokines in 0.1% (v/v) diluted serum can be detected, indicating that AMSMEIFA has good sensitivity. Generally, the concentration levels of five cytokines in the sera of CRC patients are higher than that in the sera of HDs (as shown in Fig. 3). The analysis revealed substantial variations in the concentration levels of the five cytokines between the sera of the CRC patients and sera of HDs. This result is in good agreement with that in previous reports.^22,43–45 Remarkably high concentrations of IL-6 were detected in the blood serum of patients suffering from CRC. This empirical finding supports previously published investigations that confirm the essential contribution of IL-6 to the processes of colorectal tumor formation and the progression of inflammatory cascade reactions.^46,47 The results indicate that AMSMEIFA has great potential for the liquid biopsy of CRC through profiling cytokines.

Fig. 3 Determination of (a) IL-2, (b) IL-6, (c) TNF-α, (d) IFN-γ and (e) IL-1β in the serum samples collected from 25 CRC patients and 10 HDs by AMSMEIFA. The concentrations of cytokines were estimated using the calibration curves of the corresponding cytokine in reaction buffer plus 10% serum, which are shown in Fig. 2a–e.

Machine learning

LASSO regression was employed to construct a classification model for the diagnosis of CRC based on the serum concentrations of five cytokines because it is an efficient modeling technique enabling both coefficient shrinkage and feature selection. A total of 35 serum samples was partitioned into the training set (the serum samples from 17 CRC patients and 7 HDs) and testing set (the serum samples from 8 CRC patients and 3 HDs). The regression coefficients of the model are provided in Fig. S7. The model effectiveness was evaluated by analysis of the ROC curve. As shown in Fig. 4a, the AUC is 0.92, which is better than that required for clinical diagnosis (0.8). The confusion matrix of the testing set reveals 90.9% classification accuracy, suggesting the diagnostic reliability of AMSMEIFA. Thus, the results highlight the promising application of AMSMEIFA as a valuable auxiliary diagnostic method for CRC in clinical settings.

Fig. 4 (a) ROC curve for the diagnosis of CRC and (b) confusion matrix indicating the accuracy of the training set based on AMSMEIFA.

Conclusions

In summary, we have developed an analytical high-performance AMSMEIFA capable of simultaneously detecting five cytokines in quintuplicate, which processes 12 distinct samples (3 μL each) per chip. This assay achieves remarkably low LODs at the sub-pg mL⁻¹ level and exhibits a broad dynamic range spanning 5 orders of magnitude through the systematic combination of the high-throughput property of the antibody microarray platform, MEF property of GNRs, and abundant antibody immobilization sites of P(GMA-HEMA). The clinical application of AMSMEIFA has been successfully demonstrated by profiling the cytokines in serum samples from CRC patients, which achieves 90.9% classification accuracy for CRC patients versus HDs under ML (LASSO regression)-assistance. Although only five cytokines have been measured for the auxiliary diagnosis of CRC in this proof-of-principle experiment, the as-developed AMSMEIFA can be easily modified to function as a highly multiplexed cytokine profiling system capable of performing holistic patient disease state assessment, such as continuously monitoring the occurrence and development of tumors and the cytokine storm of immunotherapy, thereby offering substantial clinical value for early disease detection, therapeutic intervention, and prophylactic healthcare strategies.

Author contributions

Wanyu Zhang: methodology, data analysis, writing – original draft. Shasha Li: methodology, data analysis. Xudong Sun: data analysis. Zhenxin Wang: writing – review and editing, supervision, project administration.

Conflicts of interest

There are no conflicts to declare.

Data availability

The data supporting this article have been included as part of the supplementary information (SI). Supplementary information is available. See DOI: https://doi.org/10.1039/d5an01041a.

Acknowledgements

The authors would like to thank the Science and Technology Development Plan Project of Jilin Province (No. SKL202302030).

References

1. N. Caronni, F. La Terza, L. Frosio and R. Ostuni, Trends Immunol., 2025, 46, 403–415.
2. M. Xie, K. Yuan, Y. Zhang, Y. Zhang, R. Zhang, J. Gao, W. Wei, L. Jiang, T. Li, Y. Ding, L. Wang, Y. Lin, C. C. Wong and J. Yu, Cancer Cell, 2025, 43, 1885–1901.e10.
3. C. Lee, H. Lee, H. Cho, S. Kim, I. Choi, Y.-S. Hwang, H. Jeong, H. Jang, S. Pak, D.-S. Hwang, I.-H. Han and H. Bae, Cancer Commun., 2022, 42, 345–349.
4. E. L. Kuan and S. F. Ziegler, Nat. Immunol., 2018, 19, 366–374.
5. S. Srivatsa, M. C. Paul, C. Cardone, M. Holcmann, N. Amberg, P. Pathria, M. A. Diamanti, M. Linder, G. Timelthaler, H. P. Dienes, L. Kenner, F. Wrba, G. W. Prager, S. Rose-John, R. Eferl, G. Liguori, G. Botti, E. Martinelli, F. R. Greten, F. Ciardiello and M. Sibilia, Gastroenterology, 2017, 153, 178–190.e10.
6. S. Yin, H. Cui, S. Qin and S. Yu, Biomed. Pharmacother., 2023, 166, 115355.
7. M. J. Waldner, S. Foersch and M. F. Neurath, Int. J. Biol. Sci., 2012, 8, 1248–1253.
8. D. E. Johnson, R. A. O'Keefe and J. R. Grandis, Nat. Rev. Clin Oncol., 2018, 15, 234–248.
9. L. M. Sedger and M. F. McDermott, Cytokine Growth Factor Rev., 2014, 25, 453–472.
10. P. Szlosarek, K. A. Charles and F. R. Balkwill, Eur. J. Cancer, 2006, 42, 745–750.
11. D. C. Fajgenbaum and C. H. June, N. Engl. J. Med., 2020, 383, 2255–2273.
12. J. Tang, C. Ji, X. Lu, H. Cao, Y. Ling, Y. Wu, L. Qian, Y. He, B. Song and H. Wang, Anal. Chem., 2024, 96, 9684–9692.
13. S. Gaba, F. Deng and U. Jain, TrAC, Trends Anal. Chem., 2025, 192, 118318.
14. J. S. Cognetti, M. T. Moen, M. G. Brewer, M. R. Bryan, J. D. Tice, J. L. McGrath and B. L. Miller, Lab Chip, 2023, 23, 239–250.
15. C. Huang, D. Li, J. Liu, S. Hou, W. Yang, H. Wang, J. Wang, Z. Wang, F. Li, Z. Hao, S. Huang, X. Zhao, P. Hu and Y. Pan, Adv. Funct. Mater., 2024, 34, 2309447.
16. H. Chu, X. Hu, C.-Y. Lee, A. Zhang, Y. Ye, Y. Wang, Y. Chen, X. Yan, X. Wang, J. Wei, S. He and Y. Li, Biosens. Bioelectron., 2023, 232, 115301.
17. N. Cennamo, A. Piccirillo, D. Bencivenga, F. Arcadio, M. Annunziata, F. D. Ragione, L. Guida, L. Zeni and A. Borriello, Talanta, 2023, 256, 124284.
18. J. Gao, X. Zhu, Y. Long, M. Liu, H. Li, Y. Zhang and S. Yao, Anal. Chem., 2024, 96, 1795–1802.
19. P. Gaikwad, N. Rahman, R. Parikh, J. Crespo, Z. Cohen and R. M. Williams, ACS Appl. Mater. Interfaces, 2024, 16, 27102–27113.
20. Y. Ouyang, W. Xie, T. Tang, X. Su, S. Xiao, Z. Liu, M. Li, M. Wu, J. Liu, J. Luo and P. Chen, Chem. Eng. J., 2024, 496, 154004.
21. J. Borowczak, K. Szczerbowski, M. Maniewski, A. Kowalewski, M. Janiczek-Polewska, A. Szylberg, A. Marszałek and Ł. Szylberg, Biomedicines, 2022, 10, 1670.
22. M. Krzystek-Korpacka, D. Diakowska, B. Kapturkiewicz, M. Bębenek and A. Gamian, Cancer Lett., 2013, 337, 107–114.
23. N. R. West, S. McCuaig, F. Franchini and F. Powrie, Nat. Rev. Immunol., 2015, 15, 615–629.
24. R. Huang, C.-H. Gao, W. Wu and R.-P. Huang, in Antibody Arrays: Methods and Protocols, eds. K. C. Whittaker and R.-P. Huang, Springer US, New York, NY, 2021, pp. 123–128.
25. N. G. Gumanova, D. K. Vasilyev, N. L. Bogdanova, Y. I. Havrichenko, A. Y. Kots and V. A. Metelskaya, Biochem. Biophys. Res. Commun., 2022, 631, 55–63.
26. M. Chen, X. Yang, M. Yang, W. Zhang, L. Li and Q. Sun, Pathol., Res. Pract., 2019, 215, 1033–1037.
27. L. Koelman, O. Pivovarova-Ramich, A. F. H. Pfeiffer, T. Grune and K. Aleksandrova, Immun. Ageing, 2019, 16, 11.
28. J. Sun, Z. Li, Y. Sun, L. Zhong, J. Huang, J. Zhang, Z. Liang, J. Chen and L. Jiang, Nano Res., 2018, 11, 953–965.
29. C. Joyce, S. M. Fothergill and F. Xie, Mater. Today Adv., 2020, 7, 100073.
30. R. L. Siegel, N. S. Wagle, A. Cercek, R. A. Smith and A. Jemal, CA-Cancer J. Clin., 2023, 73, 233–254.
31. U. Ladabaum, J. A. Dominitz, C. Kahi and R. E. Schoen, Gastroenterology, 2020, 158, 418–432.
32. D. J. Robertson, D. K. Rex, O. Ciani and M. F. Drummond, Gastroenterology, 2024, 166, 758–771.
33. Y. S. Miguel, J. Demb, M. E. Martinez, S. Gupta and F. P. May, Gastroenterology, 2021, 160, 1997–2005.e3.
34. Y. Zhang, Y. Wang, B. Zhang, P. Li and Y. Zhao, Biomed. Pharmacother., 2023, 163, 114786.
35. F. Colotta, P. Allavena, A. Sica, C. Garlanda and A. Mantovani, Carcinogenesis, 2009, 30, 1073–1081.
36. A. Mantovani, P. Allavena, A. Sica and F. Balkwill, Nature, 2008, 454, 436–444.
37. M. Jian, X. Sun, S. Li, H. Wang, H. Zhang, X. Li, Y. He and Z. Wang, Anal. Chem., 2024, 96, 7353–7359.
38. P. Czajka-Francuz, T. Francuz, S. Cisoń-Jurek, A. Czajka, M. Fajkis, B. Szymczak, M. Kozaczka, K. P. Malinowski, W. Zasada, J. Wojnar and J. Chudek, Rep. Pract. Oncol. Radiother., 2020, 25, 867–875.
39. D. Kobelt, C. Zhang, I. A. Clayton-Lucey, R. Glauben, C. Voss, B. Siegmund and U. Stein, Front. Immunol., 2020, 11 DOI:10.3389/fimmu.2020.00980.
40. A. L. Ray, K. L. Berggren, S. R. Cruz, G. N. Gan and E. J. Beswick, Int. J. Cancer, 2018, 142, 1702–1711.
41. D. Hong, E.-J. Jo, D. Bang, C. Jung, Y. E. Lee, Y.-S. Noh, M. G. Shin and M.-G. Kim, ACS Nano, 2023, 17, 16607–16619.
42. N. A. Johdi, L. Mazlan, I. Sagap and R. Jamal, Cytokine, 2017, 99, 35–42.
43. W. Li, F. Chen, H. Gao, Z. Xu, Y. Zhou, S. Wang, Z. Lv, Y. Zhang, Z. Xu, J. Huo, J. Zhao, Y. Zong, W. Feng, X. Shen, Z. Wu and A. Lu, Front. Immunol., 2023, 14 DOI:10.3389/fimmu.2023.1175513.
44. J. W. Park, H. J. Chang, H. Y. Yeo, N. Han, B. C. Kim, S.-Y. Kong, J. Kim and J. H. Oh, Br. J. Cancer, 2020, 123, 610–618.
45. M. Tertis, P. I. Leva, D. Bogdan, M. Suciu, F. Graur and C. Cristea, Biosens. Bioelectron., 2019, 137, 123–132.
46. T. Rich, P. F. Innominato, J. Boerner, M. C. Mormont, S. Iacobelli, B. Baron, C. Jasmin and F. Lévi, Clin. Cancer Res., 2005, 11, 1757–1764.
47. Y.-C. Chung and Y.-F. Chang, J. Surg. Oncol., 2003, 83, 222–226.

---