Prediagnostic Circulating Inflammation-related Biomarkers and Gastric Cancer: A Case-Cohort Study in Japan

Abstract

Gastric cancer is preceded by a chronic inflammatory process. Circulating levels of inflammation-related markers may reveal molecular pathways contributing to cancer development. Our study evaluated risk associations of gastric cancer with a wide range of systemic soluble inflammation and immune-response proteins. We performed a case-cohort analysis within the JPHC Study II, including a subcohort of 410 participants selected randomly within defined age and sex groups, and 414 individuals with incident gastric cancer. Ninety-two biomarkers were measured in baseline plasma using proximity extension assays. Gastric cancer multivariable hazard ratios were calculated for two to four quantiles used as ordinal variables of each biomarker by Cox proportional hazards regression models with age as the time metric. Of 73 evaluable biomarkers, three (CCL11, CCL20 and IL17C) were associated with increased gastric cancer risk and two (CCL23 and MMP1) with reduced cancer risk (P_trends <0.05). However, no association was statistically significant after a false discovery rate correction. This study largely expands the range of inflammation molecules evaluated for gastric cancer risk but failed to identify novel associations with this neoplasia.

1. INTRODUCTION

In the carcinogenic process of the most common form of gastric cancer (GC), mucosal colonization by Helicobacter pylori infection induces chronic inflammation that may variably progress to atrophy, intestinal metaplasia, dysplasia and cancer.[1] In addition, there is strong evidence that regular use of aspirin[2] and other non-steroidal anti-inflammatory drugs reduces the risk of GC.[3]

Regeneration of injured gastric mucosa involves a complex interaction of signaling pathways and regulators. Circulating levels of effector molecules may be informative as markers of either local activity or systemic effects and may have potential utility in terms of etiology and clinical management. We therefore evaluated associations of inflammation-related proteins with GC risk in a prospective study.

2. MATERIALS AND METHODS

2.1. Study Population

The Japan Public Health Center-based Prospective Study (JPHC Study) Cohort II is an ongoing study of middle-aged adults enrolled in 1993–94, including 23,335 who donated blood at enrolment.[4] Participants were followed through December 2010. For this case-cohort analysis, a subcohort of 410 participants was selected by age- and sex-stratified random sampling to ensure comparability with the cases. A total of 414 incident GC cases (International Classification of Diseases for Oncology, 3^rd edition, codes C16.0-C16.9) were identified among blood donors, including 23 among members of the subcohort.

This study was performed in accordance with the Declaration of Helsinki. The protocol of the JPHC Study was approved by the Japanese National Cancer Center (IRB No. 2004–059). Informed consent was obtained from all participants.

2.2. Laboratory Methods

Heparinized plasma samples (1 μL analysis volume) were tested for levels of 92 proteins by multiplex proximity extension assay (Proseek Multiplex Inflammation I panel, Olink Bioscience, Uppsala, Sweden) by the kit manufacturer using a Fluidigm Biomark reader (Fluidigm Corporation, USA), as previously described.[5] Relative levels were calculated from Ct values with corrections for assay variation and expressed as normalized protein expression (NPX) on a logarithmic scale.

Samples were randomly intermixed on testing plates, including 60 masked replicates to assess reproducibility. All biomarker levels had between batch coefficients of variation ≤10%. All or nearly all samples (90% or more) had levels below the lower limit of detection (LLOD) for artemin, fibroblast growth factor 5, glial cell line-derived neurotrophic factor, interferon gamma, interleukin 1a (IL1a), IL15RA, IL20, IL20RA, IL22RA1, IL24, IL2RB, IL13, IL2, IL33, IL4, leukemia inhibitory factor, neurturin, tumor necrosis factor and thymic stromal lymphopoietin. These 19 markers were excluded from analysis.

2.3. Statistical Analysis

The NPX levels of the 73 biomarkers detected in at least 10% of samples were grouped into two to four quantiles based on their distributions within the subcohort. A priori, the measurements were categorized as i) quartiles for biomarkers with <25% levels below LLOD (n=65), ii) tertiles for biomarkers with 25–50% levels below LLOD (n=0), iii) undetectable, < median or ≥ median for biomarkers with 50–75% levels below LLOD (n=5), and iv) undetectable or detectable for biomarkers with 75–90% levels below LLOD (n=3).

To account for the subcohort sampling scheme, Cox proportional regression models with a baseline hazard stratified by age-group and sex were used to calculate hazard ratios (HRs) and 95% confidence intervals for associations of each biomarker (as ordinal variables) with GC risk. Age was used as the model time metric. Models were adjusted for study area, and well-established risk factors, tobacco smoking, family history of GC, and salty/preserved foods intake.[6] Missing data on smoking was modeled as an independent category. False discovery rate (FDR) corrected P_trend values were also calculated to account for multiple testing. Markers with uncorrected P_trend <0.05 were further explored for nonlinear associations of the quantile categories using the lowest quantile as reference group. We performed stratified analyses by latency from blood collection to cancer diagnosis (<5 vs. ≥5 years) and anatomic subsite (proximal vs. distal). P_interaction for the latency analysis was calculated by testing for equality between the two time intervals of risk associations per change in marker quantile, censoring at 5 years for the earlier interval and having follow-up start at 5 years for the later interval. To compare the associations between proximal and distal cancers, we fit two separate regression models to the whole cohort, one for each site-specific outcome [7]. For the proximal model, distal cancer was a censoring event and vice versa. Differences between association parameters for distal and proximal cancers were then assessed by using a Wald test. Under the assumption that H. pylori infection is a necessary cause for gastric cancer, we also restricted to H. pylori seropositive subcohort members as a sensitivity analysis. Statistical analyses were conducted using SAS version 9.4 (SAS Inc, Cary, NC) and R (version 3.4.3) software.

3. RESULTS

The median time from blood collection to GC diagnosis was 8.1 years (interquartile range: 4.2–12.7 years). Compared with the subcohort members, participants who later developed GC had higher frequencies of H. pylori seropositivity, GC family history, cigarette smoking and salty foods consumption (Supplementary Table 1).

Figure 1 shows the 73 evaluable markers and their multivariable associations with risk of GC per quantile increase. Levels of CCL11, CCL20 and IL17C were positively associated (P_trend <0.05) with GC risk, with HRs (per quantile) ranging from 1.15 to 1.69, while levels of CCL23 and MMP1 were inversely associated, with HRs (per quantile) ranging from 0.78 to 0.82. Table 1 shows GC associations with individual quantiles of the five markers. None of these associations remained statistically significant after FDR correction.

Figure 1.

Estimated gastric cancer hazard ratios (HRs) and 95% confidence intervals (95% CIs) per quantile increase of biomarker levels. Age-, sex-stratified Cox proportional hazards models were adjusted for sex, study area (six public health areas), smoking habits (never, former, current < 20 and current ≥ 20 cigarettes/day, unknown), family history of gastric cancer (yes vs. no), and salty/preserved foods intake (0 vs. > 0 g/day). Biomarker levels were analyzed by quartiles unless annotated by superscript numerals as follows: ¹two categories (undetectable vs. detectable levels); ²three categories (undetectable, < median vs. ≥ median).

Table 1.

Circulating inflammation-related biomarkers statistically significantly associated with gastric cancer risk within JPHC Study Cohort II

Marker	Gastric cancer cases	HR1* (95% CI)	HR2† (95% CI)	Uncorrected / FDR-corrected Ptrend
CCL11 / Eotaxin-1
Per quantile		1.23 (1.07–1.42)	1.21 (1.05–1.40)	0.007 / 0.17
Quartile 1	83	1.0	1.0
Quartile 2	90	1.32 (0.85–2.06)	1.32 (0.85–2.07)
Quartile 3	109	1.51 (0.98–2.34)	1.48 (0.96–2.30)
Quartile 4	132	1.92 (1.24–2.99)	1.85 (1.18–2.90)
CCL20 / MIP3A
Per quantile		1.16 (1.02–1.32)	1.15 (1.01–1.32)	0.031 / 0.45
Quartile 1	89	1.00	1.0
Quartile 2	105	1.39 (0.92–2.10)	1.42 (0.94–2.16)
Quartile 3	107	1.35 (0.89–2.05)	1.31 (0.85–2.01)
Quartile 4	113	1.65 (1.09–2.49)	1.67 (1.10–2.53)
CCL23 / MIP3 / MPIF1
Per quantile		0.80 (0.69–0.91)	0.79 (0.69–0.91)	0.0008 / 0.06
Quartile 1	127	1.0	1.0
Quartile 2	116	0.83 (0.56–1.23)	0.82 (0.56–1.22)
Quartile 3	83	0.56 (0.37–0.85)	0.54 (0.35–0.82)
Quartile 4	88	0.54 (0.35–0.83)	0.54 (0.35–0.83)
IL17C / Cytokine CX2
Undetectable	337	1.0	1.0
Detectable	77	1.68 (1.13–2.51)	1.70 (1.13–2.56)	0.011 / 0.20
MMP1
Per quantile		0.84 (0.73–0.96)	0.82 (0.72–0.95)	0.006 / 0.17
Quartile 1	112	1.0	1.0
Quartile 2	119	0.93 (0.62–1.39)	0.89 (0.59–1.35)
Quartile 3	90	0.65 (0.43–0.99)	0.64 (0.42–0.97)
Quartile 4	93	0.63 (0.41–0.96)	0.59 (0.38–0.91)

Hazard ratios (HRs) and 95% confidence intervals (CIs).

Abbreviations: CCL= C-C Motif Chemokine Ligand, MIP= Macrophage Inflammatory Protein, MMP= Matrix Metalloproteinase, MPIF= Myeloid Progenitor Inhibitory Factor.

^*HR1 was adjusted for age (continuous), sex and study area (6 public health areas)

^†HR2 was adjusted for covariates used in HR1, smoking habits (never, past, current [<20 and ≥20 cigarettes/day]), family history of gastric cancer (yes vs. no), and salty/preserved foods intake (0 vs. >0 g/day).

Neither stratified analysis by latency (<5 vs. ≥5 years) nor restriction to H. pylori seropositive cohort members revealed any significant associations at a significance level of p<0.05 (data not shown). HR for proximal and distal cancers were similarly null (data not shown).

4. DISCUSSION

Despite the well-known role of chronic inflammation in gastric carcinogenesis, plasma levels of inflammatory proteins were not associated with GC risk in our long-term prospective investigation. Our findings contrast with previously reported modest associations based on plasma samples from the first JPHC for CRP and SAA (494 case-control pairs),[8] and in the Shanghai Health Cohort studies for IL8 among men (180 cases vs. 358 controls),[9] and IL6 among women (141 cases vs. 282 controls).[10] Notably, the Shanghai Men’s Cohort has a large proportion of current smokers (>50%), and both Shanghai Cohort Studies have very high H. pylori seroprevalence (>95%). Lack of replication might be also due to differences in study design and sample size.

In a previous case-cohort study using essentially the same sample set from the JPHC II, we measured a panel of 64 inflammation-related biomarkers by Luminex bead-based assays.[11] All marker-GC associations were null, including 31 overlapping proteins. Taking both studies together, analysis of 106 total markers suggests that a wide-range of common inflammation-related markers, including pro- and anti-inflammatory cytokines, growth factors, soluble receptors and angiogenesis factors, are not predictive of GC in this high-risk Asian population.

Strengths of our study include using a validated and sound testing method, prediagnostic biospecimens, large numbers of cancer cases, and adjustment for relevant risk factors. A limitation is that our biomarker measurements are based on a single plasma sample taken on average 8 years prior to cancer diagnosis and do not assess temporal variation.

In conclusion, this prospective study failed to identify associations of circulating inflammation markers with risk of GC. Our findings add to understanding of the complex role of inflammation in carcinogenesis. Further investigation of these markers is warranted to evaluate associations with early-stage disease and gastric preneoplastic conditions that could potentially guide prevention and treatment approaches.