Mendelian randomization analysis of C-reactive protein on colorectal cancer risk

Abstract

Background

Chronic inflammation is a risk factor for colorectal cancer (CRC). Circulating C-reactive protein (CRP) is also moderately associated with CRC risk. However, observational studies are susceptible to unmeasured confounding or reverse causality. Using genetic risk variants as instrumental variables, we investigated the causal relationship between genetically elevated CRP concentration and CRC risk, using a Mendelian randomization approach.

Methods

Individual-level data from 30 480 CRC cases and 22 844 controls from 33 participating studies in three international consortia were used: the Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO), the Colorectal Transdisciplinary Study (CORECT) and the Colon Cancer Family Registry (CCFR). As instrumental variables, we included 19 single nucleotide polymorphisms (SNPs) previously associated with CRP concentration. The SNP-CRC associations were estimated using a logistic regression model adjusted for age, sex, principal components and genotyping phases. An inverse-variance weighted method was applied to estimate the causal effect of CRP on CRC risk.

Results

Among the 19 CRP-associated SNPs, rs1260326 and rs6734238 were significantly associated with CRC risk (P = 7.5 × 10^–4, and P = 0.003, respectively). A genetically predicted one-unit increase in the log-transformed CRP concentrations (mg/l) was not associated with increased risk of CRC [odds ratio (OR) = 1.04; 95% confidence interval (CI): 0.97, 1.12; P = 0.256). No evidence of association was observed in subgroup analyses stratified by other risk factors.

Conclusions

In spite of adequate statistical power to detect moderate association, we found genetically elevated CRP concentration was not associated with increased risk of CRC among individuals of European ancestry. Our findings suggested that circulating CRP is unlikely to be a causal factor in CRC development.

Key Messages

• Meta-analyses of observation studies reported a moderate association between elevated C-reactive protein (CRP) concentration and colorectal cancer (CRC) risk; however, whether the association is causal is unclear.

• In this largest study to date, we had adequate statistical power to assess causal relationship between circulating CRP concentration and CRC risk, using Mendelian randomization analysis.

• We found that genetically elevated CRP concentration was not associated with increased risk of CRC among individuals of European ancestry, suggesting that CRP is unlikely to play a causal role in CRC development.

• No evidence of genetically elevated CRP-CRC association was observed in subgroup analyses stratified by other risk factors.

Introduction

Chronic inflammation plays a role in the pathogenesis of colorectal cancer (CRC).¹ Meta-analyses of observational studies have shown that a one-unit (mg/l) increase in log-transformed high-sensitivity C-reactive protein (CRP), a common biomarker for low-grade chronic inflammation, was associated with 12% higher risk of CRC.^2,3 The association was stronger among men than women, and was stronger in colon cancer than rectal cancer. Although observational studies support a role for CRP in CRC development, they are susceptible to potential bias by unmeasured confounders, such as older age,⁴ adiposity,⁵ tobacco smoking,^6,7 lower physical activity,⁸ and lower use of non-steroidal anti-inflammatory drugs (NSAIDs).⁹ Observational studies are also susceptible to reverse causality, in which elevated CRP concentrations are due to immune response and inflammation induced by premalignant or preclinical tumour growth.^10–12

Mendelian randomization analysis, by taking advantage of the random assortment of genetic alleles during gamete formation, is less susceptible to confounding or reverse causality.¹³ Because genetic variants are distributed randomly at conception, they are generally unrelated to environmental risk factors, and temporally precede both risk factors and the disease process. The heritability of CRP was estimated to range between 25% and 40%, suggesting a role of genetic factors in baseline CRP concentrations.¹⁴ Several studies have used CRP-related genetic variants as a proxy of lifelong CRP concentrations on CRC risk, but reported inconsistent findings. A nested case-control study found genetically elevated CRP concentration, based on seven single nucleotide polymorphisms (SNPs) in the CRP gene, was associated with higher CRC risk.¹⁵ Another case-control study found a tagSNP in the CRP gene to be associated with higher risk of colon cancer, and another SNP associated with lower risk of rectal cancer.¹⁶ Other studies using SNPs within the CRP gene did not find associations between CRP and CRC risk.^17–19 In a prospective cohort study, Prizment et al.²⁰ reported an association between a weighted CRP genetic risk score, based on 20 CRP-associated SNPs identified in a meta-analysis of genome-wide association studies (GWAS) studies,¹⁴ and CRC risk, corroborating a causal role for CRP in colorectal carcinogenesis. However, the cohort had limited statistical power due to relatively small sample size (7603 participants including 205 CRC cases). In addition, most previous studies assumed population homogeneity and did not adjust for population stratification, which could bias the results. Furthermore, other SNPs recently found to be associated with CRP concentration were not included.²¹

Findings from previous human genetic studies have been inconsistent and have had insufficient power to assess a moderate causal relationship between CRP and CRC risk. In this study, we aimed to investigate whether CRP plays a causal role in CRC risk, using genetic variants that were previously reported to be significantly associated with circulating CRP concentration as instrumental variables (IVs), using the largest study for such analysis to date.

Methods

Study participants

We used epidemiological and genetic data from 30 480 CRC cases and 22 844 controls from 33 participating studies in three international CRC consortia: the Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO), the Colorectal Transdisciplinary Study (CORECT) and the Colon Cancer Family Registry (CCFR). Full details have been published previously,^22,23 and the demographic characteristics of study participants are summarized in Supplementary Table 1, available as Supplementary data at IJE online. In brief, 10 644 cases and 10 729 controls were included from GECCO from nested case-control studies in eight cohorts and six case-control studies. Further, 19 836 cases and 12 115 controls were included from CORECT from nested case-control studies in seven cohorts, nine case-control studies and three case-series studies. Nested case-control studies from CCFR participated as individual studies in GECCO and/or CORECT. There was no overlap of participants between studies. Participants with non-European ancestry were excluded. Informed consent was given by all participants, and studies were approved by their respective institutional review boards.

Assessment of outcomes and environmental variables

Invasive CRC cases (International s pathology reports, death certificates or record linkage. Age at diagnosis, cancer subsites and stages were obtained from medical records and cancer registries. Patients with Lynch syndrome and other syndromic causes were excluded. Controls were selected based on study-specific eligibility and matching criteria. Case-series studies only contributed cases to this study.

Demographic and environmental factors were self-reported at either in-person interview or via structured self-administered questionnaires, based on each study. A multistep, iterative data harmonization procedure was applied, and was described previously.²² Age was defined as age at CRC diagnosis for cases, or age at selection for controls. Body mass index (BMI; kg/m²) was categorized as normal (18.5–24.9), overweight (25–29.9), and obese (≥30). Participants with BMI <18.5 were excluded. Smoking status was defined as never and ever smokers. Regular use of any NSAIDs, aspirin or non-aspirin NSAIDs was defined as binary (yes/no). Family history of CRC was defined as CRC occurring in any first-degree relative. History of endoscopy included both sigmoidoscopy and colonoscopy.

Genotyping

Details on genotyping and imputation have been reported previously.²⁴ In brief, DNA was mostly obtained from blood samples, with some from buccal swabs. Several platforms (the Illumina HumanHap 300 k, 240 k, 550 k and OncoArray 610 k BeadChip Array system, or Affymetrix platform) were used for genotyping.^25,26 Samples were excluded on the basis of sample call rate ≤97%, heterozygosity, unexpected duplicates or relative pairs, gender discrepancy and principal component analysis (PCA) outlier of HapMap2 CEU cluster. SNPs were excluded on the basis of inconsistency across platforms, call rate <98%, and out of Hardy-Weinberg equilibrium (HWE) in controls (P  0.0001).²⁵ SNPs were imputed from the 1000 Genome Project reference panel if not directly genotyped, and restricted by imputation accuracy (R² > 0.3).

Instrumental variables

All the selected SNPs came from two resources: 18 SNPs that had been previously used as IVs¹⁴ and nine SNPs from more recent findings among participants of European ancestry²¹ (summarized in Table 1). SNPs in association with CRP concentration at the genome-wide significance threshold of P < 5 × 10^–8 were selected. We checked independence between the 27 selected SNPs using linkage disequilibrium (LD) analysis. If two SNPs were in LD (R²> 0.2), the SNP with the smaller P-value was included in the final SNP set, and the other excluded. We also conducted a GWAS-catalogue search for SNPs that were associated with CRP ( P < 5 × 10^–8) among participants of European ancestry, had reported the estimated effect sizes and standard errors and were not in LD with the selected SNPs. No additional SNPs were identified. Altogether, 19 SNPs were included in the final IV set. The allele associated with higher CRP level was coded as risk allele, and the other allele was coded as baseline allele for all SNPs.

Table 1.

Association of genome-wide significant loci with CRP concentrations in previous studies

SNP	Chr	Position	Gene	EA	BA	EAF	R2	βa	SE	P-value	Study	LD exclusionb
rs2794520	1	159 678 816	CRP	C	T	0.67	1.01	0.160	0.006	2.0 × 10−186	Dehghan et al.14	No
rs4420638	19	45 422 946	APOC1	A	G	0.83	0.84	0.236	0.009	8.8 × 10−139		No
rs1183910	12	121 420 807	HNF1A	G	A	0.68	1.00	0.149	0.006	2.1 × 10−124		No
rs4420065	1	66 161 461	LEPR	C	T	0.62	0.98	0.090	0.005	3.5 × 10−62		No
rs4129267	1	154 426 264	IL6R	C	T	0.60	0.99	0.079	0.005	2.1 × 10−48		No
rs1260326	2	27 730 940	GCKR	T	C	0.42	1.01	0.072	0.005	4.6 × 10−40		No
rs6734238	2	113 841 030	IL1F10	G	A	0.41	1.00	0.050	0.006	1.8 × 10−17		No
rs12239046	1	247 601 595	NLRP3	C	T	0.62	0.99	0.047	0.006	1.2 × 10−15		No
rs9987289	8	9 183 358	PPP1R3B	A	G	0.08	1.00	0.069	0.011	3.4 × 10−13		No
rs10521222	16	51 158 710	SALL1	C	T	0.95	0.97	0.104	0.015	8.5 × 10−13		No
rs10745954	12	103 483 094	ASCL1	A	G	0.52	0.99	0.039	0.006	1.6 × 10−11		No
rs12037222	1	40 064 961	PABPC4	A	G	0.23	0.99	0.045	0.007	6.4 × 10−11		No
rs1800961	20	43 042 364	HNF4A	C	T	0.97	0.88	0.088	0.015	2.2 × 10−9		No
rs13233571	7	72 971 231	BCL7B	C	T	0.89	1.00	0.054	0.009	3.6 × 10−9		No
rs340029	15	60 894 965	RORA	T	C	0.62	0.97	0.032	0.006	4.1 × 10−9		No
rs4705952	5	131 839 618	IRF1	G	A	0.25	0.96	0.042	0.007	1.3 × 10−8		No
rs2847281	18	12 821 593	PTPN2	A	G	0.60	0.99	0.031	0.006	2.2 × 10−8		No
rs6901250	6	117 114 025	GPRC6A	A	G	0.32	0.99	0.035	0.006	4.8 × 10−8		No
rs2075650	19	45 395 619	TOMM40	A	G	0.86	1.00	0.220	0.020	1.83 × 10−38	Kocarnik et al.21	Yes
rs1205	1	159 682 233	CRP	C	T	0.67	1.01	0.170	0.010	1.03 × 10−31		Yes
rs1800947	1	159 683 438	CRP	C	G	0.94	0.85	0.300	0.030	3.1 × 10−25		No
rs2650000	12	121 388 962	HNF1A	C	A	0.65	1.00	0.120	0.010	2.62 × 10−23		Yes
rs2228145	1	154 426 970	IL6R	A	C	0.60	0.99	0.100	0.010	1.47 × 10−18		Yes
rs780094	2	27 741 237	GCKR	T	C	0.41	1.00	0.100	0.010	1.53 × 10−16		Yes
rs7310409	12	121 424 861	HNF1A	G	A	0.60	1.00	0.180	0.030	1.57 × 10−10		Yes
rs6857	19	45 392 254	PVRL2	C	T	0.84	0.97	0.230	0.040	2.07 × 10−10		Yes
rs429358	19	45 411 941	APOE	T	C	0.86	0.96	0.240	0.040	2.41 × 10−10		Yes

Chr, chromosome; EA, effect allele; BA, baseline allele; EAF, effect allele frequency; SE, standard error.

^aβ coefficient represents one unit increase in the natural log-transformed CRP (mg/l) per copy increment in the effect allele. Outcome is CRP concentration.

^bSNPs that are in linkage disequilibrium (LD; r²> 0.2) with other SNPs are excluded in the analysis.

Three basic assumptions are made in Mendelian randomization: (i) the genetic marker is robustly associated with the exposure; (ii) the genetic marker is independent of the outcome, given the exposure and confounders of the exposure-outcome association (i.e. the genetic marker has no pleiotropic effect through pathways other than the exposure),;and (iii) the genetic marker is independent of factors that confound the exposure-outcome relation.²⁷ The first assumption was met, since we only included SNPs that were significantly associated with CRP concentrations in GWAS. The second assumption could not be tested directly because CRP measures were not available in our study, but sensitivity analyses were performed for global pleiotropic effects. The third assumption was tested by evaluating the association between SNPs and each potential confounder of the CRP-CRC association among controls. No evidence of violation of this assumption was observed. In addition, if multiple IVs are combined into a single estimate by the inverse-variance weighted (IVW) method, a further assumption is made that the variants provide independent information (i.e. not in LD).²⁸ Furthermore, the statistical association between the risk factor and a valid IV should be strong enough to provide unbiased and precise estimates in finite samples.²⁹ The estimated CRP variance explained by the selected SNPs (R²) was ∼5%. Given the sample size of 53 325 subjects and 19 instruments in our study, the estimated F-statistic was 147.66,^29–31 suggesting strong instruments for the Mendelian randomization analysis.

Statistical analysis

We performed the Mendelian randomization analysis to estimate the causal effect of CRP on CRC risk using inverse-variance weighted (IVW) method, by summarizing SNP-CRP associations from literature and estimating SNP-CRC associations in our study population. Assuming all the prior assumptions previously stated are met, genetic variant $k,(k=1\dots K)$ is associated with an observed $X_k$ mean change in the risk factor per additional variant allele with standard error ${\sigma }_{Xk}$ and an observed $Y_k$ log-odds change in the outcome per allele with standard error ${\sigma }_{Yk}$. Assuming additive effects of SNPs on CRP concentrations, an IVW estimate of the causal effect combining the ratio estimates and standard errors of single SNPs can be computed as²⁸:

$${\widehat{\beta }}_{IVW}=\frac{{{\displaystyle \sum }}_k{X}_k{Y}_k{\sigma }_{Yk}^{-2}}{{{\displaystyle \sum }}_k{X}_k^2{\sigma }_{Yk}^{-2}}$$

and the approximate standard error will be $se\left({\widehat{\beta }}_{IVW}\right)=\sqrt{\frac{1}{{\sum }_k{X}_k^2{\sigma }_{Yk}^{-2}}}$.

The mean change in log-transformed CRP concentration (mg/l) per variant allele and its standard error ($X_k$ and ${\sigma }_{Xk}$) were obtained from previous studies,^14,21 and the effect size of genetic variants on CRC risk were estimated within our study populations. We used logistic regression models to estimate the association between each genetic variant and CRC risk in GECCO and CORECT separately, adjusting for age, sex, genotyping phase and principal components. The estimates from GECCO and CORECT were then combined into a summary causal estimate using fixed-effect meta-analysis if there was no heterogeneity. Random-effect meta-analysis would be used otherwise.

Exploratory stratified analyses were carried out using an a priori list of CRC risk factors and the same regression models, including sex, BMI, smoking, NSAID use, aspirin use, family history of CRC and history of endoscopy. In addition, we evaluated differences by cancer subsites and stages. We also performed sensitivity analysis using Egger regression³² for global pleiotropic effect.

Power calculation

Based on the methods described by Burgess,³³ our sample size of 30 480 CRC cases and 22 844 controls has an estimated 99.4% power to detect the previously estimated causal effect size of CRP [odds ratio (OR) = 1.19]²⁰ at a significance level of 0.05, assuming the SNPs explain a total of 5% variance of CRP based on previous estimates.¹⁴ Alternatively, we have 82.5% power to detect a minimal odds ratio of 1.12^2,3 at a significance level of 0.05, given our sample size.

Results

The mean age of participants was 63.4 years [standard deviation (SD) = 10], and 50.7% were male (Supplementary Table 1, available as Supplementary data at IJE online). A total of 27 SNPs were identified and their associations with CRP concentration are summarized in Table 1. The imputation accuracy (R²) ranged between 0.84 and 1.0. The estimated associations between the 19 SNPs and CRC risk are shown in Figure 1. In pooled analysis combining GECCO and CORECT estimates, rs1260326 (T/C) was associated with higher risk of CRC (P = 7.5 × 10^–4), and rs6734238 (G/A) was associated with lower CRC risk (P = 0.003). No other SNP was statistically significantly associated with CRC.

Figure 1.

Associations between 19 SNPs and colorectal cancer risk. (A) SNP-CRC association in GECCO; (B) SNP-CRC association in CORECT; (C) SNP-CRC association in GECCO & CORECT combined.

Using the 19 SNPs as IVs, we found that one unit increase in the log-transformed genetically elevated CRP concentration (mg/l) was associated with a 4% higher risk of CRC (OR = 1.04; 95% CI: 0.97, 1.12; Table 2); however, the association was not statistically significant (P = 0.256). No heterogeneity was observed between the two consortia (P-heterogeneity = 0.509).

Table 2.

Mendelian randomization estimates of the causal effect of genetically elevated CRP and colorectal cancer risk

Study	n Cases	n Controls	OR (95% CI)a,b	P-value	P-hetc
GECCO	10 644	10 729	1.07 (0.96, 1.20)	0.217
CORECT	19 836	12 115	1.02 (0.93, 1.12)	0.654
Combinedd	30 480	22 844	1.04 (0.97, 1.12)	0.256	0.509

^aFinal set includes 19 SNPs identified from both studies, excluding the SNPs that are in linkage disequilibrium (r²> 0.8).

^bInverse-variance weighted method was used to estimate causal effect of genetically elevated CRP and CRC risk, and corresponding 95% CIs. Odds ratios represent the change in odds of colorectal cancer risk with one unit increase in the log-transformed genetically elevated CRP concentration (mg/l).

^c P-het is P-value for heterogeneity of differences between GECCO and CORECT estimates.

^dFixed-effects meta-analysis was used to combine estimates from GECCO and CORECT.

Genetically elevated CRP concentration was not associated with CRC risk in any of the subgroups defined by sex, BMI, smoking, NSAID use, aspirin use, family history of CRC or history of endoscopy (Table 3). The strength of associations between genetically elevated CRP concentration and CRC risk was similar between subgroups.

Table 3.

Mendelian randomization estimates of the causal effect of genetically elevated CRP and CRC risk by subgroups

Subgroups	GECCO			CORECT			Combined b
	Cases/controls	OR a (95% CI)	P-value	Cases/controls	OR a (95% CI)	P-value	Cases/controls	OR a (95% CI)	P-value	P-het c
Sex
Male	5027/4940	1.12 (0.95, 1.32)	0.172	10 854/6181	1.03 (0.91, 1.17)	0.653	15 881/11 121	1.06 (0.96, 1.18)	0.229	0.431
Female	5617/5789	1.03 (0.88, 1.20)	0.717	8913/5934	1.01 (0.89, 1.16)	0.859	14 530/11 723	1.02 (0.92, 1.13)	0.710	0.876
BMI
Normal	3249/3875	1.18 (0.98, 1.43)	0.087	3672/3139	0.99 (0.82, 1.20)	0.934	6921/7014	1.08 (0.95, 1.24)	0.245	0.205
Overweight	3959/3935	0.99 (0.82, 1.18)	0.872	4637/3503	0.97 (0.82, 1.16)	0.779	8596/7438	0.98 (0.86, 1.11)	0.754	0.936
Obese	2216/1757	1.01 (0.78, 1.31)	0.921	2772/1748	1.19 (0.93, 1.51)	0.171	4988/3505	1.10 (0.92, 1.31)	0.286	0.388
Smoking
Never	4612/5107	1.12 (0.95, 1.33)	0.169	3031/2983	1.02 (0.87, 1.19)	0.836	10 139/9279	1.07 (0.95, 1.20)	0.269	0.399
Ever	5855/5654	1.03 (0.89, 1.20)	0.670	7025/4515	1.08 (0.92, 1.28)	0.342	11 425/9789	1.06 (0.94, 1.18)	0.338	0.681
NSAID use
Yes	2847/3722	1.04 (0.85, 1.27)	0.732	3031/2983	1.05 (0.86, 1.29)	0.635	5878/6705	1.04 (0.90, 1.20)	0.563	0.926
No	5855/5924	1.03 (0.89, 1.19)	0.693	7025/4515	1.02 (0.88, 1.19)	0.790	13 478/10 439	1.03 (0.92, 1.14)	0.639	0.936
Aspirin use
Yes	2182/2882	1.02 (0.81, 1.28)	0.880	2518/2448	1.07 (0.86, 1.34)	0.538	4700/5330	1.05 (0.89, 1.23)	0.582	0.751
No	7034/6695	1.05 (0.92, 1.21)	0.472	6938/4949	1.03 (0.89, 1.20)	0.676	13 972/11 644	1.04 (0.94, 1.15)	0.418	0.852
Non-aspirin NSAID use
Yes	1192/1539	0.97 (0.71, 1.34)	0.864	680/766	1.05 (0.68, 1.62)	0.831	1872/2305	1.00 (0.77, 1.29)	0.990	0.784
No	7681/7682	1.03 (0.91, 1.18)	0.635	9226/6623	1.00 (0.88, 1.14)	0.972	16 907/14 305	1.02 (0.93, 1.11)	0.721	0.752
Family history of CRC
Yes	1715/1312	0.88 (0.64, 1.21)	0.425	2023/1089	0.92 (0.68, 1.25)	0.610	3738/2401	0.90 (0.72, 1.12)	0.358	0.821
No	8299/7835	1.08 (0.95, 1.23)	0.247	8867/6819	1.00 (0.88, 1.14)	0.960	17 166/14 654	1.04 (0.95, 1.14)	0.399	0.427
History of endoscopy
Yes	2842/3809	1.11 (0.90, 1.37)	0.321	7792/3086	1.00 (0.84, 1.19)	0.986	10 634/6895	1.04 (0.91, 1.19)	0.538	0.437
No	5774/4826	1.09 (0.93, 1.28)	0.293	1581/4105	1.15 (0.90, 1.47)	0.272	7355/8931	1.11 (0.97, 1.27)	0.139	0.730

^aInverse-variance weighted method was used to estimate causal effect of genetically elevated CRP and CRC risk, and corresponding 95% CIs. Odds ratio represents the change in odds of colorectal cancer risk with one unit increase in the log-transformed genetically elevated CRP concentration.

^bFixed-effects meta-analysis was used to combine estimates from GECCO and CORECT.

^cP-het is P-value for heterogeneity of differences between GECCO and CORECT estimates.

We also stratified by CRC subsites and stages (Table 4). Genetically elevated CRP concentration was not associated with any subsite of CRC. There was an association between genetically elevated CRP concentration (mg/l) and distant CRC (OR = 1.19; 95% CI: 1.00, 1.42; P = 0.049), but not for local or regional CRC.

Table 4.

Mendelian randomization estimates of the causal effect of genetically elevated CRP and CRC risk by subgroups

Subgroups	GECCO			CORECT			Combinedb
	Cases/controls	ORa (95% CI)	P-value	Cases/controls	ORa (95% CI)	P-value	Cases/controls	ORa (95% CI)	P-value	P-hetc
Subsite
Colon	7662/10 729	1.13 (1.00, 1.27)	0.062	11 964/12 115	1.02 (0.92, 1.13)	0.748	19 626/22 844	1.06 (0.98, 1.15)	0.131	0.199
Proximal	4180/10 729	1.17 (0.98, 1.32)	0.084	5815/12 115	1.01 (0.89, 1.14)	0.907	9995/22 844	1.06 (0.96, 1.17)	0.226	0.215
Distal	3343/10 729	1.12 (0.96, 1.32)	0.155	5448/12 115	1.03 (0.90, 1.17)	0.651	8791/22 844	1.07 (0.96, 1.18)	0.213	0.412
Rectal	2780/10 729	0.92 (0.77, 1.11)	0.385	6617/12 115	1.06 (0.93, 1.20)	0.410	9397/22 844	1.01 (0.91, 1.12)	0.867	0.237
Stage
Local	2653/10 729	1.11 (0.93, 1.32)	0.270	2780/12 115	1.04 (0.88, 1.24)	0.624	5433/22 844	1.07 (0.95, 1.21)	0.263	0.652
Regional	5002/10 729	1.09 (0.94, 1.25)	0.254	7484/12 115	1.04 (0.91, 1.18)	0.586	12 486/22 844	1.06 (0.96, 1.16)	0.242	0.630
Distant	1118/10 729	1.26 (0.98, 1.63)	0.069	1206/12 115	1.13 (0.89, 1.43)	0.320	2324/22 844	1.19 (1.00, 1.42)	0.049	0.524

^aInverse-variance weighted method was used to estimate causal effect of genetically elevated CRP and CRC risk, and corresponding 95% CIs. Odds ratio represents the change in odds of colorectal cancer risk with one unit increase in the log-transformed genetically elevated CRP concentration (mg/l).

^bFixed-effects meta-analysis was used to combine estimates from GECCO and CORECT.

^c P-het is P-value for heterogeneity of differences between GECCO and CORECT estimates.

In sensitivity analysis, we observed no association between genetically elevated CRP and CRC risk using the two SNPs from the CRP gene only. Our results persisted using other Mendelian randomization methods (Supplementary Figure 1 and Supplementary Table 2, available as Supplementary data at IJE online). We also tested for global pleiotropic effect using Egger regression (Figure 2). None of the intercepts was significant (P > 0.05), suggesting no global violation of pleiotropic assumptions.

Figure 2.

Scatter plots of SNP-CRP and SNP-CRC associations for 19 SNPs. (A) GECCO (MR Egger intercept P-value = 0.442); (B) CORECT (MR Egger intercept P-value = 0.224); (C) GECCO & CORECT combined (MR Egger intercept P-value = 0.159).

Discussion

In this large multi-consortium study, we did not find evidence for an association between genetically elevated CRP concentrations and CRC risk among participants of European ancestry. No association was found in subgroups stratified by CRC risk factors. Our results suggest that circulating CRP does not play a causal role in colorectal carcinogenesis.

Our estimate of the CRP-CRC association is smaller than the 12% found in meta-analyses of prospective observational studies that used measured CRP concentrations,^2,3 suggesting that the association between measured CRP concentrations and CRC risk may be partially due to confounding. Our findings are different from the only previous study that used GWAS-identified SNPs as IVs for assessing the relationship between CRP and CRC risk.²⁰ Prizment et al. found a statistically significant 19% higher risk in CRC with a one-unit (mg/l) increment of the log-transformed CRP concentration, but our analysis suggested only a modest non-significant effect size of 4%. The sample size of the previous study was small, with 205 CRC cases diagnosed in 7603 participants. We used a much larger sample size of 30 480 CRC cases and 22 844 controls for adequate statistical power to test for moderate causal association. In addition, two SNPs that were not statistically significantly associated with CRP concentrations in GWAS (P-value > 5 × 10^–8)¹⁴ were included in previous analysis.²⁰ In comparison, we did not include these two SNPs, but rather included one additional SNP that was recently found to be associated with CRP concentration in a large consortium (P-value < 5 × 10^–8)²¹ and was independent of the previous 18 SNPs. Last, there is possibility that the SNPs associated with CRP were also associated with other inflammation-related traits, and led to a spurious positive association between CRP and CRC in the previous analyses.

Our findings are consistent with most prospective cohort studies that reported no causality using multiple SNPs as IVs.^17–19 However, a nested case-control study found that a 2-fold higher genetically determined CRP concentration (mg/l), based on seven SNPs in the CRP gene, was associated with higher CRC risk.¹⁵ However, the effect of genetically elevated CRP was not significantly attenuated after adjusting for measured CRP concentrations, indicating a potentially pleiotropic effect of selected SNPs which could lead to biased estimates of the SNP-CRP relationship in the first stage.

Chronic inflammation is a key predisposing factor in colorectal neoplasia.¹ It has been suggested that chronic inflammation creates a microenvironment that stimulates inflammatory cells to release reactive oxygen and nitrogen species, which could lead to malignant DNA alteration,³⁴ and increase the production of inflammatory cytokines that promote tumour growth.³⁵ As a biomarker of low-grade inflammation, CRP has been proposed to play a role in colorectal carcinogenesis. CRP was found to be a major serum leptin-interacting protein that directly inhibited the binding of leptin to its receptors and its ability to signal in vitro, which resulted in leptin resistance and obesity in vivo.³⁶ Lower concentrations of leptin and circulating adiponectin were also found in patients with CRC and adenomas as compared with controls,³⁷ suggesting the possibility of interaction between CRP and leptin in colorectal carcinogenesis. However, a case-control study reported no association between circulating CRP concentration and pathological measures of colonic inflammation.³⁸ Mendelian randomization analysis also found that CRP concentration itself was unlikely to be a causal factor in coronary heart disease,³⁹ although persistent inflammation was found to be a contributor.⁴⁰ Similar to coronary heart disease, it is possible that chronic inflammation promotes colorectal carcinogenesis through inflammatory mediators other than CRP.

Our study has several strengths. It is the largest study to investigate causality between CRP and CRC risk using genetic variants, and has adequate statistical power to detect a moderate association. It is also the first to explore whether effects differed between subgroups stratified by other CRC risk factors, and cancer subsites and stages. Since we have environmental factors measured in most of the participating studies, we were able to test the assumption that the genetic variants were not associated with confounders of CRP and CRC. In addition, we used a comprehensive set of GWAS-identified genetic variants. Taking advantage of the random assortment of alleles, our results from Mendelian randomization should be less susceptible to confounding and reverse causality compared with observational studies. Therefore, our results provide stronger evidence of non-causality on this topic. Furthermore, we adjusted for principal components, which accounted for potential confounding by population stratification.

There are some limitations. First, two Mendelian randomization assumptions could not be fully tested. Therefore, potential violations of the assumptions cannot be ruled out. Although we tested the assumption of no association between genetic variants and confounders, there are possible unmeasured confounders. We also performed diagnostic tests for the assumption of no pleiotropic effects. Second, we only investigated genetically elevated CRP in relation to CRC risk. Since we included studies from multiple countries, the association between genetic variants and CRP concentrations may be influenced by various environmental factors. Furthermore, there is possibility of selection bias. Compared with the average ages in the GWAS of CRP (range: 31–76),¹⁴ our study participants were slightly older but remained within the range of the GWAS samples. In Mendelian randomization studies, there is also possibility of survivor bias⁴¹ where cases survived long enough and controls remained cancer-free until study recruitment in population-based case-control studies. However, most of our studies were cohorts, and we only included incident cases whose dates of diagnosis were relatively close to the recruitment in case-control studies. Last, our results may not be generalizable to race/ethnicity groups other than European, in which the associations between genetic variants and CRP concentrations and CRC risk may be different.

In summary, we found that genetically elevated CRP concentration was not associated with increased risk of CRC among participants of European ancestry. Our findings do not support a causal role of CRP in CRC risk.

Funding

GECCO (Genetics and Epidemiology of Colorectal Cancer Consortium) is supported by the National Cancer Institute (NCI), National Institutes of Health (NIH), U.S. Department of Health and Human Services (U01 CA137088; R01 CA059045; U01 CA164930). ASTERISK was funded by a Hospital Clinical Research Program (PHRC-BRD09/C) from the University Hospital Center of Nantes (CHU de Nantes) and supported by the Regional Council of Pays de la Loire, the Groupement des Entreprises Françaises dans la Lutte Contre le Cancer (GEFLUC), the Association Anne de Bretagne Génétique and the Ligue Régionale Contre le Cancer (LRCC). COLO2&3 (Hawai’i Colorectal Cancer Studies 2 & 3) is supported by the NIH (R01 CA60987). DACHS (Darmkrebs: Chancen der Verhutüng durch Screening) was supported by the German Research Council (BR 1704/6–1, BR 1704/6–3, BR 1704/6–4, CH 117/1–1, HO 5117/2–1, HE 5998/2–1, KL 2354/3–1, RO 2270/8–1 and BR 1704/17–1), the Interdisciplinary Research Program of the National Center for Tumor Diseases (NCT), Germany, and the German Federal Ministry of Education and Research (01KH0404, 01ER0814, 01ER0815, 01ER1505A and 01ER1505B). DALS (Diet, Activity and Lifestyle Survey) is supported by the NIH (R01 CA48998 to P.A.S). HPFS (Health Professionals Follow-up Study) is supported by the NIH (P01 CA055075, UM1 CA167552, R01 CA137178, R01 CA151993, R35 CA197735, K07 CA190673 and P50 CA127003), NHS (Nurses’ Health Study) by the NIH (R01 CA137178, P01 CA087969, UM1 CA186107, R01 CA151993, R35 CA197735, K07 CA190673 and P50 CA127003) and PHS (Physician’s Health Study) by the NIH (R01 CA042182). MEC (Multiethnic Cohort Study) is supported by the NIH (R37 CA54281, P01 CA033619 and R01 CA63464). OFCCR (the Ontario Registry for Studies of Familial Colorectal Cancer) was supported by NIH (U01 CA074783; see CCFR section above), and additional funding toward genetic analyses of OFCCR was supported by a GL2 grant from the Ontario Research Fund, Canadian Institutes of Health Research and a Cancer Risk Evaluation (CaRE) Program grant from the Canadian Cancer Society Research Institute. PLCO (Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial) is supported by the Intramural Research Program of the Division of Cancer Epidemiology and Genetics and by contracts from the Division of Cancer Prevention, National Cancer Institute, NIH, DHHS. PMH-CCFR (Postmenopausal Hormone Study-Colon Cancer Family Registry) is supported by the NIH (R01 CA076366 to P.A.N.). VITAL (VITamins And Lifestyle) is supported by the NIH (K05 CA154337 to E.W). WHI (Women’s Health Initiative) is supported by the National Heart, Lung, and Blood Institute, National Institutes of Health, U.S. Department of Health and Human Services through contracts HHSN268201100046C, HHSN268201100001C, HHSN268201100002C, HHSN268201100003C, HHSN268201100004C and HHSN271201100004C. X.W. and E.W. were also supported by the NCI (R25 CA094880). CORECT (the Colorectal Transdisciplinary Study) is supported by the NCI under RFA # CA-09–002 as part of the GAME-ON consortium (US NIH, U19 CA148107) with additional support from the NCI grants (R01 CA81488 and P30 CA14089), the National Human Genome Research Institute at the US NIH (T32 HG000040) and the National Institute of Environmental Health Sciences at the US NIH (T32 ES013678). The content of this manuscript does not necessarily reflect the views or policies of the National Cancer Institute or any of the collaborating centres in CORECT, nor does mention of trade names, commercial products or organizations imply endorsement by the US Government or CORECT. CCFR (the Colon Cancer Family Registry) is supported by grant UM1 CA167551 from the US NCI and through cooperative agreements with members of the CCFR and principal investigators of the Australasian Colorectal Cancer Family Registry (US NIH, U01 CA074778 and U01/24 CA097735), University of South California Consortium Colorectal Cancer Family Registry for Colon Cancer Studies (US NIH, U01/U24 CA074799), the Mayo Clinic Cooperative Family Registry for Colon Cancer Studies (US NIH, U01/U24 CA074800), Ontario Registry for Studies of Familial Colorectal Cancer (US NIH, U01/U24 CA074783), Seattle Colorectal Cancer Family Registry (US NIH, U01/U24 CA074794) and the University of Hawaii Colorectal Cancer Family Registry (US NIH, U01/U24 CA074806). The Colon CFR Illumina GWAS was supported by NCI/NIH grant U01 CA122839 and R01 CA143237 to G.C. CPSII (the Cancer Prevention Study-II Nutrition Cohort) is funded by the American Cancer Society. MECC was supported by the NIH, U.S. Department of Health and Human Services (R01 CA81488 to S.B.G. and G.R. The MCCS cohort recruitment was funded by VicHealth and Cancer Council Victoria, GALEON: FIS Intrasalud (PI13/01136). The MCCS was further supported by Australian NHMRC grants 509348, 209057, 251553 and 504711, and by infrastructure provided by Cancer Council Victoria. The NFCCR (Newfoundland Colorectal Cancer Registry) was supported by an Interdisciplinary Health Research Team award from the Canadian Institutes of Health Research (CRT 43821), the NIH, U.S. Department of Health and Human Serivces (U01 CA74783) and National Cancer Institute of Canada grants (18223 and 18226). The Kentucky study was supported by the US NCI R01 CA136726, and the Clinical Investigator Award from Damon Runyon Cancer Research Foundation (CI-8). The Spain study was supported by Instituto de Salud Carlos III, co-funded by FEDER funds – a way to build Europe – grants PI14–613 and PI09–1286. SEARCH was supported by the Cancer Research UK (C490/A16561). The Sweden-Wolk was supported by grants from the Swedish Research Council/Infrastructure grant, the Swedish Cancer Foundation and Karolinska Institute’s Distinguished Professor Award to A.W. The ATBC Study is supported by the Intramural Research Program of the U.S. NCI and by U.S. Public Health Service contract HHSN261201500005C from the NCI, Department of Health and Human Services. The ColoCare-Heidelberg and the ColoCare-Seattle studies were funded by the US NIH (grants 2P30CA015704–40, R01 CA189184 and U01 CA152756), the Matthias Lackas-Foundation, the German Consortium for Translational Cancer Research, and the EU TRANSCAN initiative. ESTER_VERDI was supported by grants from the Baden-Württemberg Ministry of Science. The work at MSKCC (Memorial Sloan Kettering Cancer Center in New York) was supported by the Robert and Kate Niehaus Center for Inherited Cancer Genomics and the Romeo Milio Foundation. D.B. is also supported by University of Melbourne Research at Melbourne Accelerator Program (R@MAP) and NHMRC R.D. Wright Career Development Fellowship. M.S. was supported by the American Cancer Society (Grant number MRSG-17–220–01 – NEC to M.S.), by the 2017 AACR-AstraZeneca Fellowship in Immuno-oncology Research (Grant Number 17–40–12-SONG to M.S.) and by the U.S. National Institutes of Health (NIH) grants [K99 CA215314 to M.S.].