• 7. N-acetyltransferase Genotype
  
• 8. Smoking
  
• 9. Measures of Meat Exposure
  
• 10. Confounding Factors for Multivariate Analysis
  
• 11. Summary
  

7.a. NAT1 Genotype

The putative fast NAT1 allele NAT1*10 has been postulated as increasing the risk of colorectal neoplasia but many of the published papers have failed to show a significant result for NAT1 alone (Table 8).  Neither the Leeds nor the collaborative data showed any significant effect of fast (NAT1*10) acetylator genotype (Table 25, Table 27) or of any specific allele (Table 26, Table 28).

The only study that demonstrated a significant association with NAT1*10 allele was by Bell et al.²¹⁰.  They reported a 1.9 times increased risk in colorectal cancer patients and 2.5 times risk in Dukes' C classification colorectal cancer patients.  It may be therefore that the importance of the NAT1*10 allele is in the adenoma progression to carcinoma and in the dissemination of carcinoma and not in the formation of adenomatous polyps.

The reason why a significant effect of NAT1 alone might be masked is that it is just a single step in what is a multistep event (see 10. Confounding Factors for Multivariate Analysis).

7.b. NAT2 Genotype

Fast NAT2 acetylator status alone was marginally associated with colorectal adenomas in the Leeds data (Table 29, c² = 4.13 p = 0.042).  The collaborative data failed to show this (Table 32, c² = 0.41 p = 0.52)

This overall null result would be consistent with what has been reported in the literature where the association with colorectal cancer and NAT2 phenotype has been shown in several studies (Table 10) but the link between putative fast NAT2 genotypes and colorectal neoplasia has been less convincing (Table 11).

Similarly to the NAT1 results above, it may be that fast NAT2 genotype is more important in progression of adenomas to carcinoma rather than adenoma formation itself or the influence of NAT2 is masked by the contribution of additional polymorphic enzymes (see 10. Confounding Factors for Multivariate Analysis).

7.c. NAT1 / NAT2 Genotypes

The hypothesis for a gene-gene interaction between fast NAT1 and fast NAT2 is an attractive one but there has not been any convincing evidence to confirm this from the literature^210,211,214.

Having looked at the effects of combining putative fast genotypes of NAT1 AND NAT2 compared to fast NAT1 OR fast NAT2 it was found that there was a significantly increased risk within the Leeds data of having an adenoma with either fast NAT1 OR fast NAT2 (Table 35, c² = 7.23 p = 0.007).  Therefore, subsequent analysis combining the NAT1 / NAT2 genotypes looked at fast NAT1 OR NAT2.

Biologically, this could be explained in that the NAT enzymes catalyse the formation of carcinogens from procarcinogens and this would either be as a first pass effect through the liver upon absorption from the upper gastrointestinal tract with subsequent systemic recirculation (NAT2) OR as a direct effect of colonic enzyme activity (NAT1).  Fast activity for both was uncommon and did not seem to convey any greater risk then fast for NAT1 or NAT2.  Put another way, the absence of fast acetylation at both NAT1 and NAT2 is protective.

The collaborative data did not support the finding in Leeds and showed no significant difference in adenoma risk for fast NAT1 or fast NAT2 (Table 36).  The data for the group from Portsmouth was almost the opposite from the Leeds data (i.e. the group of fast NAT1 or fast NAT2 was over-represented in the control group as opposed to the case group) but this was not significant and may just represent random variation around an overall null effect.  This would be in keeping with a similar flexible sigmoidoscopy based study by Probst-Hensch et al.²¹⁴ in 1996 that showed no significant NAT1 / NAT2 gene - gene interaction.

Failure to show a significant difference in combining genotypes of one group of metabolic enzymes may just represent that there are many factors that contribute to the overall effect such as substrates and additional metabolic genes involved in eventual carcinogen metabolism or that the differences are small and therefore the sample size would need to be very large in order to achieve significance (see 10. Confounding Factors for Multivariate Analysis).

8.a. Smoking

The effect of smoking has been shown to significantly increase the formation of colorectal adenomas in the left colon in Leeds (Table 37 - Table 40) and in the collaborative data for all 3 centres (Table 41 - Table 44).  This is true for both current and ex-smokers though the latter seem to have a slightly smaller risk (Table 38, Table 42) - this might be due to the overall consumption of cigarettes or it is possible that some adenomas might regress after giving up smoking.

In addition, there is a trend with longer duration of smoking increasing the adenoma risk (Table 40, Table 44).  There was an overall 1.71 and 2.59 times increased risk in Leeds and the collaborative data respectively for those who had smoked for greater than 35 years compared to non-smokers.

There were differences in the effect of smoking between the sexes in Leeds with women demonstrating a statistically significant increased susceptibility to cigarettes (LR test for interaction between smoking and sex: c² = 10.1, p = 0.006) but this was not true of the collaborative data (LR test for interaction between smoking and sex: c² = 3.72, p = 0.16).

Our data is in keeping with most of the published literature on smoking and colorectal adenomas (1.c.iii. Environmental Factors - Smoking).  The influence of gender on the increased adenoma prevalence attributable to smoking however has not been clearly defined.  Giovannucci et al. in a large cohort study of American men and women^162,163 showed that the effect of prior smoking was greater in the male population as opposed to females as in our study.  It has been postulated that smoking is a surrogate marker for other factors such as diet that may in part contribute to absolute risk¹⁶⁴.  In the study by Giovannucci et al., correction for age, intake of saturated fat, dietary fibre, folate, alcohol, body mass and family history of colorectal cancer was undertaken which may in part explain these differences.  Within the confines of the sample size in our study and the simplistic nature of the food frequency questionnaire it was not possible to do this with our data - the data from the whole of the FlexiScope Trial (40,000 volunteers) will be analysed to investigate further the role of smoking.

8.b. Smoking and NAT1 Genotype

It is recognised that smoking is associated with an increased risk of adenoma formation (1.c.iii. Environmental Factors - Smoking) and the N-acetyltransferase enzymes (NAT1 and NAT2)  are involved in the activation of tobacco-derived procarcinogens and carcinogens.  Furthermore, the NAT1*10 allele has been found to increase the risk of lung cancer in smokers²⁷³.  The same association however has not been shown for colorectal cancer^214,245,274

Within the Leeds data there was no evidence to suggest a significant relationship of NAT1 acetylator status with smoking and adenoma risk (6.b.i. Leeds Smoking and NAT1 Genotype).  The collaborative data, however, consistently demonstrated higher odds ratios for adenoma risk when comparing slow NAT1 with fast NAT1 in the presence of different levels of cigarette exposure (Table 46) (6.b.ii. Collaborative Smoking and NAT1 Genotype).  These differences did not reach significance but may represent a trend.  Obviously, it is dangerous to read too much in to this observation.  The suggestion may be therefore that there is some marginal evidence to support the plausible biological theory that fast NAT1 increases adenoma risk in the presence of high exposure to aromatic amine tobacco-derived carcinogens.  The group at highest risk appears to be those females who have smoked for longer than 35 years (odds ratios 2.91 and 5.28 for slow and fast NAT1 respectively) though the reason behind this sex difference is unclear.

Obviously, if there is a risk attributable to fast NAT1 then this increase is only small as it would otherwise be clearly demonstrated from the number of observations presented herein.  It may be that the action of N-acetlytransferase is only one step in the metabolism of potential harmful compounds (Table 3) and that it will be the combination of genotypes of the multitude of metabolic enzymes and substrates that will be more important (see 10. Confounding Factors for Multivariate Analysis).

8.c. Smoking and NAT2 Genotype

There is little evidence to support a significant association of NAT2 genotype and smoking on the risk of colorectal adenomas (1.c.iv. N-acetyltransferases - NAT genotype, colorectal neoplasia and smoking) and the two largest studies determined that the main risk factor for adenomas was smoking and not NAT2 genotype^234,239(Table 13).

Our data supports this finding in that in neither the Leeds data nor the collaborative data is there a significant effect for NAT2 genotype but smoking is a strong risk factor as above (6.a. Smoking).

It has been suggested that the effect of smoking is not necessarily to provide carcinogens as a substrate for NAT2 but that it induces CYP1A2 (another enzyme involved in xenobiotic metabolism) and therefore enhances the effect of this enzyme in conjunction with NAT2 in the metabolism of red meat derived carcinogens²⁷⁴.  Therefore, we would not expect to see a direct effect of NAT2 and smoking.

8.d. Smoking and NAT1 / NAT2 Genotypes

There was no significant effect of the combined fast NAT1 or fast NAT2 genotype and smoking on adenoma risk

The literature shows few significant results when analysing the interaction of NAT1 / NAT2 genotype and smoking.  Our findings mirror a recent paper by Tiemersma et al.²⁴⁵ albeit they looked at colorectal cancer, in that neither NAT1 nor NAT2 individually or in combination modified the association of smoking and colorectal neoplasia.

One potential difficulty of such subgroup analysis is it reduces the number of observations as with each variable the subgroup under scrutiny is getting smaller.  In our study the total number of matched pairs was 887 which is considerably larger than any other study that has looked at NAT1, NAT2, smoking and adenomas but despite this some of the analyses have small groups e.g. fast NAT1 or fast NAT2 in females who had smoked for greater than 35 years had only 29 observations.

Heterocyclic amine exposure from well cooked red meat is thought to be a consistent factor in the aetiology of colorectal neoplasia (

1.c.iii. Environmental Factors - Red Meat).

Our data showed no significant increased risk of adenoma for overall meat consumption and red meat consumption including regression analysis for liking well-cooked red meat.  This was true in both the Leeds data and collaborative data overall.

One criticism of the data collection is that food frequency questionnaires can introduce errors when compared to more accurate methods of dietary data collection²⁷⁵.  The food frequency questionnaire used was based on the European prospective study of diet and cancer (EPIC) questionnaire^270,276 but was deliberately truncated to minimise form filling so as not to reduce attendance to the FlexiScope trial by putting people off.  In this respect then there may be inaccuracies within the data because the questions were deliberately broad (Appendix 12: Food Frequency Questionnaire).  It is planned to assess this by comparison of the dietary data held on those volunteers who attended for screening who were already taking part of the EPIC.

Also, the British diet is relatively homogeneous in that there were only small differences in average consumption with most people being clustered in the moderate consumption group with few outliers.  This would tend to make it difficult to show a significant difference without very large numbers.

9.a. Measures of Meat Exposure and NAT1 Genotype

The data from Leeds (7.a.i. Leeds Measures of Meat Exposure) showed a significant interaction between NAT1 and number of portions of red meat per week (LR test for interaction between NAT1 and red meat: c² = 8.7, p = 0.03) but there was no interaction seen with the other measures of heterocyclic amine exposure (meat portions and liking for well-cooked meat). 

The collaborative data however failed to show any significant interaction between NAT1 and any measure of meat exposure.  Even in a conditional logistic regression model investigating the interaction of NAT1 phenotype, red meat consumption and a liking for meat "well cooked" (i.e. the greatest heterocyclic amine exposure) there was no significant finding (c² = 12.7, p = 0.12).

Interpreting these results, it would seem unlikely that there is a significant interaction between NAT1 genotype and red meat consumption in Leeds that would not otherwise be apparent in the other centres.  Logically, therefore, the Leeds result is probably spurious.  Looking at the raw data (Table 59) then there are odds ratios below unity in each of the slow NAT1 groups who eat more than 2 portions of red meat per week - this is at odds with the proposed mechanism of neoplasia where one would expect increasing risk with increasing red meat consumption.  The collaborative data does not show any significant difference in the either slow or fast NAT1 acetylator status for increasing red meat consumption and no interaction between NAT1 and red meat.  Therefore it is unlikely that there is a true effect in Leeds for NAT1 and red meat.

9.b. Measures of Meat Exposure and NAT2 Genotype

There was no interaction between NAT2 fast putative phenotype, measures of meat exposure and adenoma risk in either the Leeds or collaborative data.

Much of the literature on NAT2, measures of meat exposure and colorectal neoplasia have focused on colorectal cancer and none of the studies have shown a positive interaction for colorectal adenomas alone (Table 12).  Although the study from Lang et al.⁷⁶ combined colorectal cancers and adenomas together, the association of NAT2 phenotype, red meat and colorectal neoplasia was only significant when the fast NAT2 phenotype was combined with fast CYP1A2 phenotype.

It may be that the N-acetylation performed by NAT2 is more important in producing carcinogens that cause the genetic mutations that occur later in the adenoma-carcinoma sequence (1.b.iii. The adenoma / carcinoma sequence) and therefore are responsible for the transformation of adenomas in to cancer and not greatly in the aetiology of adenomas themselves (see 10. Confounding Factors for Multivariate Analysis).

As was highlighted by the study by Lang et al.⁷⁶ that investigated the combined effects of fast NAT2 and fast CYP1A2, it may not be the effect of a single polymorphic gene for a metabolic enzyme but rather the combination of genes that is important.

9.c. Measures of Meat Exposure and NAT1 / NAT2 Genotypes

Recent studies have failed to demonstrate a significant interaction between NAT1 / NAT2 genotype, measures of heterocyclic amine exposure from meat and colorectal neoplasia^244,245,274.  Chen et al. in 1998²¹¹ demonstrated that among the group of men aged over 60 who were rapid acetylators for both NAT1 and NAT2, consumption of >1 serving of red meat per day was associated with a relative risk of colorectal cancer of 5.82 (95% CI, 1.11-30.6) compared with consumption of < or = 0.5 serving per day (P, trend = 0.02).  But this high risk group consisted of 7 cases and 3 controls only.

Our data showed no interaction between fast NAT1 or fast NAT2 putative phenotype, measures of meat exposure and adenoma risk in either the Leeds or the collaborative data.  Potential reasons for this are dealt with in chapter 10. Confounding Factors for Multivariate Analysis.

There are several reasons why the hypotheses regarding NAT genotype and environmental risk in the form of smoking and heterocyclic amine exposure may fail to be proved but might still be valid:

• Sample size - subgroup analysis with multiple variables reduces the number of observations in each group and therefore reduces the power of the analysis (see 6.d. Smoking and NAT1 / NAT2 Genotypes). See also 10.a. Sample Size and Power below.
• Food frequency questionnaire - it is well recognised that food frequency questionnaire's can be inaccurate compared to food diaries (see 7. Measures of Meat Exposure)
• Additional polymorphic genes - N-acetyltransferase 1 and 2 are but 2 genes in a complex system of detoxification(see Table 3).  It may be that the combination of several or many genes truly dictate the absolute risk attributable to high risk diets
• Adenoma / carcinoma sequence - In keeping with the model for carcinogenesis (Page 44), it may be that the role of NAT1 and NAT2 acting on smoking or meat-derived heterocyclic amines as a substrate is to produce carcinogens that cause progression of adenomatous polyps to carcinomas and therefore little or no effect will be seen in a study on adenomas.
• The primary hypothesis is incorrect

A recent meta-analysis by Houlston and Tomlinson²⁷⁷, demonstrated no strong evidence from the combined data in the literature except for the positive association of fast NAT2 determined by phenotyping and colorectal cancer.  Despite this they did not reject the primary hypothesis that inter-individual differences in the risk of colorectal neoplasia may be in part attributable to polymorphic variation in a wide variety of genes that code for metabolic enzymes.  Instead, study design, large sample size, multiple gene analysis and adequate matching of cases and controls were cited as the way forward.

10.a. Sample Size and Power

The power calculations for this research were based on single exposure variables e.g. NAT2 genotype and were performed on the Leeds dataset alone (2.f. Collaboration with other centres).  Collaboration with 2 other centres increased the sample size and therefore the power to look at these variables but no calculations were performed prior to this research to assess the power to combine exposure variables and to look for interaction.

Using NAT2 fast / slow genotype and 'ever smoked' as an example of 2 binary statistics analysed in our data, the power was calculated using Power Program v3.0.0 software downloaded from The Division of Cancer Epidemiology and genetics website (www.dceg.cancer.gov).

In the example below 2 values for theta (OR for interaction; the proportional interaction parameter at unit increases in each exposure) were used to highlight the loss in power when the interaction term is small.  All other parameters were taken from our data:

Case / control pairs = 887          Adenoma incidence = 14%

Smokers = 61%                           Fast NAT2 = 39%

Power = 91% if theta = 2.0

Power = 51% if theta = 1.5

A calculation of sample size was then performed to give 80% power for theta of 1.5:

Case / controls pairs = 1790

It can be seen therefore that in order for a study to investigate small interactions between 2 variables the sample size needs to be considerable.  Similarly, the more variables that our analysed the greater the sample size required to test for interactions between multiple variables.

10.b. The Future of Polymorphism Research

As I have shown in 10.a. Sample Size and Power, the failure to demonstrate significant gene-environment interactions, both in our study and other published series, may in part be due to a lack of power.  Only 2 studies^234,236 with regard to NAT genotype have had more than 1000 cases and controls and it seems likely that future studies that wish to explore gene-gene and gene-environment interactions should look to recruit many thousands of cases and controls.

The field of polymorphism research is an exciting one with great potential to further our knowledge of the biology of cancer and also identify high risk groups for disease.  There is a need for a collaboration on an even bigger scale than the research presented herein.

Such a collaboration would:

• Enable very large sample sizes to ensure the power to detect small differences and gene-gene and gene-environment interactions
• Allow careful study design to ensure adequate control groups
• Ensure fidelity in polymorphism analysis due to the experience and expertise gained from performing genotyping in a large numbers
• Be a definitive work that would refute or support smaller studies that may be subject to type 1 error and publication bias

Ultimately, there may come a day when enough is known of our genotype to reliably predict disease risk according to the lifestyle we choose to lead.  This knowledge may result in behaviour modification (e.g. through not smoking or dietery restriction) or it would help identify at risk groups that could be offered screening.