Parent of origin effects on age at colorectal cancer diagnosis
Genomic imprinting refers to a parent-of-origin specific effect on gene expression. At least 1% of genes in the human genome are modulated in this manner. We sought evidence for genomic imprinting in colorectal cancer by studying the ages at diagnosis in the offspring of 2,061 parent–child pairs in which both parent and child were affected by nonsyndromic colorectal cancer. Families were ascertained through the colon Cancer Family Registry [http://epi.grants.cancer.gov/CFR/] from both population-based and clinic-based sources. We found that the affected offspring of affected fathers were on average younger than offspring of affected mothers (55.8 vs. 53.7 years; p = 0.0003), but when divided into sons and daughters, this difference was driven entirely by younger age at diagnosis in daughters of affected fathers compared to sons (52.3 years vs. 55.1 years; p = 0.0004). A younger age at diagnosis in affected daughters of affected fathers was also observable in various subsets including families that met Amsterdam II Criteria, families that did not meet Amsterdam Criteria, and in families with documented normal DNA mismatch repair in tumors. Imprinting effects are not expected to be affected by the sex of the offspring. Possible explanations for these unexpected findings include: (i) an imprinted gene on the pseudoautosomal regions of the X chromosome; (ii) an imprinted autosomal gene that affects a sex-specific pathway; or (iii) an X-linked gene unmasked because of colonic tissue-specific preferential inactivation of the maternal X chromosome.
The transformation of a normal cell into a malignant cell is accompanied by a host of genetic alterations, which, in the final analysis, results in a deviation from the normal gene dosage for critical growth-regulating genes. Gene dosage aberrations can occur by multiple mechanisms such as gene mutations, deletions, duplications, chromosomal losses and gains, and epigenetic dysregulation. One mechanism for regulation of normal gene dosage is genomic imprinting, which refers to a parent-of origin-specific control of gene expression. Based on studies in both mice and humans, it has been demonstrated that biallelic expression is not the norm for all genes.1, 2 For some genes or gene clusters, monoallelic expression is the rule, and the second allele is silenced epigenetically. The expressed gene is normally inherited from a parent of 1 gender while the silenced gene is normally inherited from the parent of the other gender. This parent-of-origin specific effect on gene activity is genomic imprinting.
Silencing of gene expression may be caused by methylation and probably other epigenetic mechanisms. Well studied examples of genomic imprinting include the paternal uniparental disomy for chromosome 11q15 in Beckwith-Wiedemann syndrome and maternal uniparental disomy in Russell-Silver syndrome; the deletion of chromosome 15q11.2 that results in Prader-Willi syndrome if of paternal origin and Angelman syndrome if of maternal origin; and clinical expression of autosomal dominant hereditary paraganglioma, which results in tumors only if inherited through the paternal germline. The Genomic Imprinting Website, http://www. geneimprint.com, reported 10 human chromosomes (numbers 1, 6, 7, 8, 11, 14, 15, 18, 19 and 20) that contain imprinted genes. Though important, only ∼1% of autosomal genes are currently suspected or known to have parent-specific monoallelic expression.3 Luedi et al. have recently published on an informatics-based search method that suggests that imprinted genes may be more common than previously appreciated when assessed by the usual functional studies, identifying 154 new genes that are predicted to be imprinted.4
Given the number of genes that are imprinted and the number of genes implicated in colorectal carcinogenesis plus an observation that many imprinted genes are related to control of cell growth, it is plausible that some imprinted genes are related to risk for colorectal cancer. If this is the case, then there may be a differential risk to the offspring of a father with colorectal cancer as compared to that of a mother with colorectal cancer. In transmitting genes to offspring, the expression of normal genes or those carrying mutations or normal functional variants may be different because of the genomic imprinting through the germline. Specific genes have been evaluated for this type of effect. Loss of imprinting (LOI) of the insulin-like growth factor 2 (IGF2) gene on chromosome 11 has been reported as a somatic change in CRC, with characteristic clinical and genetic features,5 as well as in normal colonic mucosa of 30% of individuals with CRC but in only 10% of healthy individuals.6 Subsequently, LOI in lymphocytes was also reported to be associated with positive family history of CRC, presence of adenomas in the colon and CRC6 raising the possibility that this could be an inherited risk factor for CRC.
To our knowledge, a large study of affected parent-affected offspring pairs with nonsyndromic CRC has been reported only once. Tsai et al. looked at 475 parent–offspring pairs in 308 pedigrees.7 They reported no difference in age of onset in the offspring compared to the parents and found no difference in parent gender on age at diagnosis in the offspring either. They concluded that there was no evidence of genetic anticipation in either HNPCC or non HNPCC families. In this study, we sought to look at age of onset of CRC as a potential indirect indicator of a parent-of-origin effect in colorectal carcinogenesis by evaluating parent–offspring pairs in which both had colorectal cancer.
Material and Methods
CRC cases came from the Colon Cooperative Family Registry (Colon CFR), an NCI-supported consortium established in l997 to create a multinational comprehensive collaborative infrastructure for interdisciplinary studies in the genetic epidemiology of colorectal cancer. Detailed information about the Colon CFR can be found at http://epi.grants.cancer.gov/CFR/, as well as being described in detail by Newcomb et al.8 CFR registries that were used to identify eligible cases were at The Fred Hutchinson Cancer Research Center, University of Hawaii Cancer Research Center, Cancer Care Ontario, Australasia CFR, University of Southern California Consortium, and Mayo Clinic. These sites use various strategies for recruitment such that the entire spectrum of CRC risk is represented. This study included data from participants recruited both from population-based sources and clinic-based sources. Online Supporting Information 1 shows the details of the recruitment criteria for case ascertainment for both the population- and clinic-based sites.
Important note: because of the over recruitment of young onset cases in both the clinic-based series and some of the population-based series in the CFR, this study is not designed to assess intergenerational anticipation and should not be used in this way. The CFR study designs, however, should not bias the ages at diagnosis in the offspring of mothers versus the offspring of fathers, so this is what was studied.
In all, 2,628 parent–offspring pairs with CRC were identified. Those known to be carriers of deleterious gene mutations (Lynch Syndrome or MYH-associated polyposis or Familial Adenomatous Polyposis) were excluded. In addition, those who were part of an Amsterdam Criteria triad9, 10 and whose tumor showed high level of microsatellite instability or has loss of expression for 1 or more of the DNA mismatch repair genes were excluded as probable Lynch Syndrome cases.11 Nine individuals were diagnosed below Age 18, and these were excluded as this likely reflected either a syndrome or reporting error. Age at diagnosis was not available for 225 offspring (123 sons; 102 daughters) and for 336 parents (160 fathers; 176 mothers), so these also were not included. There were 2,061 remaining parent–offspring pairs, representing 1,473 unique pedigrees, included in this analysis. Additionally, families in which there was only 1 parent–child pair were analyzed again separately (n = 1,105), to see if the multiple-affected sibling families were biasing the overall results (perhaps representing unrecognized Mendelian syndromes).
All study participants had provided written informed consent for use of their data and all sites had received approval by the respective institutional research review boards.
Where available, we recorded subsite of CRCs. Subsite was available on both members of 561 parent–offspring pairs. Sites were defined as right/proximal colon (C180-185), left/descending colon (C186-187), or rectal (C199, C209). Detailed data is available as Supporting Information V, VI, VII.
For comparison, we looked at the average ages at CRC diagnosis in the C-CFR for those cases whose parents did not have CRC. The mean age for males was 55.99 years and for females was 55.45 years (Supporting Information VIII).
Statistical analyses were performed using SAS 9.1.3 (SAS Institute, Cary, NC). Differences in the colorectal cancer diagnosis mean age was compared between various groups viat-tests; all p-values were 2-sided. Different group comparisons included: (i) unique parental ages (mothers vs. fathers), (ii) offspring ages (mothers vs. fathers), (iii) difference in diagnosis age between offspring and parent (mothers vs. fathers), (iv) age of parent from families with and without multiple pairs of affected ones (mothers vs. fathers), (v) age of mother (with vs. without multiple pairs), (vi) age of father (with vs. without multiple pairs), (vii) age of female offspring (mothers vs. fathers), (viii) age of male offspring (mothers vs. fathers), (ix) age of mother's offspring (female vs. male) and (x) age of father's offspring (female vs. male).
When comparing unique parental ages and ages of parents from families with and without multiple pairs of affected ones, mothers and fathers were included only once since multiple parent–offspring pairs may have existed in 1 family.
Table 1 shows the overall results of comparing age at CRC diagnosis in 2,061 parent–child pairs. The offspring of fathers were on average younger, but when divided into sons and daughters, this difference was being driven by younger age at diagnosis in daughters of affected fathers compared to all other pairings. The same finding held true for families that met Amsterdam Criteria (Table 2), families with documented normalcy of DNA mismatch repair in tumor (Table 3), in families that failed to meet the Amsterdam Criteria (Table 4). In families with known deficiency of MMR in tumor, this trend was not seen but the numbers were very small (n = 71 pairs; data not shown). These analyses were repeated using only families with 1 affected offspring to see if higher order multiple case families were driving this trend. Although the numbers were smaller (n = 1,105 pairs), the younger age at diagnosis of affected daughters of affected fathers persisted as statistically significant (shown in online Supporting Information II–IV).
Table 1. Comparison in age at diagnosis in 2,061 parent–child pairs with colorectal cancers
Table 2. Comparison in age at diagnosis in 749 parent–child pairs with colorectal cancers in which Amsterdam II criteria was fulfilled
Table 3. Comparison in age at diagnosis in 635 parent–child pairs with colorectal cancers in which DNA mismatch repair (MMR) status was documented as normal in one or both members, using MSI plus IHC in 483 and IHC-only in152
Table 4. Comparison in age at diagnosis in 1,312 parent–child pairs with colorectal cancers in which Amsterdam II criteria was not fulfilled
With regard to tumor site, no single category was completely driving the younger age at diagnosis in the daughters of affected fathers. Statistically suggestive differences (p ≤ 0.05) were noted for multiple groups (see Supporting Information V, VI, VII). The only subset that rose to real statistical significance was that among parents with ascending CRC, the daughters of affected mothers were older than the daughters of affected fathers (66.9 years vs. 55 years [p = 0.007]). A close second was that among parents with rectal cancers, the daughters of affected mothers were again older than the daughters of affected fathers (54.1 years vs. 41.3 years [p = 0.01]).
In this study of 2,061 pairs, selected from the Colon Cancer Family Registry,8 we sought to look at age of onset in affected offspring of affected mothers versus affected fathers as a potential marker for parent of origin effect on development of CRC. These data showed a similar age at diagnosis of CRC among all offspring of affected mothers and this was indistinguishable from age of onset in the sons of affected fathers (study design prohibits looking for anticipation in this dataset). Unexpectedly, the daughters of affected fathers were, on average 3 years younger at diagnosis than all other offspring. Parent-of-origin effects relate to the gender of the parent, not the offspring. Here we have a parent of origin effect (father) affecting age of onset in a gender-specific manner (daughters younger at diagnosis). There is not an obvious explanation based on ascertainment biases that might explain this observation. The ages at diagnosis of the mothers and fathers of sons and daughters affected with CRC were comparable. Although we do not have the data to prove it, it seems improbable that daughters of affected fathers were simply screening earlier or more diligently than all other offspring of affected parents.
Curiously, this sort of finding has been reported previously, albeit in the context of extended Lynch Syndrome kindred in Newfoundland (Lynch Syndrome families were excluded to the best of our ability in this study). Among 12 families with a founder mutation in MSH2, females who inherited the mutation from their fathers had an increased risk of developing colorectal cancers (relative risk 2.5, p = 0.05) and of dying of cancer (relative risk 2.7. p = 0.04), compared with females who inherited the founder mutation from their mother.12 These results, combined with our current study support the existence of a mechanism for modification of colorectal cancer risk in a parent of origin specific manner that is operative in Lynch syndrome as well as in nonsyndromic familial CRC.
What biological factors could explain this aggregation of findings? Several hypotheses could be formulated. One might involve nonrandom X inactivation in colonic tissue. The primary genetic difference between sons and daughters of fathers is that daughters inherit an X chromosome from their fathers, while sons do not. In each cell, 1 of the 2 X chromosomes of females is lyonized (undergoes inactivation). As selection of which X chromosome undergoes inactivation is thought to be a random event, there is equal probability overall of having cells with a maternal X inactivated or a paternal X inactivated. We could not determine from medical literature if this was true in normal colonic tissue, as there has been no reason to investigate this phenomenon. It is conceivable that there could be preferential inactivation of a maternal X chromosome in colonic tissue, so if father carried an X-linked predisposition allele on the X chromosome, the daughters might be at increased risk, developing cancer at a younger age.
A second hypothesis might relate to genomic imprinting of certain genes on the X chromosome itself. In reality, not all of the inactive X chromosome is inactivated: there are 2 pseudoautosomal regions near the telomeres on each chromosomal arm, and the inactivation center is active, transcribing the XIST gene, resulting in the spread of inactivation along most of the inactive X chromosome. Thus, the genes that are not involved in that major inactivation process would be candidates for genomic imprinting just as occurs on the autosomes. To our knowledge, parent of origin expression differences in the pseudoautosomal genes have not been reported. There is, however, a small and intriguing body of literature suggesting that this idea is plausible. Stemkens et al. noted phenotypic differences in men with Klinefelter Syndrome (47,XXY) when the 2 Xs were both of maternal origin compared with men in which 1 of the X chromosomes was of paternal origin: among 54 cases, impaired speech and motor development and increased body size was more common in those with a paternally inherited X suggesting the possibility of X chromosomal imprinting.13 There are conflicting reports on whether or not there are phenotypic differences in women with Turner Syndrome (45,X karyotype) according to whether the retained X is of maternal or paternal origin.14–18 In another study it was noted that male-to-female transsexuals have a significant excess of maternal aunts vs. uncles. The authors hypothesized there may be genes on the X chromosome that escape inactivation but may be imprinted: the first generation would be characterized by a failure to erase the paternal imprints on the paternal X chromosome. Daughters in the next generation would produce sons that are XpY or XmY. Since XpY expresses Xist, due to failure to erase the paternal imprint from generation 1, the X chromosome is inappropriately silenced and half of the sons are lost at the earliest stages of pregnancy. The authors go on to speculate further about how this relates to gender identification. Several imprinted genes on the X-chromosome have been identified in animals (listed at www.otago.ac.nz/IGC) including mouse, Drosophila and sheep, but relevance to humans is unknown. Imprinting of human X chromosomal genes still remains an unproven possibility.
A third possible explanation for our observation is that there may be imprinted genes on the autosomal chromosomes that differentially affect expression in males versus females. Certainly there are a number of gender-related genes that are autosomal (e.g., those that direct expression of estrogenic and androgenic steroids, or receptors for steroids) that could be affected by imprinting resulting in gender-specific differences in offspring. Peters and Robson19 have reported now that micro-RNAs are included in the growing number of genes that are imprinted and expressed differently if maternally versus paternally inherited providing yet another explanation for the observed data in this study.
In conclusion, a large study of parent-offspring pairs affected with CRC, in which efforts were made to exclude single gene disorders, found the affected daughters of affected fathers were, on average, younger than affected sons of affected fathers and affected daughters and sons of affected mothers. This pattern is not consistent with our current understanding of genomic imprinting effect unless it involved the X-inactivation process somehow or if there was imprinting of the pseudoautosomal region of the X or X-chromosomal effects upon expression of sex-related autosomal genes. Similar results have been noted among Lynch Syndrome kindreds, suggesting the presence of a mechanism that is not disease specific. These results need confirmation in an independent dataset before being considered potentially clinically relevant.
The content of this manuscript does not necessarily reflect the views or policies of the National Cancer Institute or any of the collaborating centers in the CFRs, nor does mention of trade names, commercial products, or organizations imply endorsement by the US Government or the CFR.