Abstract
PMS2-associated Lynch syndrome is characterized by a relatively low colorectal cancer penetrance compared with other Lynch syndromes. However, age at colorectal cancer diagnosis varies widely, and a strong genetic anticipation effect has been suggested for PMS2 families. In this study, we examined proposed genetic anticipation in a sample of 152 European PMS2 families.
The 152 families (637 family members) that were eligible for analysis were mainly clinically ascertained via clinical genetics centers. We used weighted Cox-type random effects model, adjusted by birth cohort and sex, to estimate the generational effect on the age of onset of colorectal cancer. Probands and young birth cohorts were excluded from the analyses. Weights represented mutation probabilities based on kinship coefficients, thus avoiding testing bias.
Family data across three generations, including 123 colorectal cancers, were analyzed. When compared with the first generation, the crude HR for anticipation was 2.242 [95% confidence interval (CI), 1.162–4.328] for the second generation and 2.644 (95% CI, 1.082–6.464) for the third generation. However, after correction for birth cohort and sex, the effect vanished [HR = 1.302 (95% CI, 0.648–2.619) and HR = 1.074 (95% CI, 0.406–2.842) for second and third generations, respectively].
Our study did not confirm previous reports of genetic anticipation in PMS2-associated Lynch syndrome. Birth-cohort effect seems the most likely explanation for observed younger colorectal cancer diagnosis in subsequent generations, particularly because there is currently no commonly accepted biological mechanism that could explain genetic anticipation in Lynch syndrome.
This new model for studying genetic anticipation provides a standard for rigorous analysis of families with dominantly inherited cancer predisposition.
Introduction
Lynch syndrome is the most common cause of hereditary colorectal cancer, accounting for 3%–5% of all colorectal cancers diagnosed annually (1). The underlying cause is a heterozygous pathogenic germline variant in one of the mismatch repair genes: MLH1, MSH2 (EPCAM), MSH6, or PMS2. The latter gene is associated with a lower estimated penetrance and thus a markedly lower incidence of cancer. However, PMS2 families show phenotypic variability, with very wide differences in age at colorectal cancer diagnosis. While the mean age of onset for colorectal cancer for PMS2-associated Lynch syndrome is around 60, some PMS2 carriers develop colorectal cancer as early as 23 (2–4). Several external and internal modifiers have been suggested as possible explanations, one of which, genetic anticipation, has been the subject of much debate (5–9). The phenomenon of genetic anticipation is clearly defined in genetic disorders involving trinucleotide repeats such as Huntington's disease, where expansion of the repeat in subsequent generations is a clear precursor of disease (10). However, a mechanism of this type has not been described in Lynch syndrome, which in fact requires a second somatic hit for mismatch repair (MMR) deficiency to occur. A single germline mutation in one of the MMR genes does not confer haploinsufficiency (11).
Nevertheless, genetic anticipation in Lynch syndrome and other dominantly inherited cancer predisposition syndromes has been reported by several groups. If a genetic anticipation effect could indeed be confirmed, it would be of clinical utility in the development of individually tailored surveillance schemes. The only report of genetic anticipation in PMS2-associated Lynch syndrome families found a very strong effect (anticipation of 7.3 years per subsequent generation; ref. 12). However, sample size in that study was small, including only 12 PMS2 families. In the same study, carriers of pathogenic germline variants in other MMR genes showed only small or absent anticipation effects (12). By investigating a much larger cohort of 152 families, our aim was to reassess the possibility of genetic anticipation in PMS2-associated Lynch syndrome.
Materials and Methods
Description of the cohort
Pedigree data on European families carrying a segregating pathogenic PMS2 variant were originally collected from clinical genetic departments between 2009 and 2012, as described previously (4). Further families were collected between 2012 and 2017 and an extensive description is available elsewhere (ten Broeke and colleagues, 2018, in-press at Journal of Clinical Oncology). The PMS2 families included originated from the Netherlands, Norway, Germany, Sweden, Denmark, and Spain. Data collection was approved by the local ethical review board (Leiden University Medical Center Ethics Review Board, protocol ID: P01.019). This dataset consisted of clinically ascertained families where variant analysis was initiated due to (histologic) prescreening by IHC and/or microsatellite instability, usually because a family met Bethesda criteria (13). Data collection from patient records included demographic data, family pedigrees, age and location of cancer diagnosis, polypectomy, and hysterectomy if applicable. When available, clinical and pathologic diagnoses were confirmed using patient records.
Statistical analysis
The outcome of interest was age at first diagnosis of colorectal cancer. The follow-up time was defined as the time elapsed from birth till the first colorectal cancer diagnosis or censoring. Censoring occurred on the basis of last known other cancer diagnosis, death, or administrative censoring at age of last contact with the family, whichever occurred last. Family members with biallelic PMS2 mutations were excluded from the analysis given the severe and markedly different phenotype of these constitutional mismatch repair deficiency (CMMRD) patients. Genetic anticipation was estimated as the effect of generation on a person's hazard for cancer diagnosis, using a shared gamma frailty proportional hazard model:
where tij is the age at first diagnosis of colorectal cancer or the age at censoring for member j in family i, λ0 (tij refers to the baseline hazard, which is left completely unspecified (Cox-type model), β = (β1, β2 contains the main effects of interest, the regression coefficient of second and third generation Z = (Z1, Z2, taking the first oldest generation of each family as reference and u > 0 refers to an unobserved random effect (frailty) shared by the members of the same family. This unobserved heterogeneity shared within families was assumed to follow a gamma distribution (normal frailty was also checked as a sensitivity analysis). γ contains the effect of person-specific covariates X included in a second adjusted analysis, namely sex and year of birth.
Because not all family members were tested for PMS2 variants, mutation probabilities based on kinship coefficients were used as analytic weights to avoid possible testing bias and increase efficiency. Specifically, the weight for individual j of family i, wij = P(mutation|family history of mutation is given by the kinship coefficient between individual j and the closest family member with observed mutation. Mutation probabilities are included as case weights in the corresponding penalized score function provided in the R package survival (14). Remaining ascertainment bias was controlled by excluding the probands and focusing on individuals born before 1950, so that all included individuals were at risk for at least 65 years, hence avoiding potential bias due to right truncation. Statistical significance was established at 5%.
Results
A description of the cohort is given in Tables 1 and 2. The analysis included 637 family members with 123 colorectal cancers (Table 1), divided over three generations (Table 2). After weighing, the estimated number of mutation carriers in the sample is 360. Results of the Cox-type random effects model are given in Table 3, which shows increased HRs in the crude analysis (HR = 2.24; 95% CI = 1.16–4.33 for the second generation and HR = 2.64, 95% CI = 1.08–6.46 for the third generation, respectively). After correction for gender and birth cohort, HR size decreased (half of the crude effect) and was no longer statistically significant (as the corresponding CIs included 1). The adjusted analysis showed a strong effect of year of birth (HR = 1.05; 95% CI = 1.02–1.07), equaling a roughly 5% increase of risk for every year toward the present time. These results suggest that the estimated anticipation effect in the crude analysis is strongly confounded by birth cohort and that the apparent effect of generation is mainly explained by secular trends in colorectal cancer diagnosis. The use of normal random effects instead of gamma provided very similar results in terms of genetic anticipation, sex, and birth-cohort effects (results available in Supplementary Table S1).
Number of families | 152 |
Family members included | 637 |
Mutation status | |
100%a | 176 |
(50%–100%) | 282 |
(25%–50%) | 158 |
(12.5%–25%) | 21 |
Colorectal cancer | |
Number | 123 |
Mean age (SD) | 69.58 (12.94) |
Median age (IQR) | 71 (62–77) |
Number of families | 152 |
Family members included | 637 |
Mutation status | |
100%a | 176 |
(50%–100%) | 282 |
(25%–50%) | 158 |
(12.5%–25%) | 21 |
Colorectal cancer | |
Number | 123 |
Mean age (SD) | 69.58 (12.94) |
Median age (IQR) | 71 (62–77) |
NOTE: Probands were excluded from the analysis.
Abbreviation: IQR, interquartile range.
aConfirmed and obligate carriers.
Generation . | Number . | Median (IQR) . |
---|---|---|
1 | 153 | 1912 (1902–1924) |
2 | 399 | 1927 (1918–1938) |
3 | 85 | 1943 (1937–1950) |
Generation . | Number . | Median (IQR) . |
---|---|---|
1 | 153 | 1912 (1902–1924) |
2 | 399 | 1927 (1918–1938) |
3 | 85 | 1943 (1937–1950) |
Abbreviation: IQR, interquartile range.
. | Crude analysis . | Adjusted analysisa . |
---|---|---|
. | HR (95% CI) . | HR (95% CI) . |
Generation 1 | Reference | Reference |
Generation 2 | 2.24 (1.16–4.33) | 1.30 (0.65–2.62) |
Generation 3 | 2.64 (1.08–6.46) | 1.07 (0.41–2.84) |
. | Crude analysis . | Adjusted analysisa . |
---|---|---|
. | HR (95% CI) . | HR (95% CI) . |
Generation 1 | Reference | Reference |
Generation 2 | 2.24 (1.16–4.33) | 1.30 (0.65–2.62) |
Generation 3 | 2.64 (1.08–6.46) | 1.07 (0.41–2.84) |
aAdjusted for gender and year of birth.
Discussion
The occurrence of genetic anticipation in Lynch syndrome has been a subject of considerable debate and gene-specific effects have been offered as an explanation. After correction for birth cohort, our analysis found no evidence of anticipation in a very large cohort of PMS2-associated Lynch syndrome families. A rise in colorectal cancer incidence as well as lower age at diagnosis in recent decades in the general population has been observed previously (15–17). Reasons for this might include better detection with more sensitive screening methods, lifestyle factors, population-based screening protocols and increased life-expectancy. These factors could also play a role in patients with Lynch syndrome. Other factors that could cause a false genetic anticipation signal that are specific to Lynch syndrome, and other dominantly inherited cancer predisposition syndromes in general, involve the genetic diagnostic process. For example, after identification of the proband, presymptomatic family members are tested and subsequently screened if they carry the PMS2 variant. This might lower age at diagnoses of indolent tumors, which might not have presented itself otherwise. An alternative explanation for false genetic anticipation effect may be that colorectal cancer diagnosis in older generations may have been underreported.
Analysis of dominantly inherited cancer predisposition is potentially influenced by several forms of bias. First, clinically ascertained families are accompanied by a selection bias, as they were selected due to their compliance with clinical selection criteria and are therefore often severely affected, that is, many family members with (colorectal) cancer or an unusually low age at diagnosis. A problem arises when the phenotype is not caused by the pathogenic PMS2 variant alone, but is affected by other modifying factors. This is especially problematic for PMS2, as selection based on, for example, the Bethesda guidelines is influenced by criteria for classic Lynch families involving mainly pathogenic MLH1 or MSH2 variants. In the case of PMS2 variants, it is well documented that variants are at most only moderately penetrant (2, 4), suggesting that PMS2 families selected on the basis of these criteria alone will include many relatively severely affected members. However, due to universal screening for mismatch repair deficiency in all colorectal cancers below age 70 in most Western countries, a rise in unselected PMS2 carriers is expected (18).
A second form of bias that should be considered is testing bias due to the fact that people affected with (colorectal) cancer (at a young age) are more likely to be tested for the presence of a PMS2 variant. Probands (i.e., the first person in the family with a confirmed pathogenic germline PMS2 variant) are the most notable example of this, and all probands were therefore excluded from our analysis. Moreover, we also used analytic weights to model mutation probabilities. For example, first-degree relatives of a confirmed carrier that were not tested were given a weight of 0.5, whereas second-degree relatives had a weight of 0.25. This approach also helped improve the power of the analysis.
Although there is no clear biological rationale for genetic anticipation in Lynch syndrome, alternative explanations besides birth cohort have been proposed in other studies. It is generally accepted that families with Li-Fraumeni syndrome (which strongly predisposes to several forms of cancer) exhibit anticipation that cannot be explained by a birth-cohort effect (19, 20). A recent whole-genome sequencing study of germline DNA in 13 Li-Fraumeni syndrome cases did not find increased DNA copy-number variations, suggesting that CNVs do not mediate the genetic anticipation effect. The authors proposed an alternative model explaining apparent anticipation in which variants from the noncarrier parent influence tumorigenesis in the offspring of TP53 mutation carriers with late onset of cancer (21). In other words, parents with relatively late onset might have offspring that are more prone to tumorigenesis due to inheritance of specific risk increasing variants from the noncarrier parent. Similar mechanisms may also influence cancer age of onset and thus explain variability within families and birth cohorts in Lynch syndrome. Another suggested biological mechanism involves telomeres. Retrospective studies have identified shorter telomeres in colorectal cancer cases versus controls, arguing that shorter telomeres cause chromosomal instability and might therefore lead to cancer. Indeed, shortening of telomeres was also observed in peripheral blood in Lynch syndrome patients affected with colorectal cancer, compared with nonaffected mutation carriers (22). This finding has not been replicated in prospective studies, suggesting that the shortening of telomeres might be the result of the cancer process rather than a causative factor (23).
Ours is not the first study to report bias in anticipation analysis due to birth-cohort effects. Similar results have been found in other genetic syndromes, including a study by Guindalini and colleagues in BRCA1/2 families (24). This study corrected for various types of bias by excluding probands, including mutation probabilities and correcting for birth-cohort. Our analysis followed similar principles and incorporated additional flexibility in the specification of the regression model. Our model is semiparametric, because the baseline hazard is left completely unspecified and is therefore more flexible than the model used by Guindalini and colleagues, which was based on a parametric specification of the underlying time-to-event data distribution (25). Moreover, we have allowed for a more flexible, nonlinear effect of generation, considering two possibly different effects for second and third generations with respect to the first, oldest generation. Previous reports have relied on a linear and perhaps too stringent specification of the anticipation effect. We also used gamma random effects in our main analyses and checked the impact of random effect specification by also considering normal random effects. The results regarding anticipation and birth-cohort effect remained the same. Normal random effect modeling of hazard was previously used by von Salome and colleagues in a study in which the authors reported strong genetic anticipation in 12 PMS2 families (12). However, cohort effects were not considered and a linear specification was assumed for the generation effect. Daugherty and colleagues also used a Cox-type hazard regression method to study anticipation in lymphoproliferative tumors, but adopted a less flexible approach because random effects were not considered and hence family-specific effects could not be captured (26). Nevertheless, these authors also identified a confounding effect of secular trends on apparent anticipation effects of generation.
Regression strategies have previously been shown to be preferable over hypothesis testing based on parent–child pairs (9). Because our regression strategy is flexible, it is possible to reasonably reflect the underlying structure of the data while still getting interpretable results and preserving sufficient power. Boonstra and colleagues have reported genetic anticipation in Lynch syndrome based on an alternative specification that allowed for family-specific anticipation effects (random slopes; ref. 27). Such specification is flexible because it allows for a specific effect of generation in each family, although the effect is linear within families. We have introduced flexibility in a different manner, by allowing for a nonlinear fixed anticipation effect, which is less dependent in the chosen parametric family on random effects. Moreover, Boonstra and colleagues did not directly estimate cohort effects based on the sample, but inferred them from external cancer incidence registries (not specific for Lynch syndrome) on the basis of a piecewise (5-year knots) linear hazard assumption (27). Misspecification in this step may have introduced bias in the estimated anticipation effect. Despite our efforts to account for possible bias in our analysis strategy, the retrospective nature of our data is still a limitation of our study. Similarly, in an effort to avoid ascertainment bias, we excluded some data, leading to a reduction in power. Models that can accommodate right truncated data should be developed and used in this field. A last limitation is that the weights that were used to estimate the probability of carrying the familial PMS2 mutation only took into account degree of kinship, but not the presence of a cancer phenotype, for example, colorectal cancer. Including this factor in the weigh calculation is complicated, given the complex pedigree structure. Moreover, recent work by our own group suggests that the lifetime risk for colorectal cancer is only 2–3 times increased compared with the general population (28), which may cause misspecification of PMS2-associated colorectal cancer as this cancer also occurs frequently in the general population.
In conclusion, after correction for birth cohort, our study did not confirm previous findings of genetic anticipation in patients with PMS2-associated Lynch syndrome. Therefore, anticipation cannot be used in individual risk estimation. Given the large phenotypic variability in patients with Lynch syndrome, future studies should focus on other potential modifiers.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Authors' Contributions
Conception and design: S.W. ten Broeke, M. Rodríguez-Girondo, I. Bernstein
Development of methodology: S.W. ten Broeke, M. Rodríguez-Girondo
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.W. ten Broeke, M. Suerink, S. Aretz, I. Bernstein, G. Capellá, C. Engel, E.B. Gomez-Garcia, L.P. van Hest, M. von Knebel Doeberitz, K. Lagerstedt-Robinson, T.G.W. Letteboer, P. Moller, T.A.M. van Os, M. Pineda, N. Rahner, M.J.W. Olderode-Berends, J. von Salomé, H.K. Schackert, L. Spruijt, V. Steinke-Lange, A. Wagner, C.M.J. Tops, M. Nielsen
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.W. ten Broeke, M. Rodríguez-Girondo, C.M.J. Tops
Writing, review, and/or revision of the manuscript: S.W. ten Broeke, M. Rodríguez-Girondo, M. Suerink, S. Aretz, I. Bernstein, G. Capellá, E.B. Gomez-Garcia, L.P. van Hest, K. Lagerstedt-Robinson, T.G.W. Letteboer, P. Moller, T.A.M. van Os, M. Pineda, N. Rahner, M.J.W. Olderode-Berends, H.K. Schackert, L. Spruijt, A. Wagner, M. Nielsen
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Suerink, C. Engel, M. von Knebel Doeberitz, L. Spruijt
Study supervision: M. Nielsen
Other (contribution of clinical data to prove the basic hypotheses): M. von Knebel Doeberitz
Acknowledgments
This work was supported by the Dutch Cancer Society (grant number UL-2012-5515, to by Dr. Maartje Nielsen).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.