Hypothetical model for c.1100delC-associated breast cancer progression (I, III)

Hallmarks (Figure 2),

¹⁴

C.1100delC itself accounts for both ‘Genome instability and mutation’

and ‘Persisting cell death’, possibly mostly via decreased CHEK2 protein expression.

1, 28, 189, 202, 209, 216

According to the genomic analysis of c.1100delC carrier tumors, the loss of GBP genes on 1p21.3-22.2 could be a novel driver candidate of c.1100delC-associated breast cancer, contributing possibly to three cancer hallmarks and one enabling characteristic, namely ‘Avoiding immune destruction’, ‘Inducing angiogenesis’, ‘Deregulating cellular energetics’, and ‘Tumor-promoting inflammation’.

^283-290

Furthermore, increased activity of olfactory receptor signaling and the non-canonical WNT pathway are presented here as potential candidates promoting the fifth hallmark of ‘Activating invasion and metastasis’.

302, 309, 312

In terms of the remaining hallmarks related to proliferation and ‘Enabling replicative immortality’, c.1100delC carrier tumors do not seem to markedly differ from non-carrier tumors. Estrogen is the most prominent growth factor in mammary tissue and deregulated signaling via estrogen receptor downstream pathways probably is the most important event contributing to ‘Sustaining proliferative signaling’

and ‘Evading the growth suppressors’ in both c.1100delC and non-carrier breast cancers.

^{53, 94}

6.1.7 Is p.(I157T) ‘the first hit’ for germline mutation carriers (II)?

The results from differential gene expression analysis of p.(I157T) carrier tumors could be interpreted to be related to the invasive growth pattern typical for lobular breast cancer, where the normal contacts between epithelial cells sustaining the tissue integrity has been lost. CDH1 expression was lower in p.(I157T) carrier tumors than in non-carrier tumors (II: Figure 3).

Similarly, expression of a gene sets previously reported to respond to CDH1 knock-down (II:

Table S6,

ONDER_CDH1_TARGETS_2_UP,ONDER_CDH1_TARGETS_2_DN

) was consistent in p.(I157T) carrier tumors, confirming that the CDH1 activity in these tumors was reduced. Other gene sets enriched at high or low expression in p.(I157T) vs. non-carrier tumors suggested that the p.(I157T) tumors would have gone through epithelial-to-mesenchymal transition. However, this could have been a reflection of the discohesive growth pattern caused by the loss of CDH1 activity because if p.(I157T) was associated with more aggressive and invasive breast cancer, it should have been seen also in associations with higher grade and worse patient survival, which was not the case here (II: Table 1, Table 3). Lastly, the top genes with higher expression in p.(I157T) included a significantly elevated number of collagen genes, suggesting a higher degree of stromal contamination in the p.(I157T) tumors than in non-carrier tumors, which is another feature associated with lobular breast cancer and, on the other hand, speaks against increased invasiveness of the p.(I157T) tumors.

^{314, 315}

In conclusion, p.(I157T) is associated with diagnosis of lobular breast cancer, and molecular features characteristic of lobular breast cancer are present also in many p.(I157T) carrier tumors with other histologic diagnoses (II).

Lobular breast cancer, the most common ‘special’ histological breast cancer subtype, is

characterized by small, round, unattached cells invading the stroma alone or in single files. Loss

of CDH1 function has been recognized as an early event in lobular breast cancer and suggested as

a characteristic feature aiding in diagnosis of borderline cases.

^315-317

Lobular breast cancer is

largely driven by the lifelong cumulative exposure to female sex hormones. The reproductive risk

factors have a stronger association with lobular breast cancer than with ductal cancer. For example, changes in the use of post-menopausal hormone replacement therapy induced respective fluctuations in the incidence of lobular breast cancer at the same time as the incidence of ductal breast cancer remained essentially constant. Furthermore, lobular breast cancer patients have on average three years higher age at diagnosis than patients diagnosed with ductal cancer, but larger tumor size and more advanced stage.

³¹⁸

This is possibly caused by the fact that lobular tumors are hard to detect by palpation or mammography,

^{315, 318}

giving the tumors more time to develop before detection. Alternatively, the difference in diagnosis age could be explained if the tumor initiating and driving events took place at an older age, whereby the nature of the events would be influenced by the intrinsic biology of the mammary gland at involution favoring the lobular phenotype.

It is tempting to speculate that the differences between c.1100delC and p.(I157T) are mainly related to the age at which the tumor driver events take place. P.(I157T) carriers are diagnosed at an older age (57.9 years) than c.1100delC carriers (54.3 years; II: Table I), and possibly certain types of driver events are more likely to take place in premenopausal than postmenopausal mammary glands. Furthermore, because the effect of p.(I157T) on CHEK2 function is milder, it is possible that the accumulation of somatic mutations is slower in p.(I157T) carrier cells and the total loss of CHEK2 function is likely to take place later than in c.1100delC carrier cells. However, the data included in this work do not support this hypothesis. First, the association between p.(I157T) and lobular breast cancer or molecular-level lobular features cannot be explained by age because the p.(I157T) carriers did not differ from non-carriers either in the BCAC dataset or in the gene expression dataset by age at diagnosis. On the other hand, the c.1100delC carriers in Study I were on average older than the non-carriers. Furthermore, the six risk variants with nominally significant associations with c.1100delC-associated breast cancer were not associated with earlier diagnosis age or premenopausal breast cancer.

⁹

The evidence presented in this work suggests different models for c.1100delC- and

p.(I157T)-associated breast cancer and different roles for CHEK2 in the development of breast cancer for

c.1100delC and p.(I157T) carriers. C.1100delC is likely to be a true initiating factor in the

mutation carrier breast cancers. Compromised CHEK2 activity as a result of lowered expression

causes increased risk of somatic mutations and occurrence of further neoplastic events. Whether

the preneoplastic CHEK2-deficient cells ever develop into a full-blown tumor is possibly

primarily decided by interactions between the preneoplastic cells and the innate and adaptive

immune system. By contrast, the loss of CHEK2 might not be an early or even late event in

p.(I157T) carrier tumors, the cancer being driven by other factors in most cases. As a matter of

fact, a normal level of CHEK2 protein expression was previously detected in four out of five

examined invasive breast cancers from p.(I157T) carriers.

²

Loss of CDH1 function as a result of

large-scale deletion or point mutation has been suggested as an early event in lobular breast

cancer.

^315-317

Interestingly, loss of 22q12.1 covering the CHEK2 gene has been reported by four

studies to be a recurrent event in lobular breast cancer, suggesting that loss of CHEK2 could bring

a growth advantage to lobular neoplasia.

^319-322

This could explain the unexpectedly high number

of p.(I157T) carriers among patients diagnosed with lobular cancer, but raises the question, why

is lobular cancer rare among carriers of c.1100delC and other truncating mutations?

²²⁴

One

possible answer could be the order of events: if CHEK2 function were lost first, consequent loss

57

observation that CDH1 and genes involved in apical junction had higher expression in c.1100delC carrier than in non-carrier tumors. Whereas if CDH1 is lost first, there could be an increased advantage in slightly compromised fidelity in control of cell cycle checkpoints and spindle assembly as a result of p.(I157T) or loss of CHEK2, so that these would be targeted by positive selection.

6.2 Survival of breast cancer patients carrying germline CHEK2 mutations (II) Our analyses on the BCAC large international dataset indicated that the overall or breast cancer-specific survival of p.(I157T) carriers did not differ from survival of non-carriers. However, the risk of locoregional relapse and risk of second breast cancer were marginally increased especially in multivariate models adjusted for conventional clinico-pathological prognostic factors (II: Table 3). Compared with c.1100delC carriers, p.(I157T) carriers had better prognosis irrespective of the analysis endpoint, in agreement with a previous study analyzing the survival of c.1100delC carriers and non-carriers in essentially the same dataset.

²¹⁸

6.2.1 Increased mortality associated with c.1100delC

The analysis endpoints used in Study II are not necessarily connected (II: Table 3), although all refer to adverse events occurring after and presumably as a consequence of breast cancer. Death of any cause is an imperfect estimator of breast cancer-associated mortality since primary breast cancer is rarely a life-threatening disease. Mortality is increased only after metastasis to vital organs. However, in many BCAC studies, death of any cause is the only unbiased estimate available. Breast cancer-associated death or occurrence of distant metastasis would be the best measures of disease severity, but these records were comprehensively provided only by a subgroup of BCAC studies; thus, the number of patients included in these analyses was substantially lower than in the overall analysis, hindering the search for significant associations.

There are at least three important mechanisms mediating survival associations of germline mutations, the most important being intrinsic tumor aggressiveness. Subtyping, grading, and staging all aim to measure the aggressiveness in terms of proliferation and invasiveness.

37, 61, 83, 97

If a mutation predisposes to a particularly aggressive type of cancer, the mutation would also be associated with poor patient prognosis, as in the case of BRCA1, PALB2 and FANCM, whose risk mutations increase the risk of triple-negative breast cancer.

169, 170, 323

Apparent survival association could also be an outcome of either good or poor response to adjuvant therapy. For example, breast cancer patients carrying germline BRCA1 mutations have been reported to have good response to treatment with PARP inhibitors, but poor response to taxane therapy.

^324-326

In order to avoid bias caused by the availability of different adjuvant regimens at the time of cohort recruitment, the treatment choice should be included in retrospective studies of BRCA1-associated patient survival.

A third important factor influencing patient survival via tumor invasiveness and metastatic potential is the control of local and systemic microenvironments, where the immune response plays a crucial role.

32, 33, 327

The survival analyses suggesting that c.1100delC would be associated with increased mortality of

breast cancer patients were adjusted for phenotypic features measuring tumor aggressivity.

²¹⁸

Furthermore, the proportion of poor prognosis subtypes (Luminal B, Basal, and Her2) was lower

in c.1100delC carriers than in non-carriers (II: Table 1). Therefore it seems unlikely that an

especially aggressive tumor phenotype would be cause of reduced survival of breast cancer patients carrying c.1100delC.

The effect associated with treatment choice was not addressed in our study. However, this should be taken into careful consideration when proceeding with analyses on CHEK2-associated breast cancer patient survival in future studies now that the evidence is accumulating for sensitivity and resistance associated with BRCA1.

324-326, 328

Carriers of predisposing germline CHEK2 mutations would be expected to have a similar response to treatment because of the related functions of CHEK2 and BRCA1 in DNA double-strand break repair and in regulation of spindle assembly.

^28,

189, 200

A verified association between CHEK2 and treatment outcome would have implications for a wider spectrum of breast cancer patients than just the mutation carriers since CHEK2 has been reported to be lost in about 20% of unselected breast cancers.

²¹⁶

It is noteworthy that the relatively high frequency of loss of CHEK2 expression in non-carrier tumors could be a confounding factor in treatment outcome analyses, diluting possible effects associated with the germline mutations.

In an ideal situation, the analyses would be adjusted with an immunohistochemical measure of CHEK2 expression in patients’ tumors. Previously, CHEK2 mutation carrier tumors have been reported to have an unfavorable response to anthracycline-based neoadjuvant chemotherapy in a small study including three patients.

³²⁹

On the other hand, a recent study found no difference in survival of c.1100delC carriers treated with anthracycline-based and non-anthracycline-based chemotherapy.

³³⁰

In this study, however, the impact of taxanes was not taken into account. All in all, retrospective studies on treatment outcome should factor in the treatment choice and carry out parallel functional experiments to clarify the mechanisms causing enhanced or reduced patient survival.

No reports on the frequency of infiltrating lymphocytes or the tumor immunogenicity related to germline CHEK2 mutations have been published to date. The findings presented in this thesis (I) suggest that the loss of the GBP gene cluster on 1p21.3-22.2 would complement CHEK2 deficiency associated with germline c.1100delC in development of breast cancer. The loss could possibly regulate the tumor microenvironment, affecting tumor-promoting inflammation and immune surveillance, and thus, contributing to reduced survival of c.1100delC carriers.

6.2.2 CHEK2 mutations and increased risk of local recurrence or new primary tumors Compared with distant metastasis, the causal relation of locoregional relapse or second breast cancer to breast cancer mortality is less clear. These two analysis endpoints could have been intertwined in our analyses, because ‘second breast cancer’ included ipsilateral cases, some of which may have been only local recurrences. Furthermore, some of the events classified as

‘locoregional relapse’ could have actually been new primaries since genomic analyses confirming

or excluding clonality had not been performed. Both c.1110delC and p.(I157T) were associated

with increased risk of second breast cancer. The hazard ratios were consistent with the primary

risks associated with these mutations (II: Table 3),

^3-5

and it is unlikely that the mutations would

cause any surplus risk of second breast cancer.

²¹⁹

The marginally significant finding that carriers

of any of the two CHEK2 mutations had increased hazard of locoregional relapse was in agreement

with a previous report studying c.1100delC in patients treated with breast-conserving surgery and

radiotherapy.

³³¹

The adverse effect of ionizing radiation on CHEK2 mutation carriers had been

suggested also earlier,

³³²

and poor treatment outcome could at least partly explain the observed

59

increase in risk of local relapse. However, this topic would need to be addressed in future studies with more complete records on treatment history.

6.2.3 p.(I157T), lobular carcinoma, and patient survival warrant further research Lobular breast cancer has been associated with reduced long-term survival,

^{333, 334}

and loss of CDH1 expression has been suggested as an independent adverse prognostic factor in breast cancer.

^335-338

With this background, it was surprising that p.(I157T) was not associated with reduced survival, even though it was associated with lobular breast cancer in the BCAC dataset and lowered CDH1 activity in the gene expression analysis. However, in the referenced studies, the poor long-term prognosis of lobular cancer patients was visible only after ten years following the diagnosis, and our analyses may have failed to capture this effect due to an overly short follow-up or a high proportion of prevalent cases among subjects at risk at the later time-points. On the other hand, in the studies of non-lobular tumors, CDH1 downregulation possibly represented a marker for epithelial-to-mesenchymal transition,

³¹⁶

whereas in Study II low CDH1 activity was more likely to be an indication of infiltrating lobular-type growth pattern, as discussed above.

6.3 Common genetic variants in breast cancer risk prediction (III, IV) 6.3.1 PRS could be used in risk stratification of c.1100delC carriers (III)

Higher values of a polygenic risk score (PRS) based on 74 common predisposing variants were associated with increased risk of breast cancer for CHEK2 c.1100delC carriers. The effect size was comparable to previous more stable estimates made in a much larger group of breast cancer cases.

¹⁰

When PRS was used to stratify c.1100delC carriers into categories of high and low lifetime risk, 20% of carriers at highest risk were estimated to have upwards of 30% lifetime risk.

Correspondingly, for 20% of carriers at lowest risk the lifetime risk would be comparable to the population average, about 10%. Further extrapolation of the model suggested that for 10% of carriers at the high end of the PRS distribution the lifetime risk would exceed 40%, which has been considered as the threshold of the high-risk category in Finland.

In Finland, the relevance of c.1100delC in genetic counseling is attributable to its high carrier frequency, 1.4%.

¹

Ten percent of carriers in the high-risk group means 0.14% of all women and about 40 women of each annual birth cohort.

³³⁹

As a comparison, BRCA1 mutations have about 0.2% carrier frequency in many Western populations and BRCA2 mutations slightly higher, about 0.4-0.5% frequency.

³⁴⁰

Thus, using CHEK2 mutation analysis as a part of risk estimation for women with a positive family history would be well founded, especially in combination with the PRS.

6.3.2 No epistatic interaction exists between c.1100delC and the common variants (III) In pairwise interaction analyses of c.1100delC and common predisposing variants, we did not detect any deviation from the assumed multiplicative model (III: Table S4) suggesting that the risk effect associated with the common variants would be roughly the same for c.1100delC carriers and non-carriers. The common variants included in Study III were estimated to explain about 14%

of the disease heritability and the most recent novel loci an additional 4%, which together with the 20% associated with mutations in high- and moderate-risk genes summed up to about 38%.

^9,

153

The ‘missing heritability’ has been suggested to be partly attributable to an epistatic interaction

between loci.

157, 158, 341

According to a hypothetical ‘limiting pathway model’, genetic variation

affecting any single pathway would increase the risk in a linear fashion, whereas parallel variation

in another pathway would confer a rapid increase in risk.

¹⁵⁸

Despite a systematic search, no significant epistatic effect associated with breast cancer predisposition has been discovered thus far.

^{342, 343}

Furthermore, fine mapping and functional studies of the established risk loci have indicated that the common variants per se contribute to breast cancer predisposition by regulating the activity of nearby genes.

261-263, 344-346

However, even though epistasis appears to be rare in breast cancer genetics, this does not mean that there is none. Studies on c.1100delC- and p.(I157T) carrier tumor genomes presented in this work could assist in building models for tumorigenesis associated with these mutations. Furthermore, the discovered driver regions and genes could serve as candidates in further analyses of genetic risk modifiers.

6.3.3 PRS explains part of the increased familial risk (IV)

Our results confirm the hypothesis that some part of familial clustering of breast cancer can be explained by aggregation of common variants with low individual effect sizes.

⁸

The PRS values of both healthy and affected members of breast cancer families were elevated relative to the general population level (IV: Table 1). Furthermore, the magnitude of risk (OR 1.55 [1.26–1.91]

per unit standard deviation) associated with the PRS within breast cancer families, when comparing healthy and affected individuals, was very similar to estimates made between unselected cases and population controls (ORs 1.47 [1.38-1.62] in IV and 1.55 [1.52- 1.58] in Mavaddat et al.

⁸

).

Earlier, the risk associated with a PRS of 22 common variants was studied in Australian breast cancer families. They reported OR 1.88 [0.99-3.25] when comparing the highest quartile against the lowest quartile.

³⁴⁷

The 22 variants known then, in 2011, were estimated to explain about 8%

of excess familial risk of breast cancer,

^{348, 349}

whereas with the 75 variants the proportion of explained heritability rose to 14%.

⁹

Addition of novel predisposing variants enhanced the discriminatory potential of PRS since we reported a comparable risk effect (OR 1.88 [0.93-3.78]) for women at the 80-90 percentile of PRS distribution compared with the average PRS. Now that an additional 65 loci have been confirmed in Oncoarray analyses to be significantly associated with breast cancer predisposition, their incorporation into PRS is expected to improve it further.

¹⁵³

However, although all predisposing variants even those below the significance threshold were included, a large proportion of breast cancer heritability remains unexplained and the PRS incomplete.

In Study IV, BOADICEA score measuring the family history of breast cancer was weakly

In Study IV, BOADICEA score measuring the family history of breast cancer was weakly

In document Genetic modifiers of CHEK2-associated and familial breast cancer (sivua 55-0)