PMC:4564992 / 28626-43932 JSONTXT

{"project":"2_test","denotations":[{"id":"26320892-22738121-2052542","span":{"begin":91,"end":93},"obj":"22738121"},{"id":"26320892-22138821-2052543","span":{"begin":255,"end":257},"obj":"22138821"},{"id":"26320892-23269371-2052543","span":{"begin":255,"end":257},"obj":"23269371"},{"id":"26320892-11315092-2052544","span":{"begin":3057,"end":3059},"obj":"11315092"},{"id":"26320892-21181895-2052545","span":{"begin":4377,"end":4379},"obj":"21181895"},{"id":"26320892-21181895-2052546","span":{"begin":5922,"end":5924},"obj":"21181895"},{"id":"26320892-21181895-2052547","span":{"begin":7576,"end":7578},"obj":"21181895"},{"id":"26320892-24571439-2052548","span":{"begin":8422,"end":8423},"obj":"24571439"},{"id":"26320892-24571439-2052549","span":{"begin":11304,"end":11305},"obj":"24571439"},{"id":"26320892-23297363-2052550","span":{"begin":12961,"end":12963},"obj":"23297363"},{"id":"26320892-23297363-2052551","span":{"begin":15304,"end":15306},"obj":"23297363"}],"text":"Results\nHere, we present results using the latest versions of our programs PREMIM and EMIM,21 which have been updated to incorporate haplotype estimation when modeling parent-of-origin effects. PREMIM calls (if requested) the software package SHAPEIT2,31,32 and EMIM then incorporates the estimates of parent of origin of alleles provided by SHAPEIT2 into its own multinomial modeling procedure to increase power.\n\nPower for Case-Parent Trios\nThe power to detect a maternally inherited imprinting effect (i.e., in which the allele inherited from the mother increases disease risk) with 1,500 simulated case-parent trios is shown in Figure 1. Powers are presented for varying values of the imprinting parameter Im (representing the risk factor conferred by the maternal risk allele) with or without additionally conditioning on (i.e., allowing for) child-genotype effects. In each case, the top line shows the power of EMIM when using known parent of origin (calculable here because this is simulated data), the middle line shows the power of EMIM when using haplotype estimation in SHAPEIT2, and the lower line shows the power when using the previous version of EMIM. It can be seen that haplotype estimation increases the power and provides a level of power that is not too far from the maximum power achievable when using the known parent of origin of alleles.\nWe investigated the use of different options within SHAPEIT2 to try to improve the accuracy of estimation of haplotypes and thus of parent of origin. For trios, it was found that no options provided higher power than using the most basic Markov chain Monte Carlo (MCMC) options in SHAPEIT2 (“−burn 1 −prune 1 −main 1”) and the default options for the size of the SNP window (2Mb) and the states (“−window 2 −states 100”) (results not shown), and these options provided adequate control of type I error (Figure 2, Figure S2). We also investigated sampling from the posterior distribution of haplotypes within SHAPEIT2 in order to allow for haplotype uncertainty, but this did not improve either power or type I error in comparison to just using the most likely haplotype configuration (results not shown).\nThe type I error obtained with the new versions of PREMIM and EMIM was compared with that obtained with previous versions of PREMIM/EMIM and with that obtained when parent of origin is known. All three approaches gave the same levels of type I error and no signs of inflation (Figure S2). Quantile-quantile (Q-Q) plots of the test statistics from the new versions of PREMIM and EMIM (derived from 5,000 simulation replicates under the null hypothesis, each consisting of 1,500 case-parent trios and 200 SNPs) are shown in Figure 2. The results shown are for tests of maternally inherited imprinting effects, maternally inherited imprinting effects conditional on child-genotype effects, maternally inherited imprinting effects conditional on maternal-genotype effects, and maternally inherited imprinting effects conditional on both child and maternal effects. The observed genomic control36 inflation factors are all approximately 1.0, as expected for a well-calibrated test.\n\nPower for Case-Mother Duos\nIn many studies, genotype data for case subjects and their mothers are available, but no genotype information is available for the father. We used the same approach as used above for case-parent trios to investigate the powers to detect maternally inherited imprinting effects and maternally inherited imprinting effects conditional on child effects, given genotype data for 1,500 case-mother duos. Results are shown in Figure 3. In general, the powers are seen to be lower than when using case-parent trios because there is less information available to determine the parent of origin of the child’s alleles.\nThe top lines of both plots in Figure 3 show the power of EMIM when using known parent of origin (calculable here because this is simulated data). It is interesting to note that the powers are slightly less than those seen with case-parent trios, even though in both scenarios we have the same number of cases and perfect information on the parent of origin of the alleles of interest. This loss of power is most likely due to there being less information available to estimate the nuisance parameters (the minor allele frequency and the parental mating parameters μ1,…,μ6, see Ainsworth et al.20).\nThe middle lines of both plots in Figure 3 show the power of EMIM when incorporating haplotype estimation, with the cell counts either unadjusted or adjusted by PREMIM (see Material and Methods for description of adjustment procedure). It can be seen that the use of estimated parent of origin through haplotype estimation provides a substantial increase in power in comparison to that of the previous version of EMIM (the bottom line of both plots), which effectively “averages” over parent of origin. The greater improvement in power from haplotype estimation seen for case-mother duos in comparison to that for case-parent trios is due to the larger proportion of families in which the parent of origin of alleles is ambiguous (for case-mother duos, only the mother and child must be heterozygous for parent of origin to be unobserved, whereas for case-parent trios, all three individuals must be heterozygous). The power when using adjusted estimates (see Material and Methods) shows a slight improvement over the power when using unadjusted estimates, owing to the fact that the estimation of counts of duos falling into the different parent-of-origin categories is improved through the adjustment procedure.\nFigure S3 shows the type I error of the different EMIM tests, and dashed lines show the expected FWERs if the eight SNPs in the simulated haplotype are (conservatively) considered to be independent. The type I error of the previous version of EMIM (labeled “unknown”), which has been extensively evaluated by Ainsworth et al.,20 acts as a guide to the “correct” type I error rate, allowing for dependency between the SNPs. The type I error when using SHAPEIT2 with fast parameter options is very inflated. This is illustrated further in the Q-Q plots (Figure 4) in which the genomic control inflation factor λ = 1.653. However, the inflation factor can be reduced to 1.052 when using PREMIM’s adjustment procedure. The improvement obtained from adjustment when using the default (slower) parameter options in SHAPEIT2 is not as pronounced: the inflation factor is 1.146 for the unadjusted analysis and reduces to 1.061 when the counts are adjusted (see Figure 4).\n\nUtility of Paternal Data for Detection of Maternally Inherited Imprinting Effects\nAlthough case-mother duos are a more commonly used unit than case-father duos, in real studies, both types of duos might be collected. In Figure S4, it can be seen that, somewhat counter-intuitively, when performing a test of the parameter Im in the presence of genuine maternally inherited imprinting effects, more power is gained from a sample of case-father duos than from a sample of case-mother duos. (Both types of duos provide less power than case-parent trios.) This is true whether or not haplotype estimation with SHAPEIT2 is performed. This observation can be explained by the observation that, to detect maternally inherited imprinting effects, we are only interested in determining whether or not the child inherited a risk allele from the mother. The expected proportion of case-mother duos in which the parent of origin can be determined unambiguously can be calculated from column five of Table 3 of Ainsworth et al.;20 a similar calculation can be performed for case-father duos. Performing this calculation, we find that, provided the allele frequency of the maternally transmitted allele that increases disease risk is \u003c 0.5 (i.e., the “risk” allele is the minor allele), a higher proportion of case-father duos than case-mother duos allows unambiguous determination of the parent of origin (data not shown). If the major allele is the risk allele, this is reversed, and it is the case-mother duos that provide the greater power. Similarly, for a paternally inherited imprinting effect, provided the risk allele is the minor allele, it is the case-mother duos that provide the higher power; if instead the risk allele is the major allele, case-father duos provide higher power.\n\nSLI Data\nWe re-analyzed an updated version of the dataset of Nudel et al. (2014),6 who had presented evidence for a maternally inherited parent-of-origin effect on chromosome 5 and a paternally inherited parent-of-origin effect on chromosome 14. Re-analysis of the chromosome 5 data in PREMIM and EMIM without using haplotype estimation in SHAPEIT2 gave a minimum p value of 1.29 × 10−7 at rs10447141, very similar to that seen in the original analysis by Nudel et al. (p value = 1.16 × 10−7). Using PREMIM and EMIM with haplotype estimation in SHAPEIT2 gave a less significant p value of 6.18 × 10−5 at the same SNP. Plots of the two analyses, with and without use of SHAPEIT2 to estimate the parent of origin of alleles, are shown in Figure 5 and show a considerable decrease in significance of the most significant p values. The p values in the implicated region when using estimated parent of origin now provide only weak evidence of association as a result of a maternally inherited imprinting effect.\nRe-analysis of the SLI data on chromosome 14 without use of haplotype estimation in SHAPEIT2 gave a minimum p value of 2.29 × 10−8 at rs4280164, very similar to that seen in the original analysis by Nudel et al. (p value = 3.74 × 10−8). Using PREMIM/EMIM with haplotype estimation in SHAPEIT2 gave a less significant p value of 1.32 × 10−7 at the same SNP. Plots of the two analyses, with and without estimated parent of origin, are shown in Figure 6. These plots show a general decrease in significance of the most significant p values when using haplotype estimation in SHAPEIT2, although the results (see bottom plots of Figure 6) do still provide positive evidence of association due to a paternally inherited imprinting effect.\nTo investigate the cause of the decreases in significance seen in the SLI study when incorporating haplotype estimation, we examined the counts of trios and duos falling into the various categories used in EMIM’s multinomial modeling procedure (see Table 1). Cells 9 (for trios) or 4 (for duos) correspond to the ambiguous categories in which all individuals are heterozygous; when using haplotype estimation in SHAPEIT2, these cells are decomposed into cells 9a and 9b, or 4a and 4b, respectively, in which parent of origin has been (probabilistically) determined. For SNP rs10447141 on chromosome 5, there are 16 ambiguous case-parent trios, which result (when using haplotype estimation) in estimates of six case subjects receiving the risk allele from the father and ten from the mother. There were also ten ambiguous case-mother duos, resulting in estimates of 1.99 case subjects receiving the risk allele from the father and 8.01 from the mother, and four ambiguous case-father duos, resulting in estimates of 3.02 cases receiving the risk allele from the father and 0.98 from the mother. Overall, this resolves the parent of origin in an additional 30 families in comparison to the original analysis of Nudel et al.,6 giving 11.01 new receipts from the father and 18.99 from the mother. This increased number of receipts from the mother might be expected to result in a stronger maternally inherited imprinting effect than seen originally; however, the “risk” allele in this case actually decreases risk, giving an odds ratio of Im = 0.326 in the original analysis and Im = 0.494 when using estimated parent of origin. Therefore, the overall effect is now weakened by having more new receipts from mothers than from fathers. Similar observations can be made for the other SNPs in this region. It therefore seems probable that the original result could represent a statistical false positive due to stochastic sampling variation, which has been better resolved with the addition of 30 new observations. So, although the overall decrease in significance might seem disappointing, it corresponds to the use of more information (an additional 30 families contributing to the analysis) and thus should be considered a more reliable result.\nThe reduction in significance for the paternally inherited imprinting effect on chromosome 14 can also be explained by examining the parent-of-origin resolved cell counts (7.01 new receipts from the father and 15.99 from the mother, see Table 1). The fact that an additional 23 families are contributing to the analysis means that this should again be considered a more reliable result in comparison to the original analysis. However, in this case, the evidence for the presence of a paternally inherited imprinting effect remains relatively strong (p value = 1.32 × 10−7).\n\nTOF Data\nWe also re-analyzed the data of Cordell et al. (2013),35 testing for a paternally inherited imprinting effect on chromosome 12 and a maternally inherited imprinting effect on chromosome 13. Without using estimated parent of origin, the analysis on chromosome 12 gave a minimum p value of 2.10 × 10−8 at rs11065987; using estimated parent of origin gave a p value of 4.16 × 10−7 at the same SNP. On chromosome 13, the most significant SNP, rs7982677, gave p values of 9.54 × 10−7 and 6.97 × 10−6 when not estimating and estimating parent of origin, respectively. Plots of these analyses, with and without using estimated parent of origin, are shown in Figures S5 and S6. Again we see a general decrease in the significance of the most significant results when using haplotype estimation, weakening the evidence that these genomic regions genuinely harbor imprinting effects.\nTable 2 shows counts of the number of trios or duos falling into the ambiguous categories and their resolution when using haplotype estimation with SHAPEIT2 for the TOF data. For SNP rs11065987 on chromosome 12, it can be seen that, for case-parent trios and case-father duos, the number of risk alleles inherited from the father and mother are approximately equal; however, for case-mother duos, many more risk alleles are estimated to be inherited from the mother: 10.007, as compared to 1.993 from the father. This results in the initial estimated paternally inherited imprinting effects odds ratio Ip = 1.662 reducing to 1.555, thus decreasing the significance (and increasing the p value). Similarly, for SNP rs7982677 on chromosome 13, it can be seen that more alleles are inherited from the father than from the mother for all trios and duos, resulting in an initial estimated maternally inherited imprinting effect odds ratio Im = 1.577 reducing to 1.504. Again, although the significance is decreased when using haplotype estimation, it corresponds to the use of more information (an additional 43 families on chromosome 12 and an additional 29 families on chromosome 14) and thus should be considered the more reliable result. We note that the decrease of significance seen in this dataset with respect to testing imprinting effects (Ip or Im) has no impact on the significance of associations due to the case subject’s own genotype, findings which have in any case already been replicated in independent cohorts.35"}