Locating markers on the physical genome:

Physical positions of 10,202 SNPs used in the Shifman study were mapped to the mouse build 37 genome by BLAT (http://genome.ucsc.edu/cgi-bin/hgBlat) using the 100-bp flanking sequence on both sides of each SNP. Of 10,202 total SNPs, 10,196 were mapped uniquely to build 37, and 6 SNPs were mapped to more than one location; the multiply mapped SNPs were omitted from the revised map (see supporting information, Table S1).

Nomenclature for Mit markers varies across the online databases such as MGI, University of California at Santa Cruz (UCSC), and NCBI. We have identified Mit markers based on the unique MGI primer pair ID (PPID). We obtained 7080 Mit markers corresponding to 7052 PPIDs in the MGI database (ftp://ftp.informatics.jax.org/pub/reports/PRB_PrimerSeq.rpt). A single PPID may be associated with more than one Mit marker symbol, and these synonymous symbols are grouped together in our database. All marker symbols associated with the same PPID were consolidated and tabulated. There were some marker names associated with more than one primer pair, and these are flagged in our database.

Of the 7052 PPIDs, 45 did not have a primer sequence listed. The remaining 7007 unique primer pairs were mapped to the NCBI Build 37 mouse genome using the in silico PCR (isPCR) software (Hinrichs et al. 2006) with a minimum primer length of 18 bp and a minimum match size of 15 bp. The maximum product length allowed is 4000 bp, but primer pairs with amplicon lengths >1000 were not considered an alignment. In this way, we mapped 6049 primer pairs to unique positions on the NCBI mouse build 37 genome, and 144 primer pairs mapped to more than one genomic location (see Table S2). For the remaining 814 primer pairs, we again used the online version of isPCR (http://genome.ucsc.edu/cgi-bin/hgPcr?command=start), but reduced the minimum primer length to 15 bp. Using this method, we mapped 31 primer pairs to one location and 3 primer pairs to more than one location. The final remaining 780 primer pairs were deemed “unmappable” (see Table S3). They are flagged in our database, and we have retained historical genetic map positions. For each marker in the database, it is noted which of the above methods was used in mapping the primer pair. In total, we mapped 6080 primer pairs corresponding to 6388 Mit marker names.

Genetic map positions are relative. To anchor the genetic map, we assigned 0 bp in the physical map to 0 cM in the genetic map. The genetic position of the most proximal SNP was calculated on the basis of its megabase position using the average recombination rate (centimorgans/megabase) for that chromosome. We note that physical coordinates up to 3 Mb in the mouse reference sequence are placeholders for unknown or unsequenced DNA near the centromere of each chromosome.

Placing the Mit markers on the new genetic map:

We assigned male, female, and sex-averaged centimorgan positions to the Mit markers using linear interpolation based on their physical positions relative to the Shifman SNPs. Given the high density of SNPs, more sophisticated methods of interpolation should not provide any improvement. A total of 16 Mit markers at the distal ends of chromosomes 5, 8, 15, 16, 18, and X did not have flanking SNPs, and we assigned these markers to genetic map positions on the basis of their physical position by extrapolation using the chromosome-wide average recombination rate.

If available, the previous genetic map position is listed in the Mit marker database. While it is not possible to interpolate the unmapped Mit markers into the new genetic map, it may be necessary to assign updated genetic map positions for the purpose of reanalyzing historical data. We manually assigned new genetic map positions to unmapped markers on the basis of their MGI centimorgan position. This was done by determining the average base-pair positions of mapped markers with the same MGI centimorgan position and then assigning the updated centimorgan position of the mapped marker(s) to the unmapped marker at hand. In updating the QTL archive with the revised Shifman map, we assigned non-Mit markers to genetic map positions in this manner. In some cases, we found that the genotyping data from a cross are consistent with this assignment and have retained the marker; in other cases, there appear to be discrepancies and the suspect marker has been removed. We ran quality control checks on the genotypes in the archival data, but we did not revise the analysis of QTL positions.

Mit markers that aligned to multiple chromosomal positions in build 37 that were <1 Mb apart were interpolated on the basis of the average physical position of the alignments. Markers with multiple alignments to chromosomal positions >1 Mb apart or on more than one chromosome were assigned centimorgan positions in the same manner as the unmapped markers and on the basis of their MGI position. In the database, any map position not determined by the interpolation of one definitive physical position to the Shifman SNPs is flagged as unreliable.

Data sets used QTL for reanalysis:

Genotypes and phenotype data were obtained from five F₂ mouse mapping crosses: B6xCAST (Beamer et al. 1999), B6xC3H (Beamer et al. 2001; Koller et al. 2003), B6x129 (Ishimori et al. 2004), NZBxRF (Wergedal et al. 2006), and NZBxSM (Ishimori et al. 2008). The Animal Care and Use Committee approval and detailed genotyping and phenotyping methods can be found in the original articles. The primary phenotype of interest here is bone mineral density (BMD). For B6xC3H and NZBxRF, average cortical bone thickness at the mid-diaphysis (bone geometry) and ultimate load at failure (strength) were also examined. These traits share some but not all QTL with BMD in the original studies.

Primary QTL analysis:

All QTL analyses were done using the R/qtl software package (Broman et al. 2003) (version 1.09-43; http://www/rqtl.org/). Phenotypic outliers were removed prior to QTL analyses. Genotype data were also examined and obvious errors were corrected in a parallel manner for both analyses. Particular attention was paid to marker-order diagnostics using the recombination fraction plot function in R/qtl (plot.rf). If a marker-order issue was identified, the issue was corrected. In all cases, the phenotype trait data were transformed to correct for any skewing of the data using the van der Waerden normal score method (Lehmann and D'Abrera 1988). For each data set and for each map version, a single-locus mainscan for QTL was performed and LOD scores were calculated at 2-cM intervals across the genome using the “imp” or imputation method in R/qtl (Sen and Churchill 2001). Thresholds for significant and suggestive QTL were determined on the basis of 1000 permutations of the data (Churchill and Doerge 1994). A QTL was considered to be suggestive if the LOD score exceeded the 37% threshold and significant if it exceeded the 95% threshold. These thresholds were chosen as widely accepted cutoffs for suggestive and significant QTL (Lander and Kruglyak 1995). The goal of this analysis was to examine the effect of the new genetic map on single QTL, so no pairwise genome scans were run and no attempt to fit the multi-locus model was done.

The NZBxSM cross was the only cross to have data available from both male and female animals. For this cross, sex was considered as both an additive and an interactive covariate. Each model was examined for the effect of the new map on the QTL.

Mit marker map:

There are 7380 distinct Mit marker symbols associated with 7052 primer pairs. Conflicts and ambiguities in marker symbols between MGI, UCSC, and NCBI were resolved as described in materials and methods. Of the 7052 primer pairs, 6080 mapped to one unique genomic location, and these were assigned centimorgan positions on the basis of the interpolation in the revised Shifman map; 147 primer pairs mapped to more than one genomic location, and 780 primer pairs could not be mapped to the build 37 mouse genome using any of the methods described above. Forty-five PPIDs do not have known primer sequences according to the online databases. Markers that map to more than one location are not recommended for use in future studies. However, the status of these markers may be revised in light of future releases of the mouse genome assembly.

The relationships between physical and genetic map positions of Mit markers as reported in the MGI map and the revised Shifman map are shown in Figure 3. The most important difference is that markers in the revised Shifman map are consistent with the physical order of markers along each chromosome.

Open in a separate window

Figure 3.—

Physical and genetic positions of markers. Genetic marker positions (in centimorgans) are plotted against their physical positions (in megabases) for the MGI genetic map (red) and the revised Shifman map (black). Marker-ordering errors in the MGI map are indicated by non-monotone fluctuations. In contrast, the curves for the revised map are smooth and monotone.

Impact of the revised Shifman map on QTL localization:

Incorrect order of markers in the MGI map represents a substantial problem for mapping and identification of QTL. To investigate this further, we examined the effect on QTL mapping of changing to the revised Shifman map in five previously published data sets. First, a genome scan for primary QTL was done using the marker position information provided with the original data set (i.e., used in the published analysis). Second, an analysis was done using the new genetic map. Care was taken to ensure that the only difference in the analyses was the genetic map. For example, the removal of phenotypic outliers and phenotypic data normalization was identical in both analyses. As such, the effect of changing the genetic map on QTL mapping could be discerned independently of the different QTL mapping algorithms used in the original analyses.

In total, 78 QTL were examined in this study. Half of these QTL were not affected by the change in the genetic map. For the remainder, a variety of changes were noted (summarized in Table 2). For 28% of the QTL, no change in the shape of the QTL peak was identified, but the location of the peak shifted by ≥5 cM. We found 15 QTL with substantial changes (Table S4). For 6 QTL, a different marker closest to the peak was identified. For 9 QTL, a new shape for the QTL peak was noted (Figure 4). An example of such a peak shape change is presented in Figure 4, A and B. In this example, the third and fourth markers are reversed in order when comparing the two maps. This switch in marker order was not detected using standard marker quality control checks during either analysis. The LOD score associated with the peak was different in 2 of the QTL for which a change in peak shape was noted. Specifically, for one QTL, the change in LOD resulted in a failure to identify the QTL in the analysis utilizing the traditional map. In summary, for 15 of 78 QTL (19%), a mapping issue was identified that could impact the identification of the underlying gene.

Open in a separate window

Figure 4.—

QTL With changes in peak shape. Seven QTL were found where the peak shape was altered due to subtle differences between the original and the revised Shifman maps. (A) The QTL for femoral BMD on chromosome 12 in the NZBxRF cross appears as a double peak when analyzed using the Mouse Genome Database (MGD)/traditional genetic map (dashed line), suggesting the presence of two closely linked QTL. When reanalyzed using the new genetic map, this double peak is collapsed into a single peak (solid line). (B) This double peak is the result of two flipped neighboring markers. Seven markers were typed in this cross for chromosome 12. The markers are placed on the centimorgan scale (center line) in relation to the MGD/traditional map (left) and the new genetic map (right). Note the difference in spacing of the markers when comparing the two maps. The other QTL with a change in peak shape were found for: (C) B6xCAST on chromosome 2, (D) B6xC3H on chromosome 2 for vBMD, (E) NZBxRF on chromosome 4 for vBMD, (F) B6xCAST on chromosome 14, (G) B6xCAST on chromosome 18, and (H) NZBxSM on chromosome 19.

The authors thank Ben King, Randy von Smith, Stephen Grubb, Nazira Bektassova, and Keith Sheppard for advice and assistance with informatics and database tools. We thank Ken Paigen, Janan Eppig, Petko Petkov, Thomas Gridley, and Lucy Rowe for critical discussion and review. This work was supported by the National Institutes of Health with grants GM070683 (G.A.C.), GM074244 (K.W.B.), HL077796 (B.P.), 5T32HD07065 (C.L.A.-B.), and AR053992 (C.L.A.-B.); by the National Institute of General Medical Sciences Centers of Excellence in systems biology grant GM076468 (G.A.C.); and by the Federal Ministry for Education and Research, Germany (National Genome Research Network project 01GS0829, G.A.B.). The Shared Scientific Services were supported in part by a Basic Cancer Center Core Grant (CA34196) from the National Cancer Institute to the Jackson Laboratory.