We analyzed the distribution of recognition sequences for 4,060 restriction enzyme combinations (Figure 1) on human chromosome 7 (Materials and methods). From this, we identified five restriction enzyme combinations of potential utility for FPP. All five combinations included HindIII and EcoRI, and one of: BclI/BglII/PvuII, BalI/BclI/BglII, NcoI/PvuII/XbaI, Bcl/NcoI/PvuII, or BglII/NcoI/PvuII. Each of these combinations represented at least 99.98% of the chromosome7 sequence in restriction fragments of sizes that are generally accurately determined using our BAC clone fingerprinting method. Ultimately, we selected the combination HindIII/EcoRI/BglII/NcoI/PvuII for its desirable cut site distribution, ease of use in the laboratory and our favorable experience with the high quality of fingerprints from these enzymes.

To demonstrate that fingerprint patterns are sufficiently complex to uniquely identify genomic intervals, we devised in silico simulations to determine specificities of fingerprint fragments and patterns and to align virtual clones with simulated rearrangement breakpoints to the reference genome sequence.

We computed the fragment specificity for a given fragment as the fraction of fragments in the genome that are experimentally indistinguishable in size (Materials and methods). Figure 2 shows the specificity for an individual HindIII fragment of a given size in the human genome (hg17), and depicts the practical specificity where experimental sizing error is used to determine whether fragment sizes can be distinguished. Our sizing error depends on fragment size (Figure 3), effectively dividing the sizing range into approximately 380 unique bins. Also depicted is the case of exact sizing, where fragments are considered indistinguishable only if their sizes are identical. Although exact sizing is not possible in the laboratory, we include the case of exact sizing here because it represents the theoretical best possible performance of FPP with the enzymes we selected, and because it helps to contrast FPP's practical performance.

This analysis revealed that HindIII fingerprints with approximately 15 fragments exhibit a high degree of specificity, as only approximately 1.5% of the genome cannot be uniquely distinguished using patterns composed of this number of fragments. This high specificity results from accurate experimental fragment sizing, and from the fact that the length of genomic repeats is generally much shorter than restriction fragments. Therefore, a specific combination of adjacent fragment sizes represents a relatively unique event in the human genome.

To evaluate the accuracy and sensitivity of actual fingerprint alignments, we performed an in silico study (Materials and methods), in which we computationally generated virtual clones containing simulated genomic rearrangement breakpoints and used these fingerprints as inputs into the alignment algorithm. Figure 4 illustrates the sensitivity and positional accuracy of the mapping of these synthetic clones as a function of the number of digests and segment size. When a single HindIII fingerprint digest is used, we successfully aligned 50% of 35 kb segments. This cutoff size can be decreased to 25 kb if two digests are used (HindIII/EcoRI) and to 16 kb if five digests are used (HindIII/EcoRI/BglII/NcoI/PstII). The number of digests used has a large impact on the smallest alignable segment size due to the fact that the positions of cut sites of distinct enzymes are generally uncorrelated and that the individual digest patterns can be aligned independently and used together to increase sensitivity. Figure 4 suggests the number of digests that would be required to detect 90% of rearrangements of a certain size. For example, if we wish to identify a breakpoint in 90% of simulated cloned rearrangements, then the shortest rearrangements that can be detected for 1, 2, 3, 4 and 5 digests are 60, 45, 34, 28, and 25 kb, respectively. Stated differently, one can be 90% certain that when using 5 enzymes, a segment of length 25 kb within a BAC would be sufficient to identify the BAC as bearing a genome rearrangement.

Figure 4 shows the median distance between the left and right edges of the alignment and known segment spans for segments of varying sizes. While the values for 10 kb segments are difficult to interpret because of relatively few successful alignments, the error is otherwise constant for segment sizes and depends primarily on the number of digests. The error is 3.0 kb for an alignment based on a single digest and drops to 1.7 kb when two digests are used. When the number of digests is increased to 5, the error drops as low as 700 base-pairs (bp).

With knowledge gained from our simulations, we sought to apply FPP to a test set of 493 BAC clones derived from the MCF7 breast cancer cell line. Each clone was fingerprinted and aligned to the genome with FPP, and the results of the alignments were compared to alignments performed using BAC end sequences (Materials and methods, Additional data file 2). Alignments were evaluated based on their size and number, with multiple alignments indicating identification of a candidate rearrangement. We were able to obtain FPP alignments for 487/493 of the clones. On average, we were able to map 88% of a clone's fingerprint fragments to the genome, and 90% of clones had more than 72% of their fingerprint fragments mapped. Table 1 summarizes FPP and ESP rearrangement detection and Table 2 shows a detailed comparison of rearrangement detection for clones that had an FPP alignment that indicated a breakpoint. The positional accuracy of FPP alignments is shown in Table 3.

Because ESP uses BAC end sequences that produce data for only the ends of clones, ESP has limited capacity to localize the locations of rearrangement breakpoints within clones. To investigate the precision of FPP in defining the position of breakpoints within BACs, we used clone alignments spanning regions of chromosomes 1, 3, 17 and 20 that contained known breakpoints. We selected these regions because of the enriched coverage provided by our test clone set. The breakpoint position was determined to be the average FPP alignment position with the error given by the standard deviation of the alignments. Additional data file 2 shows the layout of these breakpoints in the MCF7 genome and all FPP and ESP alignments for clones in these regions. Additional data file 3 expands several of the regions from Additional data file 2, and illustrates the relative position of FPP and ESP alignments. Additional data file 6 further increases the detail shown in Additional data file 2, depicting restriction maps and fragment matching status within each clone alignment for all five enzymes. We found 51 breakpoints in 118 unique clones (Table 4). We tested the presence of breakpoints in three clones using PCR (Table 5), and demonstrated the presence of PCR products (Figure 5) to verify fusions within the clone's insert of regions non-adjacent in the reference genome sequence.

To demonstrate that FPP can resolve complex rearrangements, we closely examined the FPP results for clone 3F5. In the original MCF7 ESP analysis, Volik et al. 25 concluded that the shotgun sequence assembly of this clone is highly rearranged and composed of five distant regions of chromosomes 3 and 20 (3p14.1, 20q13.2, 20q13, 20q13.3 and 20q13.2). Our FPP analysis generally recapitulated the shotgun sequencing results - out of the five distinct insert segments found by sequencing, we detected four (Figure 6; detailed fingerprint alignments are shown in Additional data file 4; individual restriction fragment accounting is shown in Additional data file 5). The fifth segment, sized at 4,695 bp based on alignment of the clone's sequence to the reference genome, lacked the fragment complexity to confidently identify it by FPP. This small segment includes only two entire restriction fragments (marked with asterisks in the following list of intersecting fragments) in the restriction map of our enzyme combination (HindIII, 1 fragment (7.4 kb); EcoRI, 3 fragments (7.2 kb, 0.9 kb*, 8.5 kb); BglII, 2 fragments (4.1 kb, 8.6 kb); NcoII, 3 fragments (2.0 kb, 1.9 kb*, 6.2 kb); PvuII, 2 fragments (5.8 kb, 13.1 kb)).

Fingerprints provide a representation of the entire length of a clone's insert and, thus, are capable of mapping genome rearrangements internal to the clone insert that do not involve the ends of the clone. We identified 17 such small-scale candidate aberrations, and validated 4 of these using PCR (Table 6, Figure 7). PCR analysis of clone 12G17 yielded an amplicon approximately 400 bp smaller than expected, which supports the observation that experimental fragments were approximately 300 bp smaller than expected in this area. The fingerprint results are consistent with a hypothesis of a loss of a 313 bp SINE element evident in the genome sequence for this region. PCR analysis of clone 15O22 indicated an insertion of approximately 560 bp relative to the reference genome sequence. The experimental fragments nearest to the unmatched in silico fragments in this clone's fingerprints are all about 300 bp larger than expected. The results are consistent with a hypothesis of increased copy number of Alu (300 bp) or SINE (100 bp) elements evident in the genome sequence of this region.