Using Oxford Gene Technologies 44K aCGH arrays, giving a resolution of about 100 base-pairs (bp), we analyzed 300 isolates that had an unstable Lac+
phenotype, in comparison with a reference DNA sample from the parental strain FC40 
. The 300 unstable Lac+
isolates consisted of 284 new isolates from day 7 of adaptive mutation experiments and 16 isolates that were reported before 
. These were included to determine whether there was complexity in the events that had not been detected by our previous method of outward PCR 
. Indeed, two of the 16 carried an additional event identified by aCGH. We also studied 180 cultures derived from Lac−
cells taken from lactose minimal medium starvation plates on day 5 (this is equivalent to day seven Lac+
colonies, because point-mutant Lac+
take 2 days to form visible colonies). We also analyzed 60 day seven Lac+
point mutant colonies, and include these with the Lac−
isolates from starvation plates as a control of 240 stressed isolates that do not carry lac
-amplification. In addition, we studied 60 single cell Lac−
colonies that had not been stressed. No change in copy number was found in any of 300 non-amplified samples.
All 300 unstable Lac+
isolates were found to carry amplification at lac
. The mean copy number at lac
was 69.3+/−22.9 (mean +/− SD). The mean length of 298 amplicons was 22.7 kilobase pairs (kb). These same isolates, when grown in lactose minimal medium (to maintain selection for amplification) were found to have about twice the amount of F′-borne chromosomal sequence than sequences that were only present on the chromosome (1.88+/−0.30-fold). There was no increase in copy number of F′ sequences in previously stressed Lac−
cells grown in glycerol minimal medium or in previously stressed Lac+
point mutant cells grown in lactose minimal medium (0.99+/−0.10-fold). We sequenced the amplification junctions of 40 amplicons. We found that all had microhomology at the junction sequence. Sixteen of the 40 were located in REP sequences 
. The sequences of the 40 amplification junctions are shown in Table S1
In 300 lac
-amplified isolates, we identified 28 events that changed chromosomal structure in addition to the amplification at lac
(9.3%). The positions of some of these changes on the standard map of E. coli
are shown in . The difference in the occurrence of other events in amplified isolates compared with zero in the 240 stressed control samples is highly significant (p
0.0001; Fisher's exact test). Using the Peto Odds Ratio we can estimate the odds ratio (OR
6.7) and a corresponding 95% confidence interval ranging from 3.1 to 14.1 
. Some of these additional events changed the copy number at lac
, and might therefore have played a role in lac
-amplification. Other events do not appear to offer a growth advantage to Lac−
cells on lactose minimal medium, and therefore represent other events occurring in the same cells as lac
-amplification. We tested whether one inversion affected the rate of amplification by measuring amplification in a derivative that had lost amplification. Figure S2
shows that the inversion had no effect on rate.
Distribution of structural changes in E. coli genome.
Events related to lac-amplification
The most common complexity was an inverted duplication embedded in the amplified region (, PJH1490). This was found in 16 of 300 amplified isolates (5.3%) (). The same configuration was found to be common in the study by Kugelberg et al. with the Lac assay in Salmonella enterica 
. In all 16 cases, the lac
region was included in the embedded duplication. Detailed study of these events showed that the embedded inverted duplications vary in size from 5.2 to 42.6 kb. Two novel junctions were found in each case. The junctions showed microhomology of 3 to 30 bp (). We interpret these events as two inverted template switches that generate an inverted triplication, followed by unequal crossing-over that generates the amplified array (, see Discussion
Complex rearrangements on the F′128 revealed by oligonucleotide aCGH and confirmed by PCR and DNA sequencing.
Sequences of junctions of complex events for those events for which there was more than one junction.
Model for the formation of complex structures by template switch events.
We identified two other inverted regions that generated a distinct pattern on aCGH data where part of the amplicon appears to be detached from the rest on the map of the parental strain based on the standard map of E. coli (PJH39 and PJH2122) (one example, PJH39, is indicated in by an open arrow). When the map is corrected to include this inversion, the amplicon is seen to be contiguous. These events show only two novel junctions, the right end of the inversion and the amplification being the same junction. We therefore regard the inversion and the duplication as parts of the same event, and explain them below as a pair of inverted template switches followed by unequal crossing over ().
Another event of the same type, PJH2058, that did not involve inversion or duplication of lac (apart from the amplification) is shown in . There is a short sequence within amplicon that is present in 2-fold less copy number than the rest of the amplicon (open arrow in ). This can also be explained by 2 switches, but neither of them is inverted (, see below).
A very large tandem duplication (about 300 kb) was found in an isolate (PJH1475) in which the F′-factor was integrated into the chromosome, so that part of the F′ including lac, and part of the chromosome was duplicated (). We have confirmed the HFR status of this isolate by showing that conjugational transfer of proAB, which is on the F′-plasmid in FC40, is RecA-dependent in this isolate, whereas it would not be if it were situated on a plasmid. The duplication is flanked by IS5 sequences, and therefore was presumably formed by homologous recombination (). Similarly, the integration of the F′-plasmid occurred by homologous recombination between sequences that are in common between the chromosome and F′128, because aCGH detected no other copy number change.
Two other large duplications, PJH1477 and PJH1487, were found that included lac
and had one or both ends outside the chromosomal sequence on the F′. The junctions were not found in the IS3
elements that span chromosomal sequence on the F′-plasmid as has been observed previously 
. The same two events contained duplications within the amplified segment. The junction sequences of both duplications were found to be recalcitrant to amplification by PCR. Multiple primer pairs were used in all pair-wise orientations, but no product or only unspecific product was found. Similar results have been reported for some human non-recurrent copy number changes (e.g. 
). It is possible that these represent translocations, further unanticipated orientational complexities at the breakpoint junctions, or insertions of large genomic sequences/structures between the designed primers that do not correspond to a preconceived notion based on a reference genome sequence used for primer design. Array CGH provides copy number information, but neither positional nor orientational information. We were unable to characterize these further.
These data establish that, like in human, a significant proportion of events of chromosomal structural change that generate amplification are complex in that more than one structural change occurred, apparently within the same event. This applies to 19 of 300 events resolved by our approach (omitting large duplications that might have assisted amplification, but might not be part of the same event).
Secondary events in lac-amplified isolates
In the same sample of 300 amplified isolates, we also found six that included a chromosomal structural change that was not apparently directly involved in the amplification. None was seen in the 240 stressed control isolates. The null hypothesis that the amount of that unrelated chromosomal structural change does not differ between amplified and stressed non-amplified isolates, can be rejected (p
0.036; Fisher's exact test 
). Using the Peto Odds Ratio we can estimate the odds ratio (OR
6.2) and a corresponding 95% confidence interval ranging from 1.2 to 31.0. 
Duplications should be unstable, so it is not surprising that we saw none that did not duplicate lac and thereby provide selection for maintenance of the duplication. Four of the unselected events were deletions (1.33% of 300 events): two on the F′-plasmid and two on the chromosome. One of the deletions (PJH1474) was flanked by non-identical IS elements, and so might have occurred by homeologous recombination or alternatively might have utilized the shorter homology stretches to mediate a template switch. The other three show microhomology junctions (1 to 4 bp), and so probably happened by events similar to those generating amplification. The chromosomal deletions were 0.8 and 1.6 kb long, and are situated at about 1.4 and 1.6 megabases on the standard reference E. coli map (PJH2116 and PJH1482 respectively). Deletions of 0.2 and 7.5 kb long (PJH2030 and PJH1482 respectively) were found on the F′ at about 44 kb and 50 kb from lac respectively (). An example, PJH1474, is shown in .
We found one inversion because it made an apparent separation of the amplicon into two parts (based on the standard map) (, PJH1479). The endpoints of the inversion and the amplification are different, so we see no evidence that the events are related. The inversion presumably happened before the amplification, and the amplification then included part of the inverted region. Because most inversions would not be detected by aCGH, we searched all 300 amplified and 240 stressed control isolates for inversion within 20 kb to either side of lac by unidirectional PCR (). When PCR primers point in the same direction, there is no PCR product unless the sequence at one of the primer binding sites has been inverted. We found one further inversion in an amplified isolate (PJH1465) and none in the controls. These two inversions are described in . It is interesting that, although the exchanges were almost reciprocal, the junctions are not exactly in the same position, so that a mutation of a small deletion or insertion is made at either end of both inversions.
Inversion-associated small deletion and insertion mutations.