A careful analysis of two maize recombinant inbred lines (RILs) relative to their inbred parents revealed the presence of several hundred apparently de novo copy number variants (CNVs). These changes in genome content were validated via both PCR and whole exome-array capture-and-sequencing experiments. One hundred and eighty-five genomic regions, which overlap with 38 high-confidence genes, exhibited apparently de novo copy number variation (CNV) in these two RILs and in many instances the same apparently de novo CNV events were observed in multiple RILs. Further analyses revealed that these recurrent apparently de novo CNVs were caused by segregation of single-copy homologous sequences that are located in non-allelic positions in the two parental inbred lines. F1 individuals derived from these inbred lines will be hemizygous for each of these non-allelic homologs but RIL genotypes will contain these sequences at zero, one or two genomic loci. Hence, the segregation of non-allelic homologs may contribute to transgressive segregation. Indeed, statistical associations between phenotypic quantitative trait loci and genomic losses were observed for two of 14 tested pairs of non-allelic homologs.
In this report we document the existence of numerous non-allelic homologs in maize. These non-allelic homologs represent single-copy sequences that are present at different chromosomal locations in different individuals. A detailed analysis of CNVs was undertaken using several recombinant inbred lines (RILs) relative to their inbred parents. We detected more examples of apparently de novo CNV than expected and noted that several of these apparently de novo CNV were found in multiple RILs. Our investigations of these instances of recurrent apparently de novo CNV (Brunetti-Pierri et al., 2008; Fernandez et al., 2010; Shinawi et al., 2010; Neill et al., 2011) revealed that they were in at least most instances the result of segregation of non-allelic homologs (SNH), which generated RILs that completely lack or have extra copies of a given sequence. Finally, we provide evidence that these changes in sequence content can contribute to phenotypic variation.
Chromosomal segments that exhibit non-parental-signals
Array-based comparative genomic hybridization (aCGH) experiments were conducted using genomic DNAs from two maize inbred lines, B73 and Mo17 (Springer et al., 2009), and two of the inter-mated B73 × Mo17 (IBM) RILs (Lee et al., 2002): M0022 and M0023 (Fu et al., 2010) derived from these parental inbred lines. A careful analysis of the aCGH signals of single-copy (i.e. those mapped to single genomic positions in the B73 genome) probes revealed that 0.06% (1086/1 780 475 in RIL M0022) and 0.13% (2338/1 780 475 in RIL M0023) of probes yielded signals in the RILs that were statistically different from those of both parents (Figure 1 and Figures S1–S3 and Table S1 in Supporting information). We defined these probes as putative de novo CNV probes. Many of these probes exhibiting unique levels of hybridization signal in the RILs relative to the parental genotypes can be grouped into ‘segments’ using DNAcopy software (Experimental Procedures). There are 67 chromosomal segments in RIL M0022, and 130 in RIL M0023 (Table 1 and Table S2) that represent putative de novo CNV events in the RILs. These include both gain and loss events, indicating copy number gain and loss, respectively. The average length of these CNV segments is 2.8 kb with the longest being 53.4 kb (Figure S1). The majority (69%) of the 197 chromosomal segments do not exhibit substantial differences in aCGH signal between B73 and Mo17. Hence, these aCGH results indicate that although the corresponding probes and chromosomal segments do not exhibit CNV between the inbred parents, these sequences do exhibit CNV in the RILs. These putative de novo CNV comprise 0.009 and 0.012% of the M0022 and M0023 genomes, respectively.
Table 1. Numbers of putative de novo copy number variation (CNV) segments (genes in putative de novo CNV segments) derived from M0022 and M0023
No gain/loss in M0023
No gain/loss in M0022
To confirm and extend these results a NimbleGen whole exome-array was used to capture and sequence the genic portions of the chromosomal segments that exhibit non-parental hybridization levels (see Experimental Procedures). We separately (n = 4) captured genomic DNAs from the parental inbred lines (B73 and Mo17) and the two RILs (M0022 and M0023) using a previously published protocol (Haun et al., 2010). Then 32–37 million 40-bp paired-end Illumina reads generated from each capture were aligned to the B73 reference genome (GenBank accession no. SRA036595). Reads that uniquely mapped to the 197 chromosomal segments identified via aCGH were counted for all four genotypes. Figure 2 compares the aCGH and exome-Seq results of two chromosomal segments in both RILs. For subsequent analyses we focused on the 61 segments for which at least 30 reads were obtained from the B73 capture. Of these, 54 and 7 segments had exhibited aCGH signal losses (copy losses) and gains (copy gains) in the RILs relative to B73, respectively (Figure S4). The aCGH and exome-Seq count data for these segments are highly correlated (correlation = 0.62). In the vast majority (45/54) of signal loss segments, at least 80% fewer reads were obtained from the affected RIL than from B73. Notably, no reads were obtained from the affected RIL in 24 (of 54) chromosomal segments that exhibited signal loss in the aCGH experiments, demonstrating that these segments are completely absent from the RIL genome. Consistently, more reads were recovered from the RILs than from B73 for all seven segments that exhibited signal gains in the RILs relative to B73, indicating that the RIL genomes contain more copies of these segments than do the parental genomes (Figure S4). Similar results were observed in comparisons to Mo17 (Figure S4). These exome-Seq results confirm the existence of apparently de novo CNV for many of the chromosomal segments in the RIL genotypes.
Apparently de novo copy number variation is the result of segregation of non-allelic homologs
There are a number of potential mechanisms for de novo CNV formation including non-allelic homologous recombination (NAHR), rearrangements in the absence of extended sequence similarity associated with DNA repair by non-homologous end-joining (NHEJ) or with microhomology-mediated break-induced replication (MMBIR), contraction or expansions of variable number tandem repeats (VNTRs) and mobile element insertions (MEI) (Mills et al., 2011). Alternatively, apparently de novo CNV can be formed by segregation of single-copy sequences that are located in non-allelic positions. If two parental lines both contain a single copy of a sequence that is located at unlinked genomic positions, then the F1 will be hemizygous for each of these copies and meiotic segregation will generate F2 (or RIL) genotypes with zero, one or two copies of the sequence (Lu et al., 2012). The relatively high rate of apparently de novo CNV observed in the RILs suggested that segregation of non-allelic homologs (SNH) might be responsible.
Several lines of evidence support this hypothesis. First, we noted that the locations of copy number gains and losses in the RILs exhibit dependence on the parental origins of the chromosomal segments containing these gains or losses. Those associated with significant signal losses in the RILs are embedded within Mo17-derived chromosomal regions, while 19/20 apparently de novo CNVs with significant signal gains are embedded within B73-derived regions (Figure 1). This would be expected if the apparently de novo CNV arose via SNH. Second, 12 chromosomal segments exhibited apparently de novo CNV in both RILs, resulting in 185 non-redundant segments (Table 1 and Table S2). SNH would be expected to yield a high rate of recurrent apparently de novo CNV. To further examine the degree to which the apparently de novo CNVs are shared among RILs, PCR primers were designed based on several chromosomal segments that exhibited signal loss in at least one of the RILs and used to amplify products in 300 IBM RILs. Every one of these segments was missing in multiple RILs (6–34%) (Table S3), providing evidence that these segments exhibit frequent apparently de novo CNV origin, consistent with SNH.
The third piece of evidence that supports the role of SNH in the generation of these apparently de novo CNVs is based on mapping of these sequences in B73 and Mo17 using mate pairs (see Experimental Procedures). Using conservative criteria 40 (M0022) and 68 (M0023) of the 197 segments exist in non-allelic positions in the B73 and Mo17 genomes (Tables S4 and S5). In most instances (5621/5709) the positions of non-allelic homologs inferred by aligning the Mo17 mate pair reads to the B73 reference genome do not correspond to expectations based on the whole genome duplication event (Krzywinski et al., 2009; Schnable et al., 2009), indicating that differential losses of genes in duplicated genomes is not the dominant mechanism underlying SNH. In contrast, only two (M0022) and six (M0023) of 197 random control segments mapped to non-allelic positions in the Mo17 genome (Figure 3, Table S4). Hence, consistent with the SNH model, more than half of the sequences that give rise to apparently de novo CNV are located in non-allelic positions in the B73 and Mo17 genomes.
Impact of segregation of non-allelic homolog-derived copy number variation on phenotypic traits
The SNH model predicts that meiotic segregation will act on non-allelic homologs, resulting in novel complements of sequences (losses and gains) among progeny relative to parental haplotypes (Figure 4). Those SNH-derived CNVs that involve genes would be particularly interesting because they could result in novel genic complements in progeny relative to parents.
The maize genome sequencing project defined a set of 32 540 high-quality gene annotations that is referred to as the Filtered Gene Set (FGS, ver. 4a.53). Prior to inclusion in the FGS, gene models were rigorously filtered to remove gene fragments and sequences with similarity to transposons. Thirty-five of the observed cases of SNH-derived CNVs overlap (partially or completely) with 38 of the high-quality gene models in the FGS (Table S6). The RNA-Seq data from apices (GenBank accession no. SRA036595; see Experimental Procedures) provided evidence of expression for 24 (63%) of these genes that are affected by SNH-derived CNV (Table S6). In addition, 27 (71%) of these genes have homologs in sorghum or rice, indicating phylogenetic conservation. In combination, these lines of evidence suggest that many of the genes affected by SNH-derived CNV are probably functional.
To test whether changes in gene complement caused by SNH-derived CNV have phenotypic consequences, we collected data on a number of phenotypic traits from the ∼300 IBM RILs discussed above. Each of the 14 assayed cases of SNH-derived CNV fully or partially overlaps at least one of the high-confidence genes in the FGS. We then compared the average phenotypic trait values of RILs that did or did not experience gene loss via SNH. After controlling for multiple testing (see Experimental Procedures), losses of two of the 14 tested chromosomal intervals were significantly associated with phenotypic variation. Chromosomal interval M0022_seg30 is significantly associated with reduced cob diameter (adjusted P-value = 0.03) and kernel row number (adjusted P-value = 0.01). Similarly chromosomal interval M0022_seg15/M0023_seg22 (which includes a putative peptidyl–prolyl cis–trans isomerase gene) is associated with increased tiller number (adjusted P-value = 0.01) (Table S7).
De novo CNV has been hypothesized to arise via transposon-, recombination- and replication-mediated mechanisms (Hastings et al., 2009; Springer et al., 2009; Conrad et al., 2010; Innan and Kondrashov, 2010; Stankiewicz and Lupski, 2010; Mills et al., 2011). The association between the distributions of gains and losses of apparently de novo CNV observed in this study and the parental origins of the surrounding chromosomal segments (Figure 1) is inconsistent with transposon-mediated mechanisms acting during the several generations required to produce the RILs. Further, the high rates of recurrence of apparently de novo CNV are inconsistent with recombination- and replication-driven mechanisms because these mechanisms are reported to generate losses and amplifications at much lower rates (van Ommen, 2005; Yandeau-Nelson et al., 2006; Lupski, 2007; Turner et al., 2008). In contrast, SNH-derived CNV is not the result of active rearrangements of DNA but is instead the result of meiotic segregation acting upon transposed gene copies and, in some cases, fractionation events following whole genome duplication. Collectively, our observations suggest that SNH results in CNV for hundreds of maize loci.
The maize genome is a product of an ancient tetraploidization event and now consists of two ‘subgenomes’ (Schnable et al., 2011) having different properties, including gene expression levels. Intra-chromosomal recombination events can result in the loss of the copy of a pair of homologs from one subgenome (Woodhouse et al., 2010). Although the two subgenomes exhibit different rates of genes loss (Schnable et al., 2011), this process and others such as transposon-mediated gene duplication/transposition (Jiang et al., 2004; Lai et al., 2005) have generated numerous non-allelic homologs (Eichten et al., 2011). We have demonstrated that meiotic segregation of these non-allelic homologs generates CNV affecting hundreds of loci in the progeny of a single cross (Figure 4). The 185 detected SNH-derived CNV affect 38 high-confidence genes. Considering the stringent criteria used in this study, this frequency is likely to be an underestimate.
The SNH model exhibits similarities to the ‘reciprocal gene loss model’ first proposed by Lynch and Force (2000) to explain interspecific genomic incompatibility. This model proposed that the loss of different copies of duplicated genes in different populations would lead to gene loss in gametes from F1 individuals. This process has been demonstrated in crosses among three yeast species (Scannell et al., 2006) and between two fish species (Semon and Wolfe, 2007). It has also been shown to affect single genes in several intraspecific studies, including Drosophila (Masly et al., 2006) and Arabidopsis (Bikard et al., 2009). The SNH model differs from the reciprocal gene loss model in that it occurs intraspecifically, can generate copy number gains and can act on non-allelic homologs generated via various mechanisms. The SNH model would be expected to generate CNV in any species that contains non-allelic homologs and undergoes meiotic segregation. In maize we believe some of the non-allelic homologs arise via fractionation, but the mechanism outlined in Figure 4 can occur regardless of the mechanism by which the non-allelic homologs were originally generated. For example, with only minor modifications this mechanism could also generate CNVs in a species that contains non-allelic homologs generated via the transposition of single-copy genes (Vlad et al., 2010).
Phenotypic effects of segregation of non-allelic homolog-derived copy number variation and presence–absence variants
Although CNV has previously been associated with genetic disorders in humans (Stankiewicz and Lupski, 2010), this report provides evidence that the segregation of CNV via the SNH model can also contribute to the phenotypic variation present in crop species such as maize. This model may also shed light on transgressive segregation, i.e. the appearance of progeny from a bi-parental cross whose phenotypic values exceed those of their parents (Rieseberg et al., 1999).
The findings reported here have significant implications for the large-scale efforts under way to identify the genetic determinants of phenotypic variation in humans, model and agricultural species. This is because the genetic determinants of phenotypic variation arising via the SNH model will not be detected via traditional single marker association studies or QTL analyses. Indeed, the synergistic effects from multiple unlinked genomic loci are likely to lower the power of such traditional one-dimensional analyses. Multiple-dimensional scans that consider the synergistic effects of multiple markers on phenotypes can overcome this limitation of traditional genetic mapping approaches. Although computationally intensive, such studies are now tractable (Koesterke et al., 2011). Another concern is that if PAVs are not in linkage disequilibrium (LD) with nearby genetic markers such as SNPs, the power of association studies that rely on such markers will be reduced. It will be interesting to determine whether the direct genotyping of PAVs via CGH-based genotyping (Fu et al., 2010) or genotyping-by-sequencing approaches (Fogu et al., 2007; Huang et al., 2009; Lai et al., 2010; Andolfatto et al., 2011; Elshire et al., 2011) will uncover at least a fraction of the ‘missing heritability’ observed in genome-wide association studies (Kump et al., 2010).
Two maize inbred lines, B73 and Mo17, and two RILs, M0022 and M0023, were extracted from the IBM Syn4 population (Lee et al., 2002). The RILs used in this study were from the F7–9 generation.
Identification of putative de novo copy number variation probes
The aCGH experiments (GEO: GSE16938), data processing and statistical analyses were performed as described previously (Springer et al., 2009; Fu et al., 2010). Contrasts were performed between B73 versus Mo17, B73 versus RIL and Mo17 versus RIL. A P-value was determined for each probe from each contrast. To account for multiple testing, P-values were converted to q-values (Benjamini and Hochberg, 1995). Probes with significantly higher or lower signals in the RILs when compared with both B73 and Mo17 were termed putative de novo CNV probes. Signal loss probes were called using the criteria of q-value (RIL versus B73) <0.0001, q-value (RIL versus Mo17) <0.0001, log2(signal ratios of RIL/B73) <0 and log2(signal ratios of RIL/Mo17) <0; signal gain probes were called using the criteria of q-value (RIL versus B73) <0.001, q-value (RIL versus Mo17) <0.001, log2(signal ratios of RIL/B73) >0 and log2(signal ratios of RIL/Mo17) >0. Different q-value cutoffs were used to identify signal loss probes and signal gain probes because aCGH technology has a greater power to detect copy number losses than copy number gains (Altshuler et al., 2010).
Determination of appropriate parental control for each probe in array-based comparative genomic hybridization analyses of recombinant inbred lines
For each aCGH probe, the log2 ratios of the hybridization signals of the RIL versus B73 [log2(RIL/B)] and separately versus Mo17 [log2(RIL/M)] were calculated. A log2 ratio that is greater (or smaller) than 0 indicates that a given probe yields a stronger (weaker) signal in the RIL than in a particular parental inbred. For each aCGH probe, the smaller of absolute value of log2(RIL/B) and the absolute value of log2(RIL/M) was used to identify the presumptive parental origin (B73 or Mo17) in the RIL of the chromosomal segment from which each probe was designed. For each probe this parental hybridization value [log2(RIL/B|M] was used for the calculations plotted in Figure 1.
Segmentation of putative de novo copy number variation probes to identify putative de novo copy number variation segments
The putative de novo CNV probes were converted to 1 (copy gain compared with B73) or –1 (copy loss compared with B73). All other probes were assigned a value of 0. The converted binary data were subjected to segmentation via DNAcopy (Olshen et al., 2004) using the parameters: alpha = 0.01, nperm = 10 000, p.method = ’perm’, eta = 0.01, min.width = 3. Putative de novo CNV segments were required to contain at least three putative de novo CNV probes and a median absolute deviation (MAD) equal to 0.
Recombinant inbred line segmentation to distinguish the origin of regions from either B73 or Mo17
Probes that distinguished B73 and Mo17 were identified using the following criteria: q-value (Mo17 versus B73) <0.0001 and log2(Mo17/B73) < (−1). These probes were treated as genetic markers to genotype the RILs, most of which were grouped into Mo17-type [q-value (RIL versus B73) <0.001 and q-value (RIL versus Mo17) >0.1] or B73-type [q-value (RIL versus B73) >0.1 and q-value (RIL versus Mo17) <0.001].
To avoid the partition of a chromosomal region exhibiting the same origin into multiple segments, the array CGH genotyping results were converted to binary data (B73-type = 1; Mo17-type = 0) and merged to perform segmentation as described above. Segments smaller than 200 kb and having a mean segment value between 0.1 and 0.9 were removed from further analysis. By so doing, we excluded segments that were ambiguously assigned as being B73-type or Mo17-type and small segments that might represent mis-assemblies in the reference genome.
Gene annotation information was downloaded from http://www.maizesequence.org. This annotation set was defined as the entire set of evidence-based genes (predicted by Gramene GeneBuilder), complemented by a set of Fgenesh models. Pseudogenes, TE-encoded genes and low-confidence models were filtered out to produce the final annotation set. Coding sequence coordinates were extracted from this annotation (251 067 regions; 55.5 Mbp) and consolidated into non-overlapping regions (152 529 regions; 37.9 Mbp). These regions were padded to a minimum region size of 100 bp, and again consolidated into non-overlapping regions (151 929 regions; 39.7 Mbp). Final coordinates were offset by 35 bp to account for capture probe length overhang at the end of each region. Variable length capture probes (50- to 100-mers) were selected by tiling through each region at an average spacing of 48 bp (measured from 5′ start to 5′ start). Repeat-masking was done by generating a histogram of all 15-mers in the maize genome and removing probes with an average 15-mer frequency greater than 100. Probe uniqueness was assessed using SSAHA (http://www.sanger.ac.uk/resources/software/ssaha/), using a minimum match size of 26. No more than three matches were allowed for capture probes. The final design covers 131 469 of the 151 929 original regions, and has a total capture space of 55.4 Mbp. The Sequence Capture Developer 2.1M feature array design 100224_ZmB73_public_exome_cap_HX3 is available for purchase from Roche NimbleGen (http://www.nimblegen.com/).
Analysis of Exome-Seq data
Novoalign 2.05.31 (http://www.novocraft.com/) was used to map all reads (40 bp) to reference genomes, including the B73 reference genome (B73ref_v1), the maize mitochondrial genome (GenBank accession no. AY506529.1) and the maize chloroplast genome (GenBank accession no. X86563.2). Reads that were uniquely mapped with two or fewer mismatches (insertions and deletions was counted as mismatches) were used for further analysis. Read counts of each putative de novo CNV segment were adjusted by the addition of 1 to avoid zero values and were then used to calculate log2(RIL/B73) and log2(RIL/Mo17).
The RNA was extracted from a pool of three to six apices of 14-day-old seedlings of the inbred lines B73 and Mo17 using the Qiagen RNeasy Plant Mini Kit (catalog no. 74903, http://www.qiagen.com/default.aspx). RNA-Seq was conducted using an Illumina GAIIx instrument at the Iowa State University DNA facility following an Illumina protocol (mRNA-seq Sample Preparation Guide; http://www.illumina.com/).
Genotyping of a subset of the apparently de novo copy number variation segments on the recombinant inbred line population
Primers were designed on a subset of the apparently de novo CNV segments that fully or partially overlap with transcription- or annotation-supported genes to directly genotype >300 IBM recombinant inbred lines. A primer pair was designed for each selected segment using the B73 reference sequence. Primers were aligned to JGI 454 Mo17 reads to ensure that primers work for Mo17. This PCR program consisted of 94°C for 10 min; 35 cycles of 94°C for 30 sec, 60°C for 45 sec, 72°C for 1.5 min, and a final extension at 72°C for 10 min in a 20-μl volume. Three existing genetic markers of the IBM genetic map (Liu et al., 2009), IDP525, IDP7957 and IDP7324, were used as the PCR control primers. The IDP525 marker has been used to genotype the full set of RILs previously (Liu et al., 2009). The RILs with poor or no PCR amplification of any IDP markers or inconsistent IDP525 PCR results between the previous scores and the re-genotyping scores were not used for further analysis.
Phenotyping recombinant inbred lines
The following traits were collected from 291 IBM RILs: seedling dry weight, average kernel weight, cob diameter, cob length, cob weight, kernel count, kernel row number, total kernel weight, tiller number, ferulic acid (FA), p-coumaric acid (PCA), brace node number (BN), node number above primary ear (NA), node number below primary ear (NB) and total node number (NT). Four replicates of seedling dry weight trait data and three replications of other traits data were collected. Least square means for each genotype were estimated using the SAS code for the traits except for tiller number. Mean of tiller number for each genotype was calculated among replicates.
Phenotypic associations of segregation of non-allelic homolog-derived copy number variants
We used t-tests to test the null hypothesis that no association exists between segmental copy loss and each of the phenotypic traits. For each tested segment, RILs were divided into two groups based on the results of PCR-based genotyping, i.e. RILs with and without the expected PCR bands. The t-test assuming equal variation between two groups was conducted for each trait, generating a P-value. A permutation test was implemented to account for multiple testing for each segmental copy loss. In each permutation of a given segment, the genotyping scores of the segment were randomly shuffled among the RILs. A similar t-test was performed on the shuffled genotyping data for each trait and a P-value was obtained for each trait. The smallest P-value was then determined among the multiple P-values from the multiple trait tests. This procedure was repeated 1000 times. One thousand P-values were obtained for each of the tested segments. The original P-values of the multiple traits computed from the observed data of a given segment were compared with the 1000 P-values from the permutation test of this segment to generate adjusted P-values.
Physical mapping of Mo17 sequences involved in apparently de novo copy number variation
To determine the chromosomal location of paired reads, the mate-pair cluster mapping method was used. First, the individual reads from 5-kb Mo17 mate pairs (one lane from an Illumina/Solexa GA-IIx instrument) were separately mapped to the B73 reference genome. A mate pair was used for further analyses only if one and only one member of that mate pair uniquely mapped to an apparently de novo CNV segment (or within the 500 bp beyond either end of the segment because the actual endpoints of loss or duplication are not known with certainty) in the B73 reference genome. For each mate pair of this type, if the other member mapped within 1 Mb of the same apparently de novo CNV segment, the pair was categorized as having similar locations in the B73 and Mo17 genomes (‘locally mapped’). If the mapped location of the other member of a mate pair was >1 Mb away or on a different chromosome, the mate pair was categorized as having different locations in the B73 and Mo17 genomes (‘distally mapped’). The identification of multiple, independent distally mapped mate pairs clustered on a chromosomal region was considered evidence that the corresponding chromosomal segments in B73 and Mo17 are non-allelic. A ‘cluster’ of mate-pair reads was defined as consisting of two or fewer independent non-stacked reads that mapped within <100 kb of each other and that covered >80 bp of the CNV segment. Two sets of segments (one from M00022 and one from M0023) were randomly selected as the controls. For each RIL, the number of randomly selected segments from each chromosome was equal to the number of identified apparently de novo CNV segments on that chromosome in the corresponding RIL.
We thank Stephen Moldovan, Ho Man Tang and Marianne Smith for technical assistance and Tracy Millard and Dr Thomas J. Albert for sequencing and project support. We thank Roche NimbleGen for providing support for this study through the donation of reagents, and team-member time. The co-authors TR, DJG and JAJ recognize a competing interest in this publication as employees of Roche NimbleGen, Inc. This research was also supported by a grant from the National Science Foundation to PSS (IOS-1027527) and data generated as part of NSF grant IOS-0820610 (Mike Scanlon, PI).