Short tandem repeats (STRs) have become one ofthe most widely used genomic markers for identitytesting and gene mapping due to their high degreeof heterozygosity. Accurate genotyping of STRs byfragment sizing relies on precise relative migrationof identical alleles and prior knowledge of the spec-trum of most or all possible apparent sizes that arederived from those relative migrations. The list ofall possible alleles for a genetic locus is called anallele list. The allele list serves as a lookup tablethat will label fragments according to a namingscheme chosen by the user. In most cases, theapparent sizes will require sampling of a populationof results that are separated under the same condi-tions that will be used in the larger experiment.

The most commonly used STR loci have allelesthat are expected to increase in size by a fixed inter-val, generally between 2 and 6 nucleotides.However, exceptions to this rule are often observed.Some alleles do not adhere strictly to the commoninter-allelic spacing at their loci. The alleles that fallbetween the more regularly-spaced alleles aresometimes referred to as non-integer repeats. Inaddition, alleles at some loci display spacing pat-terns that are marginally shorter or longer thanexpected. The trend is not identical for all STR lociand thus requires inspection on a locus-by-locusbasis.

In the CEQ™ 8000 Fragment Analysis software,the process of binning compiles all of the pertinentfragment lengths from real data, estimates theirmost likely apparent sizes, and, taking size drift intoaccount, assigns integer lengths (nominal sizes) tothem. The product of the binning process is anallele list that then can be used to identify alleles

whenever amplification products of the same genet-ic locus are separated under the same conditions.

Here we describe the process of automatedallele binning using simple tools for recognizing themigration trends of STR alleles, identifying bothinteger repeat alleles and those fragments that mayrepresent non-integer repeats.

Software WizardsThe CEQ 8000 Fragment Analysis software uses anumber of software interface wizards—tools thataid the user in progressing through the necessarysteps of selecting data and entering required values.You may not proceed to the next step of a wizardif critical information is incorrect or missing.The software usually will indicate the problem areausing yellow highlights. The Binning wizard is oneof the more sophisticated wizards, consisting offour interdependent screens.

Automated Generation of a Locus TagUsing the Automatic Binning Wizard When data are plentiful, binning is the best way ofcreating an STR allele list. The observed sizes offragments are rarely, if ever, integers but highlyreproducible non-integer lengths (A-1876A: “HighlyPrecise DNA Sizing on the CEQ™ GeneticAnalysis System”). The first step of the automatedbinning process is cluster analysis—the organiza-tion of observed fragment lengths into groups.Where the clusters are tight, the mean observed

APPLICATION INFORMATION

G e n e t i c A n a l y s i s : C E Q S e r i e s

A-1932A

AUTOMATED BINNING PROCESS FOR THE GENERATION OFLOCUS TAGS

Heather Gull, Dana Campbell, and Mark DobbsBeckman Coulter, Inc.

2

sizes within each cluster represent the most likelylength that will be observed for an unknown frag-ment with the same number of nucleotides.

To start a new Binning Analysis, select thisoption from the analysis menu or by right-clickingon the Binning folder in the Analyses tab of theStudy Explorer. Each peak included in the study is acandidate for the binning process. On Screen 1, theuser simply views the data (Figure 1) in each of thedye colors, selects the dye, size range, maximumbin width, minimum number of data points per bin,repeat unit length, and an allele naming conventionbefore proceeding to the next step of the process.The wizard steps are fast and reversible so not allselections need to be correct at the outset.

Screen 2of the binning wizard normalizes thepeak heights for the fragment length range selectedand displays a prediction of the positions of the reg-ularly spaced alleles. Several options are now avail-able to validate new allele lists (Figure 2).

Trace Inspection

View the trace of any point in the Bin View scatterplot by left-clicking on it. To hide a trace that hasbeen launched, select this option from the rightmouse button menu. The bins overlaying the tracescan be turned off by unchecking the Show Binscheckbox in the Trace Views area.

Phase Shift

Shift the phase of the bins in the plus or minusdirection. The effect will be to shift all the bins insingle nucleotide incre-ments in the event thatthe software did notautomatically select thedesired peaks to buildthe allele list. Both theNominal and Apparentsizes will be shifted bythose same incrementsuntil a shift equivalentto a full repeat unit isreached, at which pointthe effect of the shiftwill be nullified.

Minimum RelativePeak Height

Change the minimumrelative peak height toexclude additionalpeaks that are used inthe cluster analysis. Thefeature is useful forexcluding peaks from

binning that are clearly not critical for building theallele list but were not previously excluded by othermeans.

Show Phantom Bins

Phantom bins are the bins that fall in between per-fect repeat allele positions. Phantom bin positionsare easily calculated by the software based on therepeat unit length and the register, or spacing, of theperfect repeats and the peaks associated with them(e.g., +A). When peaks do not fall into perfectrepeat bins, two observations lend support to theidea that they are non-integer repeats:

1) The peak cluster pattern, or signature, of theamplification product is similar to the signaturesof the perfect repeats; and

2) The candidate allele peak falls in a phantombin, suggesting that its secondary structure isconsistent with other products of the locus(see Discussion).

Phantom bins may be converted into real bins byleft-clicking on them and then selecting Create Allelefrom the right-click menu. Alternatively, one can addan allele above or below any row of the allele list usingthe right-click menu option when the cursor is overthe allele list grid. Conversely, alleles may be removedfrom the allele list using the opposite commands.

Regression Plot

The regression plot displays the linear relationshipbetween nominal sizes and apparent sizes (Figure 3).The regression plot can be accessed in the right-

Figure 1. First screen of the binning wizard.

3

click menu when the cursor is on the Bin View. Thesoftware selects nominal sizes to obtain the best lin-ear fit with the available data from the clusteranalysis. However, because the nominal sizes and

apparent sizes are user-editable, it is possible for theuser to inadvertently reduce the goodness of fit ofthe regression line. Editing a nominal size so that itis one integer nucleotide too high or too low would

Figure 2. Second screen of the binning wizard.

Figure 3. Regression Plot. The regression plot displays the nominal size versus the apparent sizes for fragmentlength clusters from the results in a study.

create a disjointed set of points at the position of theerror. The regression plot will take user edits intoaccount and reflect the reliability of the nominalsize estimates. The fit of the line is described bythree values: the nominal versus apparent size slope,the y-intercept, and the correlation coefficient (r2).Based on our observations, the slope should alwaysbe between 0.95 and 1.05, and r2 should be >0.9999.The most significant impact of a bad edit would be areduction in r2. The values are displayed above boththe regression plot and the Bin View scatter plot.

Screen 3of the binning wizard prompts the userto specify a unique Locus Tag, which is the namereferenced by the software, and a Locus Label,which is the name that the software applies to thealleles when they are identified. The two names canbe the same. The allele list is carried forward fromScreen 2.The Locus Tag tab (Figure 4) containslocus-specific information, some of which hasalready been entered in the binning process(e.g.,repeat unit length) and some of which is fordocumentation purposes only (e.g.,primer setnames and sequences). The Allele ID Criteria tab(Figures 7 and 10) provides options for interpretingthe +A artifact and discriminating against stutter.The proper use of these options is described in theCEQ 8000 Genetic Analysis System User’s Guide(Beckman Coulter PN 608315).

The final Screen 4of the wizard reviews thesource data that was used to initiate the binninganalysis. If no changes to the result set have beenmade, the source data list will be identical to theresults set list. However, the source data list willremain with the binning analysis that it gave rise to

even as the results set is modified or manipulatedfor other purposes.

Using the Allele List to IdentifyAllelesIt is important to keep two points in mind duringthe creation of an allele list using the binningprocess:

1) Not all true allele peaks are required to buildthe list; and

2) Peaks that are used to build the list are not auto-matically assigned allele IDs.

The first point is important because it enablesthe user to sample the data without including weak-er peaks that may blend with noise peaks from otherresults. In the second analysis of the data that isrequired to perform the actual allele assignmentswith the new Locus Tag, the sensitivity of peakdetection can be raised to include all alleles, strongor weak. While this two-step approach is valid forsimple and clean alleles, it is not foolproof.Complex allele patterns may require the use ofmore advanced tools of the CEQ 8000 software.Both simple and complex allele patterns are consid-ered in the examples following.

4

Figure 4. Third screen of the binning wizard, displaying the Locus Tag tab.

5

Case 1—One peak perallele, no stutter

Sample Locus D22S683Repeat Unit Length 4 (GATA)Peaks per Allele 1Conventional Stutter None

Special Binning Procedures

There were so many non-integerrepeats at this locus that it wasappropriate to set up a binninganalysis with a repeat unit lengthof 2 instead of 4.

Figure 5. D22S683 Allele Signature (heterozygote).

Figure 6. D22S683 Initial Bin View.

Figure 7. D22S683Allele ID Settings.

6

Case 2—Strong +Aconversion, weak stutter

Sample Locus D11S1984Repeat Unit Length 4 (GGAA)Peaks per Allele 2Conventional Stutter Weak

Special Binning Procedures

None.

Allele ID Notes

Conventional stutter peaks thoughsmaller than true alleles will oftenbe registered as identified allelesbecause they are full repeat unitsaway from the true alleles.

Both -A and +A forms of anallele, if they are large enough,will be identified as alleles unlessDetect +/-A is selected.

Figure 8. D11S1984 Allele Signature (Heterozygote).

Figure 9. D11S1984 Initial Bin View.

Figure 10. D11S1984 Allele ID Settings.

7

Case 3—Two peaks ofsimilar height per allele,leftmost selected as allele

Sample Locus GATA193A07Repeat Unit Length 4 (GATA)Peaks per Allele 4Conventional Stutter Weak

Special Binning Procedures

Check to ensure that the properphase has been selected by theautomatic binning software (theleftmost peak of each doublet).Launch some trace views by click-ing on points from within bins. Ifincorrect member of doublet isselected, use Phase shift tool tocorrect.

Allele ID Notes

In the case of greater than twopeaks per allele, it is not clearwhich peak is the -A form andwhich is the +A form. However,since two peaks per allele aretaller than the others (the otherscan be excluded during analysis,or after analysis using filtering),we can use the +A tools to preventthe second peak of the tall doubletfrom being called an unknownallele as indicated in the allele IDsettings.

Figure 11. GATA193A07 Allele Signature (Heterozygote).

Figure 12. GATA193A07 Initial Bin View.

Figure 13. GATA193A07 Allele ID Settings.

8

Case 4—Large number ofpeaks per allele

Sample Locus D3S2387Repeat Unit Length 4 (GATA)Peaks per Allele 7-9Conventional Stutter Unknown

Of the four cases presented,this is the most difficult becausethe allele clusters are composedof several peaks of similar height.This situation is problematic forboth the binning process and forthe identification of allele peaksafter the allele lists have beengenerated. Let’s examine the twoproblems one at a time.

1. Binning Procedures

The biggest challenge in the con-struction of the allele list fromcomplex alleles using the binningalgorithm is noise. In our exam-ple, the compiled collection offragments spans the entire locusrange with some peaks at everypossible position. To address thelarge number of peaks per allele,we consider four different strate-gies for automating the allele listconstruction.

Strategy 1: During primary dataanalysis, only those fragmentsthat are ≥ 99% the height of thesecond tallest fragment areincluded (relative peak heightthreshold = 99%).

Advantages: works when appliedto both simple and complexalleles, because the two tallestpeaks in a trace are alwaysincluded.

Disadvantages: independent analy-sis method just for binning;when multiplexing products ofthe same color in different sizeranges, weaker loci will beexcluded.

Figure 14. D3S2387 Allele Signature (Heterozygote).

Figure 15a: D3S2387 Bin View scatter plot using fragments that are >99%the height of the second tallest fragment are included (relative peak heightthreshold set to 99% during primary data analysis). Note that the Y-axisstarts at 0.1 because there are no peaks with a Relative Signal Strengthbelow this number.

9

Strategy 2: Exclude peaks ofbelow a fixed relative peakamount by applying a filter to theFragment List before the binningprocess is initiated (Figure 15b).

Advantages: offers more flexibil-ity than Strategy 1 enablingthe exclusion of any desiredrelative quantity of fragment.

Disadvantages: when multiplex-ing products of the same colorin different size ranges, weak-er loci will be excluded.

Strategy 3: Exclude data by rais-ing the minimum relative peakheight during the binning process(Figure 15c).

Advantages: includes weakerpeaks in the general analysis;excludes them only in thephase of analysis where theyare not needed; relative peakheights in binning are calcu-lated for the locus range only,to take into account loci ofdifferent intensities within thesame samples (multiplexedsamples).

Disadvantages: sometimes diffi-cult to decide where to set theminimum relative peakheight.

Figure 15b. D3S2387 Bin View scatter plot from fragments that are above aRelative Signal Strengthof 0.15 by applying a filter to the Fragment Listbefore binning.

Figure 15c. D3S2387 Bin View scatter plot from fragments where the mini-mum relative peak height was raised to 0.3 during the binning process.

10

Strategy 4: Filter on the frag-ment list to exclude all fragmentsthat were not the tallest peakswithin their own peak clusters.

Advantages: preserves the effec-tive recognition of smallerpeaks during primary analy-sis, while recognizing thetallest members within eachfragment cluster.

Disadvantages: when peak clustersoverlap, the tallest peaks with-in lower height sub-clusterswill not be recognized.

The decision regarding whichnoise reduction strategy to use willdepend on the complexity of thelocus, the differences in peakheights between shorter andlonger alleles at the same locus,the potential for overlap of peakclusters, and the peak height dif-ferences between peaks chosen asalleles and the remaining peaks.For D3S2387, Strategy 4 is per-haps the best because it selectsnearly all of the peaks that wewould select by manual inspec-tion of the traces. For the purpos-es of allele list construction, thelosses due to overlapping peakclusters are not significant.

2. The identification of allelepeaks after the allele lists havebeen generated

Allele ID Notes

After building the allele list, onenow has to deal with the problemof preventing non-allele peaks of being identified asunknown alleles. If there were no overlapping peakclusters, one could filter out all but the primarypeaks, and be left with only the identified alleles. Inthis approach, alleles that were part of overlappingpeak clusters would be lost. The only remainingapproach is to apply the Locus Tag to the data usingreasonable sensitive peak detection parameters (thedefault slope threshold = 10, relative peak heightthreshold =10% is sufficient) and none of the +Acheckboxes selected. This process will identify mul-tiple alleles per locus, only some of which are real.

The list must be reduced by selecting the allelepeaks manually.

Left click on the fragment list and chooseManually Select Peaks. When given the option toclear the fragment list, select Yes. A stacked graphof all the results in the study will be presented onthe right. Point the cursor at the apex of the tallestpeak of each cluster and left click to select it. Then,right click, and select Include. Each selected peak,including its allele ID, will be added to the list.

Finally, if the tallest peaks in the allele clustersdo not happen to coincide with the perfect repeatsin the allele list, their allele IDs will be blank in thefragment list. Sort the fragment list based on esti-

Figure 15d. D3S2387 Bin View scatter plot using fragments that were thetallest peaks within their own peak clusters.

Figure 16. D3S2387 Allele ID settings.

11

mated fragment size (nt). It should be relativelyeasy to interpret the allele IDs of the imperfectrepeats based on the surrounding alleles. Enter theseIDs into the fragment list.

Application and Management of LocusTagsWhen a Binning Analysis is saved, the allele list isfrozen as part of the specified Locus Tag. TheLocus Tag name is not editable except through theData Manager. However, the saved binning resultkeeps the original Locus Tag name, even if it isrenamed or deleted.

There are two ways to edit the allele list:

1. The binning analysis can be re-opened, editsmade, and the study saved. In this instance, theoriginal Locus Tag name is used, regardless ofwhether the Locus Tag was renamed or deletedfrom the database.

2. The Locus Tag can be accessed through theSTR Locus Tags tab of the Analysis Parametersmenu option by selecting Edit when anyAnalysis Parameter set is selected. First theLocus Tag is selected and then the Edit Locusbutton is pressed. The list of available LocusTags reflects the current elements of the data-base.

Using theRe-analyze results command toapply Locus Tags replaces the current study resultswith new results, and displaces, but does not dis-card, the old results. The consequence will be toremove from the study the first results that com-prised the source data list. It is also important toremember that the data points that are viewed in thebinning analyses are dynamic - they reflect the cur-rent state of the results set and the fragment list.Thus the application of new filters or the manualexclusion or inclusion of results or fragments willbe reflected in the Bin Views, despite the static binpositions and statistics. To force the binning analy-sis to conform to the new data, Re-bin must be exe-cuted from the Binning menu option (the Binningmenu is available only when the binning tab isviewed). Note that re-binning will undo most manu-al edits to the allele list. Re-binning is useful whenresults with contributing alleles for the locus havebeen added to the study.

The above points are all important to considerwhen deciding how to segregate results that havebeen used for binning and in the construction ofallele lists from those results that have identifiedalleles. In most cases the same data is used for both.To preserve the integrity of the binning analysis, the

best choice is to create new study from raw data,applying the new Locus Tags to the data. However,if you believe that the allele list will not require anyfuture edits, you may re-analyze the results usingAnalysis Parameters that include the new LocusTags. If re-construction of the original binninganalysis is then required, the list of source data canbe used to re-build the study from those results.

DiscussionThe most reliable method of determining longerDNA fragment lengths is by DNA sequencing.However, DNA sequencing cannot be multiplexed,and provides more information at each locus than isnormally required for identity testing or linkagestudies. Fragment sizing of end-labeled polymor-phic short tandem repeat amplicons by high resolu-tion gel electrophoresis has become a rapid alterna-tive tool for genetic analysis in mammals. Precisesizing of DNA fragments is essential in dealing withboth common alleles and rare variants that may dif-fer in length by as little as 1 nucleotide, a feat thatis well within the capabilities of the CEQ™ 8000(Application Information Bulletin A-1876A). Inaddition, many short tandem repeat sequences havebeen well characterized, their estimated sizes needto be re-evaluated whenever new separation systemsor new separation conditions are used.

We noted that when all peaks from a locus aretaken into account, the differences in apparentnucleotide length are not always identical to the dif-ferences in true size, i.e., a spacing of 1.00 realnucleotides per 1.00 observed nucleotides is notalways observed. The CEQ 8000 binning softwarequantitates the relationship in a term called thenominal versus apparent size slope. In two of theloci above (case 1 and case 3), we demonstrate aslope of ~0.97, indicating that, for every fragmentlength increase of 1 nucleotide, we observe anapparent increase of 0.97 nucleotides. The mostlikely explanation for this phenomenon is that theDNA fragments from different loci have slightlydifferent mobilities due to secondary structure thatare not corrected for by the internal size standards.If rounding or truncation of the apparent sizes wereused to assign nominal sizes, the subtle drift wouldeventually lead to errors in predicting allele lengths.The CEQ 8000 binning software uses the nominalversus apparent size slope to take the drift intoaccount. For all loci that we have examined, theobserved base change per actual base difference islinear and well behaved.

The process of automated allele binning greatlyfacilitates the development of an allele list that pre-

dicts the observed lengths of all alleles at a geneticlocus. In the case of simple allele peak signatures,allele IDs may be specified for both common andrare alleles. As illustrated in the cases with complexalleles, some degree of population sampling may berequired to establish the integer repeat peak distri-bution pattern. For most allele peak patterns, the IDnames may be applied directly during a secondanalysis step. For some complex allele patterns, theuser has the freedom to use the electropherogramtraces to manually select those peaks that representtrue alleles.

* All trademarks are the property of their respective owners.

Developing innovative solutions in genetic analysis, drug discovery, and instrument systems.

Beckman Coulter, Inc.• 4300 N.Harbor Boulevard,Box 3100 • Fullerton,California 92834-3100Sales:1-800-742-2345 • Service:1-800-551-1150 • Telex:678413 • Fax:1-800-643-4366 • www.beckmancoulter.com

Worldwide Life Science Research Division Offices:Australia (61) 2 9844-6000 Canada (905) 819-1234 China (86) 10 6515 6028 Eastern Europe, Middle East, North Africa (41) 22 994 07 07 France 01 49 90 90 00 Germany (89) 35870-0 Hong Kong (852) 2814 7431 / 2814 0481 Italy 02-953921 Japan 03-5404-8359 Mexico 525-605-77-70 Netherlands 0297-230630 Singapore (65) 6339 3633 South Africa/Sub-Saharan Africa (27) 11-805-2014/5 Spain (34) 91 3836080Sweden 08-564 85 900 Switzerland 0800 850 810 Taiwan (886) 2 2378 3456 Turkey 90 216 309 1900 U.K. 01494 441181 U.S.A. 1-800-742-2345

B2002-5072-LP-5 © 2002 Beckman Coulter, Inc. Printed in U.S.A.on recycled paper.

For Research Use Only. Not for use in diagnostic procedures.

© 2014 AB SCIEX. SCIEX is part of AB Sciex. The trademarks mentioned herein are the property of AB Sciex Pte. Ltd. or their respective owners. AB SCIEX™ is being used under license.

AB SCIEX Headquarters500 Old Connecticut Path | Framingham, MA 01701 USAPhone 508-383-7700www.absciex.com

View SCIEX products at www.sciex.comFind your local offi ce at www.sciex.com/offi ces

www.sciex.com/ce