Pierre Taberlet, Eva Bellemain, Aur

1
Genotyping errors

• Pierre Taberlet, Eva Bellemain, Aurélie Bonin,
  François Pompanon
• Laboratoire d'Ecologie Alpine, CNRS UMR 5553,
• Université Joseph Fourier, Grenoble, France

2
Genotyping errors

• Bonin A, Bellemain E, Bronken Eidesen P, Pompanon
  F, Brochmann C, Taberlet P (2004) How to track
  and assess genotyping errors in population
  genetics studies. Molecular Ecology, 13,
  3261-3273.
• Pompanon F, Bonin A, Bellemain E, Taberlet P
  (2005) Genotyping errors causes, consequences
  and solutions. Nature Reviews Genetics, in press.

3

Genotyping errors

• Definition
• Non-invasive sampling and genotyping errors
• Causes of genotyping errors
• Quantifying genotyping errors
• Consequences of genotyping errors
• How to limit genotyping errors and their impact?

4

Genotyping errors

• Definition
• Non-invasive sampling and genotyping errors
• Causes of genotyping errors
• Quantifying genotyping errors
• Consequences of genotyping errors
• How to limit genotyping errors and their impact?

5
Definition

• A genotyping error occurs when the observed
  genotype of an individual does not correspond to
  the true genotype.
• Genotyping errors can have strong consequences on
  the biological message that can be deduced from
  the data.

6
Distribution of papers on "genotyping errors"
according to their publication year

• Apparently, more and more attention is paid to
  genotyping errors.

7
Distribution of papers on "genotyping errors"
according to their subject

• Genotyping errors are a concern for some research
  field only (linkage analyses, non-invasive
  methods).
• What about the other fields using genetic tools?
  (population genetics/genomics?)

8

Genotyping errors

• Definition
• Non-invasive sampling and genotyping errors
• Causes of genotyping errors
• Quantifying genotyping errors
• Consequences of genotyping errors
• How to limit genotyping errors and their impact?

9
Non-invasive sampling and genotyping errors

• Historical aspects
• Solutions to limit genotyping errors
• Towards a quality index
• Practicals estimation of the quality index

10
Questions about the Pyrenean bear population
Papillon, Photo J.-J. Camarra, ONC, Août 1995
Geographic distribution of the brown bear in
Europe
11
Questions about the Pyrenean bear population

• Where to take bears to reinforce the endangered
  Pyrenean population?
• How many bears are left in the Pyrenees?
• How many males and females?

12
The three different sampling methods

• Destructive sampling
• Non-destructive sampling
• Non-invasive sampling

13
Destructive sampling

• The animal is killed in order to obtain the
  tissues necessary for genetic analysis.
• This sampling strategy has been used extensively
  for isozyme studies, and for mtDNA analysis
  before PCR was discovered.
• It has been abandoned by many researchers.

14
Non-destructive sampling

• The animal is often captured, and a biopsy or a
  blood sample is taken invasively.
• However, some invasive sampling strategies do not
  require catching the animal.
• For example tissues can be obtained from whales
  and some other large mammals by using biopsy dart
  guns.

15
Non-invasive sampling

• This term should be restricted to situation where
  the source of DNA is left behind and is collected
  without having to catch or disturb the animal.
• In the literature, non-destructive sampling is
  often improperly considered as non-invasive.
• Catching a mammal (or a bird) and plucking a few
  hairs (or feathers) should not be considered as
  non-invasive, but rather as non-destructive.

16
Non-invasive genetic sampling only possible via
PCR

• Mullis KB, Faloona FA (1987) Specific synthesis
  of DNA in vitro via a polymerase-catalysed chain
  reaction. Methods in Enzymology, 155, 335-350.
• Saiki RK, Gelfand DH, Stoffel S, Scharf SJ,
  Higuchi R, Horn GT, Mullis KB, Erlich HA (1988)
  Primer-directed enzymatic amplification of DNA
  with a thermostable DNA polymerase. Science, 239,
  487-491.

17
Problematic results about the census of the
Pyrenean bear population (in 1994)

• More bears than expected!
• No success when trying to replicate the results.
• Two years to understand and solve the problem.

18
Potential of non-invasive genetic sampling two
opposing point of view

• Non-invasive sampling can exploit the full
  potential of DNA analysis.
• True for mtDNA
• Dominant opinion ten years ago
• Non-invasive sampling has serious limitations.
• Many technical problems
• Possibility of genotyping errors

19
Non-invasive sampling can exploit the full
potential of DNA analysis

• Morin PA, Moore JJ, Chakraborty R, Jin L, Goodall
  J, Woodruff DS (1994) Kin selection, social
  structure, gene flow, and the evolution of
  chimpanzees. Science, 265, 1193-1201.
• Microsatellite study using hairs as a source of
  DNA.
• Males are more homozygous than females
  in general.
• Males are staying in their group more than   
  females.
• Wrong results due to more genotyping
  errors in males (mainly allelic
  dropout).

20
Non-invasive sampling has serious limitations

• Gerloff U, Schlötterer C, Rassmann K, Rambold I,
  Hohmann G, Fruth B, Tautz D (1995) Amplification
  of hypervariable simple sequence repeats
  (microsatellites) from excremental DNA of wild
  living bonobos (Pan paniscus). Molecular Ecology,
  4, 515-518.
• Taberlet P, Griffin S, Goossens B, Questiau S,
  Manceau V, Escaravage N, Waits LP, Bouvet J
  (1996) Reliable genotyping of samples with very
  low DNA quantities using PCR. Nucleic Acids
  Research, 26, 3189-3194.
• Gagneux P, Boesch C, Woodruff DS (1997)
  Microsatellite scoring errors associated with
  noninvasive genotyping based on nuclear DNA
  amplified from shed hair. Molecular Ecology, 6,
  861-868.

21
Gagneux P, Woodruff DS, Boesch C (1997) Furtive
mating in female chimpanzees. Nature, 387 (22 May
1997), 358-359.
22
Gagneux P, Woodruff DS, Boesch C (1997) Furtive
mating in female chimpanzees. Nature, 387 (22 May
1997), 358-359.

• Paternity study of the offspring of a chimpanze
  community.
• Half of the offspring did not display any allele
  inherited from an intragroup father.
• Conclusion these offspring had an extragroup
  father.
• The dataset contained allelic dropouts (paper
  retraction in 2001).

23
Scan of the Gagneux paper in Mol Ecol
24
Genotyping errors main difficulties in
non-invasive sampling

• Contamination
• Allelic dropout
• False alleles

25
Contamination

• Behind the possibility of detecting a single
  target molecule, there is also a possibility of
  detecting a single contaminant molecule.
• Working with non-invasive genetic sampling is
  similar to ancient DNA studies.

26
Genotyping errors allelic dropout

• For a heterozygous individual, only one allele is
  present in the template and/or is amplified in
  the PCR reaction.
• This error produces a false homozygote.

27
Genotyping errors false alleles

• Artifacts can be generated during the first
  cycles of the PCR reaction, and can be
  misinterpreted as true alleles.
• Very difficult to discern from sporadic
  contamination.

28
Genotyping errors example
Allele A
Brown bear Locus G10B
Allele B

• Five independent genetic typing using the same
  DNA extract (from feces).

29
Genotyping errors example
Allele A
Allele B

• Fifty independent genotyping experiments using
  the same DNA extract (from a bear feces) locus
  G10B.

30
Genotyping errors example
Seven independent experiments using the same DNA
extract from a bear feces.
31
Genotyping errors example
Seven independent experiments using the same DNA
extract from a single marmot hair.
32
Influence of the amount of template DNA
From Goossens et al., 1998
33
Allelic dropout mathematical model

• The model is restricted to the genotyping of an
  individual bearing alleles A and B at an
  autosomal locus.
• Many assumptions have been made.

34
Allelic dropout mathematical model assumptions

• The DNA extract contains equal numbers of the
  alleles A and B.
• A single target molecule can be amplified and
  detected.
• Each single target molecule has the same
  probability of being amplified.
• 100 PCRs and be performed using the DNA extract,
  and the target DNA molecules are distributed
  randomly among the 100 PCR tubes.
• If the initial proportion between alleles A and B
  (A/B or B/A) in the PCR tube is greater than or
  equal to five, then only the most common allele
  will be detected.

35
The problem of very small DNA samples simulations
Simulations for a heterozygote individual with
alleles A and B
Simulations for a heterozygote individual with
alleles A and B.
correct genotyping
correct genotyping
36
Results of the simulations
PCR product


(at least one

allele)
correct

genotyping

(both alleles)
one cell contains

about 7 picograms

of DNA
template DNA per

amplification (picograms)




Probability of correct genotyping at a
heterozygote microsatellite locus using very mall
DNA samples
DNA samples
37
Guidelines for genotyping very small DNA samples

• Multiple-tube approach.
• Navidi W, Arnheim N, Waterman MS (1992) A
  multiple-tube approach for accurate genotyping of
  very small DNA samples by using PCR statistical
  considerations. American Journal of Human
  Genetics, 50, 347-359.
• Taberlet P, Griffin S, Goossens B, Questiau S,
  Manceau V, Escaravage N, Waits LP, Bouvet J
  (1996) Reliable genotyping of samples with very
  low DNA quantities using PCR. Nucleic Acids
  Research, 26, 3189-3194.

38
Guidelines for genotyping very small DNA samples

• The guidelines are only valid under the following
  conditions.
• A single target molecule can be detected.
• The amount of template DNA is very low, in the
  picogram range, but is not accurately know.

39
Guidelines for genotyping very small DNA samples

• Confidence of 99.
• Multiple-tube approach.
• Heterozygotes an allele can be recorded only if
  it has been found at least twice.
• Homozygotes an individual can be considered as
  homozygous only if eight independent experiments
  have shown the same allele.

40
Guidelines for genotyping very small DNA samples

• How to avoid or to limit the impact of the
  multiple-tube approach?
• By estimating the amount of template DNA
• Miller C, Joyce P, Waits L (2002) Assessing
  allelic dropout and genotype reliability using
  maximum likelihood. Genetics, 160, 357-366.
• Morin P, Chambers K, Boesh C, Vigilant L (2001)
  Quantitative polymerase chain reaction analysis
  of DNA from noninvasive samples for accurate
  microsatellite genotyping of wild chimpanzees
  (Pan troglodytes verus). Molecular Ecology, 10,
  1835-1844.

41
Quantitative PCR (from Morin et al., 2001)
Relationship between the initial amount of
template DNA in the PCR and both the proportion
of PCRs with amplification product (grey squares)
and the proportion of PCRs with allelic dropout
(black circles).
42
Towards a quality index

• Goal estimate a quality index associated to each
  sample.
• This quality index should allow comparisons among
  samples, loci, and studies.
• Restricted to the situation where the
  multiple-tube approach is used.

43
Towards a quality index

• The estimation of the quality index (QI) is based
  on the analysis of the whole set of
  electropherograms produced when using the
  multiple-tube approach.
• For each locus of a given sample, a QI is
  estimated using the following steps
• Step 1 estimation of the most likely consensus
  genotype
• Step 2 estimation of the score for each repeat
• Step 3 estimation of the quality index for the
  locus

44
Towards a quality index

• Step 1 estimation of the most likely consensus
  genotype after simultaneous observation of the
  electropherograms corresponding to the different
  repeats of this locus. An allele is considered
  only if it is present at least twice among the
  different repeats.
• Step 2 estimation of the score for each repeat.
  If the electropherogram at one repeat corresponds
  to the consensus genotype, the score "1" is
  assigned, otherwise the score "0" is assigned,
  whetever the differences.
• Step 3 estimation of the QI for the locus. The
  scores assigned to each repeat are summed, and
  divided by the number of repeats.
• Step 4 estimation of the mean QI per locus and
  per individual.

45
Additional rules

• No signal is scored as "0".
• Electropherograms with an additional allele are
  scored as "0".
• If the less intense allele is less than 20 of
  the most intense allele, a score of "0" is given.

46
Quality index example 1
Multiple-tube approach, 8 repeats
47
Quality index example 2
Multiple-tube approach, 8 repeats
0
1
0
0
Step 2 score for each repeat
1
0
0
0
48
Quality index example 3
Multiple-tube approach, 8 repeats
1
1
0
1
Step 2 score for each repeat
1
0
0
1
49
Quality indexes for loci, samples, and study
Samples Samples Samples Samples Samples
1 2 3 4 5 mean
Locus 1 0.88 0.63 0.75 0.00 1.00
Locus 2 1.00 0.38 1.00 0.25 1.00
Locus 3 1.00 0.25 0.63 0.25 1.00
mean mean
50
Quality indexes for loci, samples, and study
Samples Samples Samples Samples Samples
1 2 3 4 5 mean
Locus 1 0.88 0.63 0.75 0.00 1.00 0.65
Locus 2 1.00 0.38 1.00 0.25 1.00 0.73
Locus 3 1.00 0.25 0.63 0.25 1.00 0.63
mean mean
51
Quality indexes for loci, samples, and study
Samples Samples Samples Samples Samples
1 2 3 4 5 mean
Locus 1 0.88 0.63 0.75 0.00 1.00 0.65
Locus 2 1.00 0.38 1.00 0.25 1.00 0.73
Locus 3 1.00 0.25 0.63 0.25 1.00 0.63
mean mean 0.96 0.42 0.79 0.17 1.00
52
Quality indexes for loci, samples, and study
Samples Samples Samples Samples Samples
1 2 3 4 5 mean
Locus 1 0.88 0.63 0.75 0.00 1.00 0.65
Locus 2 1.00 0.38 1.00 0.25 1.00 0.73
Locus 3 1.00 0.25 0.63 0.25 1.00 0.63
mean mean 0.96 0.42 0.79 0.17 1.00 0.67
53
Quality indexes for samples
54
Quality indexes for loci
55
Non-invasive census of the brown bears from the
Deosai National Park (Pakistan)
56
Non-invasive census of the brown bears from the
Deosai National Park (Pakistan)
57

Genotyping errors

• Definition
• Non-invasive sampling and genotyping errors
• Causes of genotyping errors
• Quantifying genotyping errors
• Consequences of genotyping errors
• How to limit genotyping errors and their impact?

58
Causes of genotyping errors

• Very diverse, complex, and sometimes cryptic
  origins.
• Grouping errors into discrete categories
  according to their causes is challenging.
• DNA sequence
• Low DNA quantity or quality
• Biochemical artifacts
• Human errors

59
"A" artifact
CGATCGTTAATCAGAATGCATACCGCA GCTAGCAATTAGTCTTACGT
ATGGCG
60
(No Transcript)
61
"A" artifact example
62
Three solutions

• Enzymatic treatment of the PCR product with T4
  DNA polymerase to remove the additional "A".
• Modification of the PCR parameters.
• Modification of the 5' end of the non-labeled
  primer.

63
Modification of the PCR parameters
64
Modification of the primer principle
65
Modification of the primer result
66
(No Transcript)
67
PCR conditions
68
Modifications reducing the "A"
69
Modifications enhancing the "A"
70
(No Transcript)
71
PCR conditions
72
Experiments
73
Original XA modified XT
Enhance the "A"
CGATCGTTAATCAGAATGCATACCGCA GCTAGCAATTAGTCTTACGT
ATGGCGT
original
CGATCGTTAATCAGAATGCATACCGCTA GCTAGCAATTAGTCTTACG
TATGGCGA
modified
74
Original XT modified XA
Reduce the "A"
CGATCGTTAATCAGAATGCATACCGCTA GCTAGCAATTAGTCTTACG
TATGGCGA
original
CGATCGTTAATCAGAATGCATACCGCA GCTAGCAATTAGTCTTACGT
ATGGCGT
modified
75
"A" artifact conclusion (1)

• Do not perform a final elongation without reason
  (this elongation enhance the "A" artifact).
• Even at 4C, the "A" is slowly added.
• Use the most simple PCR protocol at the
  beginning.
• In case of scoring difficulty, identify if the
  marker is "2-steps" or "3-steps".
• Modify the primers and the PCR protocol if
  necessary.

76
"A" artifact conclusion (2)

• To enhance the "A"
• Final elongation (up to 90 minutes)
• Put a "G" at the 5' end of the reverse primer
• Add a "PIGtail" GTGTCTT
• To reduce the "A"
• 2-step PCR without final elongation
• Put a "T" at the 5' end of the reverse primer
• Good luck with multiplexing markers!

77
DNA molecules interactions

• Cause DNA sequence flanking the marker
• No or less efficient amplification because of a
  mutation in the target primer sequence (null
  allele)
• Insertion or deletion in the amplified fragment
  (size homoplasy of different alleles)
• In heterozygous individuals, preferential
  amplification of one allele when its denaturation
  is favoured (allelic dropout)

78
Sample quality

• Cause 1 Low DNA quality or quantity
• In heterozygous individuals, amplification of
  only one allele (allelic dropout)
• In heterozygous individuals, preferential
  amplification of the shorter allele (short allele
  dominance)
• Cause 2 Contamination of the DNA extract
• Amplification of a contaminant allele (mistaken
  allele)
• Cause 3 Extract quality
• No or less efficient restriction/amplification
  due to inhibitors (allelic dropout)

79
Biochemical artifacts and equipment

• Cause 1 Low quality reagents
• Allelic dropout, mistaken alleles
• Cause 2 Equipment precision or reliability
• Allelic dropout, mistaken alleles
• Cause 3 Taq polymerase errors
• False allele
• Cause 4 Lack of specificity
• Mistaken allele
• Cause 5 Electrophoresis artifacts
• Size homoplasy of different alleles, mistaken
  alleles

80
Human factor

• Cause 1 sample manipulation
• Confusion between samples (e.g. mislabelling or
  tube mixing) (mistaken allele(s))
• Cause 2 Experimental error
• Contamination with an exogenous DNA or
  cross-contamination between samples (mistaken
  allele(s))
• Use of an inappropriate protocol (reagent
  forgotten, wrong hybridization temperature,
  primers, or concentrations of reagents) (allelic
  dropout, mistaken allele(s))
• Cause 3 Data handling
• Misreading of the profile or misidentification of
  the fluorescent peak (mistaken allele)
• Miscopying or confusion of the genotypes in the
  database (mistaken allele)
• Computing data bug in the database/analysis
  program (mistaken allele)

81

Genotyping errors

• Definition
• Non-invasive sampling and genotyping errors
• Causes of genotyping errors
• Quantifying genotyping errors
• Consequences of genotyping errors
• How to limit genotyping errors and their impact?

82
Quantifying genotyping errors

• Different estimates, based on replicates within a
  dataset, have been defined to quantify error
  rates.
• Some metrics have been proposed for specific
  errors such as allelic dropouts or false alleles.
  
• More global metrics, which take into account all
  types of detectable genotyping errors, are also
  commonly used although they have never been
  explicitly defined.

83
Quantifying genotyping errors

• First, a reference genotype must be defined as
  the genotype that minimizes the number of errors
  in the comparison among replicates. Several
  reference genotypes may exist. If only two
  replicates are performed and give contradictory
  genotypes, either one or the other can be
  considered as the reference.
• The calculation of error rates is based on the
  number of mismatches between the reference
  genotype and the replicates.

84
Quantifying genotyping errors

• n individual single-locus genotypes have been
  replicated t times.
• For diploid individuals, 2nt alleles and nt loci
  are typed and can be compared to the reference.
• Estimation of the error rates at the allelic,
  locus, multilocus, and reaction levels.

85
Quantifying genotyping errors

• Mean allelic error rate
• Mean error rate per locus
• Error rate per multilocus genotype
• Error rate per reaction

86
Mean allelic error rate ea

• The mean allelic error rate ea is the ratio
  between ma, the number of allelic mismatches, and
  2nt, the number of replicated alleles.
• For microsatellite markers, the error rate per
  allele can also be estimated for each particular
  allele to eventually point out error-prone
  alleles (for example, alleles prone to dropouts).

87
Mean error rate per locus el

• The mean error rate per locus is the ratio
  between ml, the number of single locus genotypes
  including at least one allelic mismatch, and nt,
  i.e. the number of replicated single locus
  genotypes.
• This metric can also be estimated for each
  particular locus, to help identifying the
  error-prone loci.
• As it can be compared between studies and
  samples, it should become the standard metric.

88
Error rate per multilocus genotype eobs (1)

• The observed error rate per multilocus genotype
  eobs is the ratio between mg, the number of
  multilocus genotypes including at least one
  allelic mismatch, and nt, the number of
  replicated multilocus genotypes.
• This metric is particularly informative for
  individual identification, parentage analyse or
  population size estimation.

89
Error rate per multilocus genotype eind (2)

• If genotyping errors occur independently among l
  loci (which is very unlikely), the error rate per
  multilocus genotype eind is deduced from the
  single-locus error rate ei at each locus i

90
Error rate per reaction er

• The error rate per reaction er is the ratio
  between ml, the number of single-locus genotypes
  including at least one allelic mismatch and r,
  the total number of reactions.
• This metric is equivalent to the mean error rate
  per locus when the PCR reaction involves one
  locus or to the multilocus error rate when all
  loci are amplified in a single multiplex reaction.

91
Estimation of the error rates per allele and per
locus, for four replicates (t4) of three
individuals (n3)
replicates replicates replicates replicates Reference genotype Reference genotype Reference genotype Error rate per allele Error rate per locus
1 2 3 4 Reference genotype Reference genotype Reference genotype Error rate per allele Error rate per locus
Genotyped individuals Ind 1 Al 1 A A B A A 3/8 2/4
Genotyped individuals Ind 1 Al 2 A B C A A 3/8 2/4
Genotyped individuals Ind 2 Al 1 A B B A A or B 2/8 2/4
Genotyped individuals Ind 2 Al 2 B B B B B or B 2/8 2/4
Genotyped individuals Ind 3 Al 1 A A A A A 1/8 1/4
Genotyped individuals Ind 3 Al 2 C C B C C 1/8 1/4
mean mean mean 1/4 5/12
92
Reference papers

• Bonin A, Bellemain E, Bronken Eidesen P, Pompanon
  F, Brochmann C, Taberlet P (2004) How to track
  and assess genotyping errors in population
  genetics studies. Molecular Ecology, 13,
  3261-3273.
• Hoffman J, Amos W (2005) Microsatellite
  genotyping errors detection approaches, common
  sources and consequences for paternal exclusion.
  Molecular Ecology, 14, 599-612.

93
Example of error rates

• Bonin et al. (2004)
• Bear tissues 0.008 per locus
• Bear faeces 0.019 per locus
• AFLP 0.019 to 0.026 per locus
• Hoffman and Amos (2005)
• 2000 antarctic fur seal genotyped at 9
  microsatellite loci
• 0.0013 to 0.0074 per locus
• Human errors are the most important cause

94

Genotyping errors

• Definition
• Non-invasive sampling and genotyping errors
• Causes of genotyping errors
• Quantifying genotyping errors
• Consequences of genotyping errors
• How to limit genotyping errors and their impact?

95
Consequences of genotyping errors

• Linkage and association studies
• Individual identification
• Population genetic studies

96
Linkage and association studies

• Erroneous genotypes might markedly affect linkage
  and association studies by hiding the true
  segregation of alleles.
• The impact on the results is measured by
  experimental or simulation studies and can be
  serious even for low error rates (e.g. lt 3).
• For example, in linkage studies, genotyping
  errors can affect the haplotype frequency and
  eventually lead to inflation of genetic map
  lengths.
• Error rates as low as 3 have serious effects on
  linkage disequilibrium analysis, and a 1 error
  rate can generate a loss of 53-58 of the linkage
  information for a trait locus. However, modest
  error rates might be tolerable in situations that
  do not involve rare alleles, as in QTL studies.

97
Linkage and association studies

• In association studies, because recombination is
  rare, errors mostly affect non-recombinant
  genotypes, which are then erroneously interpreted
  as being the result of recombination. Errors
  therefore decrease the power for detecting
  associations.
• The importance of the experimental design has
  also to be emphasised as it can generate errors
  that are not randomly distributed across
  phenotypes (i.e., differential errors). This can
  be the case when controls and cases are genotyped
  in different assays while investigating the
  genetic basis of a disease. Differential and
  non-differential errors can have opposite
  consequences on the rate of false positive in
  statistical tests of association.

98
Individual identification

• Genotyping errors can strongly affect individual
  identification studies that are based on
  multilocus genotypes by erroneously increasing
  the number of genotypes observed in a population
  sample.
• In census studies of rare or elusive species, the
  population size can be estimated based on the
  identified genotypes from non-invasive samples
  collected in the field (e.g., hair or faeces). In
  this context, genotyping errors can lead to a
  serious overestimate of population size.
• A 200 overestimate of population size has been
  found with a 5 error rate per locus when using 7
  to 10 loci for genotype identification (Creel et
  al., 2004). Such an overestimate obviously
  increases with the number of loci and with the
  number of samples per genotype.

99
Individual identification

• Genotyping errors also have a huge impact in
  parentage analysis, generating wrong paternity or
  maternity exclusion.
• Such information on population size and structure
  are required in conservation biology, and their
  inaccurate estimation due to genotyping errors
  could result in wrong decision in population
  management.
• In forensic DNA analyses, a false multilocus
  genotype can prevent the identification of a
  corpse or lead to erroneous identification (or
  exoneration) of criminal offenders.

100
Population genetic studies

• Most of the studies that take genotyping error
  into account in population genetics are those
  that use non-invasive samples, which are
  error-prone because of the low quality and/or
  quantity of DNA.
• However, it has been demonstrated that even with
  high quality DNA the error rate might not be
  negligible.
• The impact of genotyping errors remains largely
  unknown in this field, because very few studies
  have dealt with this topic until now.
• Genotyping errors may lead to erroneous allele
  identification or allele frequencies, resulting
  in wrong Fst estimates, false migration rates, or
  false detection of selection or population
  bottlenecks.

101
Population genetic studies

• Analyses based on allele frequencies will be less
  affected by errors than those based on individual
  identification (e.g., parentage analysis), but
  will be sensitive to sampling effects.
• The apparent low impact of scoring differences
  has been demonstrated on an AFLP data set that
  was scored by two different scientists. The two
  scorers had only 38 of the marker loci in
  common, but the same biological conclusions about
  population genetic structure was extracted from
  the data. In this study, the robustness of the
  inferred biological message was certainly due to
  the redundancy of the information contained in
  the large amount of AFLP markers (more than 200
  polymorphic loci screened by both scorers).
• Population genomics studies looking for selected
  markers among several hundred markers would be
  very sensitive to the impact of genotyping error,
  especially if the errors are population-specific.
  There is a great need for studies on the impact
  of genotyping error in this new emerging field.

102

Genotyping errors

• Definition
• Non-invasive sampling and genotyping errors
• Causes of genotyping errors
• Quantifying genotyping errors
• Consequences of genotyping errors
• How to limit genotyping errors and their impact?

103
How to limit genotyping errors and their impact?

• The worse situation arises when a scientist
  realises at the end of a study that the data were
  not reliable due to genotyping errors, and that
  the dataset is not retrievable.
• Such situations are almost never reported in the
  literature, but their occurrence is probably not
  rare.
• Therefore, it is important to take into account
  the possibility of genotyping errors when
  designing the experimental protocol.

104
How to limit genotyping errors and their impact?

• The strategy consists in demonstrating, via an
  appropriate procedure, that the data produced and
  the results obtained are reliable.
• The diversity of case studies, error causes, and
  laboratory contexts makes it impossible to
  propose a universal and simple procedure.
• As a consequence, the possible solutions to limit
  the occurrence and the impact of genotyping
  errors are case-specific.
• The optimal strategy will be determined by
  several factors, such as the biological question,
  the tolerable error rate, the sampling
  possibilities, the equipment and technical skills
  that are locally available, the financial support
  and time constraints.

105
How to limit genotyping errors and their impact?

• General recommendations
• Limiting the production of errors during
  genotyping
• Cleaning the dataset after genotyping
• Analysing data taking into account the errors
• Towards quality processes for genotyping
• Practicals establishing reliable experimental
  protocols (case studies)

106
General recommendations (1)

• A first step consists in checking that the
  genotyping experiments necessary to reach the
  scientific goal are realistic according to the
  sample quality and the technical skills available
  (bad sample quality and limited technical skills
  obviously influence the error rate).
• A second step involves carrying out a pilot study
  designed to first evaluate the theoretical error
  rate compatible with the data analysis, and then
  to estimate the real error rate based on the
  analysis of a subset of the samples.
• Finally, it is important to be aware of potential
  problems all along the experimental procedure,
  even after a successful pilot study, from
  sampling to data analysis.

107
General recommendations (2)

• Quality controls should be performed in real time
  during each step and each batch of experiments.
• They should also be diverse for being able to
  detect as many types of errors as possible. For
  example, highly reproducible errors such as null
  alleles cannot be detected by replicates, and
  require Hardy-Weinberg tests or inheritance
  studies. On the contrary, stochastic allelic
  dropouts might not be detected by Hardy-Weinberg
  tests, but by replicating the genotyping assays.
• Control procedures are costly and time consuming.
  Thus the effort for reducing the error rate must
  be adapted to the foreseeable impact of the
  genotyping errors.
• Because genotyping errors may be generated even
  with high quality standards, and because they
  cannot be all detected, efforts must be directed
  towards limiting both their production and their
  subsequent impact.

108
Limiting the production of errors during
genotyping (1)

• Given that human factors can be the main issue
  during genotype production, the most efficient
  approach is to concentrate first on minimizing
  human error.
• Only well-trained bench scientists/technicians
  should be involved, as suggested by quality
  assurance standards for forensic DNA testing
  laboratories.
• Only standardized and validated procedures should
  be used.
• Human manipulation should be reduced as much as
  possible according to the automation
  possibilities, from all handling and pipeting
  steps to allele scoring. However, for allele
  scoring, software packages are not yet
  sophisticated enough to prevent scoring errors.
  Semi-automated scoring followed by human visual
  inspection appears to be the most reliable
  procedure.

109
Limiting the production of errors during
genotyping (2)

• Limiting genotyping errors during laboratory
  experiments requires the systematic use of an
  appropriate number of positive and negative
  controls, but also requires the implementation of
  replicates for real-time error detection and
  error rate estimation.
• In every situation, even with high quality DNA,
  replicating five to 10 of the samples has been
  recommended, but the amount can vary according to
  the goal of the study and the potential impact of
  errors.
• As far as possible, these replicates have to be
  carried out blind and independently.
• This involves implementing the blind process from
  the beginning of the experiment, by carrying out
  a systematic duplication of the samples during
  sample collection. Such a procedure will not only
  allow to detect all laboratory errors, but will
  also pick up handling errors at any stage of the
  analysis. Moreover, comparing blind samples and
  original experiments will produce a fair estimate
  of the error rate.

110
Limiting the production of errors during
genotyping (3)

• When genotyping errors are highly probable, blind
  replicates are still necessary but not
  sufficient. The systematic replication of each
  genotyping assay (i.e., multiple-tube approach)
  may be required to define the consensus
  genotypes.
• There is a trade-off between the cost of the
  experiments and the reliability of the genotypes.
  
• One role of the pilot study is to determine the
  optimal number of replicates required.
• In some cases, errors can also be detected by
  replicating the genotyping process using a
  different technology such as sequencing whose
  error rates are typically lower than standard
  genotyping technologies.

111
Cleaning the dataset after genotyping (1)

• Even if all erroneous genotypes detected during
  the experiments are removed, and eventually
  corrected after re-genotyping, some undetected
  errors will certainly remain in the data set. A
  part of them can still be detected or suspected
  by looking at the concordance with independent
  data.
• The power of detecting errors by consistency with
  independent data can influence the strategy for
  limiting errors.
• It might be more efficient to retype erroneous
  genotypes detected by consistency checking than
  running a large proportion of blind replicates.

112
Cleaning the dataset after genotyping (2)

• Testing Hardy-Weinberg equilibrium is common to
  check the quality of the data, under the
  assumption that a high error rate implies
  disequilibrium. However, many other causes can
  lead to disequilibrium, including selection,
  inbreeding and population admixture.
• Moreover, just a few types of error might produce
  disequilibrium, such as null alleles and allelic
  dropouts.
• Therefore there is still a need for other
  controls and replicates for detecting errors that
  are compatible with Mendelian inheritance and
  Hardy-Weinberg equilibrium.

113
Cleaning the dataset after genotyping (3)

• Several computer programs specifically designed
  to detect potential errors are now available.
• Most of them check for Mendelian consistency
  and/or Hardy-Weinberg equilibrium, and are
  commonly used for pedigree analyses and linkage
  studies.
• Some others have been developed to track some
  kinds of errors that can be compatible with
  Mendelian inheritance or Hardy-Weinberg
  equilibrium. For example, some detect a spurious
  excess of recombinants in linkage studies and
  others focus on inconsistencies between
  replicates.

114
Cleaning the dataset after genotyping (4)

• Removing errors might not reduce bias, depending
  on the number and kind of errors detected and the
  bias each one creates.
• For instance, when correcting Mendelian-incompatib
  le genotypes by retyping or removing families in
  which they occur, the undetected errors can
  produce an excess of false positives for some
  family-based association tests. This problem has
  been addressed by developing an appropriate
  Likelihood Ratio Test based on a general genotype
  error model.
• In general, taking into account the occurrence of
  errors in the analysis is crucial, especially for
  large or error-prone data sets.

115
Example of genotyping error
116
Computer programs for detecting errors (1)
GEMINI http//pbil.univ-lyon1.fr/software/Gemini/g
emini.htm PAWE http//linkage.rockefeller.edu/paw
e/ PREST http//fisher.utstat.toronto.edu/sun/Sof
tware/Prest/ Pedcheck http//watson.hgen.pitt.edu/
register/docs/pedcheck.html PedManager http//www.
broad.mit.edu/ftp/distribution/software/pedmanager
/ MENDEL http//www.genetics.ucla.edu/software/ SI
MWALK http//www.genetics.ucla.edu/software/ Genoc
heck http//softlib.rice.edu/geno.html R/QTL http
//www.biostat.jhsph.edu/kbroman/qtl/
117
Computer programs for detecting errors (2)
CERVUS http//helios.bto.ed.ac.uk/evolgen/cervus/c
ervus.html GIMLET http//pbil.univ-lyon1.fr/softwa
re/Gimlet/gimlet.htm RelioType http//www.cnr.uida
ho.edu/lecg/pubs_and_software.htm Micro-checker ht
tp//www.microchecker.hull.ac.uk DROPOUT http//ww
w.fs.fed.us/rm/wildlife/genetics PARENTE http//ww
w2.ujf-grenoble.fr/leca/membres/manel.html PAPA h
ttp//www.bio.ulaval.ca/louisbernatchez/downloads_
fr.htm PseudoMarker http//www.helsinki.fi/tsjunt
un/pseudomarker/ TDTae ftp//linkage.rockefeller.
edu/software/tdtae2/ LRTae ftp//linkage.rockefel
ler.edu/softare/lrtae/
118
How to limit genotyping errors and their impact?
119
How to limit genotyping errors and their impact?
120
Towards quality processes for genotyping (1)

• In every scientific discipline, the reliability
  of the conclusions strongly depends on the
  quality of the data.
• For geneticists, genotyping errors may strongly
  affect the results.
• The protocol used for minimizing the occurrence
  of errors, the methods for error detection, and
  the estimated error rate should be provided for
  each study.
• With this information, it will be possible to
  assign to each genotype a quality index, allowing
  the scientific community to have a critical view
  when unexpected results are published.

121
Towards quality processes for genotyping (2)

• More and more studies, often in the context of
  international programs, generate enormous
  datasets that cannot be produced in a single
  laboratory.
• The reproducibility of genotyping becomes more
  and more important.
• Even for markers known to be robust (SNPs,
  microsatellites, AFLPs), differences may appear
  among laboratories and over time within the same
  laboratory.

122
Towards quality processes for genotyping (3)

• Expression studies using microarray experiments
  are known to be error-prone, and the scientific
  community reacted in designing strict standards
  the Minimum Information About a Microarray
  Experiment (MIAME) produces a checklist to guide
  authors and journal editors to ensure that data
  are made publicly available in a format that
  enables unambiguous interpretation and potential
  verification of the conclusion. It includes
  several steps verifying for instance experiment
  design, sample preparation, and data measurement.

123
Towards quality processes for genotyping (4)

• Genotyping errors have been identified since the
  early beginning of molecular genetics.
• Their consequences in statistical genetics were
  pointed out in 1957, and null alleles in blood
  groups have been recognised since 1938.
• They remained too often neglected in the past and
  it is clear that they merit much more attention
  according to their dramatic impact in some
  studies.
• Recently, many papers have dealt with genotyping
  errors, and it seems that the scientific
  community begin to realise their importance.

124
Towards quality processes for genotyping (5)

• The fields of ancient DNA and gene expression
  suffered a crisis of confidence, with series of
  erroneous papers published in leading journals.
  As a consequence, these two scientific
  communities were able to set up strict standards
  that promoted data quality and solved the crisis.
  
• In population genetics, the situation is
  different because only a few erroneous papers
  have been published. Therefore, this community
  has not been apparently strongly pushed to
  establish strict standards. Another explanation
  for the delay in establishing strict standards
  might be related to the complexity of the
  problems.
• According to the recent awareness about
  genotyping errors occurrence and about their
  potential impact, it can be predicted that more
  and more attention will be paid to these
  difficulties when designing experimental
  protocols and publishing results.

125
How to limit genotyping errors and their impact?

• General recommendations
• Limiting the production of errors during
  genotyping
• Cleaning the dataset after genotyping
• Analysing data taking into account the errors
• Towards quality processes for genotyping
• Practicals establishing reliable experimental
  protocols (case studies)

126
Practicals establishing reliable experimental
protocols (case studies)

• Identify the question
• Design the pilot study
• Design the sampling strategy
• Design the experimental protocol that will limit
  as much as possible the genotyping errors
• Design the data analysis process

127
Practicals establishing reliable experimental
protocols (case studies)

• A good approach is to consider that the
  experimental protocol used will produce only
  artifacts and that all the samples have been
  mixed during the process.
• The best strategy is to try to establish an
  experimental protocol that demonstrates that no
  artifacts have been produced, and that nothing
  has been mixed up during the process.

128
Phylogeography of Capercallie

• What is the status of the Pyrenean and Cantabrian
  populations?

129
Phylogeography of Capercallie

• 92 faeces samples covering the whole range (23
  localities).
• Sequencing of the mitochondrial DNA control
  region (443 bp) in both directions.
• Unexpected results that could come from tube
  mixing!

130
(No Transcript)