Best Practices:
Museum collections are increasingly being used for molecular sequencing,
yet comparative studies on the retrieval and reliability of
microsatellite genotypes from these data sources are not readily
available. Here we show that while museum specimens can recover
reliable, and important genotypes for rare, endangered and elusive
species, additional precautions must be made prior to acceptance of
genotypes.
From our data we recommend a minimum of three successful amplifications
for each marker of interest. The samples which routinely amplified
(HQMS) recovered genotypes with very similar rates of genotype
confirmation between replicates, and when compared to the tissue sample.
Poor performing samples may require additional replication compared to
better performing samples. We noticed in the LQMS that the longer the
microsatellite locus, the worse the marker amplified. This was apparent
when none of the LQMS recovered genotypes for the ~250
bp microsatellite locus GLSA-52. All of the HQMS samples recovered
reliable genotypes across replicates tested here, and for all marker
lengths and types of repeat motif.
Our data separated sample types into three categories, tissue, HQMS and
LQMS, the latter two designations were only applied after many rounds of
PCR and agarose gel visualization. During project design samples should
be evaluated so that adequate replicates of PCR can be performed to
ensure accurate genotypes. Additionally, calibration/confirmation of the
genotypes generated by GBS can be done via CE or other fragment
visualizing instruments (Fragment Analyzer, Advanced Analytical Ankeney,
IA). It is notable that genotypes may be predictably shifted from
comparison of GBS and CE methods as detailed in Barbian et al., (2018).
In order to reduce the inaccurate genotypes, optimization of PCRs should
be performed prior to GBS. The addition of various reagents has been
shown to increase specificity and reduce non-specific amplification, as
has been widely published over the past 30 years (Boleda, Briones,
Farres, Tyfield, & Pi, 1996; Robertson & Walsh-Weller, 1998; J. F.
Williams, 1989). The PCRs performed here incorporated a touchdown
protocol, which starts at a high annealing temperature (60°C) for 2
cycles of PCR before reducing to the lowest annealing temperature of
50°C for 35 cycles. Touchdown PCR was used on the museum specimens and
across loci as it was shown to effectively amplify all microsatellite
markers. Two microsatellites (GS-2 and GS-4) had Bovine Serum Albumin
(BSA) added since, during initial PCR testing, BSA improved
amplification success. GS-4 however, recovered numerous unconfirmed
genotypes, which could be related to input DNA quality, or from poor
performance in PCR. This locus in particular would benefit from
additional optimization in order to determine if non-specific
amplification could be reduced. GS-2 may have also benefited from
additional optimization as that locus recovered numerous flags from the
CHIIMP pipeline including PCR artifacts, PCR stutter and more than two
prominent sequences.
The CHIIMP pipeline worked well on our samples after modification of
published protocols (Barbian et al., 2018). We found it useful to
combine the CHIIMP genotypes with the quality data as determined by the
proportion of reads passing prinseq filtering to evaluate which samples
may be more prone to false/inaccurate genotypes. The combination of
multiple rounds of PCR, prinseq quality filtering and manual evaluation
of CHIIMP results allowed increased confidence in the genotypes
recovered by museum specimens in this study. This process is illustrated
in Figure 2, and summarized here. First, we would assign our samples as
high or low quality following multiple attempts of PCR with agarose gel
visualization. Second, based on these findings, we suggest recovering
minimally 3 successful PCR replicates prior to genotyping. If PCRs
continue to fail, optimization of each locus may be helpful, as well as
evaluation of DNA extracts for the presence of PCR inhibitors, which has
been shown to affect recovery of ancient DNA and environmental DNA
samples (Matheson et al., 2009; McKee, Spear, & Pierson, 2015;
Pontiroli et al., 2011). Once successful amplification has occurred
across all samples and markers, perform library preparation on
successfully amplified PCR products and sequence on an Illumina platform
with adequate insert length for the included microsatellites. Sequence
to a minimum depth of 1000 reads per sample per microsatellite marker.
For our data that would entail 5000 sequences per sample.
Demultiplexed data should have CutAdapt and FastQC performed in order to
run CHIIMP v 0.3.1. Simultaneously, run prinseq as a parallel analysis
to determine the overall quality of the samples. Samples with higher
proportions of low quality reads should be noted as they may be more
prone to erroneous genotypes. When alleles are recovered only in low
quality samples, it is imperative to look at the output from CHIIMP, and
determine if the differences are associated with primer sequences or
repeat elements. If primer sequence varies, manually correct the length
when the entire primer sequence would be included, and ignore primer
site size mismatches in allele calls as this is likely an artifact of
sequencing or amplification errors. Traditionally, fragment size
analysis via CE would ignore peaks outside of the expected size range
via programs like GeneMapper™ (Applied Biosystems). If an allele does
not have a priming site error, it is important to evaluate if the size
shift follows microsatellite evolutionary patterns, for example, if it
is two base pairs shifted in a dinucleotide sequence that makes
evolutionary sense. However, stutter sequences are often frame shifted
by the size of the repeat motif. While CHIIMP evaluates for stutter, and
allows filter manipulation, we found it was possibly including false
alleles due to the nature of our markers. Dinucleotide sourced
microsatellites are more challenging regarding stutter evaluation due to
the short difference in size between true alleles and stutter peaks.
(Barbian et al., 2018; O’reilly, Canino, Bailey, & Bentzen, 2000). In
order to further scrutinize stutter sequences, we calculated the
proportion of reads associated with the various alleles. If one call
makes up a very small percentage (30% the number of reads as the other
allele) it is likely a stutter peak. Further visualization via
electropherograms could illuminate this process if rampant in a locus.
Additionally, the number of reads represented by the allele can also
provide insight into whether or not the polymorphism is due to
sequencing error. By following these practices we reduced our allelic
dropout by about 3% across loci. We also had to remove many of the LQMS
genotypes as we could not be certain they were authentic. The CHIIMP
pipeline also allows for optimization and customization of commands for
recovering more strict versus lenient genotypes. Here we modified the
published parameters based on the depth of coverage of our samples.