Discovery of Genes for Improved Cellulose

1
Discovery of Genes for Improved Cellulose and
Cellulose-Extractability from Poplar Secondary
Xylem Jill L Wegrzyn1, Jennifer M. Lee2, Andrew
J. Eckert2, Charlyn J. Suarez2 Brian J.
Stanton3, Mark F. Davis4, Chung-Jui Tsai5, David
B. Neale1 1Department of Plant Sciences,
University of California at Davis, Davis,
CA 2Department of Evolution and Ecology,
University of California at Davis, Davis,
CA 3Genetic Resources Conservation Program,
Greenwood Resources, Portland, OR 4National
Renewable Energy Lab, Golden, CO 5School of
Forest Resouces, Michigan Technical University,
Hougton, MI
2
Project Objectives

• Resequence 40 candidate genes using a discovery
  panel of 15 unrelated poplar individuals
• Identify SNPs in the 40 genes using an automated
  alignment and SNP calling bioinformatics pipeline
• SNP genotype 456 poplar clones for 1536 SNPs
  (Illumina Golden Gate assay)
• Harvest wood increment cores from 2-3 ramets of
  each of the 456 poplar clones (1100 trees in
  total)
• Molecular Beam Mass Spectrometry (MBMS) analysis
  on all 1100 wood cores to develop secondary xylem
  metabolomic profiles
• Association genetics analyses to identify genes
  controlling cellulose quantity and quality
  phenotypic variation in poplar

3
Poplar Biofuels Genome ProjectProject Overview
4
Selected Candidate Genes 40 Genes highly
expressed in wood-forming tissues and associated
with lignin and cellulose biosynthesis
5
(No Transcript)
6
Primer Design and SequencingAgencourt Biosciences

• Primer Design
• mRNA sequences were used to direct custom
  software to use 1000 bp
• upstream along with intronic sequence from the
  poplar genome
• 517 primers were designed across 40 genes
• 203 non-overlapping primers were finally
  selected based on
• quality score, position, homopolymer regions
  (bioinformatic validation)
• Goal Fully re-sequence 40 candidate genes to
  facilitate SNP discovery
• between 3 and 12 amplicons/gene
• total of 202 amplicons from Agencourt
• forward and reverse sequencing

7
Candidate Genes Re-Sequenced from a Panel of 15
Unrelated Poplar Clones
DNA landmarks responsible for extraction
8
Alignment and SNP Calling PipelineChallenges in
High-Throughput SNP Identification

• Alignment
• Critical in the automation of base calls
• Commonly used Phrap (from PhredPhrap) is an
  assembler and is NOT ideal for alignments
• Many commonly used aligners work best with
  protein sequences or with a reference sequence
• Preservation of quality scores for input into SNP
  identification programs
• Speed for high-throughput programs
• Automated SNP Calls
• Reference Sequence Required
• Traditional approaches without reference sequence
  include eSNPs (human, maize, and pine)
• -Very little redundancy outside of abundant genes
• -Overall high number of false positives (single
  pass reads)
• Not specific to frequencies observed in different
  organisms
• High number of false positives in currently
  accepted methods
• Polybayes PolyPhred

9
Identification of SNPs in the 40 Candidate
GenesAutomated Alignment and SNP identification
Pipeline
Re-Sequencing data from Agencourt Initial
Processing Base Calling Sequence
Alignment SNP Identification Machine
Learning Data Storage Release
10
Base Calling and Sequence Alignment
Modified PhredPhrap allows for trimming of bases
from start and end of sequence based on trace
quality
Ace2FASTA Converts native PhredPhrap output (ace
file) into an unaligned FASTA file
ProbconsRNA Optimal DNA sequence alignment program
AlignedContig2ReadFASTA Provides single
multifasta file with all reads aligned to the
contig from PhredPhrap AND the contigs alignment
to the other contigs from probconsRNA
FASTA2Ace Converts resulting FASTA file back into
ace file for SNP Identification
11
Alignment and SNP IdentificationSNP
Identification Overview

• Examine features to improve the accuracy of SNP
  location prediction
• Utilize machine learning to apply the features
• Refine the accuracy of the learning algorithm
  through adjustments to feature representation
• Utilize the classifier against the large
  re-sequenced set to improve accuracy of SNP calls
  originating from Polybayes and Polyphred

12
Alignment and SNP IdentificationExisting SNP
Identification Software

• Polyphred
• http//droog.mbt.washington.edu/PolyPhred.html
• PolyPhred identifies potential SNPs using the
  base calls and peak information provided by Phred
  and the sequence alignments provided by Phrap
• SNP score based on base quality and sequence
  depth
• Polybayes
• http//genome.wustl.edu/tools/software/polybayes.c
  gi
• Fully probabilistic SNP detection algorithm that
  identifies SNPs based on discrepancies at a given
  location of a multiple alignment.
• SNP score is based on a Bayesian-statistical
  formulation and can take-in prior frequency
  information

13
Alignment and SNP IdentificationFeature Selection
Description Representation
Sequence Depth Continuous
Variation Type Categorical
Polybayes Score Continuous
Polyphred Score Continuous
Freq of major/minor alleles Continuous
Max quality of major/minor alleles Continuous
Local average quality Continuous
Overall average quality Continuous
Alignment Quality Continuous
14
Alignment and SNP IdentificationFeature
Representation

• Sequence depth
• - Count of number of sequences in the alignment
  at the position of variation.
• All sequences in the alignment may not overlap at
  the position of variation number is different
  from the total number of the sequences in the
  alignment
• Variation type
• Variation type can be SNP or INDEL.
• PolyBayes score
• PolyBayes program assigns a Bayesian posterior
  probability value for each called SNP using the
  frequency priors given for observing a variation
  at that position.

15
Alignment and SNP IdentificationFeature
Representation

• Polyphred score
• Polyphred assigns a score calculated primarily
  from sequence depth and quality score.
• Base frequencies
• The number of occurrences of different bases at
  the position of variation is important in
  determining a polymorphic position.
• Frequencies of the first (major allele) and the
  second (minor allele) represented as ratio to
  sequence depth.
• Relative distance
• Sequence quality at the ends of the alignment
  tends to be poor due to inherent limitations of
  current sequencing technology.
• SNP position was represented as the ratio of the
  distance in the consensus sequence from the
  closest end, or the relative distance

16
Alignment and SNP IdentificationFeature
Representation

• Sequence quality
• Variation is observed because of a poor quality
  base.
• Based on the base frequencies calculated
• maximum qualities of the major and minor alleles
• average qualities of major and minor alleles
• Alignment quality
• Misalignment of bases caused by sequence
  alignment programs sometimes result in an
  erroneous SNP call.
• In the neighborhood of the SNP (/- 10 bases) all
  the mismatches with the consensus sequence are
  given a penalty and the penalty is more if the
  mismatch is continuous

17
Alignment and SNP IdentificationSNP
Identification Datasets

• Training set for loblolly pine was composed of a
  total of 300 validated sequences.
• Divided to represent the relative percentages of
  sequence source
• Testing set is composed of 120 validated sequence
  sets
• Training set for poplar was composed of 42
  validated sequences selected at random
• Testing set is composed of a total of 30
  validated sequence sets.
• Validation manually observed FP, FN, TP, and TN
  SNP calls through observation of tracefiles in
  Consed.

18
Alignment and SNP IdentificationClassification

• GOAL Prediction
• Learn a function or set of functions that assign
  a record to one of several predefined classes.
• Decision tree C4.5 program is open-source C code
  (WEKA) - J48
• At each point in the construction of the decision
  tree, C4.5 selects the feature to test based on
  maximum information gain.
• The goal is to generate a minimum size tree that
  correctly classifies all the SNP calls in the
  training set.

19
(No Transcript)
20
Alignment and SNP IdentificationEvaluation
Criteria

• Accuracy (TP TN)/total
• Sensitivity TP/(TP FN)
• Specificity TN/(FP TN)

Evaluation J48 Polyphred Polybayes
Accuracy 93.6 76.25 78.02
Sensitivity 88.21 83.22 86.54
Specificity 98.73 N/A N/A
Evaluation J48 Polyphred Polybayes
Accuracy 94.6 79.35 80.24
Sensitivity 90.54 85.01 88.14
Specificity 97.23 N/A N/A
21
PineSAP

• PineSAP alignment improves
• Inaccuracies introduced by using Phrap to align
  sequences
• Time which would be required by using a aligner
  such as ProbconsRNA or ClustalW on its own
• PineSAP has a 98 success rate when used to align
  loblolly resequencing data.
• PineSAP identified a success list of features to
  enhance polymorphism predictions
• PineSAP obtained an overall prediction accuracy
  of 93 in SNP Identification
• PineSAP provided a full alignment and
  polymorphism detection system that can be adapted
  to specific genomes

22
Alignment and SNP IdentificationSNPs Identified

• Total of 202 amplicons
• Number of SNPs Identified - 1486
• Meet a minimum confidence score from the PineSAP
  pipeline
• Average number of SNPs/amplicon 7
• Amplicon length 600 - 700bp
• Remaining SNPs generated from 232 additional
  genes.
• Utilized an eSNP method with publicly available
  EST data and reference genome from JGI.
• Identified a total of 1,232 potential SNPs

23
Alignment and SNP IdentificationSNP Formatting

• Polyphred style output is transformed into
  Illumina style input
• -adding IUPAC codes for SNPs in flanking sequence

24
SNP GenotypingIllumina GoldenGate Assay
25
Alignment and SNP IdentificationIllumina Design
26
Alignment and SNP IdentificationSNP Selection

• All SNP and amplicon information is databased.
• SQL queries can be used to select specific SNPs
• Pair-wise comparisons of all SNPs
• Scores were assigned to each pair of SNPs in each
  amplicon, accounting for distance between the
  SNPs, Illumina score for both SNPs, and frequency
  of minor allele
• We can also use SQL queries to select SNPs and
  minimize additional SNPs in flanking sequence

27
Pyrolysis Molecular Beam Mass Spectrometry
Analysis
cell wall chemistry
lignin
hemicellulose
cellulose
28
(No Transcript)
29
Acknowledgements
Mike Davis
Chung-Jui Tsai
David Neale Jill Wegrzyn Jennifer Lee Andrew
Eckert John Liechty
Brian Stanton Rich Shuren
Funding