Validation of Identity and Ancestry SNP Panels for the Ion PGM

Transcription

1 Validation of Identity and Ancestry SNP Panels for the Ion PGM Christopher Phillips, Carla Santos, Maria de la Puente, Manuel Fondevila, Ángel Carracedo, Maviky Lareu Forensic Genetics Unit, University of Santiago de Compostela

2 Validation of Identity and Ancestry SNP Panels for the Ion PGM Christopher Phillips, Carla Santos, Maria de la Puente, Manuel Fondevila, Ángel Carracedo, Maviky Lareu Forensic Genetics Unit, University of Santiago de Compostela Walther Parson & Mayra Eduardoff, GMI Peter Schneider & Theresa Gross, UHC Runa Daniel, Dennis McNevin & Roland van Oorschot, VPFSD, Melbourne

3 Where does Ion PGM fit into the sequence explosion? Illumina HiSeq LT SOLiD Ion Torrent is designed to read comparatively short fragments at much greater coverage - more reads provide high levels of accuracy Next-generation sequencing (NGS) Massively parallel sequencing MiSeq Ion PGM

4 Considerations for evaluation of Ion PGM SNP panels Interested in the SNPs as much as the detection system, and can ask: are there good and bad performers in any one SNP set detected with Ion PGM? Does performance impact quality of results from low level DNA? For analysing degraded DNA with short amplicon SNPs there is plenty of choice (75 million variants in 1000 Genomes) - so if a SNP performs poorly, it can be easily replaced. SNPs selected for EVC prediction and the best AIMs must work optimally for the test to be informative. Are all SNPs in the multiplex equally well genotyped? What is the genotyping precision (but also, what is the true genotype)? What happens to sequence data when low-level or mixed DNA is the input? The need to align generated sequences to a reference sequence adds a new layer of complexity if context sequence features (homopolymers or indels) occur near the SNP and interfere with its secure alignment.

5 Considerations for evaluation of Ion PGM SNP panels Forensic SNP typing must get the balance right between the desired coverage : multiplex scales : the barcode samples loaded per chip. Ion PGM sequences are spread across the pit capacity in the chips - not easy to set up consistently. Are there coverage outliers? Can we set a minimum coverage limit?

6 Considerations for evaluation of Ion PGM SNP panels Forensic SNP typing must get the balance right between the desired coverage : multiplex scales : the barcode samples loaded per chip. Ion PGM sequences are spread across the pit capacity in the chips - not easy to set up consistently. Are there coverage outliers? Can we set a minimum coverage limit? As SNPs are binary and Ion PGM is very sensitive, it is important to create a secure system to distinguish mixtures from imbalanced heterozygotes.

7 Considerations for evaluation of Ion PGM SNP panels Forensic SNP typing must get the balance right between the desired coverage : multiplex scales : the barcode samples loaded per chip. Ion PGM sequences are spread across the pit capacity in the chips - not easy to set up consistently. Are there coverage outliers? Can we set a minimum coverage limit? As SNPs are binary and Ion PGM is very sensitive, it is important to create a secure system to distinguish mixtures from imbalanced heterozygotes. What are the patterns of allelic balance in the SNP set?

8 Three studies of the Ion PGM system

9 Three studies of the Ion PGM system Combined five established SNaPshot PCRs to create a pool of 136 unique amplicons with bp size range (five-fold increase in marker depth) No balancing made of individual PCRs - 10 µl of purified product (elution columns) pooled for the library preparation then sequenced with 314v1 chips. No Ampliseq Three samples, three input amounts, concordance checked with SNaPshot/Sanger

10 Three studies of the Ion PGM system Combined five established SNaPshot PCRs to create a pool of 136 unique amplicons with bp size range (five-fold increase in marker depth) No balancing made of individual PCRs - 10 µl of purified product (elution columns) pooled for the library preparation then sequenced with 314v1 chips. No Ampliseq Three samples, three input amounts, concordance checked with SNaPshot kit / Sanger Provided with prototype versions of the LT forensic Identity Panel Decided to adopt a simple NIST validation framework centred on the qualified run - here, closely matched protocols for the same control DNAs Four chip types, 16 samples, concordance checked with public data. Ampliseq

11 Three studies of the Ion PGM system Combined five established SNaPshot PCRs to create a pool of 136 unique amplicons with bp size range (five-fold increase in marker depth) No balancing made of individual PCRs - 10 µl of purified product (elution columns) pooled for the library preparation then sequenced with 314v1 chips. No Ampliseq Three samples, three input amounts, concordance checked with SNaPshot kit /Sanger Provided with prototype versions of the LT forensic Identity Panel Decided to adopt a simple NIST validation framework centred on the qualified run - here, closely matched protocols for the same control DNAs Four chip types, 16 samples, concordance checked with public data. Ampliseq

12 Three studies of the Ion PGM system Combined five established SNaPshot PCRs to create a pool of 136 unique amplicons with bp size range (five-fold increase in marker depth) No balancing made of individual PCRs - 10 µl of purified product (elution columns) pooled for the library preparation then sequenced with 314v1 chips. No Ampliseq Three samples, three input amounts, concordance checked with SNaPshot kit / Sanger Provided with prototype versions of the LT forensic Identity Panel Decided to adopt a simple NIST validation framework centred on the qualified run - here, closely matched protocols for the same control DNAs Four chip types, 16 samples, concordance checked with public data. Ampliseq A EUROFORGEN deliverable is to develop ancestry and EVC inference panels. Designed an 128-SNP ancestry set for NGS. Obtained customised primer design service of LT white glove team. Validating this SNP set performance with above framework in five labs. Ongoing

13 Three studies of the Ion PGM system Combined five established SNaPshot PCRs to create a pool of 136 unique amplicons with bp size range (five-fold increase in marker depth) No balancing made of individual PCRs - 10 µl of purified product (elution columns) pooled for the library preparation then sequenced with 314v1 chips. No Ampliseq Three samples, three input amounts, concordance checked with SNaPshot kit / Sanger Provided with prototype versions of the LT forensic Identity Panel Decided to adopt a simple NIST validation framework centred on the qualified run - here, closely matched protocols for the same control DNAs Four chip types, 16 samples, concordance checked with public data. Ampliseq A EUROFORGEN deliverable is to develop ancestry and EVC inference panels. Designed an 128-SNP ancestry set for NGS. Obtained customised primer design service of LT white glove team. Validating this SNP set performance with above framework in five labs. Ongoing

14 Combined five optimised forensic PCRs 136 unique sites SNPforID: plex: 34 Eurasiaplex: 27 Pacifiplex: 28 IrisPlex: 6 Pacifiplex 34plex AIMs IrisPlex SNPforID 52 Amplified three control DNAs with five SNaPshot PCRs QIAquick PCR purification, but no quantitation Eurasiaplex Pooled DNA end-labelled and put into library preparation Sequence coverage: 9947A 0.2 ng SNPforID 52 52plex 34plex Eurasiaplex Pacifiplex IrisPlex

15 Established forensic PCRs worked well 9947A 0.1 ng 0.2 ng 0.3 ng 007 S1 Eurasiaplex dominates top 10% coverage - PCR too strong Obtained three times more coverage than expected from 314 Chips Template amounts did not influence coverage

16 Allelic balance 9947A 007 S1 Allele counts Outlier 10-30% and 70-90% Allele Read Frequency were difficult to interpret These ranges represent ambiguous genotype designations - are these homozygotes with a large number of spurious alternative alleles or heterozygotes with pronounced allelic imbalance? 0.1 ng 0.2 ng 0.3 ng Coverage Outlier Allele Read Frequency ranges correlated with lower coverage If a small proportion of SNPs are consistently imbalanced they may be discounted from analyses

17 Genotyping concordance 97% concordance obtained comparing Ion PGM to SNaPshot - with a further 1.5% genotypes from Sanger sequencing concordant with Ion PGM. No no-calls - 3% - 1.5% 98.5% With SNaPshot With Sanger rs , rs rs one of several tricky homopolymeric alignments

18 Conclusions The PCR multiplex limits of SNaPshot can be readily expanded by combining products of multiple reactions and pooling for library preparation. 4/5 discordant genotypes in Ion PGM had coverage <13x indicating a 20x minimum could be applicable (a value suggested in other NGS studies). Although the average sequence coverage per SNP of 500x was very high, it was highly variable amongst the SNPs in each multiplex. This coverage bias was non-random: the same SNPs had consistently high/low coverage in each sample/input amount. Absence of AmpliSeq likely to affect balance. Furthermore, SNaPshot could be better balanced. Not certain if Ion PGM kit PCRs are equimolar or have molar ratios adjusted to improve balance. Imbalanced heterozygotes correlated with low coverage, but not all low coverage SNPs were imbalanced. So an Allele Read Frequency threshold (10/90-40/60) is preferred approach and needs adjusting for outlier SNPs.

19 Three studies of the Ion PGM system Combined five established SNaPshot PCRs to create a pool of 136 unique amplicons with bp size range (five-fold increase in marker depth) No balancing made of individual PCRs - 10 µl of purified product (elution columns) pooled for the library preparation then sequenced with 314v1 chips. No Ampliseq Three samples, three input amounts, concordance checked with SNaPshot /Sanger Provided with prototype versions of the LT forensic Identity Panel Decided to adopt a simple NIST validation framework centred on the qualified run - here, closely matched protocols for the same control DNAs Four chip types, 16 samples, concordance checked with public data. Ampliseq A EUROFORGEN deliverable is to develop ancestry and EVC inference panels. Designed an 128-SNP ancestry set for NGS. Obtained customised primer design service of LT white glove team. Validating this SNP set performance with above framework in five labs. Ongoing

20 The HID-SNP identity set has undergone several revisions M479 P202 L298 HID-SNP 2.3 prototype (169 SNPs) HID-Ion AmpliSeq Ancestry Panel 45 A-SNPs removed 9 Y-SNPs replaced HID-Ion AmpliSeq Identity Panel (124)

21 The HID-SNP identity set has undergone several revisions M479 P202 L298 HID-SNP 2.3 prototype (169 SNPs) HID-Ion AmpliSeq Ancestry Panel 45 A-SNPs removed 9 Y-SNPs replaced HID-Ion AmpliSeq Identity Panel (124) SNaPshot

22 The HID-SNP identity set has undergone several revisions M479 P202 L298 HID-SNP 2.3 prototype (169 SNPs) HID-Ion AmpliSeq Ancestry Panel 45 A-SNPs removed 9 Y-SNPs replaced HID-Ion AmpliSeq Identity Panel (124) Genplex SNaPshot

23 NIST validation framework chosen for Ion PGM evaluations NIST suggest a simple CE validation MTP for new STRs/kits/protocols: neg 10 ng 1 ng 0.1 ng 0.05 ng ng neg 10 ng 1 ng 0.1 ng 0.05 ng ng Challenging DNA sources SRM2391b now superseded by SRM2391c, but opted for 007 and 9947A forensic control DNAs. Added six staff and seven Coriell genomic controls

24 NIST validation framework chosen for Ion PGM evaluations NIST suggest a simple CE validation MTP for new STRs/kits/protocols: neg 10 ng 1 ng 0.1 ng 0.05 ng ng neg 10 ng 1 ng 0.1 ng 0.05 ng ng Coriell universal genomic controls Challenging DNA sources SRM2391b now superseded by SRM2391c, but opted for 007 and 9947A forensic control DNAs. Added six staff and seven Coriell genomic controls NA06994 EUR male NA07000 EUR female NA07029 male child NA18498 AFR HG00403 E ASN NA10540 OCE NA11200 AME Sensitivity 007 & 9947A plus 12th Century adna (0.45 ng) Concordance Inter/Intra-lab plus vs. online data for Coriells Mixture Male:female staff at above five ratios (duplicated) Qualifying run Closely matched Ion PGM protocols IMU/UHC/USC

25 Key analysis parameters Ion Torrent Suite (plugin: VariantCaller) used Somatic & Germline parameters Coverage number of target sequences per chip, per sample, per SNP Allele Read Frequency - how much of each allele-carrying sequence is detected and in what ratio Base misincorporation - the number of incorrect bases detected Strand bias - the ratio of sequences in each direction

26 Coverage

27 Coverage 2500 Ranked mean coverage per marker Ranked mean coverage per SNP 2000 > 300 > 400 > 500 > 1000 > > Y-SNPs A-SNPs

28 Coverage 2500 Ranked mean coverage per marker Ranked mean coverage per SNP 2000 > 300 > 400 > 500 > 1000 > > Y-SNPs A-SNPs Y-SNPs 0 > 10 > 100 > 200 > 300 > 400 > 500 > 1000 > 1500 > 5000 Increasing mean coverage per SNP Increasing mean coverage per sample A-SNPs

29 Coverage 2500 Ranked mean coverage per marker Ranked mean coverage per SNP 2000 > 300 > 400 > 500 > 1000 > > Y-SNPs A-SNPs Y-SNPs 0 > 10 > 100 > 200 > 300 > 400 > 500 > 1000 > 1500 > 5000 Increasing mean coverage per SNP Increasing mean coverage per sample A-SNPs The majority of the topmost samples were: amplified with 100pg degraded DNA too many samples loaded on a chip 20x minimum coverage limit

30 Allele Read Frequency balance

31 Allele Read Frequency balance Reference allele frequency / total allele frequency SNP Identified as ARF outliers by Børsting only Identified as ARF outliers by this study only Identified as ARF outliers by both

32 Base misincorporations

33 Base misincorporations Misincorporation as proportion of total coverage Non-specific base misincorporation (e.g. C or T in an A/G SNP) Reference or alternative allele misincorporations (e.g. low levels of A in GG homozygotes) 0.25% overall rate of misincorporation - applicable to nearly all SNPs <1.5% misincorporation of expected bases <3% misincorporation of a non-specific 3rd/4th base 6 C reads G reads A reads 4 4 T reads 2 2 Total Y-SNP sequences detected in females: 34 (in >2 million sequences from 6 analyses)

34 Strand bias

35 Strand bias rs430046, rs , rs , rs rs , rs576261, rs , rs , rs

36 IGV: rs C T A G Insertion (INS) Direction Deletion (DEL) The SNP target base calls unequivocally record a GG homozygote but sequences were generated from 355 forward strands and 2 reverse strands = strand bias)

37 IGV: rs C T A G Insertion (INS) Direction Deletion (DEL) One deletion in ~95% of sequences in the forward strand Twelve direction-based deletion sites recorded in the reverse strand, including the target SNP and two clustering SNP sites 1 and 3

38 Concordance 99.8% concordance from: inter-run (6009/6022); inter-lab (3751/3763); vs. online data (1621/1624) NN NN NG GN T is the incorrect base call (A/C SNP) T is the incorrect base call (A/C SNP) One Ion PGM discordancy in a single sample No call on either allele in Complete Genomics No call on either allele in Complete Genomics No call for 1st allele in Complete Genomics No call for 2nd allele in Complete Genomics Likely 1000 Genomes-Phase 1 error Likely 1000 Genomes-Phase 1 error

39 Concordance 99.8% concordance from: inter-run (6009/6022); inter-lab (3751/3763); vs. online data (1621/1624) NN NN NG GN T is the incorrect base call (A/C SNP) T is the incorrect base call (A/C SNP) One Ion PGM discordancy in a single sample No call on either allele in Complete Genomics No call on either allele in Complete Genomics No call for 1st allele in Complete Genomics No call for 2nd allele in Complete Genomics Likely 1000 Genomes-Phase 1 error Likely 1000 Genomes-Phase 1 error 1.2% 0.8% 2.4% 0.2% 0.3% 0.2% rs , rs , rs938283, rs and rs rs

40 Mixtures Single donors 1:1 1:9 3:1 1:3 9:1 Mixed input DNA created Allele Read Frequency balance shifts - but at 9:1 very little heterozygote displacement

41 Sensitivity 25 or 50 pg In optimum input concordance samples ~1-3 SNPs gave no-calls or dropouts All were outlier SNPs In low-level samples ~8-12 SNPs failed Only rs appeared disproportionately in failing SNPs

42 Sensitivity 25 or 50 pg In optimum input concordance samples ~1-3 SNPs gave no-calls or dropouts All were outlier SNPs In low-level samples ~8-12 SNPs failed Only rs appeared disproportionately in failing SNPs Low-level DNA Optimum input DNA Read length in bp Read length in bp

43 Sensitivity adna 450 pg quantified with Quantifiler Duo 25 PCR cycles or 25+5 library re-amplification cycles Volders site, Tyrol, Austria

44 adna RMP 1.2E-33 GlobalFiler 1.2E-28 SNPs >100 x coverage 450 pg quantified with Quantifiler Duo 25 PCR cycles or 25+5 library re-amplification cycles SNPs x coverage N / NN QUAL=0 Volders site, Tyrol, Austria

45 RMP 1.2E-33 Prototype SNP panel amplicons HID-Ion AmpliSeq Identity Panel amplicons GlobalFiler 1.2E SNPs removed from prototype set SNPs >100 x coverage SNPs have reduced amplicon sizes (by an average 57.5 bp) SNPs x coverage SNPs retain the original prototype primer designs N / NN QUAL=0

46 Identified three kinds of outlier SNPs should be removed have some outlying characteristics such as strand bias that means they should be treated with caution when looking for unusual patterns such as mixed DNA have some outlying characteristics but these do not have a detectable effect on genotyping performance rs rs rs rs rs rs rs rs rs rs rs rs rs rs rs rs rs rs rs rs rs rs rs rs rs

47 Conclusions Not all SNPs performed equally well. Five gave genotype discordancies: inter-lab and inter-run, two remain in the HID-Ion Identity Panel. Several other SNPs have outlier characteristics; those with imbalanced ARFs were also identified in Børsting s study, though two were not and three Børsting outliers had reasonably good allelic balance in our study. The 99.8% concordance is best of any SNP test comparisons made at USC. Limited experiments with low-level DNA suggest very high sensitivity. 81/169 SNPs gave >100x coverage from adna. 57 SNPs have since been shortened. Harmonising chip loading to balance the coverage with samples-per-chip was very difficult. Continues into AIM set validation - high inter-lab variability. Sought to identify ARF outliers to allow their discounting in mixtures. LT Genotyper uses Somatic parameter settings, but mixtures mimic mutation patterns, so Germline settings improve post-hoc analysis of mixed DNA. EUROFORGEN working on a mixture analysis system for Ion PGM data that works well with binary loci but requires conditioning on one contributor.

48 Three studies of the Ion PGM system Combined five established SNaPshot PCRs to create a pool of 136 unique amplicons with bp size range (five-fold increase in marker depth) No balancing made of individual PCRs - 10 µl of purified product (elution columns) pooled for the library preparation then sequenced with 314v1 chips. No Ampliseq Three samples, three input amounts, concordance checked with SNaPshot /Sanger Provided with prototype versions of the LT forensic Identity Panel Decided to adopt a simple NIST validation framework centred on the qualified run - here, closely matched protocols for the same control DNAs Four chip types, 16 samples, concordance checked with public data. Ampliseq A EUROFORGEN deliverable is to develop ancestry and EVC inference panels. Designed an 128-SNP ancestry set for NGS. Obtained customised primer design service of LT white glove team. Validating this SNP set performance with above framework in five labs. Ongoing

49 Optimising a custom SNP panel for Ion PGM 125 of 128 SNPs incorporated into the PCR multiplex: 97.5% conversion rate Concordance rates high: inter-run 99.98% (1 SNP), inter-lab 99.75% (6 SNPs) - with reasons for discordance identified in each case

50 Optimising a custom SNP panel for Ion PGM 125 of 128 SNPs incorporated into the PCR multiplex: 97.5% conversion rate Concordance rates high: inter-run 99.98% (1 SNP), inter-lab 99.75% (6 SNPs) - with reasons for discordance identified in each case Concordance from online database comparisons of Coriell DNAs: 1000 Genomes 99.74% (3 SNPs), Complete Genomics 99.79% (same SNPs)

51 Optimising a custom SNP panel for Ion PGM The six discordant SNPs in 13 genotypes were mainly homopolymeric tracts around the SNP. Need retrogressive analysis due to strand directionality. Eight no-call SNPs in 20 genotypes: 4 low coverage, 1 population-specific flanking indel, 3 did not pass variantcaller quality filters. Some population-specific low coverage seen - untracked primer site SNPs?

52 Thank You

53 Thanks to all the team at Santiago particularly Carla, Maria and Fonde. To Mayra, Walther at GMI and Theresa, Peter at UHC. To Matt Phipps, LT and David Ballard, KCL for help with data analysis.

54 Speaker was provided travel and hotel support by Thermo Fisher Scientific for this presentation, but no remuneration