USE OF MINION SEQUENCING FOR REAL-TIME AMR

DETECTION AND BACTERIAL TYPING

Senior scientist

HENRIK HASMAN

Statens Serum Institut (SSI)

Denmark

SEQUENCING OUTPUT – MINION VS MISEQ

Time (hours)

MiSeq v2 (2*250 bp)

MiSeq v3 (2*300 bp) Raw reads (Y)

Assembly (Y + 1 h)

MiSeq v2: Y = 40 h)

MiSeq v3: Y = 60 h)

Micro/Nano: Y = 17 h)

Raw reads (Real time)

Rapid assembly (X + 1 h)

Corrected assembly (X + 24 h)

X = 3-18 h50x

100x

MiSeq v2 (Micro/nano 2*150 bp)

MINION WORKFLOW – 8 HOURS TURNAROUND

Real-time bacterial typing

Real-time genotypic AMR detection

High-precision plasmid assembly

• SNP calling

• MLST?

• Serotyping

• Species verification (rMLST?)

• AMR gene groups AB classes

• AMR mutations AB classes

• Species intrinsic resistance profilling?

• Rough plasmid assembly

• Re-basecalling target reads (+ methylation)

• Re-assembly

GRAPHICAL USER INTERFACE (GUI)

Genus

AB classes detected

AB classes not detected

Genome coverage

# of circular molecules

Closely related isolates in database

Run SNP analysis

Extract plasmids

Last change (minutes)

Species

Clonal cluster

GRAPHICAL USER INTERFACE (GUI)

Genus

AB classes detected

AB classes not detected

Genome coverage

# of circular molecules

Closely related isolates in database

Run SNP analysis

Extract plasmids

Last change (minutes)

SpeciesKlebsiella n.a.

Beta-lactams (PEN)

Beta-lactams (CEF), Beta-lactams (CARBA), Amino-

glycosides, Flouroquinolones, Phenicols…….

Strain23

2,4 x

1

0

Clonal cluster CC23

GRAPHICAL USER INTERFACE (GUI)

Genus

AB classes detected

AB classes not detected

Genome coverage

# of circular molecules

Closely related isolates in database

Run SNP analysis

Extract plasmids

Last change (minutes)

SpeciesKlebsiella varicula

Beta-lactams (PEN), Beta-lactams (CEF), Flouroquinolones

Beta-lactams (CARBA), Aminoglycosides, Phenicols.

Strain23, Strain43, Strain142, Strain155

31 x

17

6

Clonal cluster CC23

MINION WORKFLOW – 8 HOURS TURNAROUND

Real-time bacterial typing

Real-time genotypic AMR detection

High-precision plasmid assembly

• SNP calling

• MLST?

• Serotyping

• Species verification (rMLST?)

• AMR gene groups AB classes

• AMR mutations AB classes

• Species intrinsic resistance profilling?

• Rough plasmid assembly

• Re-basecalling target reads (+ methylation)

• Re-assembly

AMR DETECTION – SMALL EXERCISE

DATA (pre-uploaded to CGE to save time)

1. Complete genome of AMA 1167 including the two AMR plasmids

2. Draft MinION genome of AMA 1167

3. Raw MinION reads (q8+ only) of AMA 1167

Your task is to asses if you think MinION data (raw or assembled) can be used

to

A) Predict phenotypes

B) Identify specific resistance genes (for surveillance purposes)

AMR DETECTION – BACKGROUND INFO

Methods (CGE)

• ResFinder (Complete genome vs MinION draft genome)

• KMA (Complete genome, MinION draft genome, MinION raw data)

Test strain

AMR DETECTION

ILLUMINA VS MINION DATA

Repeat area (rRNA, IS, homologue genes ect..)

1

2

3

RESFINDER COMPARISON

ResFinder 3.2 (https://cge.cbs.dtu.dk/services/ResFinder/)

Complete genome AMA 1167 (1)

https://cge.cbs.dtu.dk//cgi-

bin/webface.fcgi?jobid=5D8B012B000014A377E446C1

Draft MinION genome of AMA 1167 (2)

https://cge.cbs.dtu.dk/cgi-

bin/webface.fcgi?jobid=5D8B0306000019EF2DF7B1CE

Note:• The complete genome is perceived as the optimal data dataset. However, acquiring this

requires extensive sequencing with properly both short reads (Illumina) and long reads

(PacBio or MinION) + hand curation of “trouble” areas.

• The draft MinION genome has been polished with Racon twice, but will still suffer from

issues regarding homo-polymer’s and methylation in MinION data.

• Only (1) can identify the gyrA point mutations related to fluoroquinolone resistance

(activate “ “ on the links above to compare).

KMA COMPARISON

KMA 1.3 (https://cge.cbs.dtu.dk/services/KMA/)

Complete genome AMA 1167 (1)

https://cge.cbs.dtu.dk//cgi-

bin/webface.fcgi?jobid=5D8B1368000048A51A203627

Draft MinION genome of AMA 1167 (2)

https://cge.cbs.dtu.dk//cgi-

bin/webface.fcgi?jobid=5D8B138E000048F3EA63E221

Raw MinION reads (q8+ only) of AMA 1167 (3)

https://cge.cbs.dtu.dk/cgi-

bin/webface.fcgi?jobid=5D8B1300000047580EB71980

Note: (Compare, which genes are found and evaluate TemplateID scores)• The complete genome (1) is basically showing the same result in KMA as in ResFinder.

• The draft MinION genome (2) basically agrees with (1), but with lower identity scores (97-

99%). Note blaCMY-2 has changed to blaCMY-149.

• Raw MinION reads (3) shows good TemplateID values (99+%), but also identifies many

sub-variants. Here, the “Depth“ values can be used to identify most likely sub-variant..but

may still fail

RESFINDER – COMPLETE GENOME (1)

RESFINDER – DRAFT NANOPORE GENOME (2)

KMA – COMPLETE GENOME (1)

KMA – MINION DRAFT GENOME (2)

KMA OUTPUT – RAW MINION DATA (3)

KMA OUTPUT – RAW MINION DATA (3; SORTED)

RESFINDER 4.0 - MINION DRAFT GENOME (2)

SEQUENCING OUTPUT – MINION VS MISEQ

Raw reads (Real time)

Rapid assembly (X + 1 h)

Corrected assembly (X + 24 h)

X = 3-18 h50x

100x

100x 4x

90x 75x 50x 40x 25x 10x 5x

3x

2x

1x

’RAW’ RESFINDER HITS

Already at average

coverage of 10x almost all

AMR genes have been

found, however there is a

lot of noise.

Pink: correct hit truth!

orange: missing hit false negative

yellow: extra hits false positive

’GENE-GROUP’ RESFINDER HITS

Already at ‘cov_3’ almost all AMR genes have been found. At ‘cov_10’ there is a complete and

correct profile with no noise at all. So sul1 and sul2 ends in same group….but also blaOXA-1

and blaOXA-181, which has impact on the phenotypic predictions based on genotypes.

Simple grouping of genes by the three first

letters in the name

pink: correct hit truth!

orange: missing hit false negative

yellow: extra hits false positive

PHENOTYPE RESFINDER HITS

Gene hits converted to AMR phenotypes using the

ResFinder 4.0 database as guide.

Same conclusion:

At ‘cov_10’ there is a complete and correct profile with no noise at all.

pink: correct hit truth!

orange: missing hit false negative

yellow: extra hits false positive

PHENOTYPE RESFINDER HITS

Gene hits converted to AMR classes using the

ResFinder 4.0 database as guide.

Same conclusion:

At ‘cov_10’ there is a complete and correct profile with no noise at all.

pink: correct hit truth!

orange: missing hit false negative

yellow: extra hits false positive

MINION WORKFLOW – 8 HOURS TURNAROUND

Real-time bacterial typing

Real-time genotypic AMR detection

High-precision plasmid assembly

• SNP calling

• MLST?

• Serotyping

• Species verification (rMLST?)

• AMR gene groups AB classes

• AMR mutations AB classes

• Species intrinsic resistance profilling?

• Rough plasmid assembly

• Re-basecalling target reads (+ methylation)

• Re-assembly

SNP CALLING USING MINION DATA

Challenges

• Raw (basecalled) data from MinION has 5-10% errors.

• Assembled data still have 1-3% errors efter polishing (50-130 kbp’s)

• Standard mappers handle long reads poorly).

• Standard SNP-callers are trained/designed for Illumina short reads

with low error rate.

Possible solution

• Utilizing KMA to align MinION reads and build consensus sequences

• Use modified NDtree (or CSIphylogeny) to call relevant SNPs

• Build SNP calling (web) tool based on KMA and NDtree

ST410 CPE (E. COLI) OUTBREAK

Background

• Carbapenemase producing ST410 is an “International clone”.

• A clonal variant (CT587) has been introduced in Denmark to cause a

major outbreak (30+ patients)

Test data

• Raw Illumina data

• Raw MinION data

• Complete genome + plasmids of Danish index strain (AMA 1167).

Strain collection (N = 24)

A. 6 ST410 CT587 isolates (“Outbreak”)

B. 6 ST410 non-CT587 isolates (“Outliers”; not part of outbreak)

C. 12 non-ST410 isolates (background)

CORE GENOME MLST (CGMLST)

ST410

TEST STRAINS

ST410

Non-ST410

Outbreak-ST410

SNP TOOL – KMA-BASED (NDTREE V2.0)

KMA chooses reference

• AMA1167 identified as best reference (index isolate of outbreak)

• Future version allow own reference to be used

Ndtree v2.0 performs SNP filtering based on QC scores

• Different error settings depending on Illumina or MinION data

• Possible to prune closely positioned SNPs

Visualization through FigTree

• Output as pdf and Newick files

• KMA performs alignment of reads

CSI PHYLOGENY AS BENCHMARK

6 + 6 ST410 – CSI PHYLOGENY (ILLUMINA)

CPO20180039_S58_R1_001.sorted/1-2342

CPO20150034_S5_L001_R1_001.sorted/1-2342

CPO20150054_S3_L001_R1_001.sorted/1-2342

CPO20150014_S12_L001_R1_001.sorted/1-2342

CPO20160077_minION_S78_L555_R1_001.sorted/1-2342

CPO20160003_S1_L001_R1_001.sorted/1-2342

CPO20180100_S79_L555_R1_001.sorted/1-2342

CPO20180105_S54_L555_R1_001.sorted/1-2342

CPO20180119_S35_L555_R1_001.sorted/1-2342

CPO20180108_S36_L555_R1_001.sorted/1-2342

CPO20150011_S9_L001_R1_001.sorted/1-2342

CPO20170014_S11_L555_R1_001.sorted/1-2342

0.2

CT587

CT596

CT611

CT512

CT278

6 + 6 ST410 – CSI PHYLOGENY (ILLUMINA)

NDTREE V1.2 V2.0 (KMA BASED)

2.0

6 + 6 ST410 NDTREE V2.0 (ILLUMINA)

CT587

CT596

CT611CT512

CT278

CT527CT523

6 + 6 ST410 NDTREE V2.0 (MINION)

CT587

CT596

CT611CT512

CT278

CT527CT523

6 + 6 ST410 NDTREE V2.0 (ALL)

CT587

CT596

CT611

CT512

CT278

CT527

CT523

6 + 6 ST410 NDTREE V2.0 ILLUMINA + MINION

Illumina data is always 23 SNPs from MinION data.

All these positions relates to dcm methylation sites and can therefore be filtered

away in the analysis…or solved by using the new version of Guppy basecaller.

6 + 6 ST410 NANOPORE – OTHER REFERENCE*

*NZ_CP031653.1_Ecoli_UK_Dog_Liverpool

6 + 6 ST410 NANOPORE – DRAFT GENOME

6 + 6 ST410 MINION – NANOPORE REFERENCE

MINION WORKFLOW – 8 HOURS TURNAROUND

Real-time bacterial typing

Real-time genotypic AMR detection

High-precision plasmid assembly

• SNP calling

• MLST?

• Serotyping

• Species verification (rMLST?)

• AMR gene groups AB classes

• AMR mutations AB classes

• Species intrinsic resistance profilling?

• Rough plasmid assembly

• Re-basecalling target reads (+ methylation)

• Re-assembly

FULL FORECE - EJP AMR PROJECT 2020

FULL FORECE - EJP AMR PROJECT 2020