Making TB diagnostics a practical reality - … TB diagnostics a practical reality Zamin Iqbal [email protected] @zaminiqbal ... (human+TB) Send events to server, base call, map, decide,

Making TB diagnostics a practical reality

Zamin Iqbal

[email protected]@zaminiqbal

27 May 2016

© Copyright 2016 Oxford Nanopore Technologies 2

Clinical test for a TB patient

• Take a sample (sputum) and culture it. Is it M. tuberculosis ?

• Need to know which drugs will work

• Mix bug and drug, and see if it grows or dies

Slow ~ 2 months Multiple rounds of culture and testing


Pre-MinION timeline for a DNA-based approach

Get a sample, decode the genome once• Drug resistance is encoded in the DNA• Can look at strain/transmission


Mykrobe Predictor

• Determine species (in PHE workflow for accreditation) and antibiotic resistance.• Performance for TB matches commercial molecular assays• Sensitivity/specificity of 82.6% / 98.5 % (N=1600 Illumina data)• £5million grant (WT/Gates) to sequence/phenotype 100,000 Mtb – push up sensitivity


A long way from a same-day diagnostic

1. The biggest barrier is culture2. Sequencing time/MiSeq batch effects3. Human cells and nasal flora dominate sputum

2 weeks of culture 1-2d seq



1. The biggest barrier is culture Sequence from sputum2. Sequencing time/MiSeq batch effects3. Human cells and nasal flora dominate sputum



Sequencing TB direct from sputum

Depth of coverage of Mtb0 50 100 150

Watch out for preprint on BiorxivAntonina Votintseva



1. The biggest barrier is culture Sequence from sputum2. Sequencing time/MiSeq batch effects3. Human cells and nasal flora dominate sputum




1. The biggest barrier is culture Sequence from sputum2. Sequencing time/MiSeq batch effects MinION3. Human cells and nasal flora dominate sputum Read Until



Removing contamination via Read Until

3. Nanonet basecalls fragmentsMap fragments to see if TBReturn message to ReadUntil

Using 10-minute prep for 1D reads with R9 flow cell

1. Discard first 100 events2. Take first 250 events and send to our ReadUntil server

4. Reject read while still in pore if not TB


Results 1: in silico reanalyse data from an R9 run (human+TB)

Send events to server, base call, map, decide, send message back- all in ~10% of time it takes for read to pass through pore. (but this is biased..)

Min 1st Qu Median Mean 3rd Qu Max

Round trip 0.00 1.23 1.26 1.22 1.33 15.9

Read time in pore

2.0 7.08 11.92 20.20 22.02 4004

% 0 5.5 10.5 13 17 375


Results 2: live run this week

Mix of human DNA (fragment size >4kb) and M. tuberculosis (~1kb)

~3 or 4:1 human:TB

See if we can keep up with the flow cell, and reduce the amount of human reads


Human reads

Before we turned on Read Until


Human reads

Befire Read Until With ReadUntil


Before Read Until With ReadUntil

M. tuberculosis reads


On the verge of a revolution in TB healthcare!

Every component needs work, but..

• 0.5-1 hour pre-step in Cat3 (decontam, heat-kill)• 2.5 hours DNA extraction • 10 minutes sample prep• ?6 hours of sequencing with in-pore filter (depends on %Mtb)• Seconds of computation

From sputum to diagnosis in <1 day


Acknowledgements

Antonina Votintseva (TB extraction protocol)

Louise Pankhurst, Luke Anson (MinIONing)

Phelim Bradley, Carlos del Ojo Elias (RU server)

Matt Loose

Derrick Crook

ONT: Tim Massingham, Rich Ronan, Olly Hartwell

Making TB diagnostics a practical reality - … TB diagnostics a practical reality Zamin Iqbal [email protected] @zaminiqbal ... (human+TB) Send events to server, base call, map, decide,

Documents