Targeted sequencing from cerebrospinal fluid for rapid identification of drug-resistant tuberculous meningitis

Abstract

Mortality from tuberculous meningitis (TBM) remains around 30%, with most deaths occurring within 2 months of starting treatment. Mortality from drug-resistant strains is higher still, making early detection of drug resistance (DR) essential. Targeted next-generation sequencing (tNGS) produces high read depths, allowing the detection of DR-associated alleles with low frequencies. We applied Deeplex Myc-TB-a tNGS assay-to cerebrospinal fluid (CSF) samples from 72 adults with microbiologically confirmed TBM and compared its genomic drug susceptibility predictions to a composite reference standard of phenotypic susceptibility testing (pDST) and whole genome sequencing, as well as to clinical outcomes. Deeplex detected Mycobacterium tuberculosis complex DNA in 24/72 (33.3%) CSF samples and generated full DR reports for 22/24 (91.7%). The read depth generated by Deeplex correlated with semi-quantitative results from MTB/RIF Xpert. Alleles with <20% frequency were seen at canonical loci associated with first-line DR. Disregarding these low-frequency alleles, Deeplex had 100% concordance with the composite reference standard for all drugs except pyrazinamide and streptomycin. Three patients had positive CSF cultures after 30 days of treatment; reference tests and Deeplex identified isoniazid resistance in two, and Deeplex alone identified low-frequency rifampin resistance alleles in one. Five patients died, of whom one had pDST-identified pyrazinamide resistance. tNGS on CSF can rapidly and accurately detect drug-resistant TBM, but its application is limited to those with higher bacterial loads. In those with lower bacterial burdens, alternative approaches need to be developed for both diagnosis and resistance detection.

Keywords: CSF; Deeplex; drug resistance; targeted next generation sequencing.

Fig 1

Study flow chart demonstrating the results of Deeplex (in orange) in relation to microbial test results (in light blue). Deeplex sequencing quality was classified as: not detected (ND) for mycobacteria, – for partial reads, 1+, 2+, and 3 + for complete reads with adequate depth of coverage. Zn: Ziehl-Neelsen stain, ^a1 sample with Zn not done, ^b5 samples with Zn not done, and ^c8 samples with Zn not done.

Fig 2

Performance of Deeplex in Mtb detection in relation to other microbiological confirmed tests. (A) Mean read depth of CSF samples by Deeplex against semi-quantitative Xpert result. Box plots represent its median and 1st/3rd interquartile and dots represent individual samples. ND: Not detected. (B and C) Association between Deeplex Mtb detection and Xpert Ct-value readouts (B) or time to culture positivity (C) by logistic regression model. Black dots show the observed values. Gray dots and error bars show the mean values and 95% CI of observed quantiles. Blue line shows the predicted values from logistic regression model and the shaded area shows the 95% CI of predicted values.

Fig 3

Extent of drug-resistant mutations in 22 CSF samples by Deeplex. Mutations associated with drug resistance are specified in the cells when present. Numbers in the brackets indicate the proportions of reads carrying resistant allele. RIF, rifampin; INH, isoniazid; PZA, pyrazinamide; EMB, ethambutol; SM, streptomycin; FQ, fluoroquinolones; KAN, kanamycin; AMI, amikacin; CAP, capreomycin; ETH, ethionamide ; LZD, linezolid; BDQ, bedaquiline; and CFZ, clofazimine. A dark to light orange gradient represents for resistance mutations with allele frequencies indicated by color, while a dark to light blue gradient represents for uncharacterized mutations with allele frequencies indicated by color.