Citation: Cattamanchi A, Walter ND, Metcalfe JZ, Davis JL (2013) A Transcriptional Signature for Active TB: Have We Found the Needle in the Haystack? PLoS Med 10(10): e1001539. doi:10.1371/journal.pmed.1001539
Published: October 22, 2013
Copyright: © 2013 Cattamanchi et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: AC (K23 HL094141), NW (K12 HL090147), and JZM (K23 AI09425) are supported by career development awards from the U.S. National Institutes of Health. This work is also supported by the National Center for Advancing Translational Sciences, National Institutes of Health (UCSF-CTSI KL2 TR000143). Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NIH. None of the funders played any role in the preparation of this article.
Competing interests: J. Lucian Davis is involved in additional research concerning blood gene expression profiling and tuberculosis. The other authors have declared that no competing interests exist.
Provenance: Commissioned; not externally peer reviewed.
Linked Research Article
This Perspective discusses the following new study published in PLOS Medicine:
Kaforou M, Wright VJ, Oni T, French N, Anderson ST, et al. (2013) Detection of Tuberculosis in HIV-Infected and -Uninfected African Adults Using Whole Blood RNA Expression Signatures: A Case-Control Study. PLoS Med 10(10): e1001538. doi:10.1371/journal.pmed.1001538.
Using a microarray-based approach, Michael Levin and colleagues develop a disease risk score to distinguish active from latent tuberculosis, as well as tuberculosis from other diseases, using whole blood samples.
Analysis of whole-genome RNA expression in human clinical samples is a relatively novel approach to biomarker development. The pattern of RNA expression (i.e., transcriptional signature) can provide a “biological snapshot” of the immune response to physiological stressors, and specific disease states may produce distinct transcriptional signatures. In this week's issue of PLOS Medicine, Michael Levin and colleagues report that a blood RNA transcriptional signature can be used to diagnose active tuberculosis (TB) in high HIV/TB prevalence settings. Using blood samples from patients referred for TB evaluation at three sites in Cape Town, South Africa (cases from an outpatient TB clinic; controls from two hospitals) and one district hospital in Northern Malawi, the authors identified a minimal set of 44 transcripts that distinguished patients with TB from patients confirmed to have an alternative diagnosis. They then converted the complex expression data into a simple-to-calculate disease risk score, which was highly sensitive (93%, 95% CI [83–100]) and specific (88%, 95% CI [74–97]) for active TB.
Have We Found the Needle in the Haystack?
This landmark study advances the field in several critical ways. For the first time, a blood transcriptional signature for TB was defined by comparison with patients who have conditions that mimic TB in a high burden setting instead of with healthy controls or patients with sarcoidosis or auto-immune diseases –. The inclusion of controls for whom TB was in the list of differential diagnoses but ultimately excluded increases confidence that the blood transcriptional signature identified may be clinically relevant. Second, the authors developed the disease risk score, which provides a single measure of the degree to which an individual's RNA expression is consistent with TB. The disease risk score can be calculated by simply subtracting the summed normalized intensities of down-regulated transcripts from those of up-regulated transcripts. By eliminating the need to use complicated bioinformatics to make predictions from the RNA expression data, the disease risk score could simplify the application of transcriptional signatures in clinical settings. Finally, the authors demonstrate convincingly the high diagnostic accuracy of their blood transcriptional signature. The results were impressive in their test set (20% of enrolled patients), including in HIV-infected and smear-negative sub-populations, and in an entirely independent validation dataset published years earlier .
Although the results are promising, key questions remain. First, can the results be reproduced in a truly representative population? State-of-the-art technology and bioinformatics are critical tools for identifying prospective targets, but the rigorous application of fundamental epidemiological principles will be indispensable to advancing these technologies into the clinical arena, where it will be necessary to show their utility in truly representative populations. Levin and colleagues describe an “intention-to-test” recruitment strategy but nonetheless enrolled a highly selected patient population. TB patients generally had advanced disease (over 90% of HIV-uninfected patients [96/106] and over 75% of HIV-infected patients [83/109] were smear-positive) and 28% (207 of 751) of patients were excluded because TB status was uncertain. The control group was recruited entirely from inpatient wards and most patients had non-respiratory diseases. These factors resulted in a spectrum bias towards the extremes of disease manifestations, which is known to inflate estimates of diagnostic accuracy . Second, can a robust threshold be developed for the disease risk score that works in different settings? An inherent limitation of the microarray technology used in this study is that it provides only relative quantification of RNA expression so that intensity values are relevant only within, and not across, experiments . Ultimately, to be clinically useful, a threshold will need to be defined a priori rather than on the basis of experimental data and the selected threshold will have to be consistent across geographic settings. Finally, can a platform be developed to enable measurement of the transcriptional signature in low-income countries? A number of novel technologies for quantitative multi-channel measurement of nucleic acid targets are in development. However, the cost and difficulty of assaying a 44-transcript signature seem prohibitive far into the future. Absent a transformative technology, it is difficult to envision transcriptional profiling having a meaningful impact in parts of the world where novel TB diagnostics are most needed .
Triage Testing: A Target for Future Research
As Levin and colleagues suggest, their 44-transcript DRS may be more useful as a triage (i.e., rule-out) test  because negative predictive value is high (98%, 95% CI [96–100]), while positive predictive value is sub-optimal (66%, 95% CI [46–87]) when TB prevalence is 20%, as is common in routine settings. The concept of a triage test deserves further attention in the TB diagnostics literature. An ideal triage test rules out disease when negative and triggers further testing when positive (e.g., a mammogram for breast cancer screening). Thus, a triage test requires near-perfect sensitivity (particularly when the consequence of missing disease is high) but only moderate specificity. If rapid and inexpensive, such a test could be used to determine which patients presenting with TB symptoms require confirmatory testing (e.g., automated nucleic acid amplification testing or culture) and for TB screening as is recommended in high-risk populations including people living with HIV and household contacts of active TB cases ,.
The 44-transcript signature identified by the authors shows promise as a triage test and might be further optimized for this purpose during its further development. Indeed, it is likely that any set of host-derived biomarkers, especially if based on generic rather than antigen-stimulated immune responses, will have more difficulty achieving high specificity than high sensitivity. In this regard, the commonly used approach of selecting a diagnostic threshold that maximizes the number of correctly classified outcomes is misguided . This cannot lead to a clinically useful test if neither sensitivity nor specificity is high enough to provide meaningful rule-out or rule-in value. Future work to establish a threshold for the 44-transcript signature as a triage test should focus on maximizing sensitivity, even at the cost of decreased specificity. Furthermore, focusing on developing a triage test at the discovery stage may lead to selection of a different set of transcripts that retains 100% sensitivity while providing even better specificity.
Levin and colleagues have provided compelling proof of the concept that a blood transcriptional signature can distinguish between TB and clinical mimics in high-incidence settings. The field can now move on to asking more practical questions to determine the feasibility and optimal use for an RNA expression-based biomarker for TB in clinical settings. Finding the right signature—the proverbial “needle in the haystack”—may require additional discovery work involving smaller sets of RNA transcripts and will certainly require validation of candidate signatures in diverse settings. Further discovery and validation studies should adhere to fundamental principles of high quality diagnostic evaluations , including enrollment of consecutive patients presenting for TB evaluation in representative health facilities. Even if validated, significant technical hurdles remain to translate these important findings to rapid, inexpensive, and simple assays that can impact patient outcomes in countries where TB is most prevalent. The search continues.
Wrote the first draft of the manuscript: AC NDW JZM JLD. Contributed to the writing of the manuscript: AC NDW JZM JLD. ICMJE criteria for authorship read and met: AC NDW JZM JLD.
- 1. Berry MP, Graham CM, McNab FW, Xu Z, Bloch SA, et al. (2010) An interferon-inducible neutrophil-driven blood transcriptional signature in human tuberculosis. Nature 466: 973–977. doi: 10.1038/nature09247
- 2. Jacobsen M, Repsilber D, Gutschmidt A, Neher A, Feldmann K, et al. (2007) Candidate biomarkers for discrimination between infection and disease caused by Mycobacterium tuberculosis. J Mol Med (Berl) 85: 613–621. doi: 10.1007/s00109-007-0157-6
- 3. Koth LL, Solberg OD, Peng JC, Bhakta NR, Nguyen CP, et al. (2011) Sarcoidosis blood transcriptome reflects lung inflammation and overlaps with tuberculosis. Am J Respir Crit Care Med 184: 1153–1163. doi: 10.1164/rccm.201106-1143oc
- 4. Maertzdorf J, Weiner J 3rd, Mollenkopf HJ, Bauer T, Prasse A, et al. (2012) Common patterns and disease-related signatures in tuberculosis and sarcoidosis. Proc Natl Acad Sci U S A 109: 7853–7858. doi: 10.1073/pnas.1121072109
- 5. Whiting P, Rutjes AW, Reitsma JB, Glas AS, Bossuyt PM, et al. (2004) Sources of variation and bias in studies of diagnostic accuracy: a systematic review. Ann Intern Med 140: 189–202. doi: 10.7326/0003-4819-140-3-200402030-00010
- 6. Simon RM, Dobbin K (2003) Experimental design of DNA microarray experiments. Biotechniques Suppl: 16–21.
- 7. Jani IV, Peter TF (2013) How point-of-care testing could drive innovation in global health. N Engl J Med 368: 2319–2324. doi: 10.1056/nejmsb1214197
- 8. Lord SJ, Irwig L, Bossuyt PM (2009) Using the principles of randomized controlled trial design to guide test evaluation. Med Decis Making 29: E1–E12. doi: 10.1177/0272989x09340584
- 9. World Health Organization (2011) Guidelines for intensified tuberculosis case-finding and isoniazid preventive therapy for people living with HIV in esource-constrained settings. Geneva: WHO Press.
- 10. World Health Organization (2012) Recommendations for investigating contacts of persons with infectious tuberculosis in low- and middle-income countries. Geneva: WHO Press.
- 11. Irwin RJ, Irwin TC (2011) A principled approach to setting optimal diagnostic thresholds: where ROC and indifference curves meet. Eur J Intern Med 22: 230–234. doi: 10.1016/j.ejim.2010.12.012
- 12. Bossuyt PM, Reitsma JB, Bruns DE, Gatsonis CA, Glasziou PP, et al. (2003) Towards complete and accurate reporting of studies of diagnostic accuracy: The STARD Initiative. Ann Intern Med 138: 40–44. doi: 10.7326/0003-4819-138-1-200301070-00010