TRANSLATIONAL APPROACHES TO PREDICTING TUBERCULOSIS RISK: FROM ADMINISTRATIVE DATA TO TRANSCRIPTIONAL BIOMARKERS NICHOLAS DAVID WALTER

Transcription

1 TRANSLATIONAL APPROACHES TO PREDICTING TUBERCULOSIS RISK: FROM ADMINISTRATIVE DATA TO TRANSCRIPTIONAL BIOMARKERS by NICHOLAS DAVID WALTER BA, Middlebury College, 1994 MS, University of California, Berkeley, 2000 MD, University of California, San Francisco, 2002 A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Doctor of Philosophy Epidemiology Program 2015

2 2015 NICHOLAS DAVID WALTER ALL RIGHTS RESERVED ii

3 This thesis for the Doctor of Philosophy degree by Nicholas David Walter has been approved for the Epidemiology Program by Ivana Yang, Chair Tasha Fingerlin, Mentor Katerina Kechris, Mentor Mark Geraci Michael Strong Martin Voskuil Date: December 18, 2015 iii

4 Walter, Nicholas David PhD, Epidemiology Translational Approaches to Predicting Tuberculosis Risk: From Administrative Data to Transcriptional Biomarkers Thesis directed by Associate Professors Tasha Fingerlin and Katerina Kechris ABSTRACT Tuberculosis (TB) is an ongoing global public health crisis. In 2013, approximately 9 million people developed active TB and 1.5 million died. This dissertation addresses highpriority challenges in risk prediction, diagnosis and understanding bacterial adaptation to antibiotic exposure. An over-arching theme of this work is use of big data analysis to translate large administrative or biological datasets into novel insights into TB epidemiology. The first project utilized probabilistic methods to link a large overseas administrative dataset with US TB surveillance data. Analysis of the linked dataset provided new insights into TB risk among immigrants that have important implications for US guidelines. The second project identified blood transcriptional biomarkers for TB among US patients and systematically evaluated the generalizability of these biomarkers. The third and fourth projects evaluated how drug exposure and HIV-associated immune dysfunction alter transcriptional characteristics of the pathogen Mycobacterium tuberculosis (MTB) in patients. The form and content of this abstract are approved. I recommend its publication. Approved: Tasha Fingerlin & Katerina Kechris iv

5 ACKNOWLEDGMENTS The projects in this dissertation were made possible by funding from the following sources: Veteran s Administration Career Development Award VA 1IK2CX A1, National Institutes of Health National Heart Lung and Blood Institute, Genetics and Genomics of Lung Disease, K-12 5K12HL , Mucosal and Vaccine Research Colorado (MAVRC) at the University of Colorado Denver, Colorado Clinical and Translational Sciences Institute at the University of Colorado Denver and the Division of Pulmonary Sciences and Critical Care Medicine in the University of Colorado School of Medicine. All projects included were approved by the Colorado Multiple Institutional Review Board. v

6 TABLE OF CONTENTS CHAPTER I. Introduction... 1 Risk of reactivation of latent tuberculosis infection... 1 Blood transcriptional biomarkers for active TB... 5 Impact of drug treatment on M. tuberculosis transcriptome... 6 Impact of host immunity on M. tuberculosis transcriptome... 8 II. Persistent latent tuberculosis reactivation risk in US immigrants Introduction Methods Results Discussion Conclusions III. Blood transcriptional biomarkers for active TB among US patients: A casecontrol study with systematic cross-classifier evaluation Introduction Methods Results Discussion Conclusion IV. Transcriptional adaptation of drug-tolerant Mycobacterium tuberculosis during human tuberculosis Introduction Methods vi

7 Results Discussion Conclusion V. Mycobacterium tuberculosis adaptation to impaired host immunity in HIVinfected patients Introduction Methods Results Discussion Conclusion VI. Conclusions References Appendix A Appendix B... Error! Bookmark not defined. vii

8 CHAPTER I INTRODUCTION Tuberculosis (TB) is an ongoing global public health crisis. In 2013, approximately 9 million people developed active TB and 1.5 million died (1). This dissertation addresses high-priority challenges in risk prediction, diagnosis and understanding bacterial adaptation to antibiotic exposure. An over-arching theme of this work is use of big data analysis to translate large administrative or biological datasets into novel insights into TB epidemiology. The first project utilized probabilistic methods to link a large overseas administrative dataset with US TB surveillance data. Analysis of the linked dataset yielded new insights into TB risk among immigrants that have important implications for US guidelines. The second project identified blood transcriptional biomarkers for TB among US patients and systematically evaluated the generalizability of these biomarkers. The third project identified how drug exposure alters transcriptional characteristics of the pathogen Mycobacterium tuberculosis (MTB) in patients treated for TB. The final project evaluated the impact of human immunity on the pathogen transcriptome by comparing MTB from patients with and without HIV infection. These chapter topics are introduced in greater detail below. Risk of reactivation of latent tuberculosis infection Latent TB infection (LTBI) is the central barrier to TB eradication in the United States (US) (2, 3). Most TB infections result in LTBI rather than active TB. LTBI is a state in which an individual has immunologic evidence of TB infection but is asymptomatic and noncontagious. LTBI may remain stably contained by the immune system over the course of a lifetime or depending on a patient s clinical and epidemiologic risks may reactivate to cause active TB disease. 1

9 In contrast to much of the world, public health measures have largely interrupted transmission of TB in the US (2, 3). Currently, ~80% of TB in the US is attributable to reactivation of LTBI (4). For most persons with LTBI in the US, infection was acquired in high TB incidence countries prior to immigration. In 2014, 66.5% of TB in the US occurred among foreign-born persons and the rate of TB among foreign-born persons was 13.4-times higher than among US-born persons (5). Foreign-born persons and immigrants are therefore a high priority for public health prevention efforts. Importantly, an effective intervention is available; preventive antibiotic treatment can reduce the risk of progression from LTBI to TB by up to 90%. To effectively target LTBI screening and treatment to the populations at highest risk of developing TB, it is critical to understand how reactivation risk changes over time, especially among the foreign born. Risk of reactivation is highest within the first two years after infection but TB may also occur decades after initial infection. The risk of reactivation from LTBI to TB is modified by clinical and epidemiologic risk factors. A central challenge in estimating the risk of LTBI reactivation over time is that progression from LTBI to TB is a relatively rare event, even in immigrant populations. This makes it infeasible to quantify reactivation risk through a traditional cohort study of immigrants; the size, cost and follow-up duration is prohibitive. Estimates of LTBI reactivation risk among immigrants to the US has therefore been based on ecological studies. Ecological studies based on US surveillance data from the 1990s suggested that the TB rate among new arrivals to the US was >5-fold higher during the first 2-5 post-arrival years compared with subsequent years (6-8). Based on these data indicating that immigrants 2

10 are at high risk in the immediate post-arrival years, current US guidelines recommend LTBI treatment for immigrants within five years of arrival but not explicitly thereafter (9). A critical flaw of these ecological studies is that they do not account for the role of active pre- and post-immigration evaluation. For decades, applicants for US legal permanent residency (immigrants) have been required to undergo a pre-immigration exam and chest radiograph prior to departure from their home country. This pre-immigration radiographic screening aims to detect fibrotic scarring of the lung suggestive of TB (henceforth referred to as abnormal radiograph). Fibrotic scarring is a non-specific finding that may indicate the presence of active TB, old, healed TB that was not treated, or lung disease unrelated to TB (3). The subset of immigrants that have abnormal pre-immigration radiographs is required to undergo a post-immigration TB evaluation within months of arrival in the US (10). Postimmigration TB evaluation often detects indolent active rather than latent TB. Importantly, previous ecological studies have not stratified by normal or abnormal chest radiograph, instead providing estimates of TB risk in the combined group. This failure to distinguish between immigrants with normal and abnormal chest radiographs is problematic because risk for post-arrival TB diagnosis likely differs between immigrants with normal and abnormal chest radiographs. This has important public health implications because unstratified ecological data are the basis of current treatment guidelines. In Chapter II, entitled Persistent latent tuberculosis reactivation risk in US immigrants, we employed a novel study design to estimate TB risk over time among immigrants. Rather than an ecological design, we used probabilistic linkage methods to match pre-immigration records for 123,114 Filipino adults with subsequently-occurring TB 3

11 case reports in the US. This approach enabled us to quantify TB risk over time separately among immigrants with normal and abnormal radiographs for the first time. Our data demonstrated that immigrants with normal and abnormal chest radiographs have very different TB risk profiles. The 86% of our immigrant cohort that had normal preimmigration radiographs had a relatively low rate of TB diagnosis. Contrary to assumptions of current treatment guidelines, risk remained constant over time. In contrast, the 14% that had abnormal pre-immigration radiographs had a TB rate that was initially >25-times higher than among immigrants with normal radiographs. Collapsing these two risk strata immigrants with normal and abnormal pre-immigration radiographs into a single group replicates the (misleading) findings of previous ecological studies. This analysis suggests that the current US LTBI treatment guidelines are informed by an ecological fallacy in which inferences about risk among immigrants with normal radiographs are inferred (incorrectly) from ecological analysis of a population with both normal and abnormal chest radiographs. We found that excess risk of TB diagnosis in the immediate post-arrival period was borne exclusively by immigrants with abnormal radiographs. This is likely the result of active post-arrival TB evaluation. TB rates did not decline over time in immigrants with normal chest radiographs. These results suggest that future revisions of LTBI treatment guideline should consider extension of LTBI screening and treatment among immigrants from high TB-burden countries beyond the current first 5 years. 4

12 Blood transcriptional biomarkers for active TB Early, accurate diagnosis of TB is critical for control of the global TB epidemic. Early diagnosis not only improves the health of the individual patient being tested, but enables treatment that abrogates contagiousness, reducing transmission to others. Unfortunately, existing tests for TB have suboptimal sensitivity (11). Development of novel assays for TB is a leading World Health Organization priority (12). Most existing tests for active pulmonary TB are based on detection of the pathogen MTB in sputum. However, MTB is not always present in detectable levels and sputum is not always obtainable. MTB is rarely detectable in blood and is not found in urine. An alternative to direct pathogen detection is detection of host markers that are specific to TB. Since blood can be obtained from nearly all patients undergoing evaluation for pulmonary and/or extrapulmonary TB, a diagnostic test based on human markers in blood would be optimal. In Chapter III, entitled Blood transcriptional biomarkers for active TB among US patients: A case-control study with systematic cross-classifier evaluation, we used machine learning methods to evaluate patterns of mrna expression in blood as potential biomarkers. The conceptual basis for this study is that specific immune responses to MTB may be associated with a signature mrna pattern not seen other conditions. mrna expression profiles assayed via microarray or sequencing are understood to represent a biological snapshot of adaptation to physiologic stressors. Several studies have previously used this approach in different populations of patients with TB (13-19) and have suggested blood transcriptional signatures have promise for active TB diagnosis. Our study differs from previous studies in several critical ways. First, our design included enrollment of a comparison group of patients with community-acquired 5

13 pneumonia. This assures practical clinical relevance to our findings; a common challenge encountered by health care providers is determining whether patients presenting with cough, fever and radiographic abnormalities have TB or pneumonia. Previously published studies have included comparison groups of marginal clinical relevance. Second, our study is the first performed in US population of diverse ancestry with a different spectrum of co-morbid health conditions than encountered in previously studied African populations. Third, our study not only identified novel transcriptional signatures for TB based on transcriptional data from our study subjects, but additionally performed a rigorous analysis of the generalizability of our and previously-published classifiers. We found that the accuracy and generalizability of blood transcriptional classifiers differed depending on the comparison group used. Our classifier distinguishing TB from pneumonia was accurate in our subjects. But accuracy decreased when the classifier was applied to new populations with other non-tb disease comparisons. By contrast, our classifier distinguishing TB from LTBI was highly accurate, both in our subjects and in previously-published data from other populations. However, this second classifier type is a non-specific marker for systemic illness and does not distinguish between TB and other diseases like pneumonia. These results led us to conclude that transcriptional classifiers may have potential to improve TB care and control but additional discovery and validation studies in diverse populations with different disease references will be required to evaluate the clinical usefulness of these biomarkers. Impact of drug treatment on M. tuberculosis transcriptome In Chapter IV, entitled Transcriptional adaptation of drug-tolerant Mycobacterium tuberculosis during human tuberculosis, we harnessed a novel pathogen-targeted 6

14 transcriptional profiling method to elucidate changes in the MTB population that occur in human clinical samples during drug treatment. This project addresses a topic of the utmost clinical and public health importance drug-tolerant bacterial persister phenotypes. Slow killing of drug-tolerant persisters is considered the cause of TB treatment failure and relapse and is the central obstacle to shorter treatment regimens (20, 21). Currently, the duration of the shortest standard TB treatment regimen is 6 months. This long duration is costly and burdensome to patients and to health systems and pre-disposes to treatment non-adherence which enables further transmission and acquired drug resistance (22). Understanding MTB persisters and developing drugs that kill drug-tolerant phenotypes is therefore a key priority. Isogenic drug-susceptible MTB populations are understood to include heterogeneous sub-populations in different phenotypic states with variable sensitivity to antibiotics. MTB persisters have been defined as bacterial phenotypes capable of surviving prolonged antibiotic exposure despite a lack of resistance-conferring genetic mutations (23). The clinical and epidemiologic importance of persisters in TB cannot be overstated. Abundant epidemiologic evidence (24-27) demonstrates that treatment results in a biphasic killing pattern in human TB with rapid early killing followed by slower killing. Based on this biphasic pattern, Mitchison and others (25, 28, 29) defined the central paradigm of TB chemotherapy: an initial three to five day bactericidal phase in which metabolically-active dividing bacilli are rapidly killed followed by a sterilizing phase in which an antibiotictolerant persister subpopulation is killed more slowly. Many experimental models have evaluated the basis of MTB drug tolerance (24, 30-37). However, it is unclear whether experimental models recapitulate the physiologic state of MTB persisters in humans. Several previous studies indicate that MTB transcriptional 7

15 patterns in human clinical specimens are distinct from those observed in vitro and in murine models (38-40). This is important because antibiotic effectiveness depends on bacterial physiologic state (30, 41, 42). Until our study presented in Chapter IV, it was not technically feasible to characterize the transcriptional characteristics of the MTB population that survives prolonged antibiotic exposure. We found after 14 days of drug treatment transcriptional patterns were consistent with a shift in the MTB population to a slowly replicating, metabolically down-regulated phenotype that no longer displays signals of isoniazid-induced stress. Systematic comparison with transcriptional changes in previously-published experimental persister models identified models with greater and lesser similarity to the in vivo population. These results identify novel drug targets and reinforce the critical need for novel anti-tb drugs to target the recalcitrant drug-tolerant bacterial population, leading to more effective and shorter duration drug regimens. Impact of host immunity on M. tuberculosis transcriptome In Chapter V, entitled Mycobacterium tuberculosis adaptation to impaired host immunity in HIV-infected patients, we evaluated the influence of HIV-associated immune dysfunction on the bacterial transcriptome. HIV increases TB risk fold. TB is the leading cause of death among HIV-infected people, killing an estimated 400,000 HIVinfected persons worldwide each year (1). HIV has pervasive effects on human immunity, leading to strikingly different clinical presentations of TB among HIV-infected relative to HIV-uninfected persons, including with disseminated disease and atypical radiographic patterns (43). On a histologic level, HIV impairs formation of granulomas, the hallmark human response to TB. On a molecular level, HIV is associated with altered cytokine and 8

16 chemokine expression, changes in macrophage polarization and T cell dysfunction and depletion (43). Since MTB is exquisitely adapted to human immunity, dynamically adjusting its physiologic state to tolerate immune responses and diverse tissue micro-environmental conditions, (38, 44, 45) we hypothesized that these differences lead to different adaptations by the pathogen. Understanding how HIV alters the physiologic state of MTB in vivo is important because the phenotypic state of the pathogen affects antibiotic effectiveness (30, 41, 46-48) and may affect transmission fitness (49-51). We therefore compared MTB transcriptional adaptation in the sputa of HIV-infected and HIV-uninfected patients with untreated active pulmonary TB. After a pilot study in the Gambia revealed that MTB in HIV-infected patients had decreased expression of DosR regulon genes, we confirmed this result in a larger number of Ugandan patients. To elucidate alterations in human immunity that drive this adaptation, we queried a panel of human genes associated with granuloma formation and macrophage polarization. We found that MTB in HIV-infected patients had decreased expression of the DosR regulon, a crucial metabolic switch with immunomodulatory effects. Down-regulation of DosR expression among HIV-infected patients may be a direct result of arginase-driven decreased macrophage NO production or an indirect result of decreased hypoxic stress in poorly formed granulomas. This analysis provides a novel perspective on the complex interplay between HIV-associated immune impairment, altered tissue microenvironments and bacterial physiologic state. The over-arching theme of these four projects is translation of large-scale data including administrative data and high-throughput biological data to epidemiologic purposes that will ultimately advance public health. 9

17 CHAPTER II PERSISTENT LATENT TUBERCULOSIS REACTIVATION RISK IN US IMMIGRANTS 1 Introduction TB among foreign-born persons is a central challenge to TB elimination in the US. The percentage of TB cases occurring among foreign-born persons has increased steadily for nearly 20 years to 63% of all new US cases in 2012 (2). TB exposure is uncommon in the US and most TB among foreign-born persons results from reactivation of LTBI acquired prior to arrival (4). US policy therefore emphasizes targeted testing and treatment of LTBI after arrival (9, 52). National surveillance data in the 1990s showed TB rates >5-fold higher within five years of arrival compared with subsequent years (6-8). New arrivals from high incidence countries were hypothesized to arrive with high-risk early latency due to ongoing reexposure and reinfection with new TB strains until US arrival (53, 54). Since TB risk is highest immediately after infection, high TB rates immediately after arrival were assumed to indicate that reactivation risk declines with time in the US. US guidelines therefore target LTBI testing and treatment towards foreign-born persons in the US within five years of their arrival (9). An alternative explanation for high post-arrival TB rates is detection of imported TB disease that is already active when a person arrives in the US (55). Recent surveillance data indicate that TB rates are 4-6 times higher during year 1 than during years 2-4 (53, 56, 57). Since 1991, TB screening for adults entering to become immigrants has included a pre- 1 This chapter was previously published as: Walter ND, Painter J, Parker M, Lowenthal P, Flood J, Fu Y, Asis R, Reves R. Persistent latent tuberculosis reactivation risk in United States immigrants. Am J Resp Crit Care 2014;189:88-95 and is included with permission of the copyright holder. 10

18 immigration exam and chest radiograph at designated overseas Panel Physician Clinics. Immigrants with radiographic abnormalities undergo post-arrival follow-up evaluation by US local health agencies (10). If immigrants with abnormal radiographs are diagnosed with TB within six months of arrival, it is assumed TB was already active prior to arrival and their case is classified as imported TB (58, 59). Conversely, TB reported among immigrants whose pre-immigration exam was normal have been considered likely due to LTBI reactivation (60). TB also occurs due to post-arrival TB exposures, but the limited burden of ongoing TB transmission in the US makes this less common. Distinguishing LTBI reactivation from imported TB has important implications for LTBI treatment guidelines. Analyses of surveillance data have been unable to distinguish between LTBI reactivation and imported TB because national surveillance data prior to 2011 do not distinguish among key categories of new arrivals (i.e., immigrant, refugee, student, business traveler), and consequently, are unable to ascertain the role of pre- and post-immigration evaluation. The Philippines has a high TB incidence (260/100,000 in 2011) and is the source of a large volume of immigrants to the US (61, 62). More than 60% of Filipinos are estimated to have LTBI (63). Filipino immigrants have among the highest rates of TB diagnosis of any US immigrant group (59). In 2007, pre-immigration exam in the Philippines was enhanced to include sputum culture in addition to sputum microscopy to improve detection of prevalent TB among applicants with abnormal radiograph before departure. To evaluate reactivation and imported TB among a cohort of Filipinos applying for US legal permanent residency, we linked pre-immigration exam records with subsequently occurring cases in the California TB Case Registry. 11

19 Methods We identified a cohort of adult Filipino applicants for permanent legal residency intending to resettle in California and analyzed reported TB during the first 9 postimmigration years. Human subjects approval Institutional Review Boards at the Centers for Disease Control and Prevention (CDC), California Department of Public Health, Denver Health and the University of Texas Health Sciences Center, Houston approved this study. Pre-immigration evaluation records We obtained records for visa applicants evaluated at the only pre-immigration screening site in the Philippines, St. Luke s Medical Center Extension Clinic (SLMCEC), from January 4, 2001 to December 29, Two different screening guidelines were in place during this period (10). Briefly, prior to October 2007, applicants 15 years of age underwent a medical questionnaire, physical examination, and chest radiograph (1991 guidelines). Applicants with an abnormal radiograph provided three sputum samples for acidfast bacilli (AFB) smears but mycobacterial culture was not performed. Applicants with positive smears were required to initiate TB therapy prior to clearance. Emigration was permitted once smears were negative. Beginning on October 1, 2007, (10) sputum specimens from applicants with abnormal radiographs undergo both AFB smear and mycobacterial culture (2007 guidelines). For applicants with positive cultures, emigration is not permitted until TB is treated to completion by directly observed therapy. Applicants with negative smears and cultures are permitted to emigrate within three months of negative cultures. After US arrival, only those immigrants with abnormal radiographs are recommended to have a 12

20 follow-up evaluation to detect and treat active TB or inactive TB (defined as LTBI with an abnormal radiograph). California TB Case Registry TB cases among Philippines-born persons reported from January 1, 2001 to December 31, 2010 were ascertained from the California Department of Public Health. Linkage of data sources Using established methods, (64) we linked pre-immigration records and California TB reports using Registry Plus LinkPlus 2.0, a probabilistic record linkage program (65). Records were matched on last name, first name, date of birth, year of exam, and gender. Two reviewers manually reviewed results, applying pre-determined rules to classify comparison pairs as true or false matches including: (1) exact match in all fields, (2) mismatch in a single field, (3) first/last name transposition, or (4) obvious typographical errors. In cases of disagreement, a third independent reviewer made the final classification. Improbable matches were those whose evaluation occurred after the date of TB diagnosis. Definitions and statistical analysis Immigrants were defined as individuals who received a permanent residency visa after screening in the Philippines. Likely LTBI reactivation was TB in an immigrant with a normal pre-immigration exam and radiograph (Figure 1). Imported TB was TB reported 6 months after arrival in an immigrant with an abnormal pre-immigration radiograph. Reactivation of inactive TB was TB reported >6 months after arrival in an immigrant with an abnormal pre-immigration radiograph. Using lifetable methods, (66) we estimated annual TB rates following preimmigration exam among immigrants with (1) normal radiograph, (2) abnormal pre- 13

21 immigration radiograph evaluated with only sputum AFB smears (1991 guidelines), and, (3) abnormal pre-immigration radiograph evaluated with both sputum AFB smears and cultures (2007 guidelines). The date of pre-immigration exam was used for all rate calculations. To test whether hazard for LTBI reactivation changed with time in the US, we fit a cox regression model with year-specific effects with linear and quadratic terms to records from immigrants with normal pre-immigration radiographs. We compared the hazard for TB among immigrants with an abnormal radiograph evaluated according to 1991 and 2007 guidelines using the survival and KMsurv packages in the R statistical suite, version Results Cohort characteristics During , of 333,768 adult immigrants screened at SLMCEC, 123,114 (37%) intended to settle in California. Of these, 17,160 (14%) had an abnormal preimmigration radiograph (Table 1). Immigrants with abnormal radiographs were significantly older, more frequently male, and more frequently had self-reported diabetes, history of tobacco use, and history of pulmonary TB than those with normal radiographs During , 4,324 Filipino adults were diagnosed with TB in California. For four individuals who had two episodes of TB, the second episode was excluded. Of the 3,900 (90%) with a recorded US entry date, 1,316 (34%) reported entering during Of these, 793 (60%) were linked to a pre-immigration record at SLMCEC (Figure 1). Four additional records matched to California TB reports before the date of pre-immigration evaluation and were excluded. Concordance between two independent reviewers during manual review was >99%. During , post-arrival evaluation was completed for 14

22 77% of California immigrants with abnormal pre-immigration radiographs. Of these, 82% were completed within 6 months of arrival. TB rate among immigrants with normal pre-immigration radiographs (likely LTBI reactivation) The rate of likely LTBI reactivation among immigrants with normal pre-immigration exam and radiograph was 31.6/100,000 person-years (py) during the first 9 years in the US. This rate did not change with increasing time in the US (p-value for trend in linear and quadratic models 0.8 and 0.5, respectively) (Figure 2a). In the initial year following preimmigration exam, likely LTBI reactivation represented a small fraction of TB diagnosed in the cohort. During year one, 33 (5.8%) of 568 TB cases were likely LTBI reactivation with the remainder classified as imported or likely reactivation of inactive TB. Conversely, during subsequent years, 171 (76%) of 225 TB cases occurring in the entire cohort were likely LTBI reactivation. TB rates among immigrants with abnormal pre-immigration radiographs TB rates among immigrants with abnormal radiographs differed depending on whether pre-immigration exam included AFB smears alone (1991 guidelines) or smears and mycobacterial culture (2007 guidelines). During the initial year following pre-immigration exam, TB rates among immigrants with abnormal radiographs evaluated according to 1991 and 2007 guidelines were >100 and >25 times higher than among immigrants with normal radiographs (4,172 and 903/100,000 versus 31/100,000 py, respectively) (Figure 2b). Of 568 immigrants with TB in the year following pre-immigration exam, 480 (84.5%), and 55 (9.7%) had imported TB and reactivation of inactive disease, respectively, because they had 15

23 abnormal pre-immigration radiographs and TB was reported within 6 months (imported), or after 6 months (reactivation of inactive disease) of arrival. Among immigrants with abnormal pre-immigration radiographs, TB declined markedly after the first year following pre-immigration exam. Rates among those evaluated according to the smear-based 1991 guidelines declined 40 to 50-fold compared with the next 2-4 and 5-9 years (4,172/100,000 py versus 95/100,000 py and 82/100,000 py, respectively). Rates among those evaluated with both sputum smears and culture according to 2007 guidelines declined from 903/100,000 py in the first year following pre-immigration exam to 82/100,000 in years 2-4. Despite the decline during years 2-9, the TB rate among immigrants with abnormal radiographs was >3 times higher than among immigrants with normal radiographs (Hazard Ratio [HR] = 3.8, 95% Confidence Interval [CI] = ). Comparison with national surveillance data The TB rate in the entire California-bound cohort (normal and abnormal radiographs combined for comparison with existing surveillance data) was >10 times higher during year 1 than during years 2-4 and 5-9 (463, 44 and 42/100,000 py, respectively). This rate is similar to recent US surveillance data which estimate TB rates among new entrants from the Philippines to be 500/100,000 during year 1 and 30-40/100,000 thereafter (53, 56). Impact of adding sputum culture to pre-immigration exam (2007 guidelines) After mycobacterial culture was added to the pre-immigration exam in October 2007, 7.0% of applicants with abnormal radiographs had MTB positive sputum cultures and were therefore provided directly observed TB treatment prior to emigration. Of these, 74.8% were AFB smear negative and would not have been detected under previous screening guidelines. After October 2007, hazard for TB diagnosis in California within the first year decreased by 16

24 79% compared with persons evaluated according to the 1991 guidelines (Hazard Ratio [HR] = 0.21, 95% Confidence Interval [CI] = ). Following implementation, an average of 43 fewer imported TB cases were diagnosed each year compared with the preimplementation period (average 61/year pre-implementation versus 18/year postimplementation). In years 2-9, TB risk among immigrants with abnormal radiographs did not differ based on whether or not they underwent sputum culture at pre-immigration exam (HR =0.63, 95% CI: ). Age and TB rates Age-specific differences in TB rates between immigrants with normal and abnormal radiographs are illustrated in Figure 3. The age distribution of immigrants with abnormal radiographs is shifted rightward relative to those with normal radiographs reflecting substantially older age at entry. Among immigrants with normal radiographs, the TB rate was high in late adolescence, declined in the twenties then gradually increased to peak around age 50 (Figure 3a). By contrast, among immigrants with abnormal radiographs, rates were >4,000/100,000 py in late adolescence and declined four-fold by 35 years of age with a slow decline with increasing age thereafter (Figure 3b). Clinical characteristics of likely LTBI reactivation and imported TB Likely LTBI reactivation and imported TB differed in ways likely to influence transmission risk (Table 2). Immigrants with imported TB were significantly more likely to be diagnosed with pulmonary TB than immigrants with likely LTBI reactivation (χ 2 p-value <0.0001). Of those with pulmonary TB, immigrants with imported TB were significantly less likely to be sputum smear positive (13%) than either immigrants with likely reactivation of 17

25 inactive TB (23% smear positive, p=0.007) or likely LTBI reactivation (44% smear positive, p <0.0001). Discussion Among Filipino immigrants whose pre-immigration exam showed no evidence of active or inactive TB, the TB rate once in the US did not decline over a 9-year period, remaining approximately 25-times higher than the 2012 TB rate among US-born persons (2). Our findings are consistent with stable and persistent risk of LTBI reactivation. The perception that reactivation risk decreases over time has been based primarily on the observation that TB rates are highest immediately after arrival. In this cohort, high postarrival TB rates were entirely due to imported TB, likely detected through active post-arrival surveillance. Our observation of a sustained high risk of LTBI reactivation adds to mounting evidence that LTBI treatment should not be limited to the first 5 years after arrival in the US (59, 67). By distinguishing imported TB from likely reactivation TB, this cohort study provides critical insight into TB risk among immigrants and has important implications for clinicians and guidelines. The cause of the high TB rates consistently observed during initial post-arrival years in US national surveillance data (6, 7, 53, 56, 59) has been debated (55) but has generally been attributed to transiently increased risk of LTBI reactivation. Importantly, previous studies lacked pre-immigration exam data and were unable to estimate detection of imported TB through active post-arrival surveillance. The TB rate in our entire cohort (normal and abnormal radiographs combined) is consistent with previous surveillance estimates (53, 56) but stratification according to pre-immigration radiographic status demonstrates that excess early TB risk is entirely due to imported TB in the subgroup of 18

26 immigrants with abnormal radiographs. Immigrants with normal pre-immigration radiographs (86% of immigrants in this cohort) had no excess TB risk in the immediate postarrival years. Our observation that TB risk among immigrants with normal pre-immigration radiographs does not decline over time is consistent with cohorts of refugees and asylum seekers to low-incidence settings including Denmark, (68, 69) Norway,(70) Canada,(71) Australia, and the US (72). These studies consistently report a high burden of prevalent imported TB that was detected through active post-arrival evaluation followed by persistently elevated TB incidence in ensuing years. Our estimates are consistent with studies from decades ago which observed TB rates of approximately 40-80/100,000 among immigrants with a normal pre-immigration radiograph during the first 4-5 years after their immigration to the US (60, 71, 72). A more recent US case control study did not identify evidence of decreasing reactivation risk with time (55). Pre-immigration evaluation is designed to detect and prevent importation of active TB among applicants with abnormal radiographs. Routine LTBI testing is conducted only for immigrants with abnormal radiographs who complete post-arrival follow-up. Our findings highlight that substantial further reduction in TB among immigrants will require expanding LTBI treatment among those with normal radiographs either overseas or after arrival (52, 59, 67, 73). In this study, 86% of adult immigrants had a normal pre-immigration radiograph and this group accounted for 76% of TB diagnosed during years 2-9 (following the initial period of active detection of imported prevalent TB), likely due to progression of LTBI to active TB disease (4). Adding LTBI testing and treatment to the pre-immigration evaluation of adults was encouraged by a 2000, Institute of Medicine report (52) but change has been hampered 19

27 by concerns about feasibility, cost effectiveness, and legal authority to delay immigration for persons who do not pose an immediate public health risk (74-77). The development of interferon gamma release assays and shorter-course weekly LTBI therapy (78) could potentially make overseas LTBI evaluation and treatment feasible. Expanding LTBI testing and treatment for foreign-born persons already in the US could also reduce the occurrence of domestic TB. Key challenges are, 1) current guidelines do not recommend LTBI evaluation for foreign-born persons in the US >5 years who do not have a coexisting medical or additional epidemiological risk, and, 2) existing LTBI recommendations are not effectively implemented (79, 80). Several lines of evidence suggest that restricting LTBI evaluation to the first 5 years leaves foreign-born persons at substantial TB risk. TB rates among Filipino immigrants in this cohort remained higher beyond 5 years post-arrival than in other groups currently recommended for LTBI evaluation (such as USborn persons who are incarcerated or have medical conditions) (67). In 2011, 65% of TB cases among foreign-born persons were diagnosed >5 years after arrival (81). A recent costeffectiveness study identified the foreign-born in the US >5 years as the largest potential opportunity for domestic TB prevention (67). Finally, we note that the most desirable (and likely most cost-effective (76, 77)) means of preventing TB among the foreign-born in the US is reducing TB incidence in the countries of origin. By comparing two periods with distinct pre-immigration guidelines, our results support an earlier report (58) that enhanced screening has substantially reduced imported active TB. Before October 2007, applicants with abnormal radiographs received only AFB smears without mycobacterial culture. Due to the limited sensitivity of AFB smears, rates of TB diagnosis at post-arrival follow-up evaluation in the US were extremely high. 20

28 Implementation of mycobacterial culture dramatically reduced (but did not entirely eliminate) imported active disease. Post-arrival follow-up evaluation therefore remains an important additional opportunity for case detection (10). Pre-departure treatment of applicants with evidence of culture-negative active TB or inactive pulmonary TB is a potential strategy to further reduce imported TB among screened immigrants. Our findings reinforce existing evidence that post-arrival follow-up prevents the spread of TB in the US (58, 82, 83). Immigrants with imported TB were less frequently sputum AFB smear positive than those diagnosed later, suggesting that active follow-up of immigrants with abnormal radiographs detects TB at an earlier, less infectious stage of disease. Evaluating and treating inactive TB (LTBI in immigrants with abnormal radiographs) is an important secondary priority of post-arrival evaluation. Effective LTBI evaluation and treatment or previous TB treatment may explain why the TB rate during years 2-9 among persons with an abnormal pre-immigration radiograph was lower than in historical cohorts of patients with untreated inactive TB (84, 85). This study has several limitations. Our analysis is based on the conservative assumption that all applicants arrived and remained in California for capture of TB occurrence. Applicants who did not arrive would cause underestimation of the true rate in year 1. Persons who died or migrated out of California would cause underestimation during subsequent years. Second, our analysis considered only legal permanent residents from the Philippines but we believe results are likely to be similar for screened immigrants entering from other countries with high rates of TB. Immigrants account for only about 500,000 of over 30 million foreign-born US entrants each year. The remainder include students, business travelers, tourists and unauthorized visitors (57) who do not benefit from pre-arrival 21

29 evaluation or active post-arrival follow-up and are therefore less likely to have early detection of imported TB. Third, we do not know whether immigrants with normal radiographs were evaluated and treated for LTBI after US arrival as guidelines recommend (9). We suspect most were not since a recent national survey indicated only 11.6% of foreign-born persons in the US with a positive tuberculin skin test reported prior treatment for TB or LTBI (86). LTBI treatment would contribute to a lower rate. Finally, we have assumed that the majority of TB among adult immigrants with normal pre-immigration radiographs resulted from TB exposures in the Philippines for two reasons. The likelihood of acquiring LTBI over a lifetime in the Philippines (TB incidence is >60 times higher than in the US) is substantially higher than acquiring LTBI in the US during a shorter period. Second, >96% of TB among Asians in California occurs among foreign-born persons, suggesting domestic transmission within Asian communities is relatively uncommon. It is conceivable that some primary TB occurs due to new exposures occurring during return visits to the Philippines but 80% of Filipinos with reported TB in California had not returned to the Philippines in the two years before diagnosis (unpublished data). Disease occurring due to TB exposures in the US or on return visits to the Philippines would cause overestimation of the LTBI reactivation rate but would not change our primary conclusions about the lack of decline in LTBI risk over time and the need to extend LTBI evaluation in this population. Conclusions In a large cohort of Filipino immigrants, the rate of likely LTBI reactivation TB among immigrants with a baseline normal pre-immigration screening exam did not decline over 9 years. The very high early rate of TB frequently observed among immigrants is due to 22

30 imported active TB, likely detected as a result of active follow-up evaluation. Our findings suggest that current US guidelines should be revised to support LTBI screening and treatment among recently-arrived immigrants from high TB-burden countries beyond the current first 5 years. Table 1. Baseline characteristics of 123,114 California-bound Filipino adults during with abnormal and normal pre-immigration radiographs. Abnormal preimmigration radiograph N =17,160 Normal preimmigration radiograph N =105,954 p-value n (%) n (%) Male 7,791 (45) 41,765 (39) < Age, years (median, IQR) 59 (47-67) 35 (22-52) < Medical risks at time of overseas evaluation Diabetes mellitus a 1,976 (12) 4,247 (4) < Current smoking a 2,304 (13) 14,154 (13) NS History of smoking a 6,311 (37) 25,025 (24) < Renal insufficiency a 89 (<1) 168 (<1) < Chronic hepatitis a 28 (<1) 171 (<1) NS Malignancy a 302 (2) 581 (<1) < History of pulmonary TB a 2,693 (16) 655 (<1) < Pre-immigration evaluation regimen b CDC 1991 guidelines 11,625 (68) 81,354 (77) CDC 2007 guidelines 5,535 (32) 24,600 (23) a Self-reported at time of pre-immigration exam. b After October 1, 2007, applicants with an abnormal radiograph underwent sputum culture in addition to sputum smear. 23

31 Figure 1. Study flow. Overseas, pre-immigration TB exam records for a cohort of 123,114 California-bound Filipino applicants for permanent US residency were linked with the California TB case registry to ascertain which individuals developed TB. 24

32 Figure 2. (a) Annual rates of likely LTBI reactivation among Filipino immigrants by year after overseas pre-immigration TB exam. (b) Annual TB rates by year overseas preimmigration exam among three strata of immigrants: (1) abnormal pre-immigration radiograph evaluated according to 1991 guidelines with sputum smears only (red), (2) abnormal pre-immigration radiograph evaluated according to 2007 guidelines with both sputum smears and cultures (blue) and (3) immigrants with normal pre-immigration radiograph (likely LTBI reactivation, black) Inset table summarizes estimated annual rates during Year 1, Years 2-4 and Years 5-9. Immigrants evaluated under 1991 and 2007 guidelines had up to 9 years or 3 years of follow-up time, respectively. 25

33 Figure 3. Cumulative person-years (py) of observation (gray bars) and age-specific rate of TB diagnosis per 100,000 py (blue lines) (a) Among 105,987 Filipino immigrants with a normal pre-immigration radiograph young adults contributed most py of observation. The TB rate was high in late adolescence, dipped in the 20s and rose to peak around age 50 (b) Among 17,127 Filipino immigrants with abnormal pre-immigration radiograph elderly persons contributed the most py of observation. The TB was highest in the youngest age and declined consistently over time. Age-specific rates are smoothed with a centered weighted 10-year moving average. 26

34 Table 2: Clinical characteristics of 793 adult Filipino immigrants evaluated at SLMCEC during who were subsequently reported with TB in California. Abnormal pre-immigration radiograph Imported TB a N=480 Reactivation of Inactive TB b N=109 Normal preimmigration radiograph Likely LTBI reactivation c N=204 Primary site of disease Pulmonary 474 d 99% 99 d 91% % Pleural 1 <1% 3 3% 6 3% Lymphatic 1 <1% 3 3% 29 14% Other 1 <1% 3 3% 31 15% Missing 3 1% 1 1% 2 1% Sputum smear positive 60 d 13% 23 e 23% 60 44% Culture positive (any site) 343 d 71% 65 60% % Clinical case f % 39 36% 39 19% a TB diagnosed within 6-months of US arrival in an immigrant with an abnormal preimmigration radiograph. b TB diagnosis after 6-months of US arrival in an immigrant with an abnormal preimmigration radiograph. c TB diagnosed in an immigrant with a normal pre-immigration radiograph. d Significantly different from imported TB group (chi square p-value < 0.001) e Significantly different from imported TB group (chi square p-value < 0.01) f No molecular or culture confirmation 27

35 CHAPTER III BLOOD TRANSCRIPTIONAL BIOMARKERS FOR ACTIVE TB AMONG US PATIENTS: A CASE-CONTROL STUDY WITH SYSTEMATIC CROSS- CLASSIFIER EVALUATION 2 Introduction Early, accurate diagnosis of TB is critical for control of the global TB epidemic. However, the sensitivity of existing tests is inadequate (11). Most existing tests for active pulmonary TB are based on detection of MTB in sputum. However, the bacterium is not always present in detectable levels and sputum is not always obtainable. Since blood can be obtained from nearly all patients undergoing evaluation for pulmonary and/or extrapulmonary TB, a diagnostic test based on human markers in blood would be optimal (11). Human immune responses to MTB may lead to transcriptional patterns in blood that are not present in other conditions (87-89). Selected sets of mrna transcripts have been used in prediction models to classify patient samples as having active TB or not (13-19). In early phase studies, (90) blood transcriptional classifiers have shown promise for active TB diagnosis (18, 19, 88, 91, 92) and monitoring treatment response (93). However, to move from proof of concept to an assay that is useful for TB control, it will be necessary to confront important practical questions. First, heterogeneity in diagnostic performance between populations is a well-known barrier to development of diagnostic tests 2 This chapter was previously published as: Walter ND, Miller MA, Vasquez J, Weiner M, Chapman A, Engle M, Higgins M, Quinones AM, Roselli V, Canono E, Yoon C, Cattamanchi A, Davis JL, Phang T, Stearman RS, Datta G, Garcia BJ, Daley CL, Strong M, Kechris K, Fingerlin TE, Reves R, Geraci MW. Blood transcriptional biomarkers for active tuberculosis among patients in the United States: A case-control study with systematic cross-classifier evaluation. J Clin Microbiol 2015;IN PRESS and is included with permission of the copyright holder. 28

36 (94). Many genetic, environmental and technical factors affect transcriptional biomarkers (88, 95). Would a transcriptional classifier developed among sub-saharan Africans (19) be similarly accurate among ethnically-diverse US patients? Existing blood transcriptional classifiers for active TB vary in the transcripts that are included, the populations from which they were derived, and in the prediction models used to derive them (13-19). To develop assays that can be implemented widely to improve TB control, it is critical to ascertain whether there is a universal, generalizable transcriptional pattern characteristic of TB (88). A second important practical question concerns the clinical usefulness of different types of blood transcriptional classifiers for TB. Three distinct classifier types have been based on different reference groups. The first type has compared active TB patients with systemically ill patients with other diseases that may mimic TB (15, 19). The second type has compared symptomatic TB patients with healthy subjects with LTBI (15, 19). The third type has compared active TB patients with combined groups of systemically ill patients and healthy persons (including individuals with LTBI) (13, 16, 19). To develop transcriptional classifiers as a patient care and public health tool, it is critical to evaluate the limitations and potential clinical applications of each of these three classifier types. We sought to develop a minimally invasive blood test to accurately diagnose active TB in racially and ethnically diverse populations. Therefore, we conducted a case-control study of blood microarray among US adults with and without TB and compared the transcriptional classifiers we identified with previously-published classifiers. We identified and tested three novel transcriptional classifiers that distinguish active TB from pneumonia (type 1), LTBI (type 2), and a combined group of LTBI or pneumonia (type 3). Additionally, to assess generalizability, we validated our transcriptional classifiers in an externally-derived 29

37 cohort of sub-saharan Africans and systematically evaluated the accuracy of previouslypublished transcriptional classifiers when tested in our US patients. Finally, we assessed the limitations and potential clinical applications of the three classifier types. Methods Study design The expression in Pneumonia and Tuberculosis (epat) study was a prospective casecontrol study of HIV-uninfected adults with and without pulmonary TB in Colorado (Denver Health Medical Center) and at TB control programs in Texas (details in Figure 1 and Appendix A). Three groups were enrolled. Adults with TB were sputum acid-fast bacillus smear and MTB culture positive. Adults with community-acquired pneumonia had cough or dyspnea, fever or leukocytosis, an infiltrate on chest radiograph, and a negative QuantiFERON -TB Gold In-Tube (Cellestis) (QFT) result. Adults with LTBI had a positive QFT result and no cough, fever, weight loss or radiographic evidence of active TB. All patients were enrolled as outpatients; some were subsequently hospitalized. The Colorado Multiple Institutions Review Board and the University of Texas Health Sciences Center, San Antonio Institutional Review Board approved this project. All subjects provided written informed consent. Laboratory analysis, and data pre-processing Blood was drawn in PAXgene Blood RNA Tubes (QIAGEN). RNA was extracted using the PAXgene Blood RNA Kit (QIAGEN). Specimens with RNA Integrity Number >7 via Bioanalyzer (Agilent) were considered acceptable. We selected 109 of 139 acceptable samples for microarray at random with stratification to achieve an approximate 1:1:1 ratio between groups. RNA was hybridized to Affymetrix GeneChip Human Gene 1.1 ST array. 30

38 After quality inspection, expression data were Robust Multichip Average normalized, (96) log-transformed and adjusted for batch effect (97). Differential expression between groups was estimated using the R limma package, version with a Benjamini-Hochberg correction for multiple comparisons. Transcripts with adjusted p-value <0.01 and absolute fold change of >1.2 were considered significant. Identification and evaluation of novel expression classifiers We identified three different classifiers that distinguish active TB from pneumonia (type 1), LTBI (type 2), and a combined group of pneumonia or LTBI (type 3). Samples were randomly partitioned into training (2/3) and test (1/3) sets. Transcript (feature) selection methods are detailed in Appendix A. Briefly, support vector machines with recursive feature elimination (SVM-RFE) (98) was applied to training set samples to identify transcripts most predictive of TB [e1071 package, version 1.6.3]. To determine the number of transcripts in each classifier, we identified the point at which adding additional transcripts did not substantially improve classification (details in Appendix A and Figure A1) (99, 100). Accuracy in the training set was estimated across 20 iterations of 6-fold cross-validation. After transcript selection, a new SVM was fit to training samples and used to predict the class of test set samples (details in Appendix A). For external validation, we used publically-available data from the largest published whole blood transcriptional classifier study (19) by Kaforou. Briefly, Affymetrix probeset IDs were converted to Illumina transcript IDs. Classification accuracy for an SVM using converted transcripts was estimated in 20 iterations of 6-fold cross-validation. 31

39 Evaluation of previously-published expression classifiers We systematically reviewed the published literature for transcriptional classifiers in adult human whole blood based on machine-learning classification of whole genome microarray data (details in Appendix A). To determine if transcriptional patterns observed in previously-published classifiers were also evident in our data, we compared median expression of transcripts from each published classifier with median expression of the same transcripts in our epat patients. To test previously-published classifiers in our cohort, we converted transcript identifiers to Affymetrix probeset IDs and estimated classification accuracy in cross-validation (6-fold, repeated 20 times) using the same classification method used in respective original manuscripts. Data are available in NCBI s Gene Expression Omnibus [GSE 73408]. Results Enrollment We screened 184 adults, enrolled 136 and conducted microarray for 109 samples identified via stratified random selection (Figure 1). Similar to the demographics of TB reported in the US (101), 63% of TB patients were foreign-born, from Southeast Asia, Africa and Latin America. TB patients more commonly reported diabetes (34%) than did patients with LTBI (9%, p = 0.02). TB patients were more commonly foreign-born (63%) than pneumonia patients (33%, p=0.01). Identification of novel epat transcriptional classifiers Using pneumonia as a disease reference group, our novel epat type 1 classifier included 47 transcripts. Sensitivity was 89.8% in the epat training set and 100% in the test set (95% 32

40 Confidence Intervals in Table 2). Specificity was 87.9% in the training set and 80% in the test set. Figure 2 shows Receiver Operating Characteristic (ROC) curves. For external validation, we applied our type 1 epat classifier to Sub-Saharan African subjects with and without HIV infection (19). Among HIV-uninfected Africans, sensitivity of the epat type 1 classifier was 90.2% and specificity was 77.7% (Table 2). Area Under the Curve (AUC) was significantly lower among HIV-uninfected Sub-Saharan Africans (90.6%) than among subjects in the epat training set (95.9%) (Figure 2) (p<0.001). Among Sub- Saharan Africans with HIV infection, sensitivity was 84.8% and specificity was 76.3%. AUC was 88.6%, significantly lower than in the epat training set (p<0.001). Although the type 1 classifier was designed to distinguish TB from pneumonia, we additionally tested how it would classify LTBI samples. It distinguished TB from LTBI with 100% (95% CI: %) sensitivity but only 40% (95% CI: %) specificity. AUC was 86.0% (95%CI: %). Our novel epat TB vs. LTBI classifier included 51 transcripts. Sensitivity was 90.8% in the epat training set and 100% in the test set. Specificity was 90.4% in the training set and 81.8% in the test set. AUC was 95.9% in the training set and 98.4% in the test set. Figure 2 shows ROC curves. In external validation among Sub-Saharan African subjects without HIV infection, (19) sensitivity of the epat type 2 classifier was 90.4% and specificity was 86.4% (Table 2). AUC was 95.3%, not significantly different from the accuracy observed in US epat subjects (p=0.2) (Table 2). Among African subjects with HIV infection, AUC was 89.9%, significantly lower than among US epat subjects (p<0.001). 33

41 To test the specificity of the type 2 classifier (designed to distinguish TB from LTBI) we applied it to pneumonia and LTBI patients. The epat type 2 classifier categorized 35 (90%) of 39 pneumonia patients as TB, suggesting that it detects the presence of systemic illness rather than a signature unique to TB. When the epat type 2 classifier was applied to TB and pneumonia patients, 34 (87%) of 39 pneumonia patients were classified as TB. The transcripts differentially expressed in TB relative to LTBI are highly overlapping with the transcripts altered in pneumonia relative to LTBI (Figure 3). Of 1,611 transcripts with higher expression in TB than LTBI, 1,279 (74%) were also up-regulated in pneumonia relative to LTBI. Our final classifier included 119 transcripts that distinguished TB from a combined group of pneumonia or LTBI in epat samples. Sensitivity was 81.9% in the epat training set and 90.9% in the test set. Specificity was 79.0% in the training set and 76.9% in the test set (Table 2). The AUC was 85.6%, significantly lower than AUC for classifier types 1 (p<0.001) or 2 (p<0.0001). In external validation, the type 3 epat classifier had higher accuracy among HIVuninfected Sub-Saharan Africans than among the US epat subjects used for classifier development (AUC 91.1% and 85.6%, respectively, p<0.001 ) Among Sub-Saharan Africans with HIV infection, AUC was 85.4%, not significantly different than the epat type 3 classifier in epat patients. Evaluation of previously-published transcriptional classifiers Systematic literature review identified four published articles that developed classifiers for TB based on adult human whole blood microarray data (13, 15-17, 19). 34

42 These articles included two type 1 classifiers. Kaforou (19) compared active TB with a heterogeneous group of infections and neoplasms of the respiratory, genitourinay, and gastrointestinal tracts. When tested in US epat patients, the Kaforou classifier had an AUC of 82.9%, significantly lower than the epat TB vs. pneumonia classifier (p<0.001) (Table 2). Berry (15) compared TB with various infectious and rheumatologic diseases. The Berry classifier had an AUC of 90.0%, also significantly lower than the epat TB vs. pneumonia classifier (p<0.001). There was little overlap (4.5%) in the transcripts included in different type 1 classifiers (Table 3). To determine whether these different transcripts might nonetheless reflect the same biologic processes, we used DAVID Bioinformatics Resources (102) to quantify enrichment of Gene Ontology (GO) terms BP_Fat, CC_Fat and MF_Fat in the classifier transcript lists. This did not reveal shared GO terms (Appendix A). Finally, side-byside tornado plots provide visualization of global differences between studies in expression of transcripts included in classifiers. Figure 4a displays mean fold change for transcripts in the Kaforou type 1 classifier in Kaforou data. Figure 4b displays change in the same transcripts in epat data. The transcriptional change that distinguishes TB from OD in Africans is not clearly discernable among epat patients. Side-by-side tornado plots for the Berry classifier are included in Figure 5a-b. Our systematic review identified two type 2 classifiers (15, 19) that compared active TB to healthy controls with or without LTBI. When applied to epat data, the sensitivity and specificity of Berry (15) and Kaforou (19) type 2 classifiers were similar to that of the epat type 2 classifier (Table 2). The AUC for the Kaforou type 1 classifier in our epat data was 35

43 98.0%, significantly higher than the epat classifier in epat data (p<0.001). The AUC for the Berry type 1 classifier was 94.8%, significantly lower than the epat classifier (p<0.001). The average overlap in the list of transcripts included in type 2 classifiers was 33%, higher than for type 1 classifiers. Type 2 transcript lists were enriched for similar GO terms, including innate immune response, response to wounding and defense response (Appendix A). In contrast to the type 1 classifiers, side-by-side tornado plots showed that the transcriptional changes that distinguished TB from LTBI among Sub-Saharan African subjects in the Kaforou study were clearly observable among epat patients (Figure 4c-d). Similarly, the transcriptional changes observed in the Berry dataset were also evident in epat subjects (Figure 5c-d). Finally, to determine whether misclassification of LTBI patients as TB is random or whether the same individual subjects are consistently misclassified by different classifiers, we evaluated the frequency with which each subject was misclassified in repeated crossvalidation. In Figure 6, the colored vertical bars indicate that two of 35 LTBI patients were consistently misclassified by all classifiers, suggesting that misclassification may result from an intrinsic difference in the patient sample rather than from random error. Systematic review identified three previous Type 3 classifiers (13, 16, 19) that compared TB to a combined group of healthy/ltbi and other diseases. When tested in US epat patients, the Kaforou type 3 classifier had an AUC of 82.8%, significantly lower than the epat type 3 classifier (p<0.001). Bloom (13) compared TB with a combined group of healthy persons and pneumonia, cancer and sarcoidosis patients. When tested in US epat patients, the Bloom classifier had significantly higher AUC than the epat type 3 classifier (90.5% v 85.6%, p<0.001). Maertzdorf (16) compared patients with TB and sarcoidosis. 36

44 When tested in epat patients, the AUC was 90.4%, also significantly higher than the epat type 3 classifier (p<0.001). The average overlap in the list of transcripts included in type 3 classifiers was 4.1%. As was the case for type 1 classifiers, side-by-side tornado plots indicated that the transcriptional change observed in previous studies was not clearly discernable among epat patients (Figures 4e-f, 7a-b, 8a-b). Discussion We found that blood transcriptional classifiers accurately detected active TB among ethnically/racially diverse US adults. Our type 1 classifier was highly accurate in a common clinical challenge: distinguishing TB from pneumonia. However, the accuracy of classifiers for TB versus other diseases decreased when applied to new populations with different combinations of other diseases. Type 2 classifiers that distinguish TB from LTBI were highly accurate and generalizable across diverse populations. However, type 2 classifiers do not distinguish between TB and pneumonia. They are non-specific markers for systemic illness rather than signatures that are unique to TB. Transcriptional classifiers have potential to improve TB care and control but additional discovery and validation studies in diverse populations with different disease references are needed to identify the most predictive transcripts and generalizable signatures. Biomarker development is a multi-phase process, starting with pre-clinical exploratory studies, expanding to case-control and cohort studies and optimally culminating in randomized trials that demonstrate public health impact (90). It is essential to determine early in this process whether transcriptional classifiers are generalizable beyond the studies in which they were developed (90). In this study, we not only independently developed novel 37

45 transcriptional classifiers in a previously-unstudied population but also systematically evaluated the generalizability of our and previously-published classifiers. Type 1 classifiers address a high-priority clinical need by distinguishing active TB from other diseases that may mimic TB. We addressed a common clinical challenge: determining whether a patient with lower respiratory tract infection has TB. Our classifier was highly accurate among US patients. Accuracy diminished modestly (from AUC of 96.5% to 90.6%) when applied to Sub-Saharan African patients with different disease reference groups. Previously-published classifiers were also less-than-optimally generalizable; accuracy was lower in epat patients than reported in original manuscripts. Type 1 classifiers developed independently in different studies (with different other disease reference groups) had very little overlap in transcript sets. Analysis of GO terms did not identify shared biological processes in different classifier transcript sets. The sub-optimal generalizability represents a key challenge for identifying a robust and universal transcriptional signature for TB versus other diseases (88). Unlike type 2 classifiers that compare TB with a single reference condition (LTBI), a practically useful type 1 classifier would need to distinguish active TB from the protean range of infectious, neoplastic, and rheumatologic conditions that can mimic TB. A classifier optimized to distinguish TB from a single disease reference (such as lung cancer) may be less accurate in a different comparison (such as pneumonia). The essential next step for future studies will be identifying specific transcripts that are consistently predictive of TB versus the range of clinical mimics across different populations. This will require additional independent wholetranscriptome classifier development in additional settings with additional disease comparison groups. 38

46 Type 2 classifiers distinguish active TB from LTBI. Symptomatic active TB is associated with massive change in the blood transcriptome relative to healthy persons with LTBI, enabling accurate classification despite racial/ethnic diversity and medical comorbidities. Type 2 classifiers are remarkably accurate and generalizable across populations. When tested across different African, European and US settings, AUC was consistently ~95% or higher. Studies conducted independently in different populations identified similar transcripts, representing similar immune and inflammatory processes. Tornado plots showed that the transcriptional patterns distinguishing TB from LTBI were remarkably consistent across studies. Unfortunately, type 2 classifiers do not represent a signature that is unique to TB. The transcriptional changes observed in TB relative to LTBI overlap nearly entirely with those observed in pneumonia relative to LTBI. It is therefore unsurprising that our type 2 classifier categorized nearly all pneumonia patients as having active TB. Type 2 classifiers essentially distinguish sick from healthy. Type 2 classifiers are therefore not useful in evaluation of systemically ill patients; for sick patients, the relevant question is not whether the patient is sick or healthy but whether the patient has TB or an alternative disease process (i.e., the task of type 1 classifiers). Could the type 2 classifier be used to identify individuals with incipient, sub-clinical active TB? We found that several LTBI patients were consistently misclassified as TB in all type 2 models. It has been proposed that LTBI patients who are misclassified as having TB could be at a transitional point between LTBI and active TB (15, 103). If a molecular signature of illness precedes the development of TB symptoms (104), type 2 classifiers might have value as a screening test for incipient active TB among persons such as household 39

47 contacts of TB patients, health care workers in high TB incidence settings or HIV-infected persons. Since the type 2 classifier appears to be a non-specific marker of illness, positive screening results would lead to more intensive TB evaluation. This important potential application has yet to be tested. The third classifier type is an all purpose classifier designed to identify TB among patients that are either healthy or systemically ill. As a composite of the previous two classifier types, type 3 classifiers combine but do not resolve the challenges outlined above. As was observed in the Kaforou study, (19) our type 3 classifier was less accurate than our types 1 or 2. This study has several limitations. The limited sample size led to wide uncertainty intervals in the test set. Additionally, comparison between studies required conversion between microarray platforms, a methodological difference that tends to reduce generalizability (95). However, the use of different platforms makes the between-study consistency we observed in TB-LTBI classifiers an even stronger finding. Some transcript identifiers could be not converted because they have been retired from current annotation. Our conversion process therefore likely eliminated noise. A final obvious practical limitation is that there is currently no diagnostic platform that would make assays of blood transcriptional patterns feasible in high TB-incidence settings. Identification of an accurate, generalizable transcriptional classifier with a clear clinical application could motivate novel platform development. Conclusion Blood transcriptional classifiers are capable of accurately identifying active TB. A remaining challenge for classifiers designed to detect TB among systematically ill patients is 40

48 the breadth and heterogeneity of diseases that may mimic TB. Additional studies of systemically ill patients in diverse settings with systematic assessment of cross-study generalizability are needed. Classifiers comparing TB and LTBI do not identify a signature that is unique to TB but should nonetheless be explored as a screening tool in high risk asymptomatic or minimally symptomatic persons. Blood transcriptional assays that enable early accurate TB diagnosis could have an important impact on control of the global TB epidemic. 41

49 Figure 1. Enrollment flow diagram. 42

50 Table 1. Clinical and demographic characteristics of US patients in the epat study. Characteristic n (%) TB LTBI Pneumonia n = 35 n = 35 p-value a n = 39 p-value a Male 25 (71) 18 (51) (59) 0.3 Foreign-born 22 (63) 30 (86) (33) 0.01 Race/ethnicity Age Asian 6 (17) 14 (40) (5) 0.1 Black 2 (6) 6 (17) (13) 0.3 Latino 23 (66) 9 (26) < (41) 0.03 White 3 (9) 1 (3) (28) 0.04 Other 1 (3) 5 (14) (13) (29) 15 (43) 3 (8) (20) 13 (37) 16 (41) (29) 5 (14) 10 (26) >65 8 (23) 2 (6) 10 (26) Diabetes 12 (34) 3 (9) (23) 0.3 Asthma or COPD 6 (17) 4 (11) (28) 0.2 Current smoker 10 (29) 8 (23) (41) 0.3 a χ 2 or Fisher s Exact p-value relative to TB group. 43

51 Table 2. Accuracy of three types of whole-blood transcriptional classifiers for active TB. For each classifier type, results are shown for epat training and test sets and external validation in Sub-Saharan African patients with and without HIV infection. Additionally, results are shown for previously-published classifiers when tested in epat patients. Classifier/Set Num. samples Type 1: TB versus Pneumonia classifier epat type 1 classifier in epat patients Training Set Test Set TB = 24 pneumonia = 24 TB = 11 pneumonia = 15 External validation in African patients HIV-negative HIV-positive TB = 97 Other diseases = 83 TB = 102 Other diseases =88 Sensitivity (95% CI) 89.8 ( ) 100 ( ) 90.2 ( ) 84.8 ( ) Previously-published type 1 classifiers in epat patients Kaforou Berry TB = 35 pneumonia = 39 TB = 35 pneumonia = 39 Type 2: TB versus LTBI classifier epat type 2 classifier in epat patients Training Set Test Set TB = 24 LTBI = 24 TB = 11 LTBI = 11 External validation in African patients HIV-negative HIV-positive TB = 97 LTBI = 83 TB = 100 LTBI = ( ) 91.3 ( ) 90.8 ( ) 100 ( ) 90.4 ( ) 80.0 ( ) Previously-published type 2 classifiers in epat patients Kafrorou Berry TB = 35 LTBI = 35 TB = 35 LTBI = ( ) 89.7 ( ) Specificity (95% CI) 87.9 ( ) 80.0 ( ) 77.7 ( ) 76.3 ( ) 79.1 ( ) 73.7 ( ) 90.4 (88,4-92.4) 81.8 ( ) 86.4 ( ) 77.1 ( ) 92.4 ( ) 94.3 ( ) AUC (95% CI) 96.5 ( ) 90.1 ( ) 90.6 ( ) 88.6 ( ) 82.9 ( ) 90.0 ( ) 95.9 ( ) 98.4 ( ) 95.3 ( ) 89.9 ( ) 98.0 ( ) 94.8 ( ) 44

52 Table 2 continued Classifier/Set Num. samples Type 3: TB versus LTBI or pneumonia classifier epat type 3 classifier in epat patients Training Set Test Set TB = 24 LTBI or pneumonia = 48 TB=11 LTBI or pneumonia =26 External validation in African patients HIV-negative HIV-positive TB = 117 LTBI or OD = 146 TB = 121 LTBI or OD = 153 Sensitivity (95% CI) 81.9 ( ) 90.9 ( ) 87.7 ( ) 79.6 ( ) Previously-published type 3 classifiers in epat patients Kaforou Bloom Maertzdorf TB=35 LTBI or pneumonia =74 TB=35 LTBI or pneumonia =74 TB=35 LTBI or pneumonia = ( ) 85.7 ( ) 84.9 ( ) Specificity (95% CI) 79.0 ( ) 76.9 ( ) 80.3 ( ) 74.0 ( ) 75.7 ( ) 76.8 ( ) 86.8 ( ) AUC (95% CI) 85.9 ( ) 94.1 ( ) 91.1 ( ) 85.4 ( ) 82.8 ( ) 90.5 ( ) 90.4 ( ) 45

53 Table 3. Overlap in the transcripts included in seven previously-published classifiers and the three novel transcriptional classifiers developed in this study. Gray cells represent the number of unique transcripts in each classifier after conversion to Affymetrix probeids. Yellow shading indicates between-study overlap in classifiers for TB vs. LTBI. Pink shading indicates between-study overlap in classifiers for TB vs. other diseases (OD) or pneumonia (PNA). Blue shading indicates overlap in classifiers for TB vs. the combination of LTBI and other diseases. 46

54 Figure 2. ROC curves for three different transcriptional classifier types. TB versus pneumonia classifier in (A.) epat training set, (B.) epat test set, (C.) External validation in HIV-uninfected Sub-Saharan Africans, (D.) HIV-infected Sub-Saharan Africans. TB versus LTBI classifier in (E.) epat training set, (F.) epat test set, (G.) External validation in HIVuninfected Sub-Saharan Africans, (H.) HIV-infected Sub-Saharan Africans. TB versus pneumonia or LTBI in (I.) epat training set, (J.) epat test set, (K.) External validation in HIV-uninfected Sub-Saharan Africans, (L.) HIV-infected Sub-Saharan Africans. 47

55 Figure 3. Venn diagram of transcripts differentially expressed in TB relative to LTBI and pneumonia relative to LTBI. (A.) Transcripts with significantly lower expression in TB and pneumonia than in LTBI (Benjamini-Hochberg adjusted p-value <0.01 & fold change >1.2). (B.) Transcripts with significantly higher expression in TB and pneumonia than in LTBI. 48

56 Figure 4. Comparison of transcriptional changes in Kaforou classifiers in the Kaforou study and epat. (A.) Mean fold-change (log 2 scale) of 44 transcripts in the Kaforou TB versus other diseases classifier in the original Kaforou study. Rows represent individual transcript identifiers. Blue bars indicate transcripts with increased expression in TB in the original manuscript; red represents decreased expression in TB in the original manuscript. (B) Mean fold-change for the same 44 transcripts among epat patients with TB and pneumonia. (C) Mean fold-change for 27 transcripts in the Kaforou TB versus LTBI classifier in the original Kaforou study. (D) Mean fold-change for the same 27 transcripts among epat patients. (E) Mean fold-change for 53 transcripts in the Kaforou TB versus LTBI or other diseases classifier in the original Kaforou study (F) Mean fold change for the same 53 transcripts among epat patients. 49

57 Figure 5. Comparison of transcriptional changes in Berry classifiers in the Berry study and epat. (A.) Mean fold-change (log2 scale) of 86 transcripts in the Berry TB versus other diseases classifier in the original Berry study. Rows represent individual transcript identifiers. Blue bars indicate transcripts with increased expression in TB in the original manuscript; red represents decreased expression in TB in the original manuscript. (B) Mean fold-change for the same 86 transcripts among epat patients with TB and pneumonia. (C) Mean fold-change for 393 transcripts in the Berry TB versus LTBI classifier in the original Berry study. (D) Mean fold-change for the same 393 transcripts among epat patients. 50

58 Figure 6. Consistency of misclassification of TB and LTBI patients by previously-published TB/LTBI classifiers and our epat classifiers. Columns represent individual subjects. Colored headers indicate subjects true class: blue = TB, gray = LTBI. Green cells represent instances in which the classifier assigned the correct class with a high degree of certainty (probability of the correct class estimated as >0.66). Yellow cells represent instances in which the classifier did not have high certainty (probability of correct class estimated as between 0.33 and 0.66). Red cells represent instances in which the classifier assigned the incorrect class (misclassification) with a high degree of certainty (probability of correct class estimated as <0.33). The presence of columns of red and yellow suggest that different classifiers consistently misclassify certain individuals. 51

59 Figure 7. Comparison of transcriptional changes in Berry classifiers in the Bloom study (7) and epat. (A.) Mean fold-change (log2 scale) of 140 transcripts in the Bloom TB versus other diseases or LTBI classifier in the original Bloom study. Rows represent individual transcript identifiers. Blue bars indicate transcripts with increased expression in TB; red represents decreased expression in TB. (B) Mean fold-change for the same 140 transcripts among epat patients with TB and pneumonia or LTBI. 52

60 Figure 8. Comparison of transcriptional changes in Maertzdorf classifiers in the Maerzdorf study and epat. (A.) Mean fold-change (log2 scale) of 100 transcripts in the Maertzdorf TB versus other diseases or LTBI classifier in the original Maertzdorf study. Rows represent individual transcript identifiers. Blue bars indicate transcripts with increased expression in TB; red represents decreased expression in TB. (B) Mean fold-change for the same 100 transcripts among epat patients with TB and pneumonia or LTBI. 53

61 CHAPTER IV TRANSCRIPTIONAL ADAPTATION OF DRUG-TOLERANT MYCOBACTERIUM TUBERCULOSIS DURING HUMAN TUBERCULOSIS 3 Introduction Drug-tolerant bacterial phenotypes play a central role in human TB, a disease that kills 1.5 million people each year (1). In patients treated for drug-susceptible TB, >99% of the MTB population is killed within the first 3-5 days (25, 27, 105). Thereafter, the rate of killing decreases by >80% (25, 27, 105) indicating that surviving bacterial population is drug tolerant. Drug tolerance differs from drug resistance. Resistance results from mutations in specific genetic loci. Tolerance is a reversible phenotypic state that occurs in the absence of resistance-conferring mutations (23). Drug tolerance is a central challenge for drug development and global TB control. Most forms of TB require at least six months of treatment to eradicate slowly-killed, drugtolerant MTB populations and minimize risk of treatment failure and relapse (106, 107). This duration burdens patients and health systems and contributes to medication nonadherence, enabling further transmission and acquired drug resistance (22). Little is known about drug-tolerant MTB in human TB. Genetically homogeneous bacterial populations are hypothesized to include heterogeneous bacterial phenotypes (108, 109). Phenotypes capable of surviving prolonged antibiotic exposure despite genetic susceptibility have been called persisters (23). However, it is unknown whether there is a single MTB persister phenotype or a spectrum of drug-tolerant phenotypes. Drug tolerance is 3 This chapter was published as: Walter ND, Dolganov GM, Garcia BJ, Worodria W, Andama A, Musisi E, Ayakaka I, Van TT, Voskuil MI, de Jong BC, Davidson RM, Fingerlin TE, Kechris KJ, Palmer C, Nahid P, Daley CL, Geraci M, Huang L, Cattamanchi A, Strong M, Schoolnik GK, Davis JL. Transcriptional adaptation of drug-tolerant Mycobacterium tuberculosis during treatment of human tuberculosis. J Infect Dis 2015; 212:990-8 and is included with the permission of the copyright holder. 54

62 observed under a variety of experimental conditions (31, 32, 35, 46-48, 110) but it is unknown whether these models recapitulate drug-tolerant phenotypes in human TB (23). There is currently no marker or method for identifying distinct phenotypes within an isogenic MTB population. A better understanding of the clinically-relevant drug-tolerant population detectable in TB patients during treatment is essential to developing novel, shorter treatment regimens that will improve globaltb control and individual patient outcomes. We used multiplex qrt-pcr gene expression profiling to characterize the metabolic state and adaptive mechanisms of drug-tolerant MTB remaining in sputum after the early bactericidal phase of treatment. mrna expression provides a contemporaneous measure of bacterial physiologic state since the estimated half-life of MTB mrna is 9.5 to 50 minutes (111) and mrna is not detectable from dead bacilli (33). To identify the transition from rapid to slow killing that is the hallmark of a drug-tolerant bacterial population, we evaluated change in the abundance of MTB genomic DNA, 16S ribosomal RNA and mrna over time. We analyzed gene expression profiles, identifying change in functional gene categories that reflect the physiologic state of the drug-tolerant population. We examined the expression of genes previously associated with drug tolerance, including efflux pumps, drug-activating enzymes, drug targets, and the isoniazid stress signature. Finally, we compared MTB expression patterns in TB patients undergoing treatment with expression patterns in experimental models of drug tolerance (35, 46-48, 110). Methods Enrollment and specimen collection A random sample of consecutive adults with pulmonary TB were enrolled at Mulago National Referral Hospital in Kampala, Uganda. (Details in Appendix B). Before TB 55

63 treatment (day 0) and after 2, 4, 7, 14, 28 and 56 days, patients expectorated into a sterile cup containing guanidine thiocyanate solution for immediate RNA preservation (39). Within two hours, sputa were needle-sheared, centrifuged at 9000xg, and stored in Trizol at -80 C. Ugandan and US institutional review boards approved the study. All patients provided written informed consent. Expression profiling Total RNA was extracted using a phenol/chloroform protocol described in the Supplement. After first strand cdna synthesis, cdna underwent controlled multiplex amplification using MTB-specific flanking primers as previously described (112). Expression of 2,411 MTB transcripts was quantified via multiplex qrt-pcr with a LightCycler 480 (Roche, Indianapolis, IN) using TaqMan primers. Analysis We plotted mrna, 16S rrna, and DNA abundance over time relative to baseline values. We fitted the change in geometric mean percentage over time with 95% confidence intervals using saturated linear mixed models. mrna expression data were normalized using a minimum variance data-driven method (113). We identified altered gene expression at days 2, 4, 7, and 14 relative to day 0 via paired-t tests with Benjamini-Hochberg (114) multiple testing correction using a false discovery rate of To identify highly expressed genes, expression values for each sample were ranked from one (gene with lowest expression) to 100 (highest expression). The median percentile rank of each gene at each day was calculated. To identify bacterial functions that changed with drug treatment, we summarized the proportion of genes in previously described functional categories with significantly increased 56

64 or decreased expression from day 0 with an unadjusted p-value <0.05. A modified one-way Fisher s exact test (115) was used to identify enriched functional categories (p-value threshold <0.05). Data from days 28 and 56 were not normalized because of low transcript detection. To evaluate similarity in expression across days 14-56, we quantified between-day overlap in the genes consistently detected in >50% of samples. Non-specific changes were defined as genes that were significantly induced or suppressed both immediately after drug exposure (day 2) and at all subsequent time points. Genes significantly altered between days 2 and 14 were considered likely specific to drugtolerant bacilli. Hierarchical clustering using a Euclidean distance was performed on the median fold change from day 0 for tolerance-associated genes. A static tree cut at height 4 was used to create 5 distinct clusters. Enrichment of functional gene categories was compared between sputum and five experimental models. For sputum and each model, the ratio of genes up- and down-regulated was calculated for each functional category and log 2 transformed. Pearson s correlation coefficients were calculated between each model pair using the log 2 transformed values. Analysis used the R statistical suite, version except as noted in Supplement. MIAMEcompliant data are available at Results Seventeen Ugandan adults with drug-susceptible, sputum acid-fast bacillus smearpositive, culture-positive pulmonary TB were enrolled (Figure B1, Table B1). Nine were HIV-infected. All were treated with isoniazid, rifampicin, pyrazinamide, and ethambutol. 57

65 After two months, four remained sputum MTB-culture positive. One patient had recurrent TB 13 months after completing treatment. Change in nucleic acid abundance MTB mrna abundance declined 0.55 log/day during days 0-4, consistent with rapid killing during the early bactericidal phase (Figure 1). After day 7, the decline in mrna abundance slowed to 0.03 log/day (slowing 94% relative to day 0-4), consistent transition to a drug-tolerant population. 16S rrna and DNA declined more slowly. 16S rrna transcripts per genome equivalent of DNA declined 96-99% from day 0 to days 14 through 56, consistent with a slowing of replication. All samples, including those that were culture negative after 56 treatment days, had greater than100 MTB mrna transcripts detected, indicating the presence of a viable, but not culturable population, as previously reported ( ). MTB transcriptional alteration during treatment Treatment markedly altered the bacterial transcriptome. Relative to pre-treatment expression, the abundance of 452 (19%) transcripts increased and 741 (31%) transcripts decreased significantly during treatment days 2, 4, 7, and 14 after adjustment for multiple comparisons. Key functional categories with altered gene expression during drug treatment are shown in Figure 2 (Categories are defined in Table B2). Expression of genes associated with metabolic processes was predominantly down-regulated. Expression of genes involved in aerobic respiration, the tricarboxylic acid (TCA) cycle, and ATP synthesis decreased immediately and remained suppressed through day 14. By contrast, expression of genes associated with anaerobic respiration was not down-regulated; expression of narg and frda 58

66 increased significantly by day 14. Genes of the glyoxylate bypass, an alternative to the TCA cycle were down-regulated yet remained among the most highly-expressed genes. Expression of genes coding for ribosomal proteins and other genes such as translation elongation factor (tuf, Rv0685) was rapidly down-regulated, suggesting a global decrease in protein translation. Down-regulation of polyketide synthase gene expression suggested diminished production of cell envelope lipids. Genes coding for the ESX/Type VII system which secretes highly immunogenic virulence factors including ESAT-6 were highly expressed prior to treatment but were down-regulated with treatment, particularly at days Expression of genes central to DNA replication including DNA gyrase subunit A (gyra), subunit B (gyrb) and DNA topoisomerase I (topa) was significantly down-regulated during days Genes typically induced by stress had increased expression including: oxidative stress response, the immunogenic PE/PPE genes and insertion sequences and phages. There was induction of genes for the enduring hypoxic response, a response induced in vitro by prolonged hypoxia and other stressors. In contrast, expression of DosR regulon genes that are induced in vitro by hypoxia and other conditions that inhibit aerobic respiration was significantly down-regulated in sputum on day 14. Expression of the gene for RelA, the regulator of the starvation-associated stringent response, increased significantly on day 2 but not on subsequent days. The stringent response expression signature includes induction of specific stress-associated gene sets and suppression of ribosomal protein mrna and other genes (120). Genes classically induced during the stringent response did not have significantly increased expression in sputum during days By contrast, genes previously 59

67 described as suppressed during the stringent response did have significantly decreased expression on days 2, 4, 7 and 14. Three categories of regulators sigma factors, transcription factors and toxinantitoxin (TA) modules were markedly altered. Expression of the gene for SigA, the principal growth-associated sigma factor, was suppressed. Conversely, expression of genes for stress-associated sigma factors including SigB, SigE, SigF, SigH, SigI and SigJ increased significantly. Expression of genes coding for toxins were upregulated particularly of the VapBC family which cleave existing mrna to rapidly remodel the transcriptome in response to stress. Transcriptome analysis after day 14 was limited because mrna abundance was insufficient. However, transcripts consistently detected (present in >50% of samples) at days 14, 28 and 56 overlapped almost entirely, suggesting a stable expression pattern during the sterilizing phase of treatment. Drug-induced stress responses versus adaptations specific to the drug-tolerant population Gene expression in MTB has previously been stratified into: non-specific druginduced stress responses and transcriptional adaptations that are specific to the drug-tolerant population (35). We defined non-specific responses as genes significantly induced or suppressed immediately after drug exposure (day 2) with changes sustained at all subsequent time points. Of 104 non-specific stress response genes, 64 were consistently induced and 40 were consistently suppressed. Induced genes were statistically enriched for the enduring hypoxic response (p=0.003). Suppressed genes were enriched for metabolism genes involved in the TCA cycle, aerobic respiration and ATP synthesis (p=0.006). 60

68 To identify transcriptional adaptations specific to drug-tolerant bacilli, we identified 82 genes whose expression changed significantly between day 2 and day 14 of drug treatment (Figure 3A). Clustering by treatment day showed progressive expression change with increasing duration of treatment. Days 7 and 14 clustered closely together, suggesting transition to a stable drug-tolerant phenotype. Clustering of genes over time identified subsets with similar patterns of expression change, suggesting a coordinated role in development or maintenance of drug-tolerant bacilli (Figure 3B). Clusters A and B (35 genes) demonstrated little to no change between days 0 and 2, but significant induction by days 7 and 14. These included triacylglycerol synthases, and ABC transporter and toxin molecules. Clusters C and D (9 genes) were strongly induced immediately following antibiotic exposure (day 2) and were notable for three genes in the KAS operon required for mycolic acid synthesis. Cluster E (38 genes) was significantly down-regulated at days 7 and 14; it consisted primarily of ESX and ribosomal genes. Expression of antibiotic-associated genes We evaluated genes for proteins previously associated with drug tolerance or resistance, including efflux pumps, drug-activating enzymes, drug targets and the isoniazid stress signature. As a category, efflux pump genes were not significantly enriched following treatment initiation. However, a subset of efflux genes shown in vitro to export drugs including the Tap protein coded for by Rv1258c, baca (Rv1819c), and mmr (Rv3065) were induced >3-fold during days Decreased expression of genes coding for drug-activating enzymes katg (isoniazid), pnca (pyrazinamide), and etha (ethionamide) would be predicted to increase 61

69 drug tolerance. However, after seven to 14 treatment days, the expression of these genes was not suppressed. Drug tolerance could also result from increased expression of drug targets. Expression of the gene for rifampicin target DNA-dependent RNA polymerase (RpoB) was significantly induced at all intervals after treatment initiation. Expression of the gene for isoniazid target InhA did not change significantly. In contrast, genes for DNA gyrase subunits (GyrA, GyrB) which are targeted by fluoroquinolones (drugs these patients did not receive) had progressively decreased expression, consistent with decreased replication. Similarly, expression of the gene for bedaquiline target AtpE was suppressed at all intervals, consistent with down-regulation of aerobic metabolism. We examined the effects of drug treatment on the isoniazid stress response a set of mycolic acid synthesis and stress response genes previously shown to be induced and inversely proportional to growth rate during in vitro isoniazid exposure (30, ). After two treatment days, the isoniazid stress signature was robustly induced in sputum, including mycolic acid biosynthesis genes (fabd, acpm, kasa, kasb, accd6), an efflux gene (efpa), and an oxidative stress response gene (aphc). After seven days, this signature was no longer upregulated, consistent with the known poor sterilizing activity of isoniazid (124). Comparison with experimental models of MTB drug-tolerance To test whether existing experimental models recapitulate drug-tolerant MTB in humans, we compared gene expression in sputum with five widely-cited experimental models (35, 46-48, 110). As in sputum, all models identified down-regulation of ribosomal proteins and TCA cycle, aerobic respiration and ATP synthesis genes, indicating slowing of growth and metabolism (Figure 4). Most found down-regulation of ESX genes. All reported 62

70 up-regulation of the enduring hypoxic response, and most found up-regulation of oxidative stress responses. Notably, the oxygen-responsive DosR regulon was induced in models that applied hypoxic stress (41, 46-48) but not in sputum, the antibiotic selection model (35) or the nutrient starvation model (110). The stringent response regulator rela was induced in the nutrient starvation model, (110) the hollow-fiber granuloma model, (30) and the oxygen deprivation model (48). The initial transient induction of rela at day 2 in sputum is similar to the antibiotic selection model (35) in which rela was significantly induced after one day of drug exposure but then significantly down-regulated after 7 and 14 days. Drug-tolerant MTB in sputum had greater up-regulation of genes involved in regulatory functions TA modules, sigma factors, and transcription factors than did several models (41, 47, 110). Sputum MTB also had greater induction of PE/PPE genes and insertion sequences and phage genes than most models (35, 47, 48). Overall, expression in sputum was most similar to the antibiotic selection model (35) (Pearson correlation coefficient 0.73). Discussion This study provides the first direct characterization of drug-tolerant MTB during human TB treatment, a refractory population that dictates the components and duration of TB treatment. Gene expression of drug-tolerant bacilli suggests slowing of growth and metabolic and synthetic down-regulation. Nonetheless, the drug-tolerant population is not transcriptionally dormant. Rather, up-regulation of genes involved in stress responses, drug efflux, and regulatory functions indicate active transcriptional reprogramming. Understanding the adaptations of drug-tolerant bacilli in vivo should aid development of novel shorter TB treatment regimens. 63

71 Several lines of evidence indicate that MTB present in sputum after 14 days of treatment is drug-tolerant. Previous in vitro (35) and human data (125) indicate that seven days of drug exposure clears rapidly-killed bacilli, leaving a drug-tolerant population. Through serial sputum sampling, we identified a biphasic decline of MTB mrna characteristic of early bactericidal activity studies (25, 27, 105). The slowdown in killing after the bactericidal phase is the clinical hallmark of drug-tolerance (25, 27, 105). Whether drug-tolerant MTB phenotypes continue replication or become nonreplicating is intensely debated (23, 34, 126). Our gene expression data do not provide a direct measure of growth rate but the following observations suggest decreased replication: 16S rrna per genome equivalent of DNA declined 96-99% from day 0 to days 14-56; genes coding for ribosomal proteins were strongly down-regulated; and expression of DNA gyrases and topoisomerases was suppressed. Conversely, the continued detection of transcripts for ribosomal proteins and DNA synthesis functions suggests that replication may not cease entirely. Drug-tolerant MTB are metabolically down-regulated but are not transcriptionally dormant. Induction of multiple stress responses and of genes with regulatory functions indicates active transcriptional reprogramming in response to environmental conditions. Our data support the role of growth-inhibiting TA modules in mediating transition to and maintenance of drug tolerance through post-transcriptional regulation (127, 128). We evaluated mechanisms hypothesized to contribute to drug tolerance. Slow growth decreases drug effectiveness, (30, 46-48) likely because antibiotics generally target replication-associated functions. Drug-tolerant MTB had increased expression of genes for efflux pumps which export rifampicin and isoniazid (129, 130). This supports the proposal 64

72 that pharmacologic efflux pump inhibition might shorten treatment regimens (131). Conversely, drug tolerance is not readily explained by regulation of drug-activating enzymes. In vitro data suggest that one tolerance mechanism is suppression of katg expression leading to diminished isoniazid activation (34, 132). We did not identify down-regulation of katg in vivo. However, this does not rule out the possibility that decreased katg expression occurs in a subset of drug-tolerant bacilli or fosters the initial formation of drug-tolerance after isoniazid exposure. Drug tolerance was not readily explained by altered expression of drug targets. Induction of rpob in drug-tolerant MTB could be compensatory adaptation for rifampicin s inactivation of RNA polymerase. Alternatively, it could indicate a general stress-induced increase in transcriptional reprogramming. In either case, rpob overexpression has not been associated with resistance (133). Increased expression of the gene for isoniazid target InhA causes low-level resistance, (124) but, as in experimental models, (35, 36, 110, 121) inha was not induced in drug-tolerant MTB in sputum. Drug-tolerant MTB had decreased expression of genes for fluoroquinolone targets GyrA and GyrB and bedaquiline target AtpE. Since patients were not administered these drugs, this may reflect downregulation of replication and aerobic metabolism as part of a broader regulatory program rather than drug-specific responses. These results have clinical implications. We have shown that monitoring bacterial physiology during human TB treatment is feasible through MTB expression profiling. Practical applications including monitoring response to novel drugs or as biomarkers predictive of treatment failure or relapse should be evaluated (134). Detection of transcriptionally active bacilli after two months of treatment including in culture negative samples underscores existing evidence that culture detects a small proportion of viable 65

73 bacilli (135). Importantly, our findings indicate that standard doses of rifampicin an inhibitor of transcription initiation fail to fully interrupt bacterial gene expression. Evidence suggests higher rifampicin doses may improve killing and shorten treatment durations (136). Prior to this study it was not possible to test whether experimental models that inform current understanding of drug tolerance and are used for early-stage drug development successfully recapitulate MTB phenotypes in human TB. Gene expression of drug-tolerant MTB in sputum was most similar to an in vitro model that uses antibiotic exposure alone in the absence of hypoxia or other adverse conditions to select for drug-tolerant MTB (35). Drug-tolerant MTB in sputum did not have increased expression of the DosR regulon that is characteristic of models that use hypoxia (41, 47, 48) or hypoxic tissue conditions (46). Genes of the DosR regulon were highly expressed in sputum prior to drug exposure in our data and in a previous study, (39) but were not further induced by treatment. This indicates that drug tolerant bacilli are not necessarily in a more hypoxic environment than drug susceptible bacilli. The transient induction of rela we observed in sputum is consistent with the antibiotic selection model but not with nutrient-limited models. RelA with consequent ppgpp synthesis, may direct the initial formation of drug-tolerant phenotypes. Alternatively, it is possible that the stringent response is not robustly expressed in drug-tolerant MTB in sputum because antibiotic exposure may not alter nutrient availability. Treatment failure and relapse are surmised to result from drug-tolerant MTB persister bacilli. However, there is no direct evidence or scientific consensus regarding the nature of the MTB persister phenotype (23) or even whether a single persister phenotype exists (22, 109). The drug-tolerant MTB population we studied in sputum meets a proposed functional definition of persisters: bacteria that survive prolonged antibiotic exposure despite genetic 66

74 susceptibility (23). Nonetheless, since it is not currently possible to distinguish between bacterial phenotypes or track individual bacilli in humans, we cannot prove that the drugtolerant MTB present in sputum represent the hypothesized persister phenotype. This study has several limitations. The sample size did not permit correlation of bacterial expression with clinical outcomes. Although phenotypic drug-susceptibility testing was not performed, MTB strains profiled in this study were likely drug-susceptible since rates of primary drug resistance in Uganda are low (137) and no rifampicin resistanceconferring mutations were detected with GeneXpert MTB/RIF. Conclusion We provide the first genome-wide description of the transcriptional adaptations of drug-tolerant MTB during human TB treatment. Our data suggest a shift in the mycobacterial population to a slowly replicating, metabolically down-regulated phenotype that no longer displays signals of isoniazid-induced stress. Our findings reinforce the critical need for novel anti-tb drugs to target the recalcitrant drug-tolerant bacterial population, leading to more effective and shorter duration drug regimens. 67

75 Figure 1. Decline in MTB mrna abundance among 17 patients during the first 56 days of TB treatment. The solid line represents the geometric mean percentage of mrna abundance remaining at each day. The shaded area between dotted lines indicates 95% confidence intervals. 68

76 Figure 2. Change in expression of selected functional categories of genes after initiation of TB treatment. The percentage of genes in each category significantly up- (yellow) or down (blue) regulated at each day relative to pre-treatment expression (Day 0) is illustrated. Categories are ordered from most up-regulated (top) to most down-regulated (bottom). Category references are provided in the Online Supplement, Table S4. Abbreviations: EHR, enduring hypoxic response; IS & phages, insertion sequences and phage-related genes; ESX, ESX/Type VII secretion system; TCA, tricarboxcylic acid cycle. 69

77 Figure 3. (A) Heatmap of median expression fold-change of 82 genes differentially expressed (adjusted p-value <0.05) between the early bactericidal phase (Day 2) and sterilizing phase (Day 14). Yellow indicates increased expression relative to day 0; blue indicates decreased expression. Clustering by day (row) indicates similarity between expression patterns at different days. Clustering by gene (column) identifies five clusters of genes (A-E) with similar expression patterns over time. (B) Fold change in expression of genes in five cluster 70

78 Figure 4. Comparison of enrichment of functional gene categories between drug-tolerant MTB in sputum and selected experimental models. (A) Sputum after 14 days of drug treatment, (B) Antibiotic selection model (Betts 2011), (C) Hollow fiber artificial granuloma model (Karakousis 2004), (D) Nutrient starvation model (Betts 2004). (E) Oxygen deprivation non-replicating persistence model (Voskuil 2004), (F) Multiple-stress model (Deb 2004). 71

79 CHAPTER V MYCOBACTERIUM TUBERCULOSIS ADAPTATION TO IMPAIRED HOST IMMUNITY IN HIV-INFECTED PATIENTS Introduction TB is the leading cause of death in HIV, killing an estimated 400,000 HIV-infected persons worldwide each year (1). Nonetheless, it is poorly understood specifically how HIVassociated immunosuppression affects physiologic adaptations of the pathogen MTB. Enhanced understanding of this interaction is crucial because the phenotypic state of MTB alters antibiotic effectiveness (30, 41, 46-48) and may affect transmission fitness (49-51). HIV infection has pervasive effects on human immunity ranging from altered cytokine and chemokine expression to changes in macrophage polarization to T cell dysfunction and depletion leading to clinical, radiographic and histopathologic presentations of TB unlike those observed in immunocompetent hosts (43). Formation of granulomas the hallmark human response to TB is impaired. Unlike the finely-organized granuloma architecture of immunocompetent persons that displays a spatially-modulated gradient of T-cell and macrophage phenotypes, ( ) patients with advanced HIV develop disorganized, poorly-contained collections of polymorphonuclear cells and eosinophils with a paucity of mononuclear, epithelioid or giant cells ( ). MTB is exquisitely adapted to human immunity, dynamically adjusting its physiologic state to tolerate immune responses and diverse tissue micro-environmental conditions (38, 44, 45). In vivo microenvironments have variable ph, nutrient availability and oxygen tension. Hypoxia is encountered in the necrotic center of well-organized granulomas (146, 147) and macrophage phagosomes (51). By contrast, disorganized, poorly-contained 72

80 lesions like those formed by HIV-infected patients are not hypoxic ( ). In response to conditions that limit aerobic respiration (hypoxia, carbon monoxide or nitric oxide (NO)), MTB induces up-regulation of the DosR regulon, (150) a set of genes that is essential for survival and virulence in well-formed hypoxic granulomas (146). DosR proteins additionally appear to modulate adaptive immunity, eliciting T cell responses (151) that enhance host granuloma formation (146). Pathogen-targeted gene expression profiling may provide a molecular snapshot of dynamic host-pathogen interactions ( ) but has not been used to probe MTB adaptation to the altered immune milieu of HIV-infected human patients. Understanding bacterial phenotypes in human patients is important because the physiologic state of MTB in humans differs from phenotypes observed in vitro and in murine models (38-40, 156). We therefore compared MTB transcriptional adaptation to HIV-infected and uninfected hosts in the sputa of patients with untreated active pulmonary TB. After a pilot study in the Gambia revealed that MTB in HIV-infected patients had decreased expression of DosR regulon genes, we confirmed this result in a larger number of Ugandan patients. To elucidate alterations in human immunity that drive this adaptation, we queried a panel of human genes associated with granuloma formation and macrophage polarization. Methods Enrollment and specimen collection Adults with sputum acid-fast bacillus positive pulmonary TB were enrolled prior to antibiotic treatment at the Medical Research Council TB Diagnostic Laboratories, Fajara, The Gambia and Mulago National Referral Hospital in Kampala, Uganda. Using methods detailed elsewhere, (156) spontaneously expectorated sputum was preserved within 5 73

81 minutes in guanidine thiocyanate solution. Sputa were needle-sheared, centrifuged at 9000xg, and stored in Trizol at -80 C. All patients were offered HIV testing. Chest radiographs were evaluated for cavitation using a standardized protocol. Institutional review boards in The Gambia, Uganda and the US approved the study. All patients provided written informed consent. Laboratory analysis We selected only samples with both known HIV status and lineage 4 (EuroAmerican) spoligotype (to limit potential confounding due to lineage differences). Total RNA was extracted using a phenol/chloroform protocol described elsewhere (156). Quality assessment using a test panel targeting 24 MTB transcripts led to exclusion of a small number of samples. Gene expression profiling was performed for a random selection of eligible specimens in the pilot experiment (The Gambia) and all eligible specimens in Uganda. Our MTB expression profiling method is detailed elsewhere (156). Briefly, after first strand cdna synthesis, cdna underwent controlled multiplex amplification using Mtb-specific flanking primers. 2,179 Mtb transcripts were quantified via multiplex qrt-pcr with a LightCycler 480 (Roche, Indianapolis, IN) using TaqMan primers. To elucidate immunologic alterations that may drive differential MTB adaptation among HIV-infected patients, we also assayed a limited panel of human genes (n=234) selected because of their association with granuloma formation or macrophage polarization. Details of TaqMan primers are available at Measurement of human gene expression was possible only in Ugandan samples; samples from the pilot experiment in The Gambia had insufficient RNA. 74

82 Analysis mrna expression data were normalized using a median normalization method validated for this purpose (113). We identified altered gene expression in HIV-infected versus uninfected hosts via t-tests with Benjamini-Hochberg multiple testing correction using a false discovery rate of We summarized the proportion of genes in previously described functional categories (156) with significantly increased or decreased expression with an unadjusted p-value <0.05, using a modified one-way Fisher s exact test to identify enriched functional categories (p-value threshold <0.05). Analysis used the R statistical suite, version except as noted in Supplement. MIAME-compliant data are available at Results Enrollment and clinical characteristics In a pilot study, we assayed MTB gene expression in the sputum of 19 Gambian adults with drug-susceptible, sputum acid-fast bacillus smear-positive, culture-positive pulmonary TB. Four were HIV-infected. Thereafter, we enrolled 63 adults with sputum acidfast bacillus smear- or Gene Xpert-positive, culture-positive pulmonary TB in Uganda, 29 of whom were HIV-infected. Among HIV-infected Ugandan patients, median CD 4 count was 37 (Interquartile range 14-80) (Table 1). None were receiving highly active antiretroviral therapy at the time of TB diagnosis. Among Ugandans, radiographic cavitation was significantly less common among HIV-infected patients (n=13, 48%) than among HIVuninfected patients (n=25, 89%, p-value=0.003). 75

83 Decreased expression of MTB DosR regulon genes in HIV-infected patients Comparison of MTB expression between HIV-infected and uninfected patients in the Gambia, showed 447 (20.5%) genes had a p-value <0.05, considerably more than expected by chance alone. When repeated in Uganda, 344 (15.8%) genes had p-value <0.05. Enrichment analysis of functional gene categories (Table S1) in genes with p-value <0.05 showed that MTB expression of DosR regulon genes was significantly lower among HIV-infected patients than among HIV-uninfected patients (BH-adjusted p< in the Gambia; p= in Uganda). In the Gambia, 32 of 48 DosR regulon genes had significantly lower expression (p<0.05) among HIV-infected patients (Figure 1a-b); no DosR genes had significantly increased expression in HIV. In Uganda, 20 DosR regulon genes had significantly lower expression among HIV-infected patients; one DosR gene (hypothetical protein Rv2628) had increased expression in HIV. Expression of sentinel DosR genes was down-regulated among HIV-infected patients, including tgs1 (5.4-fold in the Gambia, 1.9- fold in Uganda), dosr (3.9-fold in the Gambia, 2.2-fold in Uganda), doss (3.1-fold in the Gambia, 1.3-fold in Uganda) and hspx (acr) (2.0-fold in the Gamiba, 2.1-fold in Uganda). In both the Gambia and Uganda, expression of the DosR regulon was the dominant change; no other gene categories changed unambiguously. Host transcriptome indicates altered immune milieu in HIV Expression of human genes associated with granuloma formation and macrophage polarization were profoundly altered among HIV-infected patients (Figure 2). Of 234 genes assayed, 87 (37.2%) had a p-value <0.05 (Table Sx); 60 remained significant after adjustment for multiple comparisons. Relative to HIV-uninfected patients, HIV-infected patients had lower expression of interferon gamma (IFN-γ) (2.4-fold, BH-adjusted p-value 0.02) and 76

84 interferon regulatory factor 1 (IRF1) (1.8-fold, BH-adjusted p-value 0.02). HIV-infected patients had higher expression of interleukin-4 (IL4) (4.0-fold, BH-adjusted p-value=0.03), interleukin-10 (2.7-fold, BH-adjusted p-value=0.002) and arginase 1 (ARG1) (3.0-fold, BHadjusted p-value=0.009). Expression of tumor necrosis factor alpha (TNF-α) and inducible nitric oxide synthase (NOS2) was unchanged (BH-adjusted p-values 0.6 and 0.2, respectively). Discussion HIV transforms the interaction between MTB and its human host. In two geographically-distinct human populations, we found that MTB in HIV-infected patients had decreased expression of the DosR regulon, a crucial metabolic switch with immunomodulatory effects. Down-regulation of DosR expression among HIV-infected patients may be a direct result of arginase-driven decreased macrophage NO production or an indirect result of decreased hypoxic stress in poorly formed granulomas. This analysis provides a novel perspective on the complex interplay between HIV-associated immune impairment, altered tissue microenvironments and bacterial physiologic state. The DosR regulon is genetic program that directs metabolic adaptation to conditions that inhibit aerobic respiration and cause reductive stress in a pathogen that is an obligate aerobe (150, 157). In immunocompetent hosts that form well-organized hypoxic granulomas, DosR is necessary to sustain persistent infection beyond the onset of adaptive immunity (146). Conversely, DosR is not required in animal models that form poorly-organized, nonhypoxic lesions similar to those observed in HIV-infected patients (147, 158). Decreased induction of the DosR regulon may be due to several interacting mechanisms that decrease NO and hypoxic stress. First, human gene expression in these 77

85 samples suggests altered macrophage polarization that may decrease NO production. Consistent with previous studies ( ), HIV-infected TB patients in this study had decreased expression of interferon-γ (IFN-γ), a principal mediator of macrophage activation in TB (163) and increased expression of interleukin-10 (IL-10), a Th2 cytokine that induces alternative (anti-inflammatory) macrophage polarization and enhances MTB survival (160, 164, 165). Furthermore, HIV-infected patients had a 3-fold increase in expression of Arg1, characteristic of pro-healing, anti-inflammatory macrophages (139). Arg1 competes with NOS2 for L-arginine, down-regulating macrophage NO production and increasing susceptibility of mice to MTB (165, 166). Dysregulation of granuloma morphology may decrease hypoxic stress, ( ) thereby decreasing MTB requirement for DosR. HIV-infected subjects in this study had profound depletion of CD4 lymphocytes, a central component of well-organized granulomas (144, 167) and had a Th2-biased cytokine environment. In addition to direct effects on macrophage polarization cited above, IFN-γ is critical to granuloma formation. Conversely, IL-10 expression inhibits formation of mature granulomas (168) and Arg1 limits T-cell proliferation, independent of its effect on NO (138). Finally, MTB encounters profoundly hypoxic conditions within macrophages (51). If the immune dysregulation associated with HIV increases the proportion of MTB that is extrarather than intracellular, the result would be decreased hypoxic stress and decreased DosR expression. These putative mechanisms leading to decreased DosR expression in HIV may be cumulative and interacting; all shape the tissue micro-environment to which MTB must adapt. 78

86 The interaction between host microenvironments and bacterial physiology is not necessarily one-sided. A number of DosR regulon proteins have strong T-cell and IFN-γ inducing capacity (151, 169) and appear to enhance granuloma formation (146, 158). Consistent with this hypothesis, DosR mutants cause less tissue pathology in mice relative to wild-type MTB despite a similar bacillary burden (170). This suggests that down-regulation of the DosR regulon may be not only a consequence but also a cause of altered host responses in HIV/TB co-infected patients. The MTB transcriptional alterations we identified among HIV-infected subjects are important because the pathogen s phenotypic state influences antibiotic effectiveness both in vitro (41, 47, 48) and in vivo (171). For example, comparison of C3H3b/FeJ mice that form human-like caseating hypoxic granulomas with Balb/c mice that form non-hypoxic lesions (172) showed different responses to clofazamine (171). Future research is needed to understand specifically how the altered tissue microenvironments that result from HIVassociated immune dysregulation affect MTB response to antibiotics in humans. This study has several limitations. Different MTB strains could have different gene expression responses. We alleviated, but did not entirely eliminate, this potential confounder by restricting our analysis to MTB of EuroAmerican lineage. Second, since our patients did not undergo lung biopsy, we could not directly assess granuloma organization and tissue oxygen tension. However, previous reports have described a strong and consistent impact of HIV on granuloma morphology ( ) and disorganized uncontained granulomas are shown in animals to be non-hypoxic ( ). Since HIV-infected subjects in this study had advanced HIV (with median CD4 = 37), we presume they had similarly deranged granuloma formation. 79

87 Conclusion MTB differentially adapts its physiologic state to the HIV-infected host. Likely driven by diminished NO stress from alternatively activated macrophages and diminished hypoxic stress in disorganized granulomas, MTB in HIV-infected patients displayed decreased expression of DosR, a bacterial response to hypoxia with immunomodulatory effects. These observations identify novel connections between host immunity, host tissue micro-environments and bacterial phenoptypes. Further investigation of determinants of MTB phenotypes during human infection is important because bacterial physiologic state influences antibiotic effectiveness. 80

88 Table 1. Demographic and clinical characteristics of TB patients in The Gambia (pilot study) and Uganda that underwent pathogen-targeted gene expression profiling. Age (median, IQR) Characteristic HIV-infected HIV-uninfected p-value The Gambia 36 (28-44) 25 (21-32) 0.4 Uganda 30 (27-38) 26 (22-32) 0.1 Female (n, %) The Gambia 3 (75%) 1 (7%) 0.02 Uganda 13 (45%) 11 (32%) 0.5 CD4+ T-cell count (median, IQR) a Uganda 37 (14-80) - - Radiographic cavitation (n, %) The Gambia b 1 (33%) 8 (67%) 0.7 Uganda c 13 (48%) 25 (89%) Light microscopy results (n, %) The Gambia Negative - Scanty / (7%) 2+ 1 (25%) 4 (27%) 3+ 3 (75%) 10 (67%) Uganda Negative 1 (7%) - Scanty / 1+ 5 (17%) 5 (15%) (38%) 9 (27) (38%) 20 (59%) a CD4 T cell count unavailable in pilot experiment in The Gambia. b Chest radiographs missing for one HIV-infected and three HIV-uninfected subjects. c Chest radiographs missing for two HIV-infected and six HIV-uninfected subjects. 81

89 Figure 1. Expression of MTB DosR regulon in the sputum of HIV-infected versus HIVuninfected TB patients in the Gambia (A) and Uganda (B), log2 scale. Red indicates decreased expression in HIV, blue indicates increased expression. 82

90 Figure 2. Differential expression of human genes associated with granuloma formation and macrophage polarization in HIV-infected versus uninfected TB patients with hypothesized consequences. Red indicates significantly decreased and green indicates significantly increased expression in HIV-infected samples. Mean fold-change in HIV-infected relative to uninfected samples is shown in parentheses. The observed decreased expression of key Th1 and increased expression of Th2 cytokines in HIV may contribute to impaired granuloma formation, diminishing hypoxic stress. Increased expression of Arg1 in HIV may compete with inos for L-arginine, decreasing NO production. Arg1 additionally inhibits granuloma formation. A number of DosR regulon proteins strongly induce T-cell and IFN-γ responses, enhancing granuloma formation. Decreased expression of the DosR regulon is therefore anticipated to further impair granuloma formation. 83

91 CHAPTER VI CONCLUSIONS These projects presented in this dissertation have developed novel practical approaches for using data science to advance public health interventions in TB. The components address several major public health challenges. The first project an analysis of LTBI reactivation risk among immigrants addresses a critical problem. A centerpiece of domestic TB control is a strategic plan for elimination of TB within the US, defined as less than 1 case per million persons per year (173). Substantial progress has been made in recent decades. For example, the US rate of TB diagnosis declined 71% from 1990 to 2014, from 10.3 to 3 per 100,000 (5). However, these gains were made primarily among US-born persons. The foreign-born share of TB has risen continuously. During these 25 years, the percentage of all TB diagnosed among foreign-born persons rose from 26% to 67%. TB among immigrants and the foreign-born is therefore considered the central obstacle to TB elimination in the US. TB in the foreign born is a particularly challenging problem because transmission of active TB has been largely interrupted within the US (2, 3). Most TB cases within US occur as a result of exposures and infections acquired prior to US arrival, (4) meaning preventive treatment of LTBI is the only effective prevention strategy. In 2013, approximately 41.3 million persons in the US were foreign-born (174). The prevalence of LTBI is approximately 10 time higher among foreign-born persons than among than US-born persons (3). To effectively target LTBI treatment towards persons at highest risk of developing TB, it is essential to have accurate epidemiologic data about TB risk among the foreign born. 84

92 The first project in this dissertation addressed this challenge using a novel study design to estimate TB risk among immigrants over time. We found that it is possible to link overseas pre-immigration screening records with domestic TB case reports, creating a retrospective longitudinal cohort. Analysis of this dataset challenges the conventional wisdom that risk of LTBI reactivation is highest in the immediate post-arrival period. Current guidelines appear to be based on ecological fallacy. Our finding that TB risk in immigrants with normal pre-immigration chest radiographs is stable over time has important implications for US TB control strategies. In order to maximize public health benefit, current guidelines emphasize targeting screening and preventive treatment to the populations at highest risk for TB. Our data suggest that newly arrived immigrants with normal chest radiographs do not have uniquely high rates of TB. Risk among newly-arrived immigrants is in fact similar to TB risk among immigrants who have been in the US longer. It suggests that targeted testing and preventive treatment should be extended to all foreign-born persons with evidence of LTBI, regardless of duration in the US. This makes the challenge of TB prevention in the US even more challenging than previously anticipated. Fifteen years ago in a perspective entitled Tide pools: what will be left after the tide has turned?, Paula Fujiwara predicted the challenge of diminishing returns of TB prevention efforts as TB incidence declined (175). Our findings are consistent this prediction. As the tide of TB in the US has receded to historically low levels, the cost of further TB prevention has grown. Since the individual risk for TB is lower in the general population of foreign-born persons than previously (erroneously) assumed in the population of newly-arrived immigrants, the number needed to treat to prevent a case of TB is higher. 85

93 This has important implications for the cost-effectiveness of TB prevention. Our findings are also important for models that estimate prevention benefits of LTBI treatment and estimate the feasibility of TB elimination in the US. Finally, the methodology developed in this project for performing transnational linkage of immigration screening data and using results for surveillance and program evaluation has been published (176). Our novel method is additionally potentially useful for case ascertainment in a clinical trial. This method has enabled submission of an R34 to plan an LTBI screening and treatment trial based at overseas pre-immigration screening sites. Our linkage method enables subjects to be followed electronically for development of TB after US arrival. The second project in this dissertation also addresses a topic of critical importance in TB control: accurate early detection of active TB. Current diagnostic tests have inadequate sensitivity, meaning missed diagnoses are common (11). The result is continued disease progression for individual patients and continuing TB transmission within the community. Therefore, development of novel assays for TB is a leading World Health Organization priority (12). Our project tested a promising approach to diagnosis classification based on blood transcriptional profiles in a population that has not previously been evaluated. Specifically, we evaluated a group of patients with ethnic heritage and co-morbidities typical of those observed among US TB patients. In addition to development of novel transcriptional signatures in this population, we systematically evaluated the generalizability to other populations and clinical utility of these biomarkers. 86

94 This study provided important insight into blood transcriptional classifiers for TB and also has more general implications for biomarker development generally. Our analysis was able to identify transcriptional patterns that accurately distinguished active TB from either LTBI or pneumonia within our patient cohort. However, we found that the reference group used in classifier development was critically important, both in terms of the transcripts included in the classifier and in terms of the practical usefulness of the classifier. For example, a classifier established by comparing TB with latent TB was unable to distinguish TB from pneumonia. Similarly, our classifier developed by comparing TB with pneumonia was less accurate at classifying TB when tested among patients with TB or cancer. And a classifier developed by other investigators based on comparison of TB and cancer was less accurate for detection of TB when tested among patients with TB or pneumonia. These results suggest that classifiers developed via comparison with a single disease reference may be excessively specific to that comparison. Clinically, the practical challenge is that a practical assay must distinguish active TB from a diverse spectrum of diseases, including pneumonia, cancer and a number of additional conditions. Therefore, while our results are encouraging because they show TB can be distinguished from pneumonia based on the blood transcriptome, our analysis also points to the need for further cross-cohort and cross-study comparison to identify classifiers that are robust and generalizable enough to be used as real world practical assays for TB. The third project in this dissertation used pathogen-targeted transcriptional profiling to evaluate transcriptional characteristics of the drug-tolerant MTB population that survives prolonged drug exposure in human sputum. The drug-tolerant persister MTB phenotypes that emerge during treatment are of the utmost public health importance because they 87

95 demand prolonged drug exposure to eradicate and are considered the root cause of treatment failure and relapse. Our novel approach to characterizing changes in the physiological state bacterial population during treatment of human TB provided new perspective on the physiologic alterations that enable drug tolerance. This preliminary proof-of-concept study leads to new questions. Can this method of pathogen-targeted transcriptional profiling in sputum be used as a pharmacodynamic measure of drug impact? Could it be used to evaluate novel drug regimens? Would different drug regimens result in different physiologic adaptations by the bacterial population (as indicated by different transcriptional patterns)? Additionally, would MTB transcriptional patterns be different in patients who are bound to relapse compared with patients who are successfully durably cured? Could MTB transcriptional patterns serve as surrogate markers for subsequent treatment failure or relapse? Lack of early surrogate markers predictive of relapse is a major challenge for TB drug development. Better surrogate markers based on molecular measures of pathogen physiology rather than culturability would accelerate the drug development pipeline by reducing the size, cost and time duration of clinical trials. These questions would need to be evaluated in a large-scale drug trial. The fourth and final project in this dissertation evaluated the impact of host immunity on the MTB transcriptome prior to treatment by comparing MTB isolated from HIV-infected and HIV-uninfected patients. It revealed that host immune milieu has an important impact on the physiologic state of the MTB population. Specifically, HIV-induced immune derangements were associated with decreased expression of genes of DosR regulon, a critical metabolic switch. 88

96 This important and novel observation should inspire additional lines of investigation. Binary classification of immune responses as HIV-infected or HIV-uninfected is a crude measure of the immune responses MTB encounters in vivo. In reality, there may be a range of immune responses among HIV-infected patients or among HIV-uninfected patients. With more nuanced phenotyping of human immune responses, it is possible that future analyses will identify important interactions between MTB and host immunity. In the aggregate, these four projects harnessed novel forms of large-scale administrative or biological data to address public health challenges in TB control. The continuing challenge will be translation from analysis to insights that will lead to practical solutions for the ancient and continuing scourge of TB. 89

97 REFERENCES 1. World Health Organization Global tuberculosis control: WHO report CDC Trends in Tuberculosis - United States, MMWR 62(11): Horsburgh CR, Rubin EJ Latent tuberculosis infection in the United States. N Engl J Med 364: Ricks PM, Cain KP, Oeltmann JE, Kammerer JS, Moonan PK Estimating the burden of tuberculosis among foreign-born persons acquired prior to entering the U.S., PLoS One 6:e CDC Trends in Tuberculosis - United States, MMWR 64(10): Zuber PLF, McKenna MT, Binkin NJ, Onorato IM, Castro KG Long-term risk of tuberculosis among foreign-born persons in the United States. JAMA 278: McKenna MT, McCray E, Onorato I The epidemiology of tuberculosis among foreign-born persons in the United States, 1986 to N Engl J Med 332: CDC Recommendations for prevention and control of tuberculosis among foreign-born persons: Report of the working group on tuberculosis among foreignborn persons. MMWR Recomm Rep 47(RR-16): CDC Targeted tuberculin testing and treatment of latent tuberculosis infection. American Thoracic Society. MMWR Recomm Rep 49(RR-6): CDC CDC Immigration Requirements: Technical Instructions for Tuberculosis Screening and Treatment Using Cultures and Directly Observed Therapy. Atlanta, GA. 11. Wallis RS, Kim P, Cole S, Hanna D, Andrade BB, Maeurer M, Schito M, Zumla A Tuberculosis biomarkers discovery: developments, needs, and challenges. Lancet Infect Dis 13: Doherty MT, Wallis RS, Zumla A, WHO- TropicalDiseaseResearch/EuropeanCommissionJointExpertConsultationGroup Biomarkers for tuberculosis disease status and diagnosis. Curr Opin Pulm Med 15: Bloom CI, Graham CM, Berry MPR, Rozakeas F, Redford PS, Wang Y, Xu Z, Wilkinson KA, Wilkinson RJ, Kendrick Y, Devouassoux G, Ferry T, Miyara M, 90

98 Bouvry D, Dominique V, Gorochov G, Blankenship D, Saadatian M, Vanhems P, Beynon H, Vancheeswaran R, Wickremasinghe M, Chaussabel D, Banchereau J, Pascual V, Ho L-p, Lipman M, O Garra A Transcriptional blood signatures distinguish pulmonary tuberculosis, pulmonary sarcoidosis, pneumonias and lung cancers. PLoS ONE 8:e Bloom CI, Graham CM, Berry MPR, Wilkinson KA, Oni T, Rozakeas F, Xu Z, Rossello-Urgell J, Chaussabel D, Banchereau J, Pascual V, Lipman M, Wilkinson RJ, O Garra A Detectable changes in the blood transcriptome are present after two weeks of antituberculosis therapy. PLoS ONE 7:e Berry MPR, Graham CM, McNab FW, Xu Z, bloch SAA, Oni T, Wilkinson KA, Banchereau R, Skinner J, Wilkinson RJ, Quinn C, Blankenship D, Dhawan R, Cush JJ, Mejias A, Ramilo O, Kon OM, Pascual V, Banchereau J, Chaussabel D, O'Garra A An interferon-inducible neutrophil-driven blood transcriptional signature in human tuberculosis. Nature 466: Maertzdorf J, Weiner J, Mollenkopf H-J, Network T, Bauer T, Prasse A, Müller-Quernheim J, Kaufmann SHE Common patterns and disease-related signatures in tuberculosis and sarcoidosis. Proc Nat Acad Sci USA 109: Lesho E, Forestiero FJ, Hirata MH, Hirata RD, Cecon L, Melo FF, Paik SH, Murata Y, Ferguson EW, Wang Z, Ooi GT Transcriptional responses of host peripheral blood cells to tuberculosis infection. Tuberculosis 91: Anderson ST, Kaforou M, Brent AJ, Wright VJ, Banwell CM, Chagaluka G, Crampin AC, Dockrell HM, French N, Hamilton MS, Hibberd ML, Kern F, Langford PR, Ling L, Mlotha R, Ottenhoff THM, Pienaar S, Pillay V, Scott JAG, Twahir H, Wilkinson RJ, Coin LJ, Heyderman RS, Levin M, Eley B Diagnosis of childhood tuberculosis and host RNA expression in Africa. N Engl J Med 370: Kaforou M, Wright VJ, Oni T, French N, Anderson ST, Bangani N, Banwell CM, Brent AJ, Crampin AC, Dockrell HM, Eley B, Heyderman RS, Hibberd ML, Kern F, Langford PR, Ling L, Mendelson M, Ottenhoff TH, Zgambo F, Wilkinson RJ, Coin LJ, Levin M Detection of tuberculosis in HIV-infected and -uninfected African adults using whole blood RNA expression signatures: A case-control study. PLoS Med 10:e Connolly LE, Edelstein PH, Ramakrishnan L Why Is long-term therapy required to cure tuberculosis? PLoS Med 4:e Zhang Y, Yew WW, Barer MR Targeting persisters for tuberculosis control. Antimicrob Agents Chemother 56:

99 22. Lipsitch M, Levin BR Population dynamics of tuberculosis treatment: mathematical models of the roles of non-compliance and bacterial heterogeneity in the evolution of drug resistance. Int J Tuberc Lung Dis 2: Balaban NQ, Gerdes K, Lewis K, McKinney JD A problem of persistence: still more questions than answers? Nat Rev Micro 11: Vandiviere HM, Loring WE, Melvin WE, Willis S The treated pulmonary lesion and its tubercle bacillus. II The death and resurrection. Am J Med Sci 232: Mitchison DA, Selkon JB The bactericidal activities of antituberculosis drugs. Am Rev Respir Dis 74: Rustomjee R, Lienhardt C, Kanyok T, Davies GR, Levin J, Mthiyane T, Reddy C, Sturm AW, Sirgel FA, Allen J, Coleman DJ, Fourie B, Mitchison DA, Gatifloxacin for TBst A Phase II study of the sterilising activities of ofloxacin, gatifloxacin and moxifloxacin in pulmonary tuberculosis. Int J Tuberc Lung Dis 12: Jindani A, Doré CJ, Mitchison DA Bactericidal and sterilizing activities of antituberculosis drugs during the first 14 Days. Am J Resp Crit Care 167: Mitchison DA Basic mechanisms of chemotherapy. Chest 76: Mitchison DA, Sturm AW The measurement of early bactericidal activity. Ballieres Clin Infect Dis 4: Karakousis PC, Williams EP, Bishai WR Altered expression of isoniazidregulated genes in drug-treated dormant Mycobacterium tuberculosis. J Antimicrob Chemother 61: Wayne LG, Hayes LG An in vitro model for sequential study of shiftdown of Mycobacterium tuberculosis through two stages of nonreplicating persistence. Infect Immun 64: Rifat D, Bishai WR, Karakousis PC Phosphate depletion: A novel trigger for Mycobacterium tuberculosis persistence. J Infect Dis 200: Hu Y, Mangan JA, Dhillon J, Sole KM, Mitchison DA, Butcher PD, Coates ARM Detection of mrna transcripts and active transcription in persistent Mycobacterium tuberculosis induced by exposure to rifampin or pyrazinamide. J Bacteriology 182:

100 34. Wakamoto Y, Dhar N, Chait R, Schneider K, Signorino-Gelo F, Leibler S, McKinney JD Dynamic persistence of antibiotic-stressed mycobacteria. Science 339: Keren I, Minami S, Rubin E, Lewis K Characterization and transcriptome analysis of Mycobacterium tuberculosis persisters. mbio Tudó G, Laing K, Mitchison DA, Butcher PD, Waddell SJ Examining the basis of isoniazid tolerance in nonreplicating Mycobacterium tuberculosis using transcriptional profiling. Future Med Chem 2: McCune RM, Tompsett R Fate of Mycobacterium tuberculosis in mouse tissues as determined by the microbial enumeration technique: I. The persistence of drug-susceptible tubercle bacilli in the tissues despite prolonged antimicrobial therapy. J Exp Med 104: Rachman H, Strong M, Ulrichs T, Grode L, Schuchhardt J, Mollenkopf HJ, Kosmiadi GA, Eisenberg D, Kaufmann SHF Unique transcriptome signature of Mycobacterium tuberculosis in pulmonary tuberculosis. Infect Immun 74: Garton NJ, Waddell SJ, Sherratt AL, Lee S-M, Smith RJ, Senner C, Hinds J, Rajukumar K, Adegbola RA, Besra GS, Butcher PD, Barer MR Cytological and transcript analyses reveal fat and lazy persister-like bacilli in tuberculosis sputum. PLoS Med 5:e Fenhalls G, Stevens L, Moses L, Bezuidenhout J, Betts JC, Helden Pv P, Lukey PT, Duncan K In situ detection of Mycobacterium tuberculosis transcripts in human lung granulomas reveals differential gene expression in necrotic lesions. Infect Immun 70: Muttucumaru DGN, Roberts G, Hinds J, Stabler RA, Parish T Gene expression profile of Mycobacterium tuberculosis in a non-replicating state. Tuberculosis 84: Lenaerts A, Barry CE, Dartois V Heterogeneity in tuberculosis pathology, microenvironments and therapeutic responses. Immunological Reviews 264: Diedrich CR, Flynn JL HIV-1/Mycobacterium tuberculosis coinfection immunology: How does HIV-1 exacerbate tuberculosis? Infect Immun 79: Schnappinger D, Ehrt S, Voskuil MI, Liu Y, Mangan JA, Monahan IM, Dolganov G, Efron B, Butcher PD, Nathan C, Schoolnik GK Transcriptional adaptation of Mycobacterium tuberculosis within macrophages: Insights into the phagosomal environment. J Exp Med 198:

101 45. Ehrt S, Rhee K Mycobacterium tuberculosis metabolism and host interaction: Mysteries and paradoxes, p In Pieters J, McKinney JD (ed), Pathogenesis of Mycobacterium tuberculosis and its Interaction with the Host Organism, vol 374. Springer Berlin Heidelberg. 46. Karakousis PC, Yoshimatsu T, Lamichhane G, Woolwine SC, Nuermberger EL, Grosset J, Bishai WR Dormancy phenotype displayed by extracellular Mycobacterium tuberculosis within artificial granulomas in mice. J Exp Med 200: Deb C, Lee C-M, Dubey VS, Daniel J, Abomoelak B, Sirakova TD, Pawar S, Rogers L, Kolattukudy PE A novel in vitro multiple-stress dormancy model for Mycobacterium tuberculosis generates a lipid-loaded, drug-tolerant, dormant pathogen. PLoS One 4:e Voskuil MI, Visconti KC, Schoolnik GK Mycobacterium tuberculosis gene expression during adaptation to stationary phase and low-oxygen dormancy. Tuberculosis 84: Eum S-Y, Kong J-H, Hong M-S, Lee Y-J, Kim J-H, Hwang S-H, Cho S-N, Via LE, Barry IIICE Neutrophils are the predominant infected phagocytic cells in the airways of patients with active pulmonary TB. Chest 137: Middelkoop K, Mathema B, Myer L, Shashkina E, Whitelaw A, Kaplan G, Kreiswirth B, Wood R, Bekker L-G Transmission of tuberculosis in a South African community with a high prevalence of HIV infection. J Infect Dis 211: Cunningham-Bussel A, Zhang T, Nathan CF Nitrite produced by Mycobacterium tuberculosis in human macrophages in physiologic oxygen impacts bacterial ATP consumption and gene expression. Proc Nat Acad Sci USA 110:E4256-E Geiter L (ed) Ending neglect: The elimination of tuberculosis in the United States. National Academy Press, Washington, D.C. 53. Cohen T, Murray M Incident tuberculosis among recent US immigrants and exogenous reinfection. Emerg Infect Dis 11: Guo H, Wu J Persistent high incidence of tuberculosis among immigrants in a low-incidence country: Impact of immigrants with early or late latency. Math Biosci Eng 8: Patel S, Parsyan AE, Gunn J, Barry MA, Reed C, Sharnprapai S, Horsburgh CR Risk of progression to active tuberculosis among foreign-born persons with latent tuberculosis. Chest 131:

102 56. Cain KP, Haley CA, Armstrong LR, Garman KN, Wells CD, Iademarco MF, Castro KG, Laserson KF Tuberculosis among foreign-born persons in the United States: Achieving tuberculosis elimination. Am J Resp Crit Care 175: Liu Y, Weinberg MS, Ortega LS, Painter JA, Maloney SA Overseas screening for tuberculosis in U.S.-bound immigrants and refugees. N Engl J Med 360: Lowenthal P, Westenhouse J, Moore M, Posey DL, Watt JP, Flood J Reduced importation of tuberculosis after the implementation of an enhanced preimmigration screening protocol. Int J Tuberc Lung Dis 15: Cain KP, Benoit SR, Winston CA, Mac Kenzie WR Tuberculosis among foreign-born persons in the United States. JAMA 300: Marks GB, Bai J, Simpson SE, Sullivan EA, Stewart GJ Incidence of tuberculosis among a cohort of tuberculin-positive refugees in Australia: Reapprasing the estimates of risk. Am J Resp Crit Care 162: United_States_Department_of_Homeland_Security Yearbook of Immigration Statistics. U.S. Department of Homeland Security, Office of Immigration Statistics, 62. World Health Organization Global tuberculosis control: WHO report Tupasi TE, Radhakrishna S, Pascual ML, Quelapio MID, Villa ML, Co VM, Sarol J, Mangubat N, Reyes AC, Sarmiento A, Solon M, Solon F, Burton L, Lofranco VS, Rostrata MPC, Mantala MJ BCG coverage and the annual risk of tuberculosis infection over a 14-year period in the Philippines assessed from the Nationwide Prevalence Surveys. Int J Tuberc Lung Dis 4: Xia Q, Westenhouse JL, Schultz AF, Nonoyama A, Elms W, Wu N, Tabshouri L, Ruiz JD, Flood JM Matching AIDS and tuberculosis registry data to identify AIDS/tuberculosis comorbidity cases in California. Health Informatics Journal 17: CDC Registry Plus Link Plus 2.0, National Program of Cancer Registries, Atlanta GA Rosner B Fundamentals of biostatistics. Brooks/Cole, Boston, MA. 67. Linas BP, Wong AY, Freedberg KA, Horsburgh CR Priorities for screening and treatment of latent tuberculosis infection in the United States. Am J Resp Crit Care 184:

103 68. Wilcke JT, Poulsen S, Askgaard DS, Enevoldsen HK, Rønne T, Kok-Jensen A Tuberculosis in a cohort of Vietnamese refugees after arrival in Denmark Int J Tuberc Lung Dis 2: Lillebaek T, Andersen AB, Dirksen A, Smith E, Skovgaard LT, Kok-Jensen A Persistent high incidence of tuberculosis in immigrants in a low incidence country. Emerg Infect Dis 8: Johnsen NL, Steen TW, Meyer H, Heldal E, Skarpaas IJK, June GB Cohort analysis of asylum seekers in Oslo, Norway, : effectiveness of screening at entry and TB incidence in subsequent years. Int J Tuberc Lung Dis 9: Wang JS, Allen EA, Enarson DA, Grzybowski S Tuberculosis in recent Asian immigrants to British Columbia, Canada: Tubercle 72: Nolan CM, Elarth AM Tuberculosis in a cohort of Southeast Asian refugees. Am Rev Respir Dis 137: Cain KP, Mac Kenzie WR Overcoming the limits of tuberculosis prevention among foreign-born individuals: Next steps toward eliminating tuberculosis. Clin Infect Dis 46: Mandalakas AM, Menzies D Is screening immigrants for latent tuberculosis cost-effective? Lancet Infect Dis 11: Schwartzman K, Menzies D Tuberculosis screening of immigrants to lowprevalence countries. A cost-effectiveness analysis. Am J Respir Crit Care Med 161: Schwartzman K, Oxlade O, Barr RG, Grimard F, Acosta I, Baez J, Ferreira E, Melgen RE, Morose W, Salgado AC, Jacquet V, Maloney S, Laserson K, Mendez AP, Menzies D Domestic returns from investment in the control of tuberculosis in other countries. N Engl J Med 353: Dasgupta K, Menzies D Cost-effectiveness of tuberculosis control strategies among immigrants and refugees. Eur Respir J 25: Sterling TR, Villarino ME, Borisov AS, Shang N, Gordin F, Bliven-Sizemore E, Hackman J, Hamilton CD, Menzies D, Kerrigan A, Weis SE, Weiner M, Wing D, Conde MB, Bozeman L, Horsburgh CR, Chaisson RE Three months of rifapentine and isoniazid for latent tuberculosis infection. N Engl J Med 365: Horsburgh CR, Goldberg S, Bethel J, Chen S, Colson PW, Hirsch-Moverman Y, Hughes S, Shrestha-Kuwahara R, Sterling TR, Wall K, Weinfurter P, 96

104 Consortium ttes Latent TB infection treatment acceptance and completion in the United States and Canada. Chest 137: Sterling TR, Bethel J, Goldberg S, Weinfurter P, Yun L, Horsburgh CR, the Tuberculosis Epidemiologic Studies Consortium The scope and impact of treatment of latent tuberculosis infection in the United States and Canada. Am J Resp Crit Care 173: CDC Reported Tuberculosis in the United States, U.S. Department of Health and Human Services, CDC, Atlanta, GA. 82. Liu Y, Painter JA, Posey DL, Cain KP, Weinberg MS, Maloney SA, Ortega LS, Cetron MS Estimating the impact of newly arrived foreign-born persons on tuberculosis in the United States. PLoS ONE 7:e Maloney SA, Fielding KL, Laserson KF, Jones W, Nguyen TN, Dang AQ, Nguyen HP, Nguyen AT, Duong TC, Vo TC, Seawright MF, O'Rourke T, Truong XL, Nguyen TN, Binkin N, Cetron MS Assessing the performance of overseas tuberculosis screening programs: A study among US-bound immigrants in Vietnam. Arch Intern Med 166: Anonymous Efficacy of various durations of isoniazid preventive therapy for tuberculosis: Five years of follow-up in hte IUAT trial. International Union Against Tuberculosis Committee on Prophylaxis. Bull World Health Organ 60: Ferebee SH Controlled chemoprophylaxis trials in tuberculosis. A general review. Bibl Tuberc 26: Bennett DE, Courval JM, Onorato I, Agerton T, Gibson JD, Lambert L, McQuillan GM, Lewis B, Navin TR, Castro KG Prevalence of tuberculosis infection in the United States population. Am J Resp Crit Care 177: Joosten SA, Fletcher HA, Ottenhoff THM A helicopter perspective on TB biomarkers: Pathway and process based analysis of gene expression data provides new insight into TB pathogenesis. PLoS ONE 8:e Maertzdorf J, Kaufmann SHE, Weiner J Toward a unified biosignature for tuberculosis. Cold Spring Harb Perspect Med Kaforou M, Wright VJ, Levin M Host RNA signatures for diagnostics: An example from paediatric tuberculosis in Africa. J Infect 69, Supplement 1:S28-S Pepe MS Evaluating technologies for classification and prediction in medicine. Stat Med 24:

105 91. Satproedprai N, Wichukchinda N, Suphankong S, Inunchot W, Kuntima T, Kumpeerasart S, Wattanapokayakit S, Nedsuwan S, Yanai H, Higuchi K, Harada N, Mahasirimongkol S Diagnostic value of blood gene expression signatures in active tuberculosis in Thais: a pilot study. Genes Immun 16: Laux da Costa L, Delcroix M, Dalla Costa ER, Prestes IV, Milano M, Francis SS, Unis G, Silva DR, Riley LW, Rossetti MLR A real-time PCR signature to discriminate between tuberculosis and other pulmonary diseases. Tuberculosis 95: Cliff JM, Lee J-S, Constantinou N, Cho J-E, Clark TG, Ronacher K, King EC, Lukey PT, Duncan K, Van Helden PD, Walzl G, Dockrell HM Distinct phases of blood gene expression pattern through tuberculosis treatment reflect modulation of the humoral immune response. J Infect Dis 207: Macaskill P, Gatsonis C, Deeks JJ, Harbord RM, Takwoingi Y Analyzing and presenting results, Cochrane handbook for systematic reviews of diagnostic test accuracy. The Cochrane Collaboration, Oxford. 95. Justice AC, Covinsky KE, Berlin JA Assessing the generalizability of prognostic information. Ann Intern Med 130: Bolstad BM, Irizarry RA, Åstrand M, Speed TP A comparison of normalization methods for high density oligonucleotide array data based on variance and bias. Bioinformatics 19: Johnson WE, Li C, Rabinovic A Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics 8: Guyon I, Weston J, Barnhill S, Vapnik V Gene selection for cancer classification using support vector machines. Machine Learning 46: Hastie T, Tibshirani R, Friedman J Elements of statistical learning: Data mining, inference and prediction, 2nd edition. Springer-Verlag, New York Krstajic D, Buturovic L, Leahy D, Thomas S Cross-validation pitfalls when selecting and assessing regression and classification models. J Cheminform 6: CDC Reported Tuberculosis in the United States, U.S. Department of Health and Human Services, CDC, Atlanta, GA Huang DW, Sherman BT, Lempicki RA Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protocols 4:

106 103. Barry CE, Boshoff HI, Dartois V, Dick T, Ehrt S, Flynn J, Schnappinger D, Wilkinson RJ, Young D The spectrum of latent tuberculosis: rethinking the biology and intervention strategies. Nat Rev Microbiol 7: Chaussabel D, Pulendran B A vision and a prescription for big data-enabled medicine. Nat Immunol 16: Davies GR, Brindle R, Khoo SH, Aarons LJ Use of nonlinear mixed-effects analysis for improved precision of early pharmacodynamic measures in tuberculosis treatment. Antimicrob Agents Chemother 50: Blumberg HM, Burman WJ, Chaisson RE, Daley CL, Etkind SC, Friedman LN, Fujiwara P, Grzemska M, Hopewell PC, Iseman MD, Jasmer RM, Koppaka V, Menzies RI, O'Brien RJ, Reves RR, Reichman LB, Simone PM, Starke JR, Vernon AA, American Thoracic Society CfDCaPatIDS American Thoracic Society/Centers for Disease Control and Prevention/Infectious Diseases Society of America: Treatment of tuberculosis. Am J Resp Crit Care 167: Gillespie SH, Crook AM, McHugh TD, Mendel CM, Meredith SK, Murray SR, Pappas F, Phillips PPJ, Nunn AJ Four-month moxifloxacin-based regimens for drug-sensitive tuberculosis. N Engl J Med 371: Balaban NQ Persistence: mechanisms for triggering and enhancing phenotypic variability. Curr Opin Genet Dev 21: Gefen O, Balaban NQ The importance of being persistent: heterogeneity of bacterial populations under antibiotic stress. FEMS Microbiol Rev 33: Betts JC, Lukey PT, Robb LC, McAdam RA, Duncan K Evaluation of a nutrient starvation model of Mycobacterium tuberculosis persistence by gene and protein expression profiling. Mol Microbiol 43: Rustad TR, Minch KJ, Brabant W, Winkler JK, Reiss DJ, Baliga NS, Sherman DR Global analysis of mrna stability in Mycobacterium tuberculosis. Nucleic Acids Res 41: Galagan JE, Minch K, Peterson M, Lyubetskaya A, Azizi E, Sweet L, Gomes A, Rustad T, Dolganov G, Glotova I, Abeel T, Mahwinney C, Kennedy AD, Allard R, Brabant W, Krueger A, Jaini S, Honda B, Yu W-H, Hickey MJ, Zucker J, Garay C, Weiner B, Sisk P, Stolte C, Winkler JK, Van de Peer Y, Iazzetti P, Camacho D, Dreyfuss J, Liu Y, Dorhoi A, Mollenkopf H-J, Drogaris P, Lamontagne J, Zhou Y, Piquenot J, Park ST, Raman S, Kaufmann SHE, Mohney RP, Chelsky D, Moody DB, Sherman DR, Schoolnik GK The Mycobacterium tuberculosis regulatory network and hypoxia. Nature 499:

107 113. Garcia B, Walter ND, Dolganov G, Coram M, Davis JL, Schoolnik GK, Strong M A minimum variance method for genome-wide data-driven normalization of qrt-pcr expression data Anal Biochem 458: Benjamini Y, Hochberg Y Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Statist Soc B 57: Hosack D, Dennis G, Sherman B, Lane H, Lempicki R Identifying biological themes within lists of genes with EASE. Genome Biol 4:R Hellyer TJ, DesJardin LE, Hehman GL, Cave MD, Eisenach KD Quantitative analysis of mrna as a marker for viability of Mycobacterium tuberculosis. J Clin Microbiol 37: Desjardin L, Perkins M, Wolski K, Haun S, Teixeira L, Chen Y, Johnson J, Ellner J, Dietze R, Bates J, Donald Cave M, Eisenach K Measurement of sputum Mycobacterium tuberculosis messenger RNA as a surrogate for response to chemotherapy. Am J Respir Crit Care Med 160: Li L, Mahan CS, Palaci M, Horter L, Loeffelholz L, Johnson JL, Dietze R, Debanne SM, Joloba ML, Okwera A, Boom WH, Eisenach KD Sputum Mycobacterium tuberculosis mrna as a marker of bacteriologic clearance in response to antituberculosis therapy. J Clin Microbiol 48: Honeyborne I, McHugh TD, Phillips PPJ, Bannoo S, Bateson A, Carroll N, Perrin FM, Ronacher K, Wright L, van Helden PD, Walzl G, Gillespie SH Molecular bacterial load assay, a culture-free biomarker for rapid and accurate quantification of sputum Mycobacterium tuberculosis bacillary load during treatment. J Clin Microbiol 49: Dahl JL, Kraus CN, Boshoff HIM, Doan B, Foley K, Avarbock D, Kaplan G, Mizrahi V, Rubin H, Barry CE The role of RelMtb-mediated adaptation to stationary phase in long-term persistence of Mycobacterium tuberculosis in mice. Proc Nat Acad Sci USA 100: Wilson M, DeRisi J, Kristensen H-H, Imboden P, Rane S, Brown PO, Schoolnik GK Exploring drug-induced alterations in gene expression in Mycobacterium tuberculosis by microarray hybridization. Proc Natl Acad Sci U S A 96: Betts JC, McLaren A, Lennon MG, Kelly FM, Lukey PT, Blakemore SJ, Duncan K Signature gene expression profiles discriminate between isoniazid-, thiolactomycin-, and triclosan-treated Mycobacterium tuberculosis. Antimicrob Agents Chemother 47: Fu LM Exploring drug action on Mycobacterium tuberculosis using Affymetrix oligonucleotide genechips. Tuberculosis 86:

108 124. Zhang Y, Jacobs Jr WR Mechanisms of drug action, drug resistance and drug tolerance in Mycobacterium tuberculosis: Expected phenotypes from evolutionary pressures from a highly sucessful pathogen, p In Kaufmann SHE, Rubin E (ed), Handbook of tuberculosis: Molecular biology and biochemistry, vol I. WILEY-VCH, Weinheim Chigutsa E, Patel K, Denti P, Visser M, Maartens G, Kirkpatrick CMJ, McIlleron H, Karlsson MO A time-to-event pharmacodynamic model describing treatment response in patients with pulmonary tuberculosis using days to positivity in automated liquid mycobacterial culture. Antimicrob Agents Chemother 57: Boshoff HIM, Barry CE Tuberculosis metabolism and respiration in the absence of growth. Nat Rev Micro 3: Ramage HR, Connolly LE, Cox JS Comprehensive functional analysis of Mycobacterium tuberculosis toxin-antitoxin systems: Implications for pathogenesis, stress responses, and evolution. PLoS Genet 5:e Fasani RA, Savageau MA Molecular mechanisms of multiple toxin antitoxin systems are coordinated to govern the persister phenotype. Proc Nat Acad Sci USA 110:E2528-E Adams Kristin N, Takaki K, Connolly Lynn E, Wiedenhoft H, Winglee K, Humbert O, Edelstein Paul H, Cosma Christine L, Ramakrishnan L Drug tolerance in replicating mycobacteria mediated by a macrophage-induced efflux mechanism. Cell 145: Gupta AK, Chauhan DS, Srivastava K, Das R, Batra S, Mittal M, Goswami P, Singhal N, Sharma VD, Venkatesan K, Hasnain SE, Katoch VM Estimation of efflux mediated multi-drug resistance and its correlation with expression levels of two major efflux pumps in mycobacteria. J Commun Dis 38: Louw GE, Warren RM, Gey van Pittius NC, McEvoy CRE, Van Helden PD, Victor TC A balancing act: Efflux/influx in mycobacterial drug resistance. Antimicrob Agents Chemother 53: Niki M, Niki M, Tateishi Y, Ozeki Y, Kirikae T, Lewin A, Inoue Y, Matsumoto M, Dahl JL, Ogura H, Kobayashi K, Matsumoto S A novel mechanism of growth phase-dependent tolerance to isoniazid in mycobacteria. J Biol Chem 287: Tupin A, Gualtieri M, Roquet-Baneres F, Morichaud Z, Brodolin K, Leonetti J- P Resistance to rifampin: At the crossroads between ecological, genomic and medical concerns. Int J Anntimicrob Agents 35:

109 134. Nahid P, Saukkonen J, Kenzie WRM, Johnson JL, Phillips PPJ, Andersen J, Bliven-Sizemore E, Belisle JT, Boom WH, Luetkemeyer A, Campbell TB, Eisenach KD, Hafner R, Lennox JL, Makhene M, Swindells S, Villarino ME, Weiner M, Benson C, Burman W Tuberculosis biomarker and surrogate endpoint research roadmap. Am J Resp Crit Care 184: Mukamolova GV, Turapov O, Malkin J, Woltmann G, Barer MR Resuscitation-promoting factors reveal an occult population of tubercle bacilli in sputum. Am J Resp Crit Care 181: de Steenwinkel JEM, Aarnoutse RE, de Knegt GJ, ten Kate MT, Teulen M, Verbrugh HA, Boeree MJ, van Soolingen D, Bakker-Woudenberg IAJM Optimization of the rifampin dosage to improve the therapeutic efficacy in tuberculosis treatment using a murine model. Am J Resp Crit Care 187: Lukoye D, Cobelens FGJ, Ezati N, Kirimunda S, Adatu FE, Lule JK, Nuwaha F, Joloba ML Rates of anti-tuberculosis drug resistance in Kampala-Uganda are low and not associated with HIV infection. PLoS ONE 6:e Duque-Correa MA, Kühl AA, Rodriguez PC, Zedler U, Schommer-Leitner S, Rao M, Weiner J, Hurwitz R, Qualls JE, Kosmiadi GA, Murray PJ, Kaufmann SHE, Reece ST Macrophage arginase-1 controls bacterial growth and pathology in hypoxic tuberculosis granulomas. Proc Nat Acad Sci USA 111:E4024- E Mattila JT, Ojo OO, Kepka-Lenhart D, Marino S, Kim JH, Eum SY, Via LE, Barry CE, Klein E, Kirschner DE, Morris SM, Lin PL, Flynn JL Microenvironments in tuberculous granulomas are delineated by distinct populations of macrophage subsets and expression of nitric oxide synthase and arginase isoforms. J Immunol 191: Marino S, Cilfone NA, Mattila JT, Linderman JJ, Flynn JL, Kirschner DE Macrophage polarization drives granuloma outcome during Mycobacterium tuberculosis infection. Infect Immun 83: de Noronha ALL, Báfica A, Nogueira L, Barral A, Barral-Netto M Lung granulomas from Mycobacterium tuberculosis/hiv-1 co-infected patients display decreased in situ TNF production. Pathol Res Pract 204: Silva AA, Mauas T, Saldiva PHN, Pires-Neto RC, Coletta RD, Graner E, Vargas PA Immunophenotype of lung granulomas in HIV and non-hiv associated tuberculosis. MedialExpress 1: Müller H, Krüger S Immunohistochemical analysis of cell composition and in situ cytokine expression in HIV- and non-hiv-associated tuberculous lymphadenitis. Immunobiology 191:

110 144. Shen J-Y, Barnes PF, Rea TH, Meyer PR Immunohistology of tuberculosis adenitis in symptomatic HIV infection. Clin Exp Immunol 72: Di Perri G, Cazzadori A, Vento S, Bonora S, Malena M, Bontempini L, Lanzafame M, Allegranzi B, Concia E Comparative histopathological study of pulmonary tuberculosis in human immunodeficiency virus-infected and noninfected patients. Tuber Lung Dis 77: Mehra S, Foreman TW, Didier PJ, Ahsan MH, Hudock TA, Kissee R, Golden NA, Gautam US, Johnson A-M, Alvarez X, Russell-Lodrigue KE, Doyle LA, Roy CJ, Blanchard JL, Khader SA, Lackner AA, Sherman DA, Kaushal D The DosR regulon modulate adaptive immunity and is essential for M tuberculosis persistence. Am J Resp Crit Care EPUB Via LE, Lin PL, Ray SM, Carrillo J, Allen SS, Eum SY, Taylor K, Klein E, Manjunatha U, Gonzales J, Lee EG, Park SK, Raleigh JA, Cho SN, McMurray DN, Flynn JL, Barry CE Tuberculous granulomas are hypoxic in guinea pigs, rabbits, and nonhuman primates. Infect Immun 76: Converse PJ, Karakousis PC, Klinkenberg LG, Kesavan AK, Ly LH, Allen SS, Grosset JH, Jain SK, Lamichhane G, Manabe YC, McMurray DN, Nuermberger EL, Bishai WR Role of the dosr-doss two-component regulatory system in Mycobacterium tuberculosis virulence in three animal models. Infect Immun 77: Aly S, Wagner K, Keller C, Malm S, Malzan A, Brandau S, Bange FC, Ehlers S Oxygen status of lung granulomas in Mycobacterium tuberculosis-infected mice. J Pathol 210: Voskuil MI, Schnappinger D, Visconti KC, Harrell MI, Dolganov GM, Sherman DR, Schoolnik GK Inhibition of respiration by nitric oxide induces a Mycobacterium tuberculosis dormancy program. J Exp Med 198: Leyten EMS, Lin MY, Franken KLMC, Friggen AH, Prins C, van Meijgaarden KE, Voskuil MI, Weldingh K, Andersen P, Schoolnik GK, Arend SM, Ottenhoff THM, Klein MR Human T-cell responses to 25 novel antigens encoded by genes of the dormancy regulon of Mycobacterium tuberculosis. Microbes Infect 8: Schnappinger D, Schoolnik GK, Ehrt S Expression profiling of host pathogen interactions: how Mycobacterium tuberculosis and the macrophage adapt to one another. Microbes Infect 8: Talaat AM, Lyons R, Howard ST, Johnston SA The temporal expression profile of Mycobacterium tuberculosis infection in mice. Proc Nat Acad Sci USA 101:

111 154. Faucher SP, Porwollik S, Dozois CM, McClelland M, Daigle F Transcriptome of Salmonella enterica serovar Typhi within macrophages revealed through the selective capture of transcribed sequences. Proceedings of the National Academy of Sciences of the United States of America 103: Tailleux L, Waddell SJ, Pelizzola M, Mortellaro A, Withers M, Tanne A, Castagnoli PR, Gicquel B, Stoker NG, Butcher PD, Foti M, Neyrolles O Probing host pathogen cross-talk by transcriptional profiling of both Mycobacterium tuberculosis and infected human dendritic cells and macrophages. PLoS One 3:e Walter ND, Dolganov GM, Garcia BJ, Worodria W, Andama A, Musisi E, Ayakaka I, Van TT, Voskuil MI, de Jong BC, Davidson RM, Fingerlin TE, Kechris KJ, Palmer C, Nahid P, Daley CL, Geraci M, Huang L, Cattamanchi A, Strong M, Schoolnik GK, Davis JL Transcriptional adaptation of drugtolerant Mycobacterium tuberculosis during treatment of human tuberculosis. J Infect Dis IN PRESS Honaker RW, Dhiman RK, Narayanasamy P, Crick DC, Voskuil MI DosS responds to a reduced electron transport system to induce the Mycobacterium tuberculosis DosR regulon. J Bacteriol 192: Voskuil MI, Schlesinger LS Toward resolving the paradox of the critical role of the DosR regulon in Mycobacterium tuberculosis persistence and active disease. Am J Resp Crit Care 191: Condos R, Rom WN, Weiden M Lung-specific immune response in tuberculosis. Int J Tuberc Lung Dis 4:S11-S Akridge RE, Oyafuso LK, Reed SG IL-10 is induced during HIV-1 infection and is capable of decreasing viral replication in human macrophages. J Immunol 153: Patel NR, Swan K, Li X, Tachado SD, Koziel H Impaired M. tuberculosismediated apoptosis in alveolar macrophages from HIV+ persons: potential role of IL- 10 and BCL-3. J Leukoc Biol 86: Boussiotis VA, Tsai EY, Yunis EJ, Thim S, Delgado JC, Dascher CC, Berezovskaya A, Rousset D, Reynes J-M, Goldfeld AE IL-10 producing T cells suppress immune responses in anergic tuberculosis patients. J Clin Invest 105: Flynn JL, Chan J, Triebold KJ, Dalton DK, Stewart TA, Bloom BR An essential role for interferon gamma in resistance to Mycobacterium tuberculosis infection. J Exp Med 178:

112 164. Murray PJ, Young RA Increased antimycobacterial immunity in Interleukin- 10-deficient mice. Infect Immun 67: Schreiber T, Ehlers S, Heitmann L, Rausch A, Mages J, Murray PJ, Lang R, Hölscher C Autocrine IL-10 induces hallmarks of alternative activation in macrophages and suppresses antituberculosis effector mechanisms without compromising T cell immunity. J Immunol 183: El Kasmi KC, Qualls JE, Pesce JT, Smith AM, Thompson RW, Henao-Tamayo M, Basaraba RJ, Konig T, Schleicher U, Koo M-S, Kaplan G, Fitzgerald KA, Tuomanen EI, Orme IM, Kanneganti T-D, Bogdan C, Wynn TA, Murray PJ Toll-like receptor-induced arginase 1 in macrophages thwarts effective immunity against intracellular pathogens. Nat Immunol 9: Law KF, Jagirdar J, Weiden MD, Bodkin M, Rom WN Tuberculosis in HIV-positive patients: cellular response and immune activation in the lung. Am J Resp Crit Care 153: Cyktor JC, Carruthers B, Kominsky RA, Beamer GL, Stromberg P, Turner J IL-10 inhibits mature fibrotic granuloma formation during Mycobacterium tuberculosis infection. J Immunol 190: Black GF, Thiel BA, Ota MO, Parida SK, Adegbola R, Boom WH, Dockrell HM, Franken KLMC, Friggen AH, Hill PC, Klein MR, Lalor MK, Mayanja H, Schoolnik G, Stanley K, Weldingh K, Kaufmann SHE, Walzl G, Ottenhoff THM, Consortium atgbft Immunogenicity of novel DosR regulon-encoded candidate antigens of Mycobacterium tuberculosis in three high-burden populations in Africa. Clin Vaccine Immunol 16: Bartek IL, Rutherford R, Gruppo V, Morton RA, Morris RP, Klein MR, Visconti KC, Ryan GJ, Schoolnik GK, Lenaerts A, Voskuil MI The DosR regulon of M. tuberculosis and antibacterial tolerance. Tuberculosis 89: Irwin SM, Gruppo V, Brooks E, Gilliland J, Scherman M, Reichlen MJ, Leistikow R, Kramnik I, Nuermberger EL, Voskuil MI, Lenaerts AJ Limited activity of clofazimine as a single drug in a mouse model of tuberculosis exhibiting caseous necrotic granulomas. Antimicrob Agents Chemother 58: Harper J, Skerry C, Davis SL, Tasneen R, Weir M, Kramnik I, Bishai WR, Pomper MG, Nuermberger EL, Jain SK Mouse model of necrotic tuberculosis granulomas develops hypoxic lesions. J Infect Dis 205: CDC A strategic plan for the elimination of tuberculosis in the United States MMWR 38(S-3):

113 174. Brown A, Stepler R Statistical portrait of the foreign-born population in the United States, , on Pew Research Center. Accessed 175. Fujiwara PI Tide pools: what will be left after the tide has turned? Int J Tuberc Lung Dis 4(12 Suppl 2):S Aiona K, Lowenthal P, Painter JA, Reves R, Flood J, Parker M, Fu Y, Wall K, Walter ND Transnational record linkage for tuberculosis surveillance and program evaluation. Public Health Rep 130: Simon R, Korn EL, McShane LM, Radmacher MD, Wright GW, Zhao Y Design and analysis of DNA microarray investigations. Springer, New York Le Dell E, Petersen M, Laan M, Van Der MJ Computationally efficient confidence intervals for cross-validated area under the ROC curve estimates. UC Berkeley DivBiostat Pap Ser Chen C, Liaw A, Breiman L Using random forest to learn imbalanced data. University of California Berkeley, 180. Cattamanchi A, Davis JL, Worodria W, Den Boon S, Yoo S, Matovou J, Kiidha J, Nankya F, Kyeyune R, Byanyima P, Andama A, Joloba M, Osmond DH, Hopewell PC, Huang L Sensitivity and specificity of fluorescence microscopy for diagnosing pulmonary tuberculosis in a high HIV prevalence setting. Int J Tuberc Lung Dis 13: Cole ST, Brosch R, Parkhill J, Garnier T, Churcher C, Harris D, Gordon SV, Eiglmeier K, Gas S, Barry CE, Tekaia F, Badcock K, Basham D, Brown D, Chillingworth T, Connor R, Davies R, Devlin K, Feltwell T, Gentles S, Hamlin N, Holroyd S, Hornsby T, Jagels K, Krogh A, McLean J, Moule S, Murphy L, Oliver K, Osborne J, Quail MA, Rajandream MA, Rogers J, Rutter S, Seeger K, Skelton J, Squares R, Squares S, Sulston JE, Taylor K, Whitehead S, Barrell BG Deciphering the biology of Mycobacterium tuberculosis from the complete genome sequence. Nature 393: Lew JM, Kapopoulou A, Jones LM, Cole ST TubercuList 10 years after. Tuberculosis 91: Shao Y, Harrison EM, Bi D, Tai C, He X, Ou H-Y, Rajakumar K, Deng Z TADB: a web-based resource for Type 2 toxin antitoxin loci in bacteria and archaea. Nucleic Acids Res 39:D606-D

114 184. Voskuil MI, Bartek I, Visconti K, Schoolnik GK The response of Mycobacterium tuberculosis to reactive oxygen and nitrogen species. Front Microbiol Rustad TR, Harrell MI, Liao R, Sherman DR The enduring hypoxic response of Mycobacterium tuberculosis. PLoS ONE 3:e Quadri LEN Biosynthesis of mycobacterial lipids by polyketide synthases and beyond. Crit Rev Biochem Mol Biol 49:

115 APPENDIX A SUPPLEMENTAL METHODS AND RESULTS FOR CHAPTER 3 Blood transcriptional biomarkers for active TB among US patients: Case-control study with systematic cross-classifier comparison epat Study Sites Consecutive subjects meeting criteria for TB and LTBI groups were enrolled in outpatient TB control programs in Colorado [Denver Metro Tuberculosis Clinic, Denver] and Texas [Fairpark Clinic (Harlingen), Brownsville Clinic (Brownsville) of Cameron County Health Department; Edinburg Pulmonary Clinic (Edinburg) of Hidalgo County Health Department; San Antonio Metropolitan Health District (San Antonio); Audie L Murphy Veterans Administration Medical Center during Consecutive subjects meeting the community-acquired pneumonia case definition were enrolled in the Denver Health Medical Center Emergency Department. Classifier development and evaluation Cross-validation for model tuning epat transcriptional classifiers were support vector machines (SVM). For each classifier type, 20 iterations of 6-fold cross-validation were performed using training set samples to select model parameters. The minimal number of transcripts (features) required was additionally treated as a parameter in model selection. Minimal number of features for each classifier As required for unbiased cross validation, feature selection was repeated de novo with each fold of cross-validation (177). For each fold, the predictive value of each transcript was quantified via Support Vector Machine with Recursive Feature Elimination (SVM- 108

116 RFE)(98). SVM-RFE began by training an SVM classifier [e1071 package, version using default settings: cost = 1, linear kernel] using all 21,734 probesets. The weight of each transcript s contribution to the SVM classifier was ranked and the least-predictive 20% of transcripts were removed. A new SVM was trained using remaining transcripts and the process was iterated until all transcripts were removed. The result was a rank list of all transcripts in order of predictive value for each fold. We identified the minimal number of transcripts required for each classifier as follows. Multiple SVM models including between the top 5 and the top 200 transcripts from the SVM-RFE rank lists were evaluated. The misclassification error in 20 iterations of 6-fold cross-validation was quantified. The figure below illustrates the misclassification rate according to the number of transcripts included in the classifier for the type 1 (TB vs pneumonia), type 2 (TB vs LTBI), and type 3 (TB vs LTBI or pneumonia) classifiers. We used an established backwards selection approach to identify the point at which removing additional transcripts substantially worsened classification (99, 100). Starting with the largest classifier (200 transcripts), we evaluated successively smaller classifiers until we identified the first point at which elimination of a single transcript resulted in a misclassification rate greater than the misclassification rate of the preceding classifier plus one standard error. Based on this analysis, the TB vs pneumonia classifier included 47, the TB vs LTBI included 51 transcripts, and the TB versus LTBI or pneumonia included 119. Importantly, this process identified the number of transcripts that would be included in each classifier. The specific transcripts included were selected as detailed in the Final transcript sets for each classifier section below. 109

117 Figure A1. Misclassification rate in cross-validation according to number of transcripts included in classifier. Results are shown for three classifier types: (A) TB vs. pneumonia, (B) TB vs. LTBI and (C) TB vs. pneumonia or LTBI. Points represent the proportion of samples misclassified in classifiers including a given number of transcripts. Dotted bands represent one standard error around the misclassification rate. The vertical dotted line indicates the point at which elimination of a single transcript resulted in a misclassification rate that exceeded the misclassification rate of the next largest classifier plus one standard error. Selection of SVM parameters Model parameters were optimized in the training set using the tune.svm function in the R package e1071, version This function performs a grid search over a range of parameters to identify the set of parameter values that minimize misclassification error. The 110

118 classification threshold was adjusted such that sensitivity and specificity were approximately equal in cross-validation. Final parameters are listed below. Final SVM parameters Weighting Kernel Cost Type 1: TB vs pneumonia None Linear Type 2: TB vs LTBI None Linear Type 3: TB vs LTBI or pneumonia 2:1 Linear Estimation of accuracy in training set After selection of model parameters, classification accuracy in the training set was estimated as follows. Six-fold cross-validation was repeated 20 times and the mean sensitivity and specificity were calculated. After confirming that sensitivity and specificity estimates from 20 iterations of cross-validation were approximately normally distributed, 95% two-sided t 19, confidence intervals were estimated. Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC) estimates were generated for Training Set Samples using the cvauc package, version for each of the 20 iterations (178). Thus, 20 cross-validated ROCs and the corresponding AUCs were produced to assess the overall accuracy. The mean and standard error of the 20 cross-validated AUCs were calculated, and the lower and upper 95% confidence limits were determined by subtracting and adding, respectively, t 19, standard errors from the mean AUC. Final transcript sets for each classifier The number of transcripts to be included in each classifier was identified through cross-validation as described above. To identify the specific transcripts that would be included in each classifier, SVM-RFE was implemented with all training set samples without cross-validation. The top 47, 51 and 119 transcripts were used for the epat type 1, 2, and 3 classifiers, respectively. 111

119 Evaluation in epat test set The three transcriptional classifiers developed above were used to predict the disease state of test set samples without modification of model parameters. Confidence intervals for sensitivity and specificity were estimated using the exact binomial methods implemented in epir, version ROC curves were produced and AUC calculated using the proc package in R (version 1.7.3). External validation of epat classifiers To validate our classifiers in an independent external population, we utilized publically-available transcriptional data available on NCBI Gene Expression Omnibus [GSE37250] from Kaforou, (19) the largest existing whole blood transcriptional classifier study of adults with TB. To accommodate differences in microarray platform, we converted our Affymetrix HuGene ST 1.1 transcript cluster identifiers for transcripts in epat classifiers to Illumina HumanHT12v4 identifiers. For the transcript identifiers currently annotated with a gene symbol, we identified Illumina identifiers with matching gene symbols. Where the original Affymetrix gene symbol did not match current Illumina gene symbols, we manually searched NCBI and other databases for synonymous gene symbols that matched. Following platform conversion, we tested epat classifiers in the Kaforou dataset with 6-fold cross-validation using an SVM with the final parameters described above. The epat type 1 classifier was tested in Kaforou subjects with TB or other diseases. The epat type 2 classifier was tested in Kaforou subjects with TB or LTBI. The epat type 3 classifier was tested in Kaforou subjects with TB or LTBI or other diseases. epat classifiers were tested separately in Kaforou subjects with and without HIV infection. Cross-validation was 112

120 repeated 20 times and sensitivity, specificity and AUC were estimated using methods identical to those described for the training set above after assessing normality Systematic review of published literature Review methods Selection Criteria: We included all studies which developed expression classifiers distinguishing active pulmonary TB from other conditions based on whole transcriptome microarray analysis of human whole blood using SML algorithms. We excluded the following types of studies: (1) studies which primarily evaluated expression change in response to TB treatment, (2) studies of transcriptional differences between patients with first-time and recurrent TB, (3) studies focused on non-tuberculosis mycobacteria or extrapulmonary tuberculosis, (4) studies of specimens other than whole blood, (5) abstracts, case reports, editorials, commentaries, and reviews. We maintained a list of excluded studies and the reasons for their exclusion. Two reviewers independently screened the accumulated citations for relevance and reviewed full-text articles using pre-specified eligibility criteria; they resolved disagreements about study selection by consensus. 113

121 Database Search Terms a PubMed b (((((((tuberculosis[title] OR TB[Title]) AND blood[title/abstract]) AND ("2005/01/01"[PDAT] : "3000"[PDAT])) NOT pbmc[title/abstract]) AND expression[title/abstract]) NOT ("vaccination"[mesh Terms] OR "vaccination"[all Fields])) NOT mononuclear[all Fields]) AND "humans"[mesh Terms] a For papers published from January 1, 2005 to the present. b Results were further refined in EndNote, searching for RNA Microarray Studies among imported PDFs. Evaluation of previously published classifiers We evaluated whether previously published transcriptional classifiers identified during systematic literature review accurately classify epat samples. First, we converted transcript identifiers from the original microarray platforms to Affymetrix HuGene ST 1.1 identifiers. Second, we used the same classification methods described in previous publications to test the classification accuracy of each previously published classifier in epat data. These steps are described in detail below. Transcript conversion to Affymetrix identifiers We abstracted lists of transcript identifiers included in previously published classifiers from the original manuscripts and converted these identifiers to Affymetrix HuGene ST 1.1 transcript cluster IDs. For the transcript identifiers currently annotated with a gene symbol, we identified Affymetrix IDs with matching gene symbols. Where the original gene symbol did not match current Affymetrix gene symbols, we manually searched NCBI and other databases for synonymous gene symbols that matched. Most published classifiers included a number of transcript identifiers that lack current biological annotation. These represent microarray probesets designed for a target that has not 114

122 subsequently been demonstrated to exist. Since they do not correspond to any known transcript, these non-annotated identifiers could not be converted to Affymetrix IDs. Classification using Kaforou classifiers We evaluated the classification accuracy of the Kaforou (19) classifiers in epat training and test sets. The Kaforou Type 1 TB vs other diseases classifier included 26 transcript identifiers after conversion to Affymetrix and was tested in epat TB and pneumonia samples. The Kaforou Type 2 TB vs LTBI classifier included 42 transcript identifiers after conversion to Affymetrix and was tested in epat TB and LTBI samples. The Kaforou Type 3 TB vs LTBI or other diseases classifier included 44 transcript identifiers after conversion to Affymetrix and was tested in epat TB and LTBI or pneumonia samples. The Disease Risk Score (DRS) developed and implemented by Kaforou was used for classification. Sensitivity, specificity and AUC were estimated based on 20 iterations of 6- fold cross-validation using methods identical to those described for the training set. Classification using Bloom classifier We evaluated the classification accuracy of the Bloom (13) classifier in our training and test sets using only the 117 transcript identifiers converted from the Bloom list. A linear SVM (the same method used by Bloom) was implemented with cost = 0.01 and 2:1 weighting to adjust for imbalanced groups. Sensitivity, specificity and AUC were estimated based on 20 iterations of 6-fold cross-validation using methods identical to those described for the training set. Classification using Berry classifiers We evaluated the classification accuracy of the Berry (15) classifiers in epat training and test sets. The Berry Type 1 TB vs other diseases classifier included 317 transcript 115

123 identifiers after conversion to Affymetrix and was tested in epat TB and pneumonia samples. The Berry Type 2 TB vs LTBI classifier included 80 transcript identifiers after conversion to Affymetrix and was tested in epat TB and LTBI samples. K-nearest neighbors (the same method used by Berry) was implemented with K=3 using the class package in R. Sensitivity, specificity and AUC were estimated based on 20 iterations of 6- fold cross-validation using methods identical to those described for the training set. Classification using Maertzdorf classifier We evaluated the classification accuracy of the Maertzdorf (16) classifier in our training and test sets using only the 70 transcript identifiers converted from the Maertzdorf list. Random forest (the same method used by Maertzdorf) was implemented using the randomforest package (version ) in R with ntree=1000, mtry =5. To adjust imbalance in group sizes, we down-sampled the larger group during training of the random forest model as recommended by random forest developers (179). Sensitivity, specificity and AUC were estimated based on 20 iterations of 6-fold cross-validation using methods identical to those described for the training set. 116

124 117

125 118

126 119

127 120

128 121

129 APPENDIX B SUPPLEMENTAL METHODS AND RESULTS FOR CHAPTER 4 Transcriptional adaptation of drug-tolerant Mycobacterium tuberculosis during human tuberculosis Enrolment and clinical evaluation From a consecutive random sample of adults (age 18 years) with cough of at least 2 weeks admitted to Mulago National Referral Hospital and enrolled in a parent TB diagnostic study, (180) we recruited patients with sputum AFB smear-positive, culture-positive pulmonary TB. We excluded from the eligible population any patient who was unable to expectorate sputum, already receiving anti-tb therapy, infected with drug-resistant MTB, too ill to walk, severely anemic, residing more than 20 kilometers from the study site, or lacking a treatment supporter. (Figure B1, Enrollment flow diagram) Resource constraints prevented us from performing gene expression profiling for all enrolled patients. Therefore, of patients who were enrolled and provided specimens, 17 were randomly selected with stratification by HIV status for the current gene expression profiling. 122

130 Figure B1. Enrolment flow diagram At enrollment, we collected demographic and clinical information using a standardized questionnaire. All patients underwent HIV antibody testing and blood CD4+ T- lymphocyte measurement if HIV-seropositive. All patients provided one specimen for molecular testing on direct sputum (GeneXpert MTB/RIF, Cepheid Diagnostics, Sunnyvale, CA), and two specimens for smear microscopy and mycobacterial culture. Adherence to TB therapy was assessed at each follow-up visit. All patients were followed for up to 2 years after their last treatment visit to adjudicate treatment outcomes. Human subjects approval Institutional Review Boards at Makerere University, Mulago Hospital, the Uganda National Council for Science and Technology, the University of California San Francisco, and the University of Colorado Denver approved this study. All participants provided written informed consent. 123

131 124

132 Laboratory methods Extraction of total RNA Total RNA was extracted using a glass matrix tube for cell lysis (Lysing Matrix B, Q Biogene) in a FastPrep FP120 instrument (Bio 101, Thermo Savant) with a speed setting of 6 for 3 x 30 seconds, with cooling on ice for 1 minute in between. After processing, chloroform was added twice, followed by separation of the aqueous and organic layers. The aqueous phase containing the RNA was removed to a fresh tube and the RNA was precipitated overnight at -20C after addition of glycoblue (Ambion), 0.1 volume of 5 M ammonium acetate, and an equal volume of isopropanol were added. The resulting RNA pellet was cleaned by three rounds of the RNeasy mini column system, interspersed with off-column DNAse treatments (Promega RQ1 DNAse). Extraction of genomic DNA Genomic DNA was extracted along with total RNA for each sample. The interphase containing the DNA and organic layers were removed to a fresh tube and a volume of guanidine thiocyanate-based buffer was added (equivalent to 0.5 volume of Trizol used in the extraction ). Samples were mixed for 10 minutes. then centrifuged at 12,000 x g for 30 minutes. at room temperature. The aqueous phase now containing the DNA was removed to a fresh tube. Glycoblue and 0.4 volume of isopronal were added. Samples were mixed by inversion and precipated for 10 minuntes. at room temperature. The resulting DNA pellet was washed twice with 70% ethanol and resuspended in nuclease-free water. qrt-pcr expression profiling Multiplex qrt-pcr was used to selectively amplify and quantify expression of 2,411MTB transcripts. This method, described in detail elsewhere, (112) is capable of 125

133 profiling low abundance MTB mrna in a mixed sample (predominantly of human mrna). Briefly, 2,411 MTB transcripts underwent controlled PCR pre-amplification using outflanking forward and reverse primers following first strand cdna synthesis. Following amplification, expression of these 2,411 MTB transcripts was quantified via 384-well format multiplex qrt-pcr with a LightCycler 480 (Roche, Basel, Switzerland). Sequences, design and validation details of PCR primer/probe sets can be found at and Statistical analysis Estimation of change in nucleic acid abundance We used the following measures of nucleic acid abundance. mrna No single mrna transcript or group of transcripts has been demonstrated to be consistently expressed in sputum during treatment. Therefore, to obtain an aggregate measure of MTB mrna abundance in each sample, we calculated the median cycle threshold (CT) value of the 100 most highly expressed transcripts in each sample. For each sample, we then quantified the percentage decrease in this value relative to the patient s pre-treatment (day 0) value. 16S rrna ribosomal gene copy number (RGCN) in each sample was estimated relative to genomic standards (10 5 or 10 6 genomes H37Rv). For each sample, RGCN was then transformed to a percentage decrease from pre-treatment RCGN. DNA genome equivalents (GE) in each sample were estimated relative to a genomic standard (10 4 genomes H37Rv). For each sample, GE was then transformed to a percentage decrease from pre-treatment GE. 126

134 To characterize average changes in relative abundance of MTB mrna, 16S rrna and DNA over time, we calculated geometric mean percentage changes from day 0 to days 2, 4, 7, 14, 28 and 56. Geometric rather than arithmetic means were used because they are less sensitive to right skewness. A saturated linear mixed model was used to obtain 95% confidence intervals for the geometric means. Models were estimating using Stata Version 13.1 (Stata Corp, College Station, TX). Data normalization Because there are no validated endogenous control (housekeeping) genes, we applied a minimum variance, data-driven normalization method described elsewhere (113) to adjust for the progressive decline in the sputum bacillary load that occurs with ongoing treatment. This method is more robust to non-detected probes than other data-driven normalization methods such as quantile normalization, (113) enabling normalization of samples with up to 50% non-detected probes. Most samples from days 28 and 56 had very low nucleic acid abundance and >50% of probes were non-detectable; data from these time-points were therefore not normalized and were analyzed separately. 127

135 128

136 129