Tuberculosis (TB) is an ancient disease caused by the airborne pathogen Mycobacterium tuberculosis (Mtb), which has infected billions of people and has killed more people than any other bacterial infectious agent (1). Current tools to combat TB, including diagnostics, vaccines and drugs, each for their own reasons, are insufficient to diminish the global TB burden. In response to this WHO has formulated consensus research priorities to accelerate the discovery and development of better TB drugs, diagnostics and vaccines (1, 2).

The current clinical TB vaccine pipeline includes more than 20 new candidates, which in pre-clinical animal models are able to reduce Mtb infection in the lung (quantified by measuring numbers of viable Mtb and/or pathology in the lung) and/or systemically (often the spleen), or TB morbidity (quantified by tissue pathology, weight loss or host survival) (1, 3). Recent human phase II studies showed that sustained Mtb infection can be prevented by BCG revaccination of Mtb uninfected individuals (as measured by interferon gamma release assay conversion), and that incidence of new TB cases can be reduced amongst persons with LTBI by the M72 TB subunit vaccine (4, 5). While these results are highly encouraging, the immune mechanisms underlying these protective effects remain poorly understood. Current understanding of the molecular and cellular basis of Mtb/human host interaction is limited, and the exact mechanisms exerting protection against TB remain largely unknown, and could range from (trained) innate to adaptive immune effector mechanisms (6). Many of the most advanced TB subunit candidates are based on Mtb proteins historically identified from in vitro cultured Mtb which were thereafter tested for antigenicity using peripheral blood cells from Mtb exposed individuals (7).

Although immune recognition of such proteins suggests previous immune priming by these antigens, it does not provide information regarding the expression of these antigens in the main organ affected by Mtb, the lung. We have hypothesized that in order for vaccine antigens to be protective (and for drug targets to be effective), a minimal requirement is that they are expressed in the lung of Mtb infected individuals, preferably over prolonged periods of time such that immune cells can recognize and eliminate infected targets cells presenting these antigens; or conversely, that TB drugs are able to act effectively on such Mtb target molecules. To address this important issue, which cannot easily be studied in humans, we recently reported a novel in vivo pulmonary Mtb gene expression analysis based on the Mtb RNA expression patterns in the lung of highly susceptible (C3HeB/FeJ) as well as resistant (C57BL/6J) mice after aerosol Mtb (Erdman) infection, from early to late time points (8). Of note, the highly susceptible C3HeB/FeJ “Kramnik” mice uniquely develop a form of lung TB that includes centrally necrotic lesions characteristic of human TB. By analysing the expression of 2068 Mtb genes [selected to represent the first genes of most predicted Mtb operons, in order to enhance representative “genome wide” coverage (9)] in their lungs during early and late phase Mtb infection, we identified upregulated Mtb genes that were designated in vivo expressed Mtb (IVE-TB) genes. A total of 50 IVE-TB candidate antigens was selected further based on information including high-level conservation among whole-genome sequenced Mtb-complex strains (n = 219) and algorithms predicting epitopes presented by HLA-class-Ia and -II. Many of the IVE-TB candidate antigens were highly conserved among the genomes of >200 human isolated Mtb strains, and were recognized by Mtb responsive human subjects, supporting the hypothesis that the IVE-TB genes found in the C3HeB/FeJ “Kramnik” mouse lung were also expressed and presented in humans, likely in the lung.

Nevertheless, formal proof that these Mtb IVE-TB genes are truly expressed in the Mtb infected human lung is lacking. Only few studies have evaluated the Mtb transcriptome by human lung infection centred approaches, using sputum and bronchoalveolar lavage (BAL) samples from pulmonary TB patients. Those few studies used mostly quantitative real-time PCR (RT-PCR) or microarray platforms (10–15). In these reports, the expression levels of Mtb genes recovered from human sputum were compared to those in in vitro cultures. Up- and down-regulated Mtb genes were reported mainly as ‘fold-changes’ compared to in vitro cultured Mtb or evaluated as quantitatively altered Mtb gene expression profiles in samples collected during TB treatment. Since a direct comparison between these ‘fold-changes’ and quantitative qPCR data was not immediately possible, we collected and reanalyzed all raw data and relative expression scores of Mtb transcripts from three studies (11, 12, 15). These human transcriptomic data sets were selected because raw data were publicly available and allowed comparative analysis. Ranking Mtb genes within each dataset allowed us to assess and combine in vivo Mtb gene expression in multiple individual studies independent of their individual design. Using this information and comparing three independent Mtb transcriptome datasets from human sputum or BAL with our previously published dataset from C3HeB/FeJ “Kramnik” mouse lungs (8), which was then complemented by validation in a recently deposited Mtb RNA-Seq dataset (GEO: GSE132354) (16) from alveolar macrophages of Mtb infected C57BL/6J mice, as well a partial validation in a Mtb RNA-Seq dataset from seven human active TB sputum samples (GEO: GSE137518) (17), we here provide a first lung-centric multi-study transcriptomic integrated dataset, which provides a novel tool likely to be useful to TB vaccine and drug target discovery.

This study represents the first lung-oriented description of Mtb transcripts from sputum and bronchoalveolar lavage (BAL) of TB patients and from the lungs of Mtb infected mice (8, 11, 12, 15). Our analysis identified considerable overlap among the most abundant Mtb transcripts in different settings, spanning information from mouse (C3H/FeJ) and human Mtb infection. The only exception were microarray based data, probably due to the lower dynamic range of that platform (discussed further below). These findings were further validated and corroborated using mouse and human sputum Mtb transcriptomes measured by RNA-Seq. In addition, the data were compared with in vitro datasets, revealing significant differences as expected.

An important observation was that many Mtb genes encoding new candidate Mtb vaccine antigens or existing drug targets were among the most highly expressed Mtb genes in the lungs of infected mice and humans. A second important observation was that our analysis revealed high expression of 35 Mtb genes that have not been studied as vaccine antigens, diagnostic or drug target molecules, suggesting that the here reported data might inform novel strategies for discovery of new antigens and drug targets. Thus, this first comprehensive lung infection centred Mtb gene expression comparison provides novel insights into the pulmonary Mtb transcriptome signature and the biological pathways involved, significantly enhances our understanding of Mtb infection biology and is useful in deciphering the in vivo expressed pulmonary Mtb transcriptome, which can guide precision development of new TB diagnostics, drugs and vaccines (7).

Assessing the Mtb transcriptome in infected human organs is challenging since airway specimens such as sputum, BAL and biopsies are not always easy to obtain. Among those, sputum collection is the only non-invasive procedure and therefore the most commonly used to investigate Mtb transcriptomes in TB (10–15). However, adequate RNA isolation from sputum is hampered by several factors such as the small size of the samples, the presence of RNases, and the contamination with other commensal or pathogenic bacteria (26). In past studies, we circumvented those limitations by analysing RNA isolated from lung of Mtb infected mice, including those of highly susceptible C3HeB/FeJ “Kramnik” mice presenting with pulmonary lesions characteristic of human TB (8, 27), using RT-PCR. Although these methods allowed the identification of a new class of in vivo expressed Mtb (IVE-TB) antigens, proven to be recognized by peripheral blood cells of Mtb infected individuals (8, 27), it was not clear whether the Mtb transcriptome signature found in the lung of “Kramnik” mice sufficiently reflected that of clinical Mtb isolates infecting the human respiratory tract. The present comparison shows a strongly positive correlation between Mtb transcriptomes signatures from infected lungs of “Kramnik” mice and three out of four published Mtb transcriptome signatures identified from sputum and BAL of TB patients ( Figure 2 ). The only Mtb transcriptome dataset that did not correlate with the mouse dataset was obtained using micro-arrays, and also did not correlate with any of the other human datasets ( Figure 2 ). There are two main factors that could explain this lack of correlation: the most likely is the technology platform used (dual microarray vs. RT-PCR: microarrays are less sensitive and quantitative than RT-qPCR and have a significantly lower dynamic range (18) although we cannot exclude a role for the geographic area of patients’ recruitment, India vs. Sub-Saharan Africa [there are different distributions of geographically restricted Mtb strains detected in the samples (28)]. The differences compared to the microarray dataset was also clear when the top 15% expressed Mtb genes of each published dataset were compared: only seven of the 90 Mtb top expressed genes which were shared in all RT-PCR datasets, were found to be highly expressed in the microarray dataset ( Figure 1B ). This discrepancy is evident also when comparing our findings to those obtained in a previous study (29) that analyzed the overlaps between upregulated genes in human and murine lungs identified from two microarray datasets (30, 31). In this meta-analysis only four genes were shared across species (Rv0316, Rv3347c, Rv3532, and Rv3706c), none of which is included in the top ranked transcripts of our analysis. Besides the fact that this data was based on microarray assays, the limited overlaps could be explained by the fact that in these studies the up-regulation of the genes is relative to the expression found in in vitro transcriptomics. Our findings differ also from another comparative analysis among Mtb transcriptomes from susceptible (I/StSnEgYCit) and resistant (C57BL/6YCit) mouse strains, which defined 209 commonly upregulated genes using the Audic-Claverie algorithm (32). However, from this list only 17 Mtb transcripts overlap with the top ranked genes found in our C3HeB/FeJ “Kramnik” dataset, a number that decreases to two if including in the comparison also the human datasets ( Data File S1 : Table S1.10 ). Therefore, we think it is crucial to use the same analysis pipeline when examining different datasets.

We also analyzed two additional in vitro Mtb transcriptome (RT-PCR based) datasets, both from log phase as well as from stationary phase cultured Mtb bacteria, and compared their gene expression profiles with those from the in vivo analyses. As expected, the in vitro log phase grown data correlated most strongly with the in vivo datasets ( Figure 2B ). However, despite the overall correlation, the in vivo pulmonary core Mtb transcript set only partially overlapped with the in vitro data, emphasizing the need for unbiased in vivo analyses to identify core Mtb transcriptome expression signatures that can inform the design of intervention tools, such as diagnostic tests, vaccines or new drugs. Of note, when taking only the top ranked transcripts from the in vitro datasets these had more overlap with the human datasets than those found between the two in vitro conditions. This finding may suggest that a combination of in vitro experimental models may translate better into the dynamic and heterogeneous state of Mtb during an in vivo infection (33).

The 90 Mtb top expressed genes were enriched for biological processes such as growth, translation, pathogenicity, protein secretion, and response to hypoxia and oxidative stress ( Figure 1C ). This trend was found also when genes were grouped according to previously described functional categories (11). In fact, the highest and most consistent expression ranks included virulence-associated Mtb genes, and were found among esx genes, ribosomal protein genes, genes related to metabolic pathways, and genes encoding proteins in response to hypoxic and oxidative stress, suggesting the presence of metabolic active bacilli in the samples investigated here ( Figure S1 ). These signatures are compatible with active Mtb infection. Additionally, most of the 90 top expressed genes within the RT-PCR dataset were found expressed in independent and unbiased mouse and human derived RNA-Seq datasets. Unexpectedly, 35 out of these 90 highly abundant Mtb transcripts encode for proteins not yet been described as Mtb antigens or drug targets ( Figure 1B ) (7, 19). This key finding underscores how our integrative transcriptomic approach helps generate new hypotheses worth future investigation.

Although based on different mouse strains (C3H vs. BL6), and time of infection (five time points spanning from two to 12 weeks vs. two weeks only), the mouse lung RT-PCR dataset showed the highest correlation with the unbiased RNA-Seq datasets ( Figure 2A ) which detected, with only one exception (Rv3619c for the mouse derived dataset)¸ all genes encoding IVE-TB antigens. The latter had been selected based on high expression in the Mtb infected lungs of “Kramnik” mice, and the present analysis further extends this to human Mtb infected lung derived samples ( Figure 3A ) (8). By contrast, many of the Mtb genes encoding antigens identified only on the basis of their recognition in human peripheral blood, mostly showed lower expression than the IVE-TB antigens in the lung-centric datasets studied here ( Figure 3B ). Based on our hypothesis this suggests that only a subset of proteins immunogenic in human blood tests might have vaccine potential, but this needs to be demonstrated formally. However, one case in point is the failure of the MVA85 phase IIb trial which showed no additional efficacy of MVA85 booster vaccination on top of BCG in preventing TB in infants (34). We found that the encoded antigen, Ag85A (Rv3804c) used in MVA85A exhibited lower expression than for example antigens contained in the successful M72 vaccine ( Figure 3A ) (5), and this may underlie Rv3804c’s poor performance as single subunit vaccine antigen. Although the different protective efficacies of subunit vaccines are difficult to compare, antigen availability and presentation to immune cells in the respiratory tract of Mtb infected subjects may well be a key factor influencing their eventual efficacy. Of note, in our analysis, most promising TB subunit-vaccine candidates, including those in current clinical trials, contained at least one protein encoded by a highly expressed Mtb gene in multiple transcriptomic datasets.

We also performed further comparative analyses on genes encoding antigens that are currently used in commercial TB diagnostics or additional candidate vaccines (7). As expected, antigens implemented in approved TB diagnostics such as ESAT-6 and CFP-10 (Rv3875c and Rv3874) were highly or moderately expressed, except TB7.7 (Rv2654) that showed only low or undetectable expression levels. T cell reactivity towards TB7.7 has been investigated in short-term T cell cultures (35, 36) and in a direct ex vivo assay (37). The latter approach failed to find evidence for antigenicity of TB7.7 in a cohort of 18 LTBI donors (37). This result was unexpected given the presence of HLA-class II binding peptides in the antigen. Here we postulate that this lack of antigenicity in fact may reflect the low expression levels of TB7.7 which we find in the respiratory tract of untreated TB patients. Together, these findings support the use of other antigens for early detection of Mtb infection such as those recently proposed to be included in ESAT-6 free immunodiagnostic tests (Rv2348, Rv3615, and Rv3865) (23), which were instead found highly expressed ( Figure 3B ).

Another interesting application of our findings may be the refinement and implementation of new mega-pools of Mtb epitopes, a first version of which has already been tested in several cohorts of Mtb infected subjects (24, 25). Our analysis revealed that 19% (n=17) of the 89 Mtb genes encoding for those epitopes were poorly expressed in vivo ( Figure S1 ). If confirmed also at a protein level, depleting new mega-pools of such peptides might help denoising the promiscuous immune reactivity found against the Mtb epitope mega-pool (24) and allow a better characterization of those peptides most likely available during infection.

A limitation of our analysis might be that differential gene expression does not by definition correlate with differential protein expression, and differential protein expression does not equal differential functionality. The analysis of large sets of genes as performed in this study, however, we think might have mitigated this risk, even though at the individual gene level exceptions may exist. Several Mtb proteomic datasets have been reported, confirming high expression of Mtb proteins known to be diagnostic or candidate vaccine antigens, but as far as we know this has not been correlated systematically to vaccine-, diagnostic- of drug-target efficacy. Furthermore, although Mtb transcript levels may not necessarily predict protein levels (38), they are helpful in understanding global transcriptome dynamics with impact on potential local antigen availability and presentation as well as drug target expression during Mtb infection. Such knowledge may help de-risking TB vaccine and drug development by selecting the most highly expressed and presented Mtb genes and proteins in the human lung during infection.

Currently there are 28 approved TB drugs which have been or are being prioritized and combined based on the sensitivity of Mtb to these drugs. Isoniazid, Rifampin (or rifapentine or Rifabutin), Ethambutol (administered for six months) and Pyrazinamide are used as first-line drugs for drug sensitive TB, while the rest are used in different regimens against multi- or extensively-drug resistant TB (MDR-TB and XDR-TB respectively) (http://bmi.icmr.org.in/tbdrugs/). An important aspect of our analysis is that it confirmed the high-medium expression of all 22 Mtb genes encoding targets of TB drugs efficacious in clinical settings ( Figures 4 , S1 ). These results indirectly validate our approach but, more importantly, support the value of using pulmonary Mtb transcriptomic datasets to select for new potential TB drug targets. Indeed, our work also reveals high expression of Mtb genes that have not been studied as drug target molecules, suggesting that the here reported approach could inform discovery of new drug targets.

Collectively, by our comparative analyses of in vivo pulmonary Mtb transcriptomes both from Mtb infected animals and humans significantly increase our understanding of in situ host-pathogen interactions in TB, and can help refining the in vivo expressed Mtb genes and proteins to accelerate and innovate TB vaccine, diagnostics and drug development. Enhanced focus on lung centric studies in TB should therefore be encouraged.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/ Supplementary Material .

The studies involving human participants were reviewed and approved by The Human Research Ethics Committee of the University of Cape Town (HREC References: 031/2012 and 568/2012). The patients/participants provided their written informed consent to participate in this study.

This project was developed and designed by MC and TO. MC performed the analyses. MC and TO interpreted the analyses and wrote the manuscript. RP-JL and RW shared the human derived RNA-Seq dataset, participated to the scientific discussion and reviewed the manuscript. All authors contributed to the article and approved the submitted version.

EC HORIZON2020 TBVAC2020, Grant Agreement No. 643381(MC, TO); EC ITN FP7 VACTRAIN project (MC, TO); NWO-TOP project Grant Agreement No. 91214038 (TO). Francis Crick Institute funded by URI grant FC0010218, Cancer Research UK (FC0010218) and Wellcome (FC0010218) (RP-JL, RW); Wellcome grant 104803 (RW); Wellcome grant 203135 (RW).

The text represents the authors’ views and does not necessarily represent a position of the funders who will not be liable for the use made of such information.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer CLA declared a past co-authorship with one of the authors to the handling Editor.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.