Abstract

   Mycobacterium tuberculosis (Mtb), the etiologic agent of tuberculosis
   (TB), remains a persistent global health challenge due to the lack of
   an effective vaccine. The only licensed TB vaccine, Bacille
   Calmette-Guerin (BCG), is a live attenuated strain of Mycobacterium
   bovis that protects young children from severe disease but fails to
   provide protection through adulthood. It is unclear why BCG provides
   incomplete protection despite inducing a robust Th1 immune response. We
   set out to interrogate mycobacterial determinants of vaccine escape
   using a functional genomics approach, TnSeq, to define bacterial genes
   required for survival in mice vaccinated with BCG, the live attenuated
   Mtb vaccine strain, ΔLprG, and in mice with Mtb immunity conferred by
   prior infection. We find that critical virulence genes associated with
   acute infection and exponential growth are less essential in hosts with
   adaptive immunity, including genes encoding the Esx-1 and Mce1 systems.
   Genetic requirements for Mtb growth in vaccinated and previously
   Mtb-infected hosts mirror the genetic requirements reported for
   bacteria under in vitro conditions that reflect aspects of the adaptive
   immune response. Across distinct immunization conditions, differences
   in genetic requirements between live attenuated vaccines and
   vaccination routes are observed, suggesting that different immunization
   strategies impose distinct bacterial stressors. Collectively, these
   data support the idea that Mtb requires genes that enable stress
   adaptation and growth arrest upon encountering the restrictive host
   environment induced by the adaptive immune response. We demonstrate
   that TnSeq can be used to understand the bacterial genetic requirements
   for survival in vaccinated hosts across pre-clinical live attenuated
   vaccines and therefore may be applied to other vaccine modalities.
   Understanding how Mtb survives vaccine-induced immunity has the
   potential to inform the development of new vaccines or adjuvant
   therapies.

   Subject terms: Live attenuated vaccines, Bacterial genomics, Bacterial
   immune evasion

Introduction

   Mycobacterium tuberculosis (Mtb), the etiologic agent of tuberculosis
   (TB), caused over 10 million infections in 2023, resulting in 1.25
   million deaths across the globe^[38]1. One reason why TB remains a
   persistent global health challenge is the lack of an effective vaccine.
   The only licensed TB vaccine, Bacille Calmette-Guerin (BCG), is a live
   attenuated strain of Mycobacterium bovis developed over 100 years ago.
   BCG is effective in preventing severe, extrapulmonary TB disease in
   infants and young children, a major cause of morbidity and mortality in
   TB-endemic regions. However, BCG provides minimal protection against
   pulmonary TB in adults, the most common form of the disease, despite
   inducing a durable memory response^[39]2.

   It is unclear why BCG-induced immunity provides incomplete protection.
   BCG elicits a robust CD4^+ T-cell response, which, via IFNγ production,
   enhances macrophage oxidative burst and phagolysosomal maturation,
   resulting in increased bacterial killing^[40]3. CD4^+ T cells also
   promote antibody production and maturation^[41]4, and recruitment and
   activation of cytotoxic CD8^+ T cells and other effector
   cells^[42]5,[43]6. While IFNγ production is clearly necessary for
   vaccine-induced protection, the frequency of IFNγ-producing T cells in
   vaccinated individuals is not predictive of disease risk, suggesting
   that additional bacterial control mechanisms are involved^[44]7. In
   contrast to the intensive studies of the immune response induced by BCG
   vaccination, mycobacterial determinants of vaccine escape remain
   unknown, further hampering the rational design of improved immunization
   strategies. Understanding how Mtb survives in the setting of
   vaccine-induced immunity could uncover bacterial vulnerabilities that
   inform the development and prioritization of new vaccines or adjuvant
   therapies.

   With the goal of uncovering such vulnerabilities, we used a functional
   genomics approach, TnSeq, to define mycobacterial genes required for
   survival in vaccinated mice and mice with immunity conferred by prior
   Mtb infection. TnSeq entails generating a genome-wide, saturated
   bacterial mutant library with a randomly integrating transposon, then
   selecting this library under a defined growth condition.
   Next-generation sequencing is used to map transposon insertion sites in
   input and output libraries, and the distribution of insertions across
   the genome is compared to identify genes with differences in insertion
   frequency. Genes with fewer transposon insertions in the output library
   are considered conditionally essential for bacterial growth, while
   genes with no change or an increase in insertion abundance are
   considered conditionally non-essential. Transposon mutagenesis has been
   used to define the genes mycobacteria require to survive under myriad
   in vitro and in vivo conditions, providing important insights into
   bacterial physiology^[45]8–[46]20. We therefore reasoned that TnSeq
   could be used to understand how Mtb survives the numerous pressures
   imposed by the pathogen-specific adaptive immune response when applied
   in the context of vaccination and post-primary infection.

   This pathogen-centric approach provided new insights into the
   physiology of Mtb under immune memory. Of note, we found that virulence
   genes required to establish acute infection, Esx-1 and Mce1, are less
   essential in hosts with immune memory. Integrative analysis of our data
   with published TnSeq datasets suggests that the cellular
   microenvironment of the vaccine-induced immune response is more akin to
   what Mtb experiences during chronic infection, when bacterial growth
   becomes constrained. Defining the bacterial genetic requirements for
   survival in vaccinated hosts across pre-clinical live attenuated
   vaccines (LAV) may provide a more nuanced understanding of correlates
   and mechanisms of protection, and lead to improved vaccines to curb
   this ongoing global health emergency.

Results

Comprehensively defining M. tuberculosis genetic requirements for survival in
BCG-vaccinated hosts

   The added immune pressure conferred by BCG vaccination is expected to
   impose substantial selection against the bacterium during infection,
   yet the genes and pathways Mtb relies on to survive in this context
   remain largely unknown. We wished to define this suite of bacterial
   genes by performing TnSeq in BCG-vaccinated mice. Prior studies in the
   C57BL/6 mouse model have found that subcutaneous BCG vaccination
   confers ~0.5–1.5 log reduction in total Mtb burden^[47]21–[48]24. To
   ensure that we would not bottleneck the complex pool of mutants in our
   transposon library with this selective pressure, we performed a
   preliminary study in mice that were aerogenically infected with a low
   dose (~100 CFU) of H37Rv. Because ongoing Mtb infection has been shown
   to impose near-sterilizing protection against secondary infection in
   non-human primates^[49]25, we reasoned that if the transposon library
   was not bottlenecked under this selective pressure, it would not be
   bottlenecked by BCG vaccination. We challenged the aerosol-infected and
   age-matched naïve C57BL/6 mice with a saturated transposon library
   generated in the widely used Mtb reference strain H37Rv^[50]11. Two and
   four weeks post-challenge, mice were sacrificed and tissue harvested
   for CFU enumeration and to recover genomic DNA from the surviving
   mutants for transposon-junction sequencing. Concomitant Mtb infection
   reduced bacterial burden by ~1.5 logs in the spleen, the site of
   transposon library selection, at these time points (Supplementary Fig.
   [51]1a). Transposon-junction sequencing and mapping of the surviving
   mutants with the Transit pre-processing software tool^[52]26 found that
   approximately one third fewer mutants survived in animals with ongoing
   Mtb infection as compared to naïve mice (Supplementary Data [53]1).
   This represents a moderate selection pressure comparable to the
   selection imposed on the input library by naïve C57BL/6 mice^[54]12,
   but less than that imposed in some antibiotic TnSeq screens^[55]27.
   Analysis of reproducibility between individual mice within each arm of
   the experiment revealed high correlation coefficients for mean
   transposon count across each gene (Spearman correlation coefficients
   0.77–0.93, Supplementary Fig. [56]1b).

   As there was no evidence of transposon library bottlenecking in mice
   with ongoing Mtb infection, we next challenged BCG-vaccinated and
   age-matched naïve C57BL/6 mice using the same approach (Fig. [57]1a).
   We found that subcutaneous BCG vaccination reduced total bacterial
   burden by ~1.5 logs at 2 and 4 weeks post-challenge (Supplementary Fig.
   [58]1c), consistent with prior studies. Selective pressure, as
   indicated by library saturation, was slightly less than that observed
   in our preliminary study, and reproducibility among animals was
   similarly high (Spearman correlation coefficients 0.76–0.93,
   Supplementary Fig. [59]1d). To identify genes that are differentially
   essential between naïve and BCG-vaccinated animals, we performed
   head-to-head comparisons of the output transposon libraries using the
   Transit resampling method^[60]28. With correction for multiple testing,
   we found that 102 genes were differentially essential 2 weeks
   post-challenge, and 64 genes were differentially essential 4 weeks
   post-challenge (adj. p value < 0.05, Fig. [61]1b, Supplementary Data
   [62]1).

Fig. 1. Mtb genes required for survival and growth in BCG-vaccinated mice.

   [63]Fig. 1
   [64]Open in a new tab

   a Schematic of the experimental strategy for the TnSeq screen. n = 4–5
   mice per group. b Volcano plots highlighting differentially essential
   genes in BCG-vaccinated mice relative to naïve mice at 2 weeks (left
   panel) and 4 weeks (right panel) post-infection.

   Given the selective pressure imposed by vaccination, we expected to
   identify genes that were more essential in immunized relative to naïve
   mice. Consistent with this expectation, 41 genes had a statistically
   significant decrease in transposon insertion count in BCG-vaccinated
   mice compared to naïve mice 2 weeks post-challenge. The set of genes
   conditionally essential for survival in vaccinated mice are distributed
   across a number of functional categories, and pathway analysis did not
   identify significant enrichment in any category (Supplementary Fig.
   [65]2a). However, among the more essential genes are katG/Rv1908c,
   which encodes a catalase-peroxidase that detoxifies reactive oxygen
   species; mosR/Rv0348, a transcriptional regulator induced by hypoxia
   and one of the most highly expressed genes during chronic infection in
   mice; virS/Rv3082c, a pH-responsive transcriptional regulator activated
   in the phagolysosome; and ctpC/Rv3270, a P-type ATPase responsible for
   metalating sodA, the major superoxide dismutase in Mtb. Altogether,
   these genes represent responses to host-imposed stresses enhanced by
   BCG vaccination, such as the CD4^+ T-cell-mediated increase in IFN-γ
   and TNF-α from infected and bystander macrophages^[66]29.

   Four weeks post-challenge, 36 genes were significantly more required in
   vaccinated as compared to naïve mice. Only one gene, the
   catalase-peroxidase katG/Rv1908c, overlapped with the more essential
   genes at the 2-week time point (Supplementary Fig. [67]2a), suggesting
   continued, enhanced oxidative stress in vaccinated hosts. Among the
   other genes most essential for growth 4 weeks post-challenge are
   rsfB/Rv3687c, an anti-anti-sigma factor that enhances the activity of
   the alternative sigma factor σ^F, which plays a central role in stress
   response and persistence during chronic infection^[68]30; cuvA/Rv1422,
   which encodes a protein involved in lipid utilization and has a
   persistence defect when deleted^[69]31; fadD6/Rv1206, an acyl-CoA
   synthetase that promotes triacylglycerol accumulation and mycobacterial
   dormancy^[70]32; and Rv2628, a hypothetical protein in the DosR
   dormancy regulon that encodes an antigen associated with latent
   infection^[71]33,[72]34. Altogether, many of these genes are linked to
   the induction and maintenance of Mtb dormancy, suggesting that Mtb’s
   ability to arrest growth is critical to surviving the memory immune
   response.

Canonical virulence genes are less required in BCG-vaccinated mice

   Unexpectedly, approximately half of the statistically significant genes
   were less essential in BCG-vaccinated mice compared to naïve mice,
   displaying an increase in transposon insertion count: 61 genes two
   weeks post-challenge and 28 genes four weeks post-challenge (Fig.
   [73]1b, Supplementary Data [74]1). These two gene sets are highly
   overlapping, and pathway analysis found that the functional category
   “virulence” was significantly over-represented at both time points
   (Supplementary Fig. [75]2b). Notably, many of the less essential genes
   belong to two systems that play key roles in Mtb pathogenesis and its
   intracellular lifestyle: the Esx-1 Type VII secretion system, and the
   Mce1 lipid uptake system (Figs. [76]1b and [77]2a, Supplementary Fig.
   [78]2b).

Fig. 2. Critical virulence genes are less essential in BCG-vaccinated mice.

   [79]Fig. 2
   [80]Open in a new tab

   a Log[2] fold-change in transposon read count for each gene in the
   Esx-1 system and Mce1 system in BCG-vaccinated mice relative to naïve
   mice. Asterisks indicate adj. p < 0.05 by the Transit resampling
   pipeline, as detailed in the “Methods.” b Mean log[2] fold-change in
   transposon read count for genes in the Esx-1 or Mce1 systems across
   defined in vitro conditions and mouse genetic backgrounds. Error bars
   represent SEM. Purple bars represent data from this study. Gray bars
   indicate data from the MtbTnDb database. c Correlation between log[2]
   fold-change in transposon read count averaged across Esx-1 or Mce1
   genes and bacterial burden across a set of genetically diverse mouse
   backgrounds. Log[2] fold-change is relative to the input transposon
   library. Each dot represents one mouse background (n = 1–6 mice per
   background).

   Both the Esx-1 and Mce1 systems are known to be essential for Mtb
   growth in vivo^[81]9,[82]35–[83]39. The Esx-1 system consists of a Type
   VII transmembrane secretion system and secreted effector proteins,
   which are critical to Mtb pathogenesis^[84]40. A key role of Esx-1 is
   to permeabilize the phagolysosomal membrane, facilitating release of
   bacterial components into the cytosol, thereby initiating a complex
   innate immune response from the host^[85]41–[86]45. The Esx-1 secretion
   system, secreted proteins, and regulatory factors are encoded across
   two unlinked genomic loci^[87]40. Genes across both loci were
   significantly less essential in BCG-vaccinated mice as compared to
   naïve mice at both time points (Fig. [88]2a, Supplementary Fig. [89]3a,
   Supplementary Data [90]1). Most genes in the Esx-1 loci that did not
   reach statistical significance showed a trend towards an increase in
   the number of transposon insertions (Fig. [91]2a, Supplementary Fig.
   [92]3a, Supplementary Data [93]1). Some short genes in the Esx-1
   system, such as esxA/Rv3875 and esxB/Rv3874, which encode the secreted
   proteins ESAT-6 and CFP-10, have few transposon insertion sites, making
   it difficult to reach statistical significance, a known limitation of
   TnSeq methodology.

   Mce1 is also a transmembrane complex and is responsible for importing
   host-derived fatty acids, an important nutrient for Mtb during acute
   infection^[94]36. The Mce1 system is encoded across a 12-gene operon,
   with seven accessory genes distributed across the genome. All 12 of the
   genes in the Mce1 operon, in addition to the accessory genes
   lucA/Rv3723, omamA/Rv0199, omamB/Rv0200, and mceG/Rv0655, were
   significantly less essential in BCG-vaccinated mice compared to naïve
   mice at one or both time points (Fig. [95]2a, Supplementary Fig.
   [96]3a, Supplementary Data [97]1).

   We note that the reduced essentiality of Esx-1 in BCG-vaccinated mice
   is relative, rather than absolute. Transposon insertions across the
   Esx-1 genes are increased in vaccinated compared to naïve mice.
   However, relative to the input library, they are decreased
   (Supplementary Fig. [98]3b). By contrast, bacteria with transposon
   insertions across the Mce1 genes are more abundant in vaccinated mice
   as compared to the input library (Supplementary Fig. [99]3b),
   suggesting an absolute growth advantage for Mce1 mutants in vaccinated
   mice.

   To validate our observation that the Esx-1 virulence system is less
   essential in vaccinated mice, we challenged groups of BCG-vaccinated
   and naïve animals with either wild-type H37Rv or an Esx-1 deletion
   mutant^[100]46 (Supplementary Fig. [101]3c). Two and four weeks
   post-challenge, animals were sacrificed and CFU enumerated from the
   spleens and lungs. To quantify the relative fitness cost of Esx-1
   disruption in naïve compared to BCG-vaccinated animals, we calculated
   the log[10] difference in bacterial burden (CFU) between naïve and
   vaccinated animals. We found that the Esx-1 deletion strain had a
   substantial growth defect in naïve mice 2 and 4 weeks post-infection,
   consistent with previous reports^[102]37,[103]38,[104]47. However, this
   growth defect was reduced in BCG-vaccinated mice in the spleen, with an
   even greater effect in the lung (Supplementary Fig. [105]3c). These
   results support our TnSeq findings that the in vivo essentiality of a
   canonical virulence system, Esx-1, is partially rescued in the context
   of BCG vaccination.

Decreased essentiality of Esx-1 and Mce1 genes in vaccinated hosts
corresponds to conditions promoting Mtb growth arrest

   What does it mean for virulence systems to be less essential in
   vaccinated mice? To gain biological insight, we evaluated the
   requirement for Esx-1 and Mce1 genes across TnSeq datasets obtained
   from the MtbTnDb database^[106]48. MtbTnDb has aggregated data from
   TnSeq experiments across numerous growth conditions, and the raw
   sequencing data have been standardized and analyzed by Transit, the
   same pipeline used here. We extracted data from MtbTnDb for wild-type
   H37Rv transposon libraries selected under defined in vitro conditions
   or wild-type and knock-out mice for comparison to our data
   (Supplementary Data [107]2). We then plotted the mean log[2]
   fold-change in transposon read count averaged across Esx-1 or Mce1
   genes for each dataset to identify conditions that phenocopy the
   bacterial requirement for these genes during vaccination (Fig.
   [108]2b). As expected, Esx-1 genes are essential in wild-type C57BL/6
   mice as compared to the in vitro selected input library (Fig. [109]2b),
   consistent with our data (Supplementary Fig. [110]3b). Esx-1 genes
   become even more essential in immune-deficient mice, including Rag,
   NOS, IFNγ, and Phox (Cybb) knock-out backgrounds. By contrast,
   conditions where Esx-1 is not essential include in vitro hypoxia, in
   vitro acid stress, and in vitro amino acid limitation.

   Across this same set of MtbTnDb conditions, we observed that Mce1 genes
   are not required when grown in vitro in media where cholesterol is the
   sole carbon source, consistent with the known role of Mce1 as a
   fatty-acid importer^[111]35,[112]36 (Fig. [113]2b). This suggests that
   bacteria surviving the immune pressure of vaccination have distinct
   metabolic activity or access to different nutrient sources. Overall,
   the rank order of conditions for Mce1 genes is similar to that for
   genes in the Esx-1 system. Like Esx-1, Mce1 genes are less essential
   under in vitro hypoxia, in vitro acid stress, and in vitro amino acid
   limitation, and more essential in Rag, NOS, IFNγ, and Phox knock-out
   mice (Fig. [114]2b). Because bacterial replication is uncontrolled in
   these immune-deficient mouse backgrounds but restricted under the
   aforementioned in vitro stress conditions, we hypothesized that Esx-1
   and Mce1 may be required for rapid bacterial growth.

   To test this hypothesis, we analyzed a large, published dataset of
   bacterial burden and TnSeq measurements collected across a panel of
   genetically diverse mice with variable capacity for Mtb
   control^[115]17. We evaluated the relationship between the essentiality
   of Esx-1 and Mce1 genes, indicated by the average log[2] fold-change in
   transposon insertion count in each gene set relative to the input
   library, and bacterial proliferation, reflected by endpoint CFU, across
   these mouse backgrounds. We found bacterial burden to be negatively
   correlated with the log[2] fold-change in transposon count for both
   Esx-1 genes (r = −0.48) and Mce1 genes (r = −0.66) (Fig. [116]2c,
   Supplementary Data [117]3). This is consistent with a model in which
   the Esx-1 and Mce1 systems are required under conditions of rapid
   bacterial replication, and the decreased essentiality of these genes in
   BCG-vaccinated mice may reflect a slow- or non-replicating state of the
   surviving bacteria.

Differential effect of immunization strategy on the essentiality of Esx-1 and
Mce1

   Given the poor performance of BCG in conferring lasting, protective
   immunity against Mtb infection and disease, significant efforts have
   been devoted toward the development of improved vaccine strategies.
   These include subunit vaccines, novel attenuated whole cell vaccines,
   and methods of delivery that promise to enhance the immune response,
   including mucosal and intravenous routes of
   administration^[118]7,[119]49. Quantitative and qualitative differences
   in immune response and degree and duration of protection have been
   described under various vaccination strategies^[120]49,[121]50, yet
   bacterial physiology under these conditions has not been interrogated.

   We therefore expanded our vaccine TnSeq challenge model to include
   additional immunization conditions (Fig. [122]3a). We included mice
   vaccinated with BCG via intravenous route (BCG-IV), which has been
   shown to confer substantially greater protection against Mtb compared
   to BCG-SQ in non-human primates^[123]51,[124]52. This improvement in
   protection has been attributed to greater numbers of antigen-specific T
   cells as well as an enhanced humoral response^[125]51,[126]53–[127]55
   In C57BL/6 mice, BCG-IV has not been shown to provide enhanced
   protection compared to BCG-SQ^[128]22,[129]23, but has been found to
   induce qualitative differences in immune response, with a role for
   trained immunity found in some, but not all studies^[130]22,[131]56. We
   also included mice vaccinated with a live, attenuated vaccine strain of
   Mtb, ΔLprG, which has been shown to confer greater protection against
   Mtb than subcutaneous BCG (BCG-SQ) in the C3HeB/FeJ mouse model of
   infection, induce a more robust cytokine response, and elicit a
   polyfunctional CD4^+ T-cell response^[132]21,[133]57. Finally, we
   included mice that had been infected via the aerosol route with H37Rv,
   then cured with combination antibiotic therapy (Mtb+cure). This
   approximates reinfection in humans and may provide insight into how Mtb
   escapes the immunity conferred by natural infection. A meta-analysis of
   epidemiological studies from the pre-antibiotic era found that prior
   Mtb infection confers a degree of protection against TB disease similar
   to that of BCG vaccination^[134]58. In mouse models, prior Mtb
   infection and cure confer approximately a log reduction in Mtb burden,
   similar to the level of protection afforded by subcutaneous BCG
   vaccination, the traditional benchmark for TB vaccine evaluation in
   pre-clinical animal models^[135]59,[136]60.

Fig. 3. Genetic requirements across distinct immunization conditions.

   [137]Fig. 3
   [138]Open in a new tab

   a Experimental strategy for the expanded TnSeq screen. n = 2–5 mice per
   group. b Heatmaps showing log[2] fold-change in transposon insertion
   count relative to naïve mice for the Esx-1 genes (left panels) and Mce1
   genes (right panels) for each immunization condition. Asterisks
   indicate adj. p < 0.05 by the Transit resampling pipeline, as detailed
   in the “Methods.” c Box plots showing log[2] fold-change in transposon
   insertion count for Esx-1 and Mce1 genes, relative to naïve mice.
   Significance determined by the Friedman test with Dunn’s multiple
   comparison to the BCG-SQ condition.

   Mice in each of these groups were challenged with the H37Rv transposon
   library, and surviving mutants were recovered and sequenced as
   described above. We found that bacterial burden was reduced by ~1.5
   logs in each arm of the experiment compared to naïve animals and was
   not significantly different across arms in the spleen, the site of
   transposon library selection (Supplementary Fig. [139]1c). While this
   suggests quantitatively similar selective pressure across the
   immunization strategies, we wanted to assess whether there were
   qualitative differences among these conditions as reflected by the
   bacterial genes essential for growth. Given the striking difference in
   essentiality of Esx-1 and Mce1 genes in BCG-SQ vaccinated compared to
   naïve animals, we first evaluated the essentiality of these systems.

   Compared to naïve mice, many genes in the Esx-1 and Mce1 loci were
   significantly less essential in the ΔLprG, BCG-IV, and Mtb+cure arms,
   consistent with our observations in BCG-SQ vaccinated animals (Fig.
   [140]3b, Supplementary Data [141]1). However, there are quantitative
   differences in the essentiality of these genes across the vaccination
   strategies. When transposon read counts are averaged across all genes
   in the Esx-1 system, there is a significant difference among the
   vaccine strategies at both time points (p = 0.0001 at 2 weeks,
   p = 0.0013 at 4 weeks, Friedman test). There is a greater log[2]
   fold-change value for the BCG-SQ arm compared to Mtb+cure (2 weeks) and
   BCG-IV and Mtb+cure (4 weeks), indicating that Esx-1 genes are less
   required in BCG-SQ vaccinated animals compared to other immune animals
   (Fig. [142]3c). Across genes in the Mce1 system, there is also a
   significant difference in essentiality by condition (p < 0.0001 at 2
   weeks, p = 0.0012 at 4 weeks, Friedman test). A similar trend was seen
   with Mce1 as for Esx-1, with a significantly greater increase in
   transposon counts in BCG-SQ as compared to Mtb+cure (Fig. [143]3c).
   Thus, despite the similar degree of protection conferred by our
   immunization strategies, there are quantitative differences among the
   surviving bacterial transposon mutants, highlighting the potential for
   TnSeq to uncover unique bacterial genetic requirements under distinct
   immune conditions.

Distinct immunization strategies impose differential selective pressures

   Given the differences in essentiality of Esx-1 and Mce1 genes across
   vaccine conditions, we wanted to more comprehensively evaluate how
   memory responses induced by distinct immunization strategies impact
   bacterial genetic requirements. In each immune condition, between 4 and
   41 genes were conditionally more essential, while between 23 and 61
   genes were conditionally less essential (Supplementary Data [144]1).
   The less essential genes were highly overlapping across conditions
   (Fig. [145]4a), and the overlap set was largely comprised of genes in
   the Esx-1 and Mce1 systems (17/23 genes at 2 weeks, 12/14 genes at 4
   weeks, Supplementary Fig. [146]4a). Beyond Esx-1 and Mce1 genes, six
   additional genes were less essential in all conditions at the 2-week
   time point, including the non-ribosomal peptide synthetase nrp/Rv0101;
   pitA/Rv0545c, a low-affinity phosphate importer; Rv0485, which
   regulates a PE/PPE pair; the biotin synthesis gene bioA/Rv1568;
   cpsA/Rv3484; and satS/Rv3311, a protein export chaperone^[147]61. At 4
   weeks, in addition to the Esx-1 and Mce1 genes, nrp/Rv0101 and
   satS/Rv3311 remain significantly less essential. Overall, genes that
   were uniquely less essential in one condition had transposon insertion
   count log[2] fold-change values that trended in the same direction
   across the other immune conditions, but did not meet criteria for
   significance, resulting in high correlation coefficients across these
   genes (Supplementary Fig. [148]4b, c). Together, this suggests a
   conserved signature in the conditionally non-essential genes under
   memory immunity, driven by the decreased requirement for the key
   virulence pathways encoded by Esx-1 and Mce1 genes and other genes
   known to be essential during acute infection, including cpsA/Rv3484 and
   nrp/Rv0101^[149]62,[150]63. However, as shown in UpSet plots, there is
   greater variability in the genes that are significantly more essential
   under individual immune conditions (Fig. [151]4a). When the set of
   genes that are more essential in at least one condition are considered,
   the correlation coefficients among the log[2] fold-change values are
   lower than for the genes that are less essential (Supplementary Fig.
   [152]4d, e). This suggests that the immune pressures experienced by the
   bacteria may vary by immunization strategy.

Fig. 4. Pathway analysis of differentially essential genes under immune
memory conditions.

   [153]Fig. 4
   [154]Open in a new tab

   a UpSet plots of more essential genes in each immunization condition
   relative to naïve mice (upper panels) and less essential genes in each
   immunization condition relative to naïve mice (lower panels). b Bubble
   heatmaps of GSEA analysis of differentially essential genes. Each
   bubble represents the enrichment of a gene set, defined by
   significantly more essential or less essential genes from an in vitro
   condition in MtbTnDb, against the ranked list of log[2](fold-change)
   ratio of the TnSeq immunization condition using the GSEA preranked
   tool. The size of the bubble corresponds to the normalized enrichment
   score (NES); blue bubbles have a positive NES, and orange bubbles have
   a negative NES.

   While resampling provides a rigorous way to identify genes with
   statistically significant differences in transposon insertion count, a
   limitation of this approach is that genes that are close to, but do not
   meet, thresholds for statistical significance are not considered. We
   have previously demonstrated that quantitative differences in
   transposon insertion count are predictive of differences in
   mycobacterial physiology, even if a threshold for statistical
   significance is not met^[155]11. Therefore, to gain more biological
   insight into our data, we performed gene set enrichment analysis (GSEA)
   using the preranked method, ranking all genes from each immunization
   arm of the study by the log[2] fold-change in transposon count relative
   to naïve animals. We then created gene sets from the defined in vitro
   stress conditions from MtbTnDb presented in Fig. [156]2b, where each
   set is comprised of genes that are significantly more or less essential
   under each in vitro condition, and looked for enrichment of these sets
   in our ranked data.

   This analysis finds that for some in vitro conditions, the immune
   conditions have similar gene essentiality profiles, as indicated by a
   similar normalized enrichment score (NES) (Fig. [157]4b). For example,
   all immune conditions have a positive NES of similar magnitude for the
   gene sets more essential for growth in cholesterol media and in minimal
   media, indicating that genes essential for growth in these conditions
   are not required by the bacterium under immune memory. This suggests
   that the metabolic milieu experienced by Mtb is similar across immune
   conditions.

   By contrast, for many of the in vitro stress conditions, including acid
   stress, hypoxia, iron limitation, nitrosative stress, and amino acid
   starvation, the NES varies both quantitatively and qualitatively (i.e.,
   positive vs. negative enrichment) across immune conditions and over
   time. For example, genes that are more essential under in vitro
   nitrosative stress are more essential in BCG-SQ vaccinated mice and
   Mtb+cure mice, as indicated by a large, negative NES at the 2-week time
   point. However, the NES for this same gene set is less negative in
   ΔLprG vaccinated mice and slightly positive in BGC-IV vaccinated
   animals at this time point. Comparing the 2- and 4-week time points,
   the NES for the nitrosative stress gene sets becomes less negative in
   BCG-SQ vaccinated mice, suggesting a reduction in nitrosative stress
   over time, and flips from negative to positive in Mtb+cure animals,
   suggesting an absence of nitrosative stress in this immune condition.
   Taken together, this analysis suggests that TnSeq can provide insight
   into how different immunization strategies achieve bacterial control.

Discussion

   Tuberculosis will remain a global health emergency until a more
   effective vaccine is available. In the quest to develop an improved TB
   vaccine, numerous studies have characterized the host response to BCG
   and new vaccine candidates. Despite intensive study, immune correlates
   of protection remain elusive, and the features that determine bacterial
   control are incompletely characterized. Here, we have investigated this
   question from the bacterial perspective, using a functional genomics
   approach to interrogate bacterial physiology under the selective
   pressure of adaptive immunity induced by vaccination or previous
   infection.

   Notably, we found that genes encoding the canonical virulence factors
   Esx-1 and Mce1 are less essential in mice with immunity induced by
   vaccination or prior Mtb infection. Esx-1 is critical to establishing
   infection, playing myriad roles including inhibition of
   phagosome-lysosome fusion; phagolysosomal permeabilization and
   subsequent stimulation of Type I interferons via the cGAS-STING pathway
   and inflammasome activation via NLRP3; and recruitment of macrophages
   to form the nascent granuloma, providing a niche for bacterial
   replication^[158]38,[159]44,[160]64,[161]65. The importance of these,
   and other, functions of Esx-1 in Mtb pathogenesis is illustrated by
   BCG, which is avirulent because it lacks Esx-1 due to a large genomic
   deletion^[162]66,[163]67. The Esx-1 secreted proteins ESAT-6 and
   CFP-10, encoded by esxA/Rv3875 and esxB/Rv3874, respectively, in
   addition to being virulence determinants, are immunodominant antigens
   in mice and people. We therefore considered whether the relative
   advantage of Esx-1 transposon mutants in BCG-vaccinated mice could
   reflect a lack of antigenic selection, since these genes are absent in
   BCG. However, our finding that Esx-1 genes are also less essential in
   mice vaccinated with the ΔLprG strain, which possesses an intact Esx-1
   system, and in mice previously infected with wild-type H37Rv is not
   consistent with antigenic escape. Rather, our data suggest that one or
   more of the physiologic roles fulfilled by the Esx-1 system are no
   longer required in hosts with adaptive immunity.

   Many of the roles played by Esx-1 are critical during the initial
   stages of infection, hinting that vaccination or prior infection
   recapitulates some aspects of the post-primary infectious milieu, when
   Esx-1 is no longer needed. Consistent with this interpretation, Mce1 is
   also important during the early stages of infection. Mce1 was
   identified as essential during acute Mtb infection in the first in vivo
   transposon mutagenesis study by Sassetti et. al.^[164]9. In this
   seminal work, it was found that the growth defect of Mce1 transposon
   mutants resolved later in infection, when Mce4 genes, which encode a
   cholesterol import system, became essential. This led to the model,
   since experimentally validated, that Mtb consumes fatty acids during
   acute infection and switches to cholesterol as a nutrient source during
   chronic infection^[165]68,[166]69. In our studies, we did not find a
   concomitant increase in the essentiality of Mce4 genes in vaccinated
   mice. In agreement with our data, a TnSeq study using a BCG transposon
   library to challenge BCG-vaccinated cattle also found that genes in the
   Mce1 system are less essential in the context of vaccination, but did
   not observe an increase in the essentiality of Mce4 genes^[167]70.
   These observations suggest that bacteria surviving the memory immune
   response may utilize a nutrient source other than cholesterol, or may
   enter a slowly replicating state where nutritional demands are reduced.

   It is known that the efficacy of antibiotics against Mtb is limited by
   the bacterium’s ability to enter a slow-growing, persistent
   state^[168]71–[169]73. Evidence suggests that persistent bacteria are
   also capable of evading the immune response during chronic
   infection^[170]74, and recent work employing a ribosomal RNA-based
   metric of bacterial replication indicates that BCG vaccination induces
   more rapid bacterial growth arrest^[171]75. Consistent with the model
   that bacteria surviving the immune memory response exist in a
   slow-growing state, our analyses of data from diverse in vitro and in
   vivo Mtb TnSeq studies found that conditions of restricted bacterial
   growth most closely phenocopy the reduced requirement for Esx-1 and
   Mce1 genes we observed in immunized mice. This agrees with a BCG
   transposon mutagenesis study conducted in a chemostat, a bioreactor
   capable of controlling bacterial growth rate via regulated inflow and
   outflow of media, which found that Mce1 genes become non-essential
   under conditions of slow growth, although the essentiality of Esx-1
   genes could not be assessed because BCG lacks an intact Esx-1
   system^[172]76. Intriguingly, our data may also suggest a functional
   link between Mce1 and Esx-1, consistent with comparative genomic
   analyses^[173]77. A plausible model for such a relationship could be
   that Esx-1, which permeabilizes the phagolysosomal
   membrane^[174]78–[175]80, facilitates access to the host-derived fatty
   acids imported by Mce1. This nutrient acquisition may be most critical
   under conditions of rapid bacterial replication, and less essential
   when the bacterial growth rate is reduced.

   Immune correlates of protection following Mtb infection and in the
   context of vaccination are still lacking^[176]50. A key finding of this
   work is that different vaccination strategies induce qualitatively
   distinct bacterial genetic requirements for survival, and these are
   different from the bacterial requirements to survive immunity conferred
   by prior Mtb infection. We interpret the differences in bacterial
   genetic requirements to reflect differences in immune pressures imposed
   by each immunization strategy. Currently, the pipeline for TB vaccine
   development includes a number of candidate LAV^[177]7,[178]49,[179]81,
   with MTBVAC being the most advanced in human clinical trials. Emerging
   data suggest that there are differences in the immunogenicity and
   efficacy of different mycobacterial LAV^[180]82. For example, it has
   been shown that BCG and ΔLprG induce different levels of
   antigen-specific T cells following vaccination in mice^[181]21. In
   particular, ΔLprG, which is more effective than BCG, was shown to
   induce more polyfunctional CD4+ T lymphocyte responses. Our data
   support the notion that LAV will also differentially sculpt the
   subsequent immunological pressures on the bacterium.

   A limitation of this work is that we investigated only one vaccine
   modality, LAV, and we did not systematically investigate the effect of
   vaccination route. Future studies are needed to determine whether other
   vaccine modalities, including vectored vaccines and mRNA vaccines, or
   alternative administration routes, such as intravenous or mucosal
   vaccination, also differentially impact Mtb physiology. Additional
   studies on the timing of induction of specific immune cell subsets
   under various vaccine conditions in conjunction with TnSeq may further
   our understanding of which mycobacterial genes are required to counter
   specific immune responses. Extended studies using targeted bacterial
   genetic mutants and reporter strains may further elucidate how Mtb
   persists in vaccinated hosts and help explain how and why reinfection
   occurs. In the long term, such studies may contribute to rational
   vaccine and adjuvant design.

   An inherent limitation of TnSeq is that a high bacterial dose must be
   delivered via the lateral tail vein for selection in the spleen. This
   dose and route of delivery is necessitated by the complex pool of over
   50,000 individual transposon mutants that would be bottlenecked by a
   more physiologic dose delivered aerogenically. However, several studies
   have shown that genetic requirements identified by TnSeq are generally
   confirmed in aerosol route infections^[182]13,[183]16,[184]17,[185]83.
   Another limitation of TnSeq is that simultaneously profiling host
   immune responses at the site of infection is not possible because the
   tissues are required for bacterial outgrowth and transposon-junction
   sequencing. Future studies evaluating longitudinal peripheral immune
   responses, including immune cell profiling, serum cytokine
   measurements, and antibody responses, may provide additional insight.
   Integrated with TnSeq data, such studies may identify the specific
   effector mechanisms that contribute to the control of Mtb. Performing
   TnSeq studies in vaccinated knock-out mice or mice treated with
   blocking antibodies to diminish the activity of critical arms of the
   adaptive immune response, such as CD4+ and CD8+ T lymphocytes or immune
   signaling, may further elucidate key bacterial virulence mechanisms
   that have evolved to counter these immune pressures.

   Altogether, this work highlights the utility of functional genetic
   screens such as TnSeq under vaccine conditions to identify bacterial
   vulnerabilities that could be targeted in vaccine design. This
   approach, when applied across diverse vaccination modalities, Mtb
   challenge strains, and host genetic backgrounds, may further illuminate
   the complexities of bacterial evasion of host immune responses and
   inform the development of more effective vaccines for tuberculosis.

Methods

Bacterial strains

   Unless otherwise noted, Mtb was cultured in Middlebrook 7H9 salts
   supplemented with 10% oleic acid-albumin-dextrose-catalase (OADC), 0.5%
   glycerol, and 0.05% Tween 80 at 37 °C with agitation, or plated on 7H10
   agar supplemented with 10% OADC, 0.5% glycerol, and 0.05% Tween 80 at
   37 °C. The saturated H37Rv transposon library was previously
   generated^[186]11 and outgrown on 7H10 agar supplemented with 10% OADC,
   0.5% glycerol, 0.05% Tween 80, 0.2% Cas-amino acids (Difco), and
   20 µg/mL kanamycin at 37 °C. BCG was obtained from the Statens Serum
   Institute and prepared as previously described^[187]21. The Esx-1
   mutant is a deletion strain of esxA/Rv3875 that disrupts the entire
   secretion system, and was generated as described^[188]46.

Animals and infections

   Six- to eight-week-old female C57BL/6 mice were purchased from Jackson
   Laboratories (Bar Harbor, ME). Infected animals were housed in
   Biosafety Level 3 facilities under specific-pathogen-free conditions at
   the Harvard T. H. Chan School of Public Health. The protocols,
   personnel, and animal use were approved and monitored by the Harvard
   University Institutional Animal Care and Use Committee.

   For BCG-SQ and ΔLprG vaccinations, animals were anesthetized with
   isoflurane using a precision vaporizer and induction chamber, then
   vaccinated by injecting 100 µL of frozen bacterial culture (2e7 CFU
   total) subcutaneously in the flank. For IV-BCG vaccination experiments,
   mice were briefly restrained in an IACUC-approved restraining tube, and
   2e7 CFU of BCG in 100 µL were injected into the lateral tail vein. Mice
   were rested for 12 weeks post-vaccination prior to infection with the
   transposon library or single-strain infections. For aerosol infections,
   animals were infected with H37Rv using a Glas-Col instrument
   (Terra-Haute, Indiana) to deliver ~100 CFU. After 4 weeks of infection,
   Mtb+cure animals were treated with rifampicin (0.2 g/L) and isoniazid
   (0.1 g/L) in the drinking water for 7 weeks. The antibiotic-containing
   water was changed weekly. Animals were rested for 1 week to ensure no
   carry-over effects of antibiotic selection on the challenge transposon
   library. Transposon library infections were performed by injecting 2e6
   CFU of frozen library in 100 µL in the lateral tail vein as described
   above. Single-strain infections with wild-type H37Rv or the Esx-1
   mutant were performed by injecting 2e6 CFU of mid-log bacterial culture
   in 100 µL in the lateral tail vein as described above. To recover
   tissue for CFU enumeration or transposon-junction sequencing, animals
   were euthanized with an overdose of isoflurane followed by cervical
   dislocation in accordance with the Harvard IACUC-approved protocol and
   AVMA guidelines for the Euthanasia of Animals: 2020 Edition.

Transposon library sequencing and analysis

   Spleens from infected animals were homogenized and plated to recover
   1e6 CFU. After 3 weeks of outgrowth, colonies were scraped and genomic
   DNA extracted. Transposon-junction libraries were prepared from the
   recovered genomic DNA essentially as previously described^[189]11.
   Libraries were sequenced on a HiSeq 2500 at the Tufts University
   Genomics Core, with 100 bp paired-end reads. Read mapping and
   statistical comparisons of read counts between conditions were
   performed using Transit v3.2.7. Reads were normalized with the TTR
   method, insertions in the central 90% of each gene were considered, and
   a LOESS correction was performed. Repetitive regions, such as PE/PPE
   genes, were excluded as previously described^[190]11; excluded genes
   are listed in Supplementary Data [191]1.

Data analysis

   Statistical tests were performed with GraphPad Prism v10.3.0. Pathway
   enrichment analysis was performed with g:Profiler^[192]84 using Sanger
   roles for functional annotation. The Sanger terms “IS1081,” “Phage
   related functions,” “PE_PGRS subfamily,” “PE Subfamily,” “IS6110,” and
   “PPE family” were excluded from analysis, as were any genes excluded
   from Transit analysis, as detailed above. GSEA was performed using the
   preranked tool^[193]85. TnSeq datasets were downloaded from MtbTnDb.app
   on January 25, 2024, and conditions and references are included in