RETRACTED ARTICLE: IspH inhibitors kill Gram-negative bacteria and mobilize immune clearance

Article abstract of DOI:10.1038/s41586-020-03074-x

Isoprenoids are vital for all organisms, in which they maintain membrane stability and support core functions such as respiration¹. IspH, an enzyme in the methyl erythritol phosphate pathway of isoprenoid synthesis, is essential for Gram-negati

Full text of DOI:10.1038/s41586-020-03074-x

Article
IspH inhibitors kill Gram-negative bacteria
and mobilize immune clearance
https://doi.org/10.1038/s41586-020-03074-x
Received: 28 January 2020
Accepted: 11 November 2020
Published online: xx xx xxxx
Check for updates
Kumar Sachin Singh1, Rishabh Sharma1, Poli Adi Narayana Reddy2, Prashanthi Vonteddu1,
Madeline Good2, Anjana Sundarrajan1, Hyeree Choi1, Kar Muthumani1, Andrew Kossenkov3,
Aaron R. Goldman4, Hsin-Yao Tang4, Maxim Totrov5, Joel Cassel6, Maureen E. Murphy2,
Rajasekharan Somasundaram2, Meenhard Herlyn2, Joseph M. Salvino2,6 & Farokh Dotiwala
1
ꢀ✉
ꢀ✉
Isoprenoids are vital for all organisms, in which they maintain membrane stability and
support core functions such as respiration1. IspH, an enzyme in the methyl erythritol
phosphate pathway of isoprenoid synthesis, is essential for Gram-negative bacteria,
mycobacteria and apicomplexans2,3. Its substrate, (E)-4-hydroxy-3-methyl-but-2-enyl
pyrophosphate (HMBPP), is not produced in metazoans, and in humans and other
primates it activates cytotoxic Vγ9Vδ2 T cells at extremely low concentrations4–6
Here we describe a class of IspH inhibitors and refne their potency to nanomolar
levels through structure-guided analogue design. After modifcation of these
.
compounds into prodrugs for delivery into bacteria, we show that they kill clinical
isolates of several multidrug-resistant bacteria—including those from the genera
Acinetobacter, Pseudomonas, Klebsiella, Enterobacter, Vibrio, Shigella, Salmonella,
Yersinia, Mycobacterium and Bacillus—yet are relatively non-toxic to mammalian cells.
Proteomic analysis reveals that bacteria treated with these prodrugs resemble those
after conditional IspH knockdown. Notably, these prodrugs also induce the expansion
and activation of human Vγ9Vδ2 T cells in a humanized mouse model of bacterial
infection. The prodrugs we describe here synergize the direct killing of bacteria with a
simultaneous rapid immune response by cytotoxic γδ T cells, which may limit the
increase of antibiotic-resistant bacterial populations.
As a first line of defence, innate immune cells—such as dendritic cells, microptosis do not develop resistance16. However, M. tuberculosis,
monocytes and macrophages—phagocytose bacteria and present the P. falciparum and the ESKAPE pathogens evade antigen presentation
bacterial antigens on their cell surface using the major histocompat- by killing antigen-presenting cells, by preventing phago-lysosomal
ibility complex7. This antigen presentation initiates the adaptive T and B fusion or by segregating themselves inside different compartments
cell immune response that clears the infected host cells and the bacteria of antigen-presenting cells18. In addition, some antibiotics impair the
within them in 6–30 days8. Antibiotics prevent bacteria from over- function of immune cells19.
whelming the body of the host, while the combined immune responses
Here we introduce a double-pronged antimicrobial strategy through
clear the bacterial infection. A group of six bacteria, known as the the use of dual-acting immuno-antibiotics (DAIAs)20,21. We focus on the
ESKAPE pathogens (Enterococcus faecium, Staphylococcus aureus, Kleb- methyl-D-erythritol phosphate (MEP) pathway for isoprenoid biosyn-
siella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa thesis, whichisessentialforthesurvivalofmostGram-negativebacteria
and Enterobacter species), is the leading cause of multidrug-resistant and apicomplexans (malaria parasites) (Fig. 1a) but is absent in humans
nosocomial infections9. In addition, multidrug-resistant strains of and other metazoans2,3. The first line of attack in the DAIA strategy
Mycobacterium tuberculosis and Plasmodium falciparum are also targets the MEP enzyme IspH, which metabolizes HMBPP into isopen-
global public health threats10,11. Rare mutations and the acquisition of tenyl pyrophosphate (IPP) and dimethylallyl pyrophosphate (DMAPP).
antibiotic-resistance genetic elements give rise to bacterial cells that IPP and DMAPP are building blocks for downstream terpenoids, which
resist antibiotics through various mechanisms, including modifica- areessentialforproteinprenylation, thesynthesisofpeptidoglycancell
tion of the antibiotic target, the secretion of inactivating enzymes, the walls and the production of quinones for respiration1,22. The Escherichia
use of drug efflux pumps and metabolic bypass12–14. It has previously coli strain CGSC 8074 (here termed ΔispH) conditionally expresses
been reported that natural killer and cytotoxic T cells deliver gran- E. coli IspH in the presence of 0.5% arabinose; however, the addi-
zymes to within bacteria or protozoan parasites, where they disrupt tion of glucose to the medium shuts down IspH expression, which
several essential systems and induce a mechanism of programmed is lethal to this strain3 (Fig. 1b). The lack of IspH causes a build-up of
pathogen death known as microptosis15–17. Bacteria that are undergoing HMBPP—a bacterial pathogen-associated molecular pattern—which
1Vaccine and Immunotherapy Center, The Wistar Institute, Philadelphia, PA, USA. 2Program in Molecular and Cellular Oncogenesis, The Wistar Institute, Philadelphia, PA, USA. 3Bioinformatics
Facility, The Wistar Institute, Philadelphia, PA, USA. 4Proteomics and Metabolomics Facility, The Wistar Institute, Philadelphia, PA, USA. 5Molsoft, San Diego, CA, USA. 6Molecular Screening and
✉
Protein Expression Facility, The Wistar Institute, Philadelphia, PA, USA. e-mail: jsalvino@wistar.org; fdotiwala@wistar.org
Nature | www.nature.com | 1

Article
***
a
c
b
25
***
CGSC 8074
WT
Non-mevalonate Mevalonate pathway
100
80
60
40
20
0
10–6
UI
HMBPP + glucose
E. coli
(MEP) pathway
Acetyl CoA
0
3
10–5
10–4
10–3
10–2
+
Glyceraldehyde-3P
+ pyruvate
2.4
10
12
6
*
*
Acetoacetyl CoA
Methyl erythritol
phosphate
γ9 TCR
0
Mevalonate
***
***
***
100
***
0
7
0.4
1.2
0.2
0
Mevalonate-5PP
100
***
HMBPP
25
57
30
0
0.2
Perforin
IPP
Glucose (%)
IspH
+
0
100
***
DMAPP
1.1
2
77
14
88
11
14
0.5%
0%
0%
0.5%
Arabinose
Glucose
Terpenoids
0.6
0
HLA-DR
Cell-wall synthesis
Ubiquinone synthesis
d
g
e
f
0.4
100
80
60
40
20
0
IspH (nM)
100
50
10
0.2
0
25
0
0
12.5
0
10
20
30
40
60
–0.002
0.002
0
1,000
[HMBPP] (μM)
2,000
Time (min)
1/[HMBPP] (μM)
P. falciparum
P. aeruginosa
M. tuberculosis
IspH (nM)
3
2
1
0
3
3
0
50
2
1
0
2
1
0
100
500
1,000
2,000
5,000
5
25
45
65
5
25
45
65
5
25
45
65
Time (min)
Time (min)
Time (min)
Fig. 1 | Testing IspH as a target for the DAIA strategy. a, IspH is an essential
enzyme of the MEP pathway (found in Gram-negative bacteria, mycobacteria
and apicomplexan parasites), in which it produces IPP and DMAPP from
HMBPP. IspH is absent from the mevalonate pathway (found in humans and
complex metazoans). b, Left and bottom, the E. coli strain CGSC 8074 produces
IspH in the presence of arabinose but not in the presence of glucose. Right,
conditional knockdown of IspH, achieved by decreasing the levels of arabinose,
reduces bacterial viability as measured by CFU assay (n = 3 biological and 3
technical replicates). Data are mean s.e.m. c, Left, human PBMCs co-infected
with wild-type E. coli or CGSC 8074 (ΔispH) E. coli were analysed for expansion
of CD3+Vγ9TCR+ (γδ) T cells after 24 h and compared to uninfected (UI) or
HMBPP-treated PBMCs (top). Gated γδ T cell populations were analysed for the
cytotoxic granule proteins granzyme A and perforin (middle) or cell surface
markers of T cell activation CD69 and HLA-DR (bottom). Data are
representative of 4 independent experiments (4 donors). Right, the percentage
of Vγ9+ T cells from the CD3+ population and the percentage of Vγ9+ T cells with
increased expression of granzyme A, perforin, CD69 and HLA-DR. Data are
mean s.e.m. ***P < 0.001; one-way ANOVA. d, Activity of IspH in the presence
of different concentrations of HMBPP, as measured by the methyl viologen
assay at 30 min. Related to Extended Data Fig. 1d, e. e, Lineweaver–Burk double
reciprocal plot showing the activity of E. coli IspH at different concentrations of
the enzyme and its substrate HMBPP. f, Activity of 50 nM E. coli IspH over time
in the presence of 1 mM HMBPP. g, The activities of purified recombinant IspH
from P. falciparum, P. aeruginosa or M. tuberculosis LytB2. For d–g, n = 3
biological replicates with 8 technical replicates. Data are mean s.e.m.
stimulates the Vγ9Vδ2 T cells to expand and produce the cytotoxic IspH from P. falciparum, P. aeruginosa and M. tuberculosis using this
proteins perforin, granulysin and granzymes, all of which are impor- methyl viologen assay (Fig. 1g).
tant for microptosis4–6. This forms the second line of attack in the DAIA
We next performed a molecular docking study using the crystal
strategy, and was demonstrated either by treating human peripheral structure of E. coli IspH (Protein Data Bank (PDB) ID 3KE8)28. The
blood mononuclear cells (PBMCs) with HMBPP + IL2 or by infecting HMBPP-binding pocket was modelled (Methods) and the atomic
them with CGSC 8074 in the presence of glucose. Under both condi- property field established (Extended Data Fig. 2a) for the automated
tions, a greater expansion of γδ T cells (Fig. 1c, top) and higher levels molecular docking of 9.6 million compounds. The 168 best-scoring
of the cytotoxic proteins perforin and granzyme A (Fig. 1c, middle) compounds (Extended Data Fig. 2b) were visually compared to HMBPP.
and the T cell surface activation markers CD69 and HLA-DR (Fig. 1c, The top 24 compounds (denoted C1–24)—that is, those with lower bind-
bottom) were observed in comparison with PBMCs infected with ing energies and atomic property field scores than HMBPP (Extended
wild-type (BL21) E. coli. An IspH inhibitor will therefore kill bacteria Data Fig. 2c)—were further evaluated in terms of their chemical and
directly, as with other antibiotics, but will also kill persistent bacteria drug-like properties; the three-dimensional conformations of the
by microptosis15,16,23,24
.
docked ligand−IspH complex were also assessed (Extended Data Fig. 3a,
Supplementary Fig. 2a). Analysis by methyl viologen assay revealed C10,
C17 and C23 as the best inhibitors of E. coli IspH—with half-maximal
inhibitory concentrations (IC50) of 9 μM, 4 nM and 85 nM, respectively
Molecular docking and biochemical activity
We purified recombinant IspH proteins from several bacterial spe- (Fig. 2a, Supplementary Table 1)—whereas the assessment of IspH activ-
cies—E. coli, M. tuberculosis and P. aeruginosa—and from the malaria ity over time showed that C17 and C23 were more stable inhibitors
parasite P. falciparum (Extended Data Fig. 1a, b). The activity of IspH of E. coli, M. tuberculosis, P. aeruginosa and P. falciparum IspH than
is coupled to a system that reduces the oxidized iron–sulfur cluster25,26 was C10 (Fig. 2b, Extended Data Fig. 3b). Although both C17 and C23
[4Fe-4S]2+. In vitro, this reduction can be achieved chemically using were found to be potent inhibitors of IspH from different pathogens
sodium dithionite-reduced methyl viologen (Extended Data Fig. 1c), (Fig. 2c, Supplementary Table 2), we tested several analogues of each
after which IspH activity is determined by the proportional change in of C10, C17 and C23 to improve their potency against purified E. coli
the UV absorbance (398 nm) of oxidized methyl viologen27. Using E. coli IspH (Fig. 2d, Extended Data Fig. 3c–e, Supplementary Table 3, Supple
-
IspH, we determined optimal concentrations of 50 nM IspH and 1 mM mentary Fig. 2b–d). C23 analogues showed substantial improvement
HMBPP and an optimal reaction time of 30 min (Fig. 1d–f, Extended (that is, lower IC50 values) in terms of E. coli IspH inhibition compared
Data Fig. 1d, e). We also measured the activities of purified recombinant with the parent compound, whereas C10 and C17 analogues did not.
2 | Nature | www.nature.com

a
b
E. coli IspH
100
50
0
C1
C2
C3
C4
C5
C6
C7
C8
C9
C17
C18
C19
C20
C21
C22
C23
C24
100
50
0
DMSO
C10
C17
C10
C11
C12
C13
C14
C15
C16
C23
C23
C17
0
1
2
3
4
5
0
50
100
Time (min)
log[Drug] (nM)
c
d
C23
C23.07
C23.20
C23.21
C23.28
C23.47
100
M. tuberculosis
100
P. aeruginosa
P. falciparum
100
50
0
100
50
0
C17
C23
50
0
50
0
0
1
2
3
4
5
0
1
2
3
4
5
0 1 2 3 4 5
log[Drug] (nM)
0
1
2
3
4
5
log[Drug] (nM)
log[Drug] (nM)
log[Drug] (nM)
e
Concentration (μM)
TPP
C23.07 + TPP
MIC90 < 62 μM
C23.20 + TPP
MIC90 < 4 μM
C23.21 + TPP
MIC90 < 4 μM
C23.28 + TPP
MIC90 < 4 μM
500
250
125
62
31
16
8
16
16
8
16
8
16
8
16
8
8
0
0
0
0
0
4
0
0
240
Time (min)
480
0
240
Time (min)
480
0
240 480
Time (min)
0
240
Time (min)
480
0
240
Time (min)
480
Fig. 2 | Inhibition of purified IspH and the killing of bacteria by IspH
inhibitors. a, Dose–response (nonlinear regression) curves for the inhibition
of E. coli IspH by compounds C1–24, determined by the methyl viologen assay.
The IC50 values (Supplementary Table 1) were calculated from the respective
curves. Data are mean s.e.m. Associated with Extended Data Fig. 1d, e.
b, Activity of E. coli IspH pre-treated with DMSO (control), C10, C17 and C23
over time. Data are mean s.e.m. Associated with Extended Data Fig. 3b.
c, Inhibition of M. tuberculosis, P. aeruginosa and P. falciparum IspH at varying
concentrations of C17 or C23. The IC50 values were calculated from the
respective curves and are given in Supplementary Table 2. Data are mean s.e.m.
d, Dose–response (nonlinear regression) curves for the inhibition of E. coli IspH
by C23 analogues, as measured by the methyl viologen assay. The IC50 values
were calculated from the respective curves and are provided in Supplementary
Table 3. Data are mean s.e.m. Associated with Extended Data Fig. 3e. e, The
killing of E. coli by TPP-linked analogues of the prodrug C23 (C23.07, C23.20,
C23.21, C23.28; Supplementary Fig. 2e) compared to TPP-treated control,
as illustrated by dynamic growth curves. The delivery of the prodrugs into
bacteria and their cleavage into the active form is shown in Extended Data
Fig. 4e, f. For a–d, e, n = 3 biological replicates and 8 technical replicates.
Compared to the IspH substrate HMBPP, the compounds C23.20 and Analysis of the lysates of prodrug-treated bacteria by mass spectrom-
C23.21—the two most potent C23 analogues—show improved binding etry detected both the delivery of the prodrug molecule C23.28–TPP
(a lower dissociation constant, KD) to purified E. coli IspH according to into E. coli and its subsequent cleavage into C23.28 and TPP (Extended
surface plasmon resonance analysis (Extended Data Fig. 4a). By testing Data Fig. 4e, f). Notably, the inhibition of E. coli IspH by C23.28 pre-
different C23 analogues, we established a structure–activity relation- vented the conversion of HMBPP to DMAPP and IPP, whereas treatment
ship (Fig. 2d, Extended Data Figs. 3e, 4b) and confirmed C23.07, C23.20, with TPP alone had no effect on this process (Extended Data Fig. 4g, h).
C23.21, C23.28 and C23.47 as the most potent inhibitors of E. coli IspH.
The levels of IspH in the E. coli strain CGSC 8074 can be regulated by
changing the amount of arabinose in the culture medium (Extended
Data Fig. 5a). Increasing IspH levels in this manner increased the dose of
C23.28–TPPthatwasrequiredtokillCGSC8074(ExtendedDataFig.5b,c).
Bacterial killing by prodrugs
Because C23 and its analogues are not bacteria-permeable, we cou- We tested several C23 derivatives on drug-resistant clinical isolates of
pled them to triphenyl phosphonium to aid in the penetration of Vibrio cholerae using the resazurin blue assay and the colony-forming
membranes29. However, because cations are efficiently effluxed out unit (CFU) assay, and determined their MIC90 values (Extended Data
of Gram-negative bacteria by transporters—such as AcrAB–TolC in Fig. 5d–f). Whereas TPP alone did not kill V. cholerae, the prodrugs
E. coli—we designed prodrugs from which the negatively charged IspH C23.20–TPP, C23.21–TPP and C23.28–TPP had the lowest MIC90 values
inhibitor would be released once inside the bacteria. We synthesized of 16 μM (8 μg ml−1), followed by C23.07–TPP (MIC90 = 125 μM; 63 μg
ester prodrugs from the C23.47 analogue by linking it to a lipophilic cat- ml−1) and C23.47–TPP (MIC90 = 63 μM; 31 μg ml−1). The MIC90 values for
ion (6-hydroxyhexyl triphenylphosphonium bromide (TPP), a lipophilic these compounds in several species of antibiotic-resistant bacteria
alcohol (ethanol) or a basic amine (3-(dimethylamino)propan-1-ol) are shown in Supplementary Table 4. In summary, the IspH inhibitor
(Supplementary Fig. 3). Similar strategies involving the use of prodrugs prodrugs had lower MIC90 values in multidrug-resistant clinical iso-
with cleavable ester bonds have been shown to facilitate drug deliv- lates of Enterobacter aerogenes, A. baumanii, P. aeruginosa, V. cholerae
ery into bacteria30. We found that the C23.47 + TPP ester was the most and K. pneumoniae than the current best-in-class antibiotics, includ-
potent against E. coli, with a MIC90 value—the minimum drug concentra- ing meropenem (a member of the carbapenem class), amikacin and
tion at which 90% bacteria are killed—of 4 μM (Extended Data Fig. 4c, d). tobramycin (aminoglycosides), ciprofloxacin (a fluoroquinolone)
We therefore focused on the TPP ester form of C23 analogues as well as ceftriaxone, cefepime and ceftaroline (third, fourth and
(Supplementary Fig. 2e). C23.20–TPP, C23.21–TPP and C23.28–TPP fifth generation cephalosporins, respectively (Fig. 3, Supplementary
showed the best activity against E. coli (MIC90 < 4 μM) (Fig. 2e). Table 5).
Nature | www.nature.com | 3

Article
a
E. aerogenes (UCI 15)
b
E. aerogenes (UCI 15)
The half-lives (t1/2) of the prodrugs C23.28–TPP and C23.21–TPP were
40 and 56 min in human plasma, 218 and 245 min in pig plasma, and
20 and 21 min in mouse plasma, respectively (Extended Data Fig. 7a).
Similarly, t1/2 values in presence of liver microsomes were 27 and
48 min (human plasma), 25 and 24 min (monkey plasma), and 24 and
41 min (mouse plasma), respectively (Extended Data Fig. 7b). The disap-
pearance of the prodrug forms coincided with the appearance of the
respective parent drugs. Although our prodrugs showed low toxicity
in the mammalian cell lines HepG2, RAW264.7 and Vero (Extended
Data Fig. 7c), lipophilic triphenylphosphonium cations are reported
Drug (μM)
500
250
125
63
31
16
8
16
6
4
2
0
31
63
4
V. cholerae (MO45)
V. cholerae (MO45)
500
250
125
63
31
16
8
6
4
2
0
to cause mitochondrial proton leak and toxicity in C2C12 myoblasts32
.
4
Furthermore, the human hERG gene (KCNH2) is a known target for
lipophilic cations such as TPP33. However, our 6-hydroxyhexyl TPP
carrier molecule and our prodrugs were neither toxic to C2C12 cells
nor caused loss of mitochondrial membrane potential (Extended Data
Fig. 7d, e). Additionally, C23.28–TPP, methyl TPP (Me-TPP) and the
carrier molecule showed tenfold higher IC50 values (5–10 μM) than
verapamil in hERG electrophysiological profiling experiments using
an automated QPatch HTX assay (Extended Data Fig. 7f).
Notably, we found that the prodrug C23.28–TPP reduces IspH levels in
E. coli and in clinical isolates of several antibiotic-resistant bacteria
(Extended Data Fig. 8a, b). We next performed proteomics analysis on
E. coli treatedwithC23.28–TPPandonCGSC8074(ΔispH)inthepresenceof
glucose. Out of 2,346 proteins, 525 showed similar changes after treatment
with C23.28–TPP and after IspH knockdown (Extended Data Fig. 8c, d).
Among the downregulated proteins, 323 (22%) were common to drug treat-
ment and conditional IspH knockdown (Extended Data Fig. 8e). Pathway
analysis34 showed enrichment of the electron transport chain complexes,
of ubiquinone, and of other pathways (Extended Data Figs. 8f–h, 9).
K. pneumoniae (1.53)
K. pneumoniae (1.53)
500
250
125
63
31
16
8
6
4
2
0
4
A. baumannii (AB5075-UW)
A. baumannii (AB5075-UW)
500
250
125
63
31
16
8
6
4
2
0
4
P. aeruginosa (MRSN 5524)
P. aeruginosa (MRSN 5524)
500
250
125
63
31
16
8
6
4
2
0
4
Dual action leads to γδ T cell response
The activation of human γδ T cells does not require epitope presenta-
tion by the major histocompatibility complex or by CD1 receptors.
Instead, the butyrophilin receptors BTN2A1 and BTN3A1 on target cells
act to detect phosphoantigens such as HMBPP35,36 and as a direct ligand
for the Vγ9Vδ2 T cell receptor, respectively37,38. The treatment of E. coli
with C23.07–TPP, similar to the conditional IspH-knockdown strain
CGSC 8074, resulted in activation of Vγ9Vδ2 T cells within 24–48 h
(Fig. 1c, Extended Data Fig. 10a), and the activated cells showed high
levels of cytotoxic markers such as perforin and granulysin, as well as
high levels of the T cell surface activation markers CD69 and HLA-DR.
We observed similar results after the treatment of Mycobacterium
smegmatis- or V. cholerae-infected PBMCs with C23.07–TPP (Extended
Data Fig. 10b). By contrast, kanamycin-treated and TPP-treated samples
did not show γδ T cell activation. Whereas E. coli and V. cholerae were
resistant to kanamycin, our prodrug C23.07–TPP could effectively kill
both bacteria (Extended Data Fig. 10c). To assess for resistance to IspH
inhibitors, we grew clinical isolates of V. cholerae and K. pneumoniae for
18 serial passages with the prodrug C23.28–TPP in the presence or in the
Fig. 3 | Analogues of the prodrug C23 have lower MIC90 values against
multidrug-resistant clinical isolates of Gram-negative bacteria than
best-in-class antibiotics. a, b, The prodrugs C23.20–TPP, C23.21–TPP and
C23.28–TPP—as well as a range of current best-in-class antibiotics—were tested
against pan-resistant or multidrug-resistant clinical isolates of E. aerogenes,
V. cholerae, K. pneumoniae, A. baumannii and P. aeruginosa using the resazurin
blue assay (a) and CFU plating (b) after 24 h treatment (n = 3 biological
replicates). For the resazurin blue assay, pink indicates bacterial growth and
blue indicates no bacterial growth. TPP was used as a negative control, and
uninfected culture medium was used as a positive control. Data are mean of
three independent experiments s.e.m., with individual data points shown.
*P < 0.05, **P < 0.01, ***P < 0.001, the remainder are not significant; two-tailed
paired Student’s t-test. Associated with Supplementary Table 5.
Specificity, mechanism and toxicity
Isoprenoids are required in Gram-negative bacteria and in M. tuberculo- absence of human PBMCs. To demonstrate the critical role of Vγ9Vδ2
sis for respiration and for synthesis of the cell wall1,31. Using a Seahorse T cell activation and expansion in the efficacy of these prodrugs, PBMCs
XF Analyzer we found that prodrug-treated E. coli cells show a signifi- depleted in γδ T cells were also used in the serial passaging. The efficacy
cant decrease in oxygen consumption rate (aerobic respiration) and of γδ T cell depletion is reflected in the lack of Vγ9Vδ2 T cell expansion
extracellular acidification rate (glycolysis) compared with untreated after 6 days of treatment with HMBPP and IL-15 (Extended Data Fig. 10d).
cells (Extended Data Fig. 6a, b). This was accompanied by increased In the absence of PBMCs, both V. cholerae and K. pneumoniae developed
levels of superoxide and hydrogen peroxide16 (Extended Data Fig. 6c, d). resistance to the prodrug C23.28–TPP as well as to conventional antibi-
Prodrug-treated bacteria showed a dose-dependent loss of membrane otics (V. cholerae to hygromycin and K. pneumoniae to streptomycin)
integrity (as assessed by SYTO 9 and propidium iodide staining) and of (Extended Data Fig. 10e, f, top). However, in the presence of human
membrane potential, whereas treatment with TPP alone had no effect PBMCs, neither V. cholerae nor K. pneumoniae developed resistance to
(Extended Data Fig. 6e–h). Scanning and transmission electron micro- C23.28–TPP (Extended Data Fig. 10e, f, bottom). Passaging V. cholerae
graphs showed spherocyte formation, cell-membrane protrusions, and and K. pneumoniae in γδ-T-cell-depleted human PBMCs significantly
defects in the cell wall and in the periplasm of prodrug-treated E. coli diminished the dual action of the prodrug, supporting the relevance of
and V. cholerae, and in conditional IspH knockdown E. coli cells (strain γδ T cells in its mechanism of action. Owing to the lack of reliable in vivo
CGSC 8074) (Extended Data Fig. 6i, j).
γδ-depleting antibodies, we used E. coli infection in NSG mice (instead
4 | Nature | www.nature.com

p
a
b
f
g
C23.07
–TPP
Spleen
Liver
Lung
Kidney
Brain
Blood
TPP
3
2
1
4
2
0
4
2
4
2
***
*
***
100
**
*
***
*
*
NS
*
10
78
0
0
4
2
0
γδ
–
+
−
+
−
+
−
+
−
+
−
+
100
***
***
*
14
26
22
63
88
96
Spleen
1.24
Liver
2.90
Lung
2.95
Kidney
Brain
Blood
0
0.61
1.01
1.22
0
0
4
2
5
100
4
4
3
*
**
0
2
55.1
74.7
71.0
7.10
15.7
22.8
0
0
4
0
***
***
100
5
4
4
Vγ9 TCR
***
3
0
***
***
Li*ve*r*
c
d
Spleen
Lung
Kidney
Brain
Blood
2
100
2
0
4
2
0
100
***
***
0
0
13
81
70
98
5
4
***
***
0
***
4
3
***
0
γδ
100
2
−
+
−
+
−
+
−
+
−
+
−
+
0
0
0
e
C23.07–TPP
TPP
E. coli
7
100
***
Vγ9
5
3
50
0
TPP
C23.07–TPP
h
i
Spleen
Lung
Liver
E. aerogenes
***
100
5
3
1
5
3
1
5
3
1
1
2
3
4
5
6
***
0
2
4
6
***
***
***
Time (d)
Time (d)
TPP
E. coli
C23.28–TPP
TPP
100
50
0
***
C23.28–TPP
Meropenem
50
0
7
5
3
Kidney
Brain
TPP
5
5
3
1
TPP
C23.28–TPP
***
C23.28–TPP
Meropenem
***
3
1
***
0
2
4
1
2 3 4 5 6
0
2
4
6
Time (d)
Time (d)
Time (d)
Fig. 4 | γδ T cell activation in prodrug-treated, bacteria-infected PBMCs
different organs at the experimental endpoint measured as CFU mg−1 (f).
g, Left, CD3+Vγ9TCR+ T cell expansion in E. coli-infected humanized mice, treated
with TPP or C23.07–TPP for five days after infection. Right, the percentage of
Vγ9+ T cells from CD3+ cells in each organ is shown. Associated with Extended
Data Fig. 11a. For d–g, n = 6 mice, 3 technical replicates. Data are mean s.e.m.
***P < 0.001, **P < 0.01, *P < 0.05; two-tailed unpaired Student’s t-test, relative to
TPP-treated mice. h, i, BALBc mice were infected with E. aerogenes, treated with
10 mg kg−1 TPP, C23.28–TPP or meropenem and monitored for survival (h)
and i, Enterobacter load (CFU mg−1) (i) (n = 19 mice, 3 technical replicates). Data are
mean s.e.m. ***P < 0.001, **P < 0.01, *P < 0.05, NS, not significant; one-way ANOVA,
relative to TPP-treated or meropenem-treated mice.
and humanized mice. a, Escherichia coli load (CFU mg−1) in the organs of NSG
mice injected with γδ depleted (γδ−) or undepleted (γδ+) human PBMCs,
infected with E. coli and treated with 1 mg kg−1 C23.28–TPP for 3 days. b, CD3+
Vγ9TCR+ T cell expansion in γδ− or γδ+ NSG mice, four days after infection.
c, Percentage of CD3+ cells that are also Vγ9+ in each organ. For a–c, n = 10 mice,
3 technical replicates. Data are mean s.e.m. *P < 0.05, **P < 0.01, ***P < 0.001,
NS, not significant; two-tailed unpaired Student’s t-test, relative to γδ− mice.
d–f, Humanized mice infected with E. coli were treated with C23.07–TPP
(d, e, top; f, left) or C23.28–TPP (d, e, bottom; f, right) and monitored daily for
survival (d), bacteraemia in terms of CFU per ml of blood (e) and E. coli load in
of humanized mice) to corroborate the role of Vγ9Vδ2 T cells in vivo. We not the αβ T cells (Extended Data Fig. 10i). Humanized mice that were
injected one group of NSG mice with human PBMCs and another group infected with E. coli and treated with prodrugs treated showed lower
with ex-vivo γδ-T-cell-depleted human PBMCs. Both groups of mice levels of bacterial CFU in circulation as well as improved survival com-
were infected with 107 E. coli cells (Fig. 4a–c) and the levels of γδ T cells pared with mice that were treated with TPP alone (Fig. 4d, e). Similarly,
were monitored by fluorescence-activated cell sorting. After infection, prodrug-treated humanized mice showed significantly lower bacte-
mice in both the γδ-T-cell-depleted and the undepleted groups were rial load and expansion of Vγ9Vδ2 T cells in several organs than their
given suboptimal doses (1 mg kg−1) of C23.28–TPP, to minimize the kill- TPP-treated counterparts (Fig. 4f, g). We confirmed both the expan-
ing of bacteria by direct antibiotic action and increase the prominence sion of Vγ9Vδ2 T cells and the lower bacterial burden in the tissues of
of the immune effect. Mice with γδ-T-cell-depleted PBMCs had 2–10-fold prodrug-treated humanized mice by immunofluorescence microscopy
higher CFU (Fig. 4a) and significantly lower levels of γδ T cells (Fig. 4b, c) (Extended Data Fig. 11). Finally, we found that the prodrug C23.28–TPP
than their counterparts with undepleted PBMCs.
was able to clear infection by a clinically isolated multidrug-resistant
As a final test, we assessed the direct bactericidal effects of IspH strain of E. aerogenes (UCI 15) and significantly improve the survival
prodrugs in C57BL/6 mice infected with V. cholerae. Mice treated with of infected BALBc mice, whereas meropenem—a current best-in-class
C23.28–TPP showed significantly lower mortality and had a lower bacte- carbapenem antibiotic—did not (Fig. 4h, i).
rial load in all organs tested compared to those treated with TPP alone
(Extended Data Fig. 10g, h). Because mouse γδ T cells do not respond
to HMBPP39,40, we used humanized mice in experiments to test the dual
Discussion
action of IspH prodrugs. After the injection of HMBPP into human- The family of antibiotics and the antimicrobial strategy that we report
ized mice, we saw rapid expansion of the human Vγ9Vδ2 T cells but here synergize direct antibiotic action with rapid immune response.
Nature | www.nature.com | 5

Article
16. Walch, M. et al. Cytotoxic cells kill intracellular bacteria through granulysin-mediated
delivery of granzymes. Cell 157, 1309–1323 (2014).
17. Dotiwala, F. et al. Killer lymphocytes use granulysin, perforin and granzymes to kill
intracellular parasites. Nat. Med. 22, 210–216 (2016).
18. Finlay, B. B. & McFadden, G. Anti-immunology: evasion of the host immune system by
bacterial and viral pathogens. Cell 124, 767–782 (2006).
19. Yang, J. H. et al. Antibiotic-induced changes to the host metabolic environment inhibit
drug efficacy and alter immune function. Cell Host Microbe 22, 757–765.e3 (2017).
20. Chiang, C. Y. et al. Mitigating the impact of antibacterial drug resistance through
host-directed therapies: current progress, outlook, and challenges. MBio 9, e01932-17
(2018).
21. Oldfield, E. & Feng, X. Resistance-resistant antibiotics. Trends Pharmacol. Sci. 35,
664–674 (2014).
22. Marakasova, E. S. et al. Prenylation: from bacteria to eukaryotes. Mol. Biol. (Mosk.) 47,
717–730 (2013).
23. Workalemahu, G. et al. Metabolic engineering of Salmonella vaccine bacteria to boost
human Vγ2Vδ2 T cell immunity. J. Immunol. 193, 708–721 (2014).
24. Dieli, F. et al. Granulysin-dependent killing of intracellular and extracellular
Mycobacterium tuberculosis by Vγ9/Vδ2 T lymphocytes. J. Infect. Dis. 184, 1082–1085
(2001).
25. Wolff, M. et al. Isoprenoid biosynthesis via the methylerythritol phosphate pathway: the
(E)-4-hydroxy-3-methylbut-2-enyl diphosphate reductase (LytB/IspH) from Escherichia
coli is a [4Fe-4S] protein. FEBS Lett. 541, 115–120 (2003).
This dual mechanism of action, an inherent feature of these com-
pounds, could delay the emergence of drug resistance15–17 Our DAIA
prodrugs are bacteria-permeable and are more effective against several
species of multidrug-resistant bacteria than the current best-in-class
antibiotics. They act specifically on IspH, show low toxicity to mam-
malian cells (specifically to myoblasts, with MIC90 values 10–100 times
higher than in bacteria) and high IC50 values against hERG channels32.
Unlike antibiotics derived from natural sources, no IspH inhibitors have
been discovered in microorganisms, so it is less likely that resistance
mechanisms—such as β-lactamases and macrolide esterases in the case
of β-lactam and macrolide antibiotics—have evolved specifically against
our prodrugs. Future experiments should investigate the potential
mechanisms of resistance to IspH inhibitors. In addition, the synergy
between γδ T cells activated by these prodrugs and other immune cells
merits further study.
Online content
26. Rohdich, F. et al. The deoxyxylulose phosphate pathway of isoprenoid biosynthesis:
studies on the mechanisms of the reactions catalyzed by IspG and IspH protein.
Proc. Natl Acad. Sci. USA 100, 1586–1591 (2003).
27. Xiao, Y., Chu, L., Sanakis, Y. & Liu, P. Revisiting the IspH catalytic system in the
deoxyxylulose phosphate pathway: achieving high activity. J. Am. Chem. Soc. 131,
9931–9933 (2009).
Any methods, additional references, Nature Research reporting sum-
maries, source data, extended data, supplementary information,
acknowledgements, peer review information; details of author contri-
butions and competing interests; and statements of data and code avail-
ability are available at https://doi.org/10.1038/s41586-020-03074-x.
28. Gräwert, T. et al. Probing the reaction mechanism of IspH protein by X-ray structure
analysis. Proc. Natl Acad. Sci. USA 107, 1077–1081 (2010).
29. Nazarov, P. A. et al. Mitochondria-targeted antioxidants as highly effective antibiotics.
Sci. Rep. 7, 1394 (2017).
30. Ortmann, R. et al. Acyloxyalkyl ester prodrugs of FR900098 with improved in vivo
anti-malarial activity. Bioorg. Med. Chem. Lett. 13, 2163–2166 (2003).
31. Søballe, B. & Poole, R. K. Microbial ubiquinones: multiple roles in respiration, gene
regulation and oxidative stress management. Microbiology 145, 1817–1830 (1999).
32. Trnka, J., Elkalaf, M. & Anděl, M. Lipophilic triphenylphosphonium cations inhibit
mitochondrial electron transport chain and induce mitochondrial proton leak. PLoS ONE
10, e0121837 (2015).
1.
Odom, A. R. Five questions about non-mevalonate isoprenoid biosynthesis. PLoS Pathog.
7, e1002323 (2011).
Jomaa, H. et al. Inhibitors of the nonmevalonate pathway of isoprenoid biosynthesis as
antimalarial drugs. Science 285, 1573–1576 (1999).
McAteer, S., Coulson, A., McLennan, N. & Masters, M. The lytB gene of Escherichia coli is
essential and specifies a product needed for isoprenoid biosynthesis. J. Bacteriol. 183,
7403–7407 (2001).
2.
3.
33. Wingrove, D. E. & Gunter, T. E. Kinetics of mitochondrial calcium transport. II. A kinetic
description of the sodium-dependent calcium efflux mechanism of liver mitochondria
and inhibition by ruthenium red and by tetraphenylphosphonium. J. Biol. Chem. 261,
15166–15171 (1986).
34. Schwartz, A. S., Yu, J., Gardenour, K. R., Finley, R. L., Jr & Ideker, T. Cost-effective strategies
for completing the interactome. Nat. Methods 6, 55–61 (2009).
35. Wang, H. et al. Butyrophilin 3A1 plays an essential role in prenyl pyrophosphate
stimulation of human Vγ2Vδ2 T cells. J. Immunol. 191, 1029–1042 (2013).
36. Sandstrom, A. et al. The intracellular B30.2 domain of butyrophilin 3A1 binds
phosphoantigens to mediate activation of human Vγ9Vδ2 T cells. Immunity 40, 490–500
(2014).
37. Rigau, M. et al. Butyrophilin 2A1 is essential for phosphoantigen reactivity by γδ T cells.
Science 367, eaay5516 (2020).
38. Karunakaran, M. M. et al. Butyrophilin-2A1 directly binds germline-encoded regions of the
Vγ9Vδ2 TCR and is essential for phosphoantigen sensing. Immunity 52, 487–498.e6 (2020).
39. Wei, H. et al. Definition of APC presentation of phosphoantigen (E)-4-hydroxy-
3-methyl-but-2-enyl pyrophosphate to Vγ2Vδ2 TCR. J. Immunol. 181, 4798–4806 (2008).
40. Mogues, T., Goodrich, M. E., Ryan, L., LaCourse, R. & North, R. J. The relative importance
of T cell subsets in immunity and immunopathology of airborne Mycobacterium
tuberculosis infection in mice. J. Exp. Med. 193, 271–280 (2001).
4. Rhodes, D. A. et al. Activation of human γδ T cells by cytosolic interactions of BTN3A1
with soluble phosphoantigens and the cytoskeletal adaptor periplakin. J. Immunol. 194,
2390–2398 (2015).
5.
Chien, Y. H., Meyer, C. & Bonneville, M. γδ T cells: first line of defense and beyond. Annu.
Rev. Immunol. 32, 121–155 (2014).
6.
Chen, Z. W. Multifunctional immune responses of HMBPP-specific Vγ2Vδ2 T cells in
M. tuberculosis and other infections. Cell. Mol. Immunol. 10, 58–64 (2013).
Alberts, B. et al. Molecular Biology of the Cell 4th edn (Garland Science, 2002).
Lieberman, J. in Fundamental Immunology 7th edn (ed. Paul, W. E.) Ch. 37 (Lippincott,
Williams & Wilkins, 2012).
7.
8.
9.
Magill, S. S. et al. Multistate point-prevalence survey of health care-associated infections.
N. Engl. J. Med. 370, 1198–1208 (2014).
10. WHO. Tuberculosis https://www.who.int/en/news-room/fact-sheets/detail/tuberculosis
(World Health Organization, 2020).
11. WHO. Artemisinin resistance and artemisinin-based combination therapy efficacy https://
www.who.int/docs/default-source/documents/publications/gmp/who-cds-gmp-2019-
17-eng.pdf?ua=1 (World Health Organization, 2019).
12. Wright, G. D. Bacterial resistance to antibiotics: enzymatic degradation and modification.
Adv. Drug Deliv. Rev. 57, 1451–1470 (2005).
13. Li, X. Z., Plésiat, P. & Nikaido, H. The challenge of efflux-mediated antibiotic resistance in
Gram-negative bacteria. Clin. Microbiol. Rev. 28, 337–418 (2015).
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
14. Wilson, D. N. Ribosome-targeting antibiotics and mechanisms of bacterial resistance.
Nat. Rev. Microbiol. 12, 35–48 (2014).
15. Dotiwala, F. et al. Granzyme B disrupts central metabolism and protein synthesis in
bacteria to promote an immune cell death program. Cell 171, 1125–1137.e11 (2017).
© The Author(s), under exclusive licence to Springer Nature Limited 2020
6 | Nature | www.nature.com

Animal Care and User Committee (IACUC). The humanized mice
(Hu-mice) were generated by R. Somasundaram in the Herlyn labora-
Methods
No statistical methods were used to predetermine sample size. The tory and transferred over to the Dotiwala laboratory. NOD/LtSscidIL-
experiments were not randomized and the investigators were not 2Rgnull (NSG) mice were inbred at The Wistar Institute under licence
blinded to allocation during experiments and outcome assessment.
from the Jackson Laboratory. For humanization, fetal liver and thymus
were obtained from the same donor (18–22 weeks of gestation). Female
NSG mice (6–8 weeks) received a thymus graft (1 mm3) in the sub-renal
Molecular docking studies
The IspH–HMBPP complex with PDB ID 3KE8 was used for the virtual capsule 24 h after myeloablation using busulfan (30 mg kg−1, intraperi-
screening28. The protein was prepared using standard automated toneally (i.p.); Sigma-Aldrich, B2635). This was immediately followed by
protocols embedded in MolSoft’s (Internal Coordinate Mechanics) the injection of autologous liver-derived CD34+ haematopoietic stem
ICM-Pro software41,42. Hydrogen atoms were added to the structure, cells (105 cells per mouse, intravenously (i.v.)) that were magnetically
and considerations were made regarding the correct orientation of sorted by microbeads conjugated with anti-human CD34 (Miltenyi,
Asn and Gln side chains, ligand and protein charges, histidine orienta- 130-046-703)46. Six to eight weeks (>50 days) later, the presence of
tion and protonation state and any crystallographic quality flags such human immune cells was monitored by multi-colour flow cytometry
as high b-factors or low occupancy. All waters and heteroatoms were using an 18-colour BD LSR II Analyzer (BD Biosciences)47. NSG mice
deleted except for the iron–sulfur complex. Virtual screening of the with human PBMCs were generated by i.v. injection of human PBMCs
MolCart chemical database (http://www.molsoft.com/screening.html, or PBMCs depleted of all γδ T cells using Anti-TCRγ/δ Microbead Kit
v.2017, containing around 9.6 million chemicals) was undertaken using (Miltenyi, 130-050-701). About 107 cells per mouse were injected every
MolSoft’s ICM-VLS software43,44. The binding site was represented by 3 days, for a total of 3 doses per mouse and the presence of human
five types of interactions to create a potential docking map: (i) van der immune cells was monitored by multi-colour flow cytometry. An equal
Waals potential for a hydrogen atom probe; (ii) van der Waals potential number of male and female C57BL/6 or BALBc mice were obtained
for a heavy-atom probe (generic carbon of 1.7 Å radius); (iii) optimized from the Jackson Laboratory and used for mouse models of Vibrio or
electrostatic term; (iv) hydrophobic terms; and (v) loan-pair-based Enterobacter infection, respectively. Mice were housed in plastic cages
potential, which reflects directional preferences in hydrogen bonding. on an ad libitum diet and maintained on a 12-h light/12-h dark cycle at
The energy terms are based on the all-atom vacuum force field ECEPP/3 22ꢀ°C at 60% humidity. Controls and experimental groups were age- and
and conformational sampling is based on the ICM biased probability genotype-matched non-littermates. Both initial infection and drug
Monte Carlo (BPMC) procedure42. This method randomly selects a treatments were administered by i.p. or i.v. routes. Infected mice were
conformation in the internal coordinate space and then makes a step monitored twice daily for survival and distress. To monitor bacteraemia,
to a new random position independent of the previous one but accord- mice were bled daily from tail nicks. At the end of the experiment mice
ing to a predefined continuous probability distribution followed by were euthanized by CO2 inhalation and their spleens, livers, kidneys,
local minimization.
lungs and brains were collected for CFU and flow cytometry analysis.
A hit list of 37,849 chemicals was obtained and this was filtered down
to a set of 168 chemicals recommended for experimental testing using
Human samples
the following criteria: (1) low van der Waals interaction energy; (2) Human PBMCs were obtained from the Human Immunology Core of the
low ICM docking score; (3) similarity between the 3D atomic property University of Pennsylvania (UPenn) under UPenn protocol 705906 (PI:
fields of the pharmacophore and the substrate45; and (4) number of Riley) ‘Pre-clinical studies of the Human Immune System’. The donors
hydrogen-bond acceptors in the phosphate-binding region.
of the PBMCs provided informed written consent for the use of their
samples. De-identified specimens were transferred to the Wistar Insti-
tute under Wistar protocol 21906321, reviewed and approved by the
Bacteria
Escherichia coli BL21(DE3) from New England Biolabs was used as a Wistar Institutional Review Board. PBMCs were washed in PBS counted
model strain. Clinical isolates of E. aerogenes (CRE) (UCI 15), K. pneu- and kept in plastic culture plates in RPMI medium containing 10%
moniae 1.53 (ST147+, CTX-M15+), Salmonella enterica subsp. enterica human serum. Human cell lines (HepG2, Vero, RAW264.7 and C2C12)
serovar Typhimurium (LT2 – SL7207), V. cholerae (M045), A. baumannii were obtained from the American Type Culture Collection (ATCC),
(BC-5), A. baumannii (AB5075-UW), P. aeruginosa (PA14 and MRSN 5524), authenticated by short tandem repeat profiling and PCR assays with
Helicobacter pylori (Hp CPY6081), Shigella flexneri (2457T), Bacillus species-specific primers and were confirmed to be free of mycoplasma
sphaericus (CCM 2177), M. tuberculosis (M. tuberculosis H37Ra) and contamination.
Yersinia pestis (KIM 10+) were obtained from BEI Resources. The condi-
tional IspH knockdown E. coli strain CGSC 8074 (ΔispH), was obtained
Antibodies
from the Coli Genetic Stock Center at Yale University. All strains were The following antibodies were used in this study.
cultured at 37ꢀ°C in their respective medium (2.5% brain heart infusion
Antibodies for western blotting and immunohistochemistry (unless
agar, Middlebrook 7H10 with OADC, Luria Bertani (LB), tryptic soy agar, otherwise mentioned, dilution: primary antibody, 1:50; secondary
5% blood agar, Columbia agar (BD Difco, BD 241830, BD 262710, BD antibody, 1:200): anti-E. coli antibody (Abcam, ab137967); anti-E. coli
244610, BD 236950; and Fisher, R01217, R02030) based on the vendors’ IspH rabbit polyclonal antibody (Genscript, generated in this study,
recommendation. LB medium with 0.5% arabinose (Sigma, A3256) was dilution 1:100,000); anti-E. coli RNA Sigma 70 mouse antibody (Bio-
used to culture the CGSC 8074 (ΔispH) strain. Changing arabinose and Legend, 63208); secondary-biotinylated rabbit anti-rat IgG (Vector
glucose concentrations (0.5–0.05%) in the LB medium enabled us to Laboratories, BA-4001); mouse IgG HRP-linked whole antibody (GE
modulate IspH protein levels in CGSC 8074. For testing the antibiotic Healthcare, NA931V); rabbit IgG HRP-linked whole antibody (GE Health-
sensitivity, bacteria were grown in RPMI medium containing 10% fetal care, NA934V); biotinylated goat anti-rabbit IgG antibody (Vector Labo-
bovine serum or human serum.
ratories, BA-1000); donkey anti-rabbit IgG AF-488 (BioLegend, 406416).
Antibodies for fluorescence-activated cell sorting (FACS; dilution
1:100): anti-CD3-PerCP-Cy5.5 (clone UCHT1, BD Biosciences, 560835);
Animal models
All studies were carried out in accordance with the recommendations anti-CD4-Alexa Fluor 700 (clone RPA-T4, BD Biosciences, 557922);
in the Guide for the Care and Use of Laboratory Animals of the National anti-CD8a-Brilliant Violet 711 (clone RPA-T8, BioLegend, 301044);
Institutes of Health (NIH). All animal experiments were performed anti-TCR vg9-FITC (clone 7A5, Invitrogen, TCR2720) (or anti-TCR vd2
according to protocols approved by the Wistar Institute’s Institutional (clone B6, BioLegend, 331402) with anti-mouse IgG-AF647 (Invitrogen,

Article
A21236)); anti-CD107a (LAMP-1)-Brilliant Violet 510 (clone H4A3, BioLe- NSG mice injected with human PBMCs were given a suboptimal
gend, 328632); anti-CD69-PE/Cy7 (clone FN50, BD Biosciences, 557745); (1 mg kg−1) dose of C23.28–TPP through the i.v. route, once a day for
anti-HLA-DR-Brilliant Violet 421 (clone L243, BioLegend, 307636); 4 days. Blood from infected mice was collected daily using tail snips
anti-CD38-Brilliant Violet 510 (clone HIT2, BD Biosciences, 563251); and analysed for bacteremia by CFU and flow cytometry for γδ T cell
anti-CD25-Alexa Fluor 647 (clone BC96, BioLegend, 302618).
expansion. After death from infection or euthanasia at the end of the
Antibodies for FACS compensation (dilution 1:200): anti CD3 mouse experiment, the spleen, liver, lungs, brain and kidneys were collected,
monoclonal PE/Dazzle 594 (BioLegend, 317346); anti CD3 mouse mono- sectioned and studied for bacterial CFU, immunohistochemistry or
clonal APC (BioLegend, 300412); anti CD3 mouse monoclonal APC Cy7 flow cytometry as indicated.
(BioLegend, 300317); anti CD3 mouse monoclonal BV711 (BioLegend,
344838); anti CD3 mouse monoclonal PE (BioLegend, 300408); anti
CD3 mouse monoclonal PE Cy7 (BioLegend, 300316).
Isolation of cells and bacteria from different organs
Samples of mouse spleen, liver, lung, brain and kidney were weighed
Anti-E. coli IspH antibody generation: the control sera (2–3 ml) were and crushed in 12-well plastic tissue culture plates using a 5-ml syringe.
collected from the ear pinna of the rabbit before the start of immuni- RBCs were lysed in RBC lysis (ACK) buffer at 37ꢀ°C and for 1 min. Cells
zation. The 200 μg of purified E. coli IspH protein was mixed with the were washed 3–5 times with MACS buffer at 4ꢀ°C. Cells were then either
KLH conjugate and Freud’s complete adjuvant and injected subcuta- analysed by flow cytometry or lysed in distilled, deionized water and
neously into the rabbit (2–4 site per rabbit) in the animal facility at serial dilutions of samples were plated for bacterial CFU on medium
Genscript. The second immunization was performed 14 days after the plates respective to the bacteria studied.
first immunization with 200 μg purified protein, KLH conjugate and
Freud’s incomplete adjuvant. One week after the second immunization,
Ex vivo infection in human PBMCs
the test sera (first test bleed) were collected from the rabbit to test the Human PBMCs were washed in medium (10% human serum RPMI
antibody titration by ELISA and western blot. The third immunization, medium supplemented with 100 U ml−1 penicillin G and 100 μg ml−1
with 200 μg purified protein, KLH conjugate and Freud’s incomplete streptomycin sulfate, 6 mM HEPES, 1.6 mM L-glutamine, 50 mM
adjuvant was performed 14 days after the first test bleed. One week 2-mercaptoethanol) then cultured in medium without penicillin or
later the second test bleed was performed, and sera were purified for streptomycin in 6-, 12-, 24- or 96-well Primaria plates (Fisher Scientific,
IgG antibodies using a protein A column. The purified IgG antibodies 08-772). Escherichia coli, V. cholerae, K. pneumoniae or M. tuberculosis
were used for the confirmation of anti-IspH antibody production by ex vivo infections were induced at a multiplicity of infection of 1:0.1, 1:1,
ELISA and western blot. After confirmation, that antibody was raised in 1:10 or 1:100. Various dilutions of 100 mM stock solutions of prodrugs
rabbits, the production bleed was performed, the sera were separated, C23.07, C23.28–TPP or TPP (control) were added to sample wells to give
and antibodies were purified using a protein A column. The purified a final working concentration range from 500 μM to 4 μM. Infected
anti-E. coli IspH rabbit polyclonal antibody was validated by western PBMC samples were analysed at 24, 48 or 72 h by flow cytometry or
blots using purified ispH protein from E. coli, P. aeruginosa, M. tuber- lysed in distilled water at different time points where indicated and the
culosis and P. falciparum. The antibody was further validated using lysates were used for CFU analysis. The Vγ9Vδ2 T cells in uninfected
lysates of A. baumannii, S. flexneri, S. enterica, V. cholerae and H. pylori. PBMCs show low initial levels of perforin, probably because of the
length of time spent in culture (up to 72 h).
Depletion of γδ T cells from human PBMCs
CFU analysis
The γδ T cells were separated from human PBMCs using anti-TCRγ/δ
Microbead Kit (Miltenyi, 130-050-701). After Ficoll separation the Bacterial cultures treated with different prodrugs or antibiotics, or
human PBMCs were washed and resuspended in RPMI medium con- lysates from infected mouse blood, tissues or infected ex vivo human
taining human serum. The cells were counted, pelleted at 300g for PBMCs were serially diluted and 50 μl was plated on bacterial culture
10 min and resuspended in 40 μl of MACS buffer for every 107 cells. The plates. The plates were incubated at 37ꢀ°C and counted after overnight
cells were incubated with 10 μl of anti-TCR γ/δ hapten-antibody per incubation (after 20 days for M. tuberculosis colonies). The CFU were
107 cells, at 4–8ꢀ°C for 10 min. After incubation, 30 μl MACS buffer and normalized per ml for blood or per mg weight for tissues. For all experi-
20 μl of MACS anti-hapten MicroBeads-FITC per 107 cells were added ments, at least three independent experiments were performed, with
followed by further incubation at 4–8ꢀ°C for 15 min. The cells were 3–8 technical replicates in each experiment.
washed with 1–2 ml of MACS buffer per 107 cells and centrifuged at 300g
for 10 min. The supernatant was removed, and the cells resuspended in
Recombinant IspH cloning and expression
500 μl MACS buffer per 108 cells. The sample was loaded on the MACS IspH gene sequences from E. coli, Pseudomonas, Plasmodium and
buffer-rinsed LS column and was kept in the magnetic field. The cells M. tuberculosis (LytB2) were optimized for expression in E. coli
in the flow through were collected and the column washed three times and synthesized by Genscript. These sequences were cloned in a
with 3 ml MACS buffer. The cells in the flow through and washes were pET24a-KAN vector and co-expressed with iron–sulfur cluster (isc)
combined, pelleted and resuspended in RPMI + 10% human serum and proteins (encoded in the pACYC184 plasmid) in Nico (DE3) cells (NEB,
counted for further experiments.
C2529H)48. Transformed Nico (DE3) cells were grown at 37ꢀ°C in Ter-
rific Broth (12 g tryptone, 24 g yeast extract, 5 ml glycerol per litre of
broth) supplemented with sterile monopotassium phosphate (23.1 g l−1),
Mouse infection studies
In experiments with Hu-mice or NSG mice injected with human PBMCs, dipotassium phosphate (125.4 g l−1), ferric ammonium citrate (35 mg l−1),
infection was induced by injecting 107 E. coli per mouse i.p. in 200 μl Dul- L-cysteine (1 mM) and the antibiotics kanamycin (50 mg l−1) and chloram-
becco phosphate buffered saline (DPBS). In experiments with C57BL/6 phenicol (35 mg l−1). At an optical density at 600 nm (OD600) of 0.6–0.7,
mice, 106 V. cholerae and in experiments with BALBc mice, 5 × 104 E. aero- IspH production was induced by adding IPTG at 1 mM concentration
genes (UCI 15) were injected i.p. After 24 h, prodrugs the C23.07–TPP or before overnight incubation at 25ꢀ°C.
C23.28–TPP (where mentioned), or just the carrier molecule TPP, (10 mg
per kg per mouse) in 1% DMSO–DPBS solution were injected i.p. (or i.v. in
IspH purification
case of E. aerogenes-infected BALBc mice) once a day for 1–2 weeks, until After IspH induction, bacteria were spun down at 6,000g and washed
mice succumbed to infection or were euthanized for tissue analysis, as three times with 50 ml degassed PBS. All subsequent steps were per-
indicated. A group of E. aerogenes-infected mice were given meropenem formed in an anaerobic glove box at 0.5 ppm O2. After the final wash,
(10 mg per kg per mouse) for comparison to a best-in-class antibiotic. the bacteria were resuspended in 20 ml degassed lysis buffer (25 mM

Tris, 1M KCl, 5% glycerol, cOmplete protease inhibitor cocktail (Sigma, compounds using a vacuum manifold maintained at approximately
4693132001), 5 mM sodium dithionite, pH 7.5). The rest of the proce- 1 Torr. All yields reported are isolated yields. Preparative reversed-phase
dure was carried out under anaerobic conditions (<0.5 ppm O2) in an high pressure liquid chromatography (RP-HPLC) was performed using
mBraun glovebox. Bacteria were lysed by freeze-thawing five or six a Gilson GX-271 semi-prep HPLC, eluting with a binary solvent system
times in liquid nitrogen. Nucleic acids were eliminated by incubating A and B using a gradient elution (A, H2O with 0.1% trifluoroacetic
with 500 units of Benzonase (Sigma E1014) at room temperature for acid (TFA); B, CH3CN with 0.1% TFA) with UV detection at 220 nm.
30 min. The lysate was spun down at 6,000g and filtered through a Low-resolution mass spectral (MS) data were obtained on a Waters
0.45-μm filter under anaerobic conditions (<0.5 ppm O2). The lysate ACQUITY QDa LC–MS mass spectrometer with UV detection at 254 nm.
was incubated for 2–3 h at room temperature with 3–5 ml Ni-NTA resin Proton nuclear magnetic resonance (1H NMR) spectra were obtained
(Qiagen, 30230) that had been equilibrated in lysis buffer. The Ni-NTA on a Bruker Avance II 400 (400 MHz) spectrometer. Chemical shifts (δ)
resin was washed with 3 column volumes of wash buffer 1 (25 mM Tris, are reported in parts per million (ppm) relative to residual undeuter-
1 M KCl, 5% glycerol, cOmplete protease inhibitor cocktail, 30 mM ated solvent as an internal reference. The following abbreviations are
imidazole, pH 7.5) and 1 column volume of wash buffer 2 (25 mM Tris, used for the multiplicities: s, singlet; d, doublet; t, triplet; q, quartet;
0.1 M KCl, 5% glycerol, cOmplete protease inhibitor cocktail, 30 mM dd, doublet of doublets; dt, doublet of triplets; td, triplet of doublets;
imidazole, pH 7.5). The protein was eluted from Ni-NTA using 15 ml tt, triplet of triplets; ddd, doublet of doublet of doublets; m, multiplet;
elution buffer (25 mM Tris, 0.1 M KCl, 5% glycerol, cOmplete protease br, broad.
inhibitor cocktail, 300 mM imidazole, pH 7.5). The eluted protein was
passed through a 5-ml bed of chitin resin to remove contaminating pro-
teins and then passed in tandem through Sepharose SP (GE Healthcare,
Synthesis of (6-hydroxyhexyl)triphenylphosphonium bromide
(TPP)
17072910) and Sepharose Q (GE Healthcare, 17051010) resin beds. The To a stirred solution of 6-bromohexan-1-ol (5.0 g, 27.61 mmol) in 70 ml
protein was eluted from the Q column using the Q column elution buffer of acetonitrile at room temperature was added triphenylphosphine
(25 mM Tris, 1 M KCl, 5% glycerol, pH 7.5) desalted using Econo-Pac (7.967 g, 30.37 mmol), and the reaction mixture was heated under reflux
10DG (Bio-Rad, 732-2010) desalting columns and concentrated using for 48 h under a nitrogen atmosphere. Completion of the reaction was
Amicon Ultra 10k spin columns.
confirmed by thin layer chromatography (TLC). The solvent was evapo-
rated under reduced pressure, the crude product was washed with
ethanol (2 × 30 ml), and the solid was dried under high vacuum without
Methyl viologen assay
All solutions were degassed by boiling before use and the assays were further purification to afford the title compound (0.95 mmol) as a white
performed under <0.5 ppm O2 in a glove box. To monitor the activity solid. The product was confirmed by 1H NMR and liquid chromatog-
of IspH, methyl viologen was used as the reducing agent. The oxidation raphy coupled to mass spectrometry (LC–MS). 1H NMR (400 MHz,
of methyl viologen (blue to colourless) was followed by measuring CDCl3) δ 7.92–7.75 (m, 9H), 7.71 (td, J = 7.5, 3.4 Hz, 6H), 3.87–3.71 (m, 2H),
the loss of absorption at 398 nm. The assay solution contained 50 mM 3.63 (t, J = 5.4 Hz, 2H), 1.77–1.56 (m, 4H), 1.51 (d, J = 2.9 Hz, 4H). Mass
Tris–HCl (pH 8), 1 mM methyl viologen and 0.5 mM sodium dithionite spectrometry: m/z: calcd for [C24H28OP]+ ([M]+), 363.19; found, 363.16
in a total volume of 100 μl in 96-well flat bottom plastic plates. Varying (Supplementary Fig. 3a).
concentrations of IspH (0–5 μM) and HMBPP (0–1.25 mM) were titrated
and optimal concentrations of 50 nM IspH and 1 mM HMBPP were used
for subsequent experiments. After reduction of methyl viologen with
Synthesis of (6-hydroxyhexyl)triphenylphosphonium bromide
esters
sodium dithionite an approximate absorbance of 3 was reached. The 4-(Naphthalen-2-yl)-4-oxobutanoic acid, 4-(naphthalen-1-yl)-
reactions were initiated by the addition of IspH. For inhibition studies, 4-oxobutanoic acid and 4-(2,5-dimethylphenyl)-4-oxobutanoic
varying concentrations of candidate drugs (1 nM–250 μM) or DMSO acid were used for the synthesis of (6-(4-(naphthalen-2-yl)-
(negative control) were added. The plates were sealed by Parafilm, 4-oxobutanoyloxy)hexyl)triphenylphosphonium bromide
incubated at 37ꢀ°C and the absorbance at 398 nm was read every 5 min (C23.20–TPP), (6-(4-(naphthalen-1-yl)-4-oxobutanoyloxy)
in a Biotek Synergy 2 plate reader. The activity was expressed as micro- hexyl)triphenylphosphonium bromide (C23.21–TPP) and
moles of HMBPP consumed per second, as measured by the decrease in (6-(4-(2,5-dimethylphenyl)-4-oxobutanoyloxy)hexyl)triphenylphos-
absorbance at 398 nm. Samples that lacked HMBPP served as a baseline phonium bromide (C23.28–TPP) respectively (Supplementary
negative control. The assay is linear with respect to time and protein Fig. 3b–d). To a stirred solution of the respective aryl-4-oxobutanoic
concentration.
acid (about 0.3 g, 1.31 mmol), (6-hydroxyhexyl)triphenylphospho-
nium bromide (0.583 g, 1.31 mmol) and N,N-dimethylpyridin-4-amine
(DMAP; 0.176 g, 1.58 mmol) in anhydrous CH2Cl2 (15 ml) at 0ꢀ°C was
Surface plasmon resonance
Approximately 30,000 RU of purified recombinant His-tagged E. coli added dicyclohexylcarbodiimide (0.271 g, 1.45 mmol) under a nitrogen
IspH was immobilized onto a Ni-NTA surface plasmon resonance (SPR) atmosphere. Then the reaction mixture was brought to room tempera-
chip activated by N-(3-dimethylaminopropyl)-N′-ethyl carbodiimide ture and stirred for 16 h. Completion of the reaction was confirmed by
hydrochloride (EDC) and N-hydroxysuccinimide (NHS). The remain- TLC. The reaction mixture was then cooled to −10ꢀ°C and the insolu-
ing binding sites were blocked with 1 M ethanolamine at pH 8.5. Test ble material was filtered off. The solid was washed with cold (−10ꢀ°C)
compounds C23.20, C23.21 and HMBPP were serially diluted to 1:3.16 CH2Cl2. The combined organic layer was then washed with aqueous
starting at 100 μM final concentration in running buffer (10 mM HEPES, 1 M HCl (15 ml), water (15 ml), saturated aqueous NaHCO3 (15 ml) and
pH 7.4, 150 mM NaCl, 0.05% Tween20, 5% DMSO) and run on a Biacore saturated aqueous NaCl (15 ml), and then dried over anhydrous Na2SO4.
T200 instrument at a flow rate of 50 μl min−1, to reduce the mass trans- The solvent was evaporated under reduced pressure and the crude
port limitation effects.
product was purified by silica gel flash column chromatography using
5–10% MeOH in CH2Cl2 to afford the title compound, (about 0.687 g,
1.05 mmol). The products were confirmed by 1H NMR and LC–MS as
General chemistry
All reactions were conducted under an inert gas atmosphere (nitrogen follows (Supplementary Fig. 3b–d):
or argon) using a Teflon-coated magnetic stir bar at the temperature
(6-(4-(Naphthalen-2-yl)-4-oxobutanoyloxy)hexyl)triphenylphospho-
indicated. Commercial reagents and anhydrous solvents were used nium bromide (C23.20–TPP): 1H NMR (400 MHz, CDCl3) δ 8.50 (s, 1H),
without further purification. Solvents were removed using a rotary 8.03–7.93 (m, 2H), 7.87 (ddd, J = 12.6, 5.5, 3.3 Hz, 7H), 7.81–7.73 (m, 3H),
evaporator, and residual solvent was removed from non-volatile 7.73–7.65(m,5H),7.64–7.49(m,2H),4.12–4.00(m,2H),3.99–3.84(m,2H),

Article
3.44 (t, J = 6.6 Hz, 2H), 2.79 (t, J = 6.6 Hz, 2H), 1.72–1.49 (m, 6H), 1.36 pressure. The resulting crude product was purified by silica gel flash
(dt, J = 15.0, 7.5 Hz, 2H). Mass spectrometry: m/z: calcd for [C38H38O3P]+ chromatography (ethyl acetate/hexane) to afford the title compound
([M]+), 573.26; found, 573.21.
(406 mg, 1.68 mmol) as a white solid. The product was confirmed by
(6-(4-(Naphthalen-1-yl)-4-oxobutanoyloxy)hexyl)triphenylphos- NMR and LC–MS as follows (Supplementary Fig. 3g): 1H NMR (400 MHz,
phonium bromide (C23.21–TPP): 1H NMR (400 MHz, CDCl3) δ 8.55–8.48 DMSO) δ 14.33 (s, 2H), 8.82 (s, 1H), 8.20 (d, J = 8.0 Hz, 1H), 8.10–7.93
(m, 1H), 8.02–7.90 (m, 2H), 7.89–7.80 (m, 7H), 7.75 (tt, J = 12.0, 5.3 Hz, (m, 3H), 7.76–7.58 (m, 2H), 7.23 (s, 1H). Mass spectrometry: m/z: calcd
3H), 7.71–7.61 (m, 6H), 7.56–7.45 (m, 3H), 4.13–4.01 (m, 2H), 3.97–3.83 for [C14H11O4]+ ([M+H]+), 243.07; found, 243.14.
(m, 2H), 3.40–3.30 (m, 2H), 2.86–2.75 (m, 2H), 1.73–1.51 (m, 6H), 1.36
(dt, J = 15.0, 7.5 Hz, 2H). Mass spectrometry: m/z: calcd for [C38H38O3P]+
Synthesis of 3-(dimethylamino)propyl-4-(naphthalen-2-yl)-
2,4-dioxobutanoate (C23.47–DAP)
([M]+), 573.26; found, 573.31.
(6-(4-(2,5-Dimethylphenyl)-4-oxobutanoyloxy)hexyl)triphenylphos- To a stirred solution of 4-(naphthalen-2-yl)-2,4-dioxobutanoic acid
phonium bromide (C23.28–TPP): 1H NMR (400 MHz, CDCl3) δ 7.87 (ddd, (100 mg, 0.41 mmol) in anhydrous CH2Cl2 (7 ml) at 0ꢀ°C was
J = 12.6, 5.2, 3.3 Hz, 6H), 7.81–7.74 (m, 3H), 7.73–7.63 (m, 6H), 7.48 (s, 1H), added 3-(dimethylamino)propan-1-ol (0.62 mg, 64 mmol), tri-
7.17 (dd, J = 7.8, 1.2 Hz, 1H), 7.10 (d, J = 7.8 Hz, 1H), 4.04 (t, J = 6.5 Hz, 2H), ethylamine (125 mg, 1.24 mmol), DMAP (65 mg, 0.54 mmol) and
3.97 – 3.84 (m, 2H), 3.18 (t, J = 6.5 Hz, 2H), 2.69 (dd, J = 11.8, 5.4 Hz, 2H), 2-chloro-1-methylpyridinium iodide (127 mg, 0.49 mmol), and stirred
2.38 (s, 3H), 2.35 (s, 3H), 1.76–1.50 (m, 6H), 1.35 (dt, J = 15.0, 7.6 Hz, 2H). for 1 h at 0ꢀ°C. Completion of the reaction was confirmed by TLC. The
Mass spectrometry: m/z: calcd for [C36H40O3P]+ ([M]+), 551.27; found, reaction mixture was diluted with cold water and the product was
551.21.
extracted with CH2Cl2 (10 ml × 2). The combined organic layers were
2,4-Dioxo-4-phenylbutanoic acid and 4-(naphthalen-2-yl)- washed with aqueous 1 M HCl (10 ml), aqueous NaHCO3 (10 ml) and
2,4-dioxobutanoic acid were used for the synthesis of (6-(2,4-dioxo brine (10 ml), and then dried over anhydrous Na2SO4, The solvent was
-4-phenylbutanoyloxy)hexyl)triphenylphosphonium bromide (C23.07– evaporated under reduced pressure and the crude product was puri-
TPP) and (6-(4-(naphthalen-2-yl)-2,4-dioxobutanoyloxy)hexyl)triphe- fied by silica gel flash chromatography (ethyl acetate/hexane) to afford
nylphosphonium bromide (C23.47–TPP), respectively (Supplementary the title compound (81 mg, 0.25 mmol) as a white solid. The product
Fig. 3e, f). To a stirred solution of the respective aryl-2,4-dioxobutanoic confirmed by NMR and LC–MS as follows (Supplementary Fig. 3h):
acid (about 200 mg, 1.04 mmol) and (6-hydroxyhexyl)triphenylphos- 1H NMR (400 MHz, CDCl3) δ 12.15 (s, 1H), 8.51 (d, J = 53.1 Hz, 1H), 8.12–7.79
phonium bromide (461 mg, 1.04 mmol) in anhydrous CH2Cl2 (15 ml) at (m, 4H), 7.73–7.46 (m, 2H), 7.25 (s, 1H), 4.46 (t, J = 5.9 Hz, 2H), 3.26 (dd,
0ꢀ°C was added triethylamine (316 mg, 3.12 mmol), DMAP (165 mg, 1.35 J = 21.8, 14.1 Hz, 2H), 2.92 (s, 6H), 2.47–2.20 (m, 2H). Mass spectrometry:
mmol) and 2-chloro-1-methylpyridinium iodide (319 mg, 1.25 mmol) m/z: calcd for [C19H22NO4]+ ([M+H]+), 328.38; found, 328.15.
and stirred for 2 h at 0ꢀ°C. Completion of the reaction was confirmed by
Synthesis of ethyl esters
TLC. The reaction mixture was diluted with cold water and the product
was extracted with CH2Cl2 (20 ml × 2). The combined organic layer was The synthetic steps were identical to those for the synthesis of
washed with aqueous 1 M HCl (15 ml), aqueous NaHCO3 (15 ml) and (6-hydroxyhexyl)triphenylphosphonium bromide esters described
brine (15 ml), and was then dried over anhydrous Na2SO4. The solvent above. Ethanol was used for esterification in place of (6-hydroxyhexyl)
was evaporated under reduced pressure, and the crude product was triphenylphosphonium bromide. To a stirred solution of the respective
purified by silica gel flash chromatography (ethyl acetate/hexane) to aryl-2,4-dioxobutanoic acid (about 100 mg, 0.52 mmol) in anhydrous
afford the title compound (about 321 mg, 0.5 mmol) as a thick liquid. CH2Cl2 (8 ml) at 0ꢀ°C was added ethanol (72 mg, 1.56 mmol), trieth-
The product was confirmed by NMR and LC–MS as follows (Supple- ylamine (158 mg, 1.56 mmol), N,N-dimethylpyridin-4-amine (DMAP;
mentary Fig. 3b-d):
(6-(2,4-Dioxo-4-phenylbutanoyloxy)hexyl)triphenylphospho-
83 mg, 0.68 mmol) and 2-chloro-1-methylpyridinium iodide (159 mg,
0.62 mmol) and stirred for 1 h at 0ꢀ°C. Completion of the reaction was
1
nium bromide (C23.07–TPP): H NMR (400 MHz, CDCl3) δ 15.29 confirmed by TLC. The reaction mixture was diluted with cold water
(s, 1H), 8.06–7.96 (m, 2H), 7.93–7.82 (m, 6H), 7.78 (dt, J = 7.3, 3.6 Hz, 2H), and the product was extracted with CH2Cl2 (10 ml × 2). The combined
7.73–7.65 (m, 6H), 7.62 (dd, J = 10.5, 4.2 Hz, 1H), 7.52 (t, J = 7.6 Hz, 2H), 7.06 organic layers were washed with aqueous 1 M HCl (10 ml), aqueous
(s, 1H), 4.33–4.23 (m, 2H), 4.01–3.88 (m, 2H), 1.84–1.53 (m, 6H), 1.49–1.33 NaHCO3 (10 ml) and brine (10 ml), and then dried over anhydrous
(m, 2H). Mass spectrometry: m/z: calcd for [C34H34O4P]+ ([M]+), 537.22; Na2SO4. The solvent was evaporated under reduced pressure and the
found, 537.31.
crude product was purified by silica gel flash chromatography (ethyl
(6-(4-(Naphthalen-2-yl)-2,4-dioxobutanoyloxy)hexyl)triphenylphos- acetate/hexane) to afford the title compound (80 mg, 0.36 mmol) as a
phonium bromide (C23.47–TPP): 1H NMR (400 MHz, CDCl3) δ 15.32 white solid. The product was confirmed by NMR and LC–MS as follows
(s, 1H), 8.55 (s, 1H), 8.05–7.97 (m, 2H), 7.92 (dd, J = 16.7, 8.4 Hz, 2H), (Supplementary Fig. 3i–l).
7.85–7.74 (m, 3H), 7.74–7.64 (m, 11H), 7.64–7.53 (m, 2H), 7.21 (s, 1H),
Ethyl 4-(naphthalen-2-yl)-2,4-dioxobutanoate (C23.20–EA): 1H NMR
4.32 (t, J = 6.5 Hz, 2H), 3.35 (dd, J = 12.5, 7.4 Hz, 2H), 1.81–1.69 (m, 2H), (400 MHz, CDCl3) δ 8.51 (s, 1H), 8.03 (dt, J = 15.2, 7.6 Hz, 1H), 8.00–7.93
1.65 (d, J = 3.8 Hz, 4H), 1.50–1.36 (m, 2H). Mass spectrometry: m/z: calcd (m, 1H), 7.88 (t, J = 8.3 Hz, 2H), 7.65–7.46 (m, 2H), 4.18 (q, J = 7.1 Hz, 2H),
for [C38H36O4P]+ ([M]+), 587.23; found, 587.32.
3.45 (t, J = 6.7 Hz, 2H), 2.82 (t, J = 6.7 Hz, 2H), 1.28 (t, J = 7.1 Hz, 3H). Mass
spectrometry: m/z: calcd for [C16H17O3]+ ([M+H]+), 257.12; found, 257.14.
1
Synthesis of 4-(naphthalen-2-yl)-2,4-dioxobutanoic acid
(C23.47)
Ethyl 4-(naphthalen-1-yl)-4-oxobutanoate (C23.21–EA): H NMR
(400 MHz, CDCl3) δ 8.51 (s, 1H), 8.04 (dd, J = 8.6, 1.7 Hz, 1H), 7.96 (t, J =
To a stirred solution of ethyl 4-(naphthalen-2-yl)-2,4-dioxobutanoate 8.4 Hz, 1H), 7.89 (t, J = 8.4 Hz, 2H), 7.65–7.48 (m, 2H), 4.18 (q, J = 7.1 Hz, 2H),
(504 mg, 1.86 mmol) in methanol (10 ml), tetrahydrofuran (10 ml) and 3.46 (t, J = 6.7 Hz, 2H), 2.83 (q, J = 6.6 Hz, 2H), 1.28 (t, J = 7.1 Hz, 3H). Mass
water (2 ml) at room temperature was added lithium hydroxide monohy- spectrometry: m/z: calcd for [C16H17O3]+ ([M+H]+), 257.12; found, 257.14.
drate (235 mg, 5.59 mmol) and the reaction mixture was stirred for 6 h at
Ethyl 4-(2,5-dimethylphenyl)-4-oxobutanoate (C23.28–EA):
room temperature. Completion of the reaction was confirmed by TLC. 1H NMR (400 MHz, CDCl3) δ 7.50 (s, 1H), 7.16 (dt, J = 23.2, 4.5 Hz, 2H), 4.16
The volatile compounds were evaporated under reduced pressure to (q, J = 7.1 Hz, 2H), 3.20 (dd, J = 8.8, 4.4 Hz, 2H), 2.81–2.64 (m, 2H), 2.44
yieldthecrudeproduct,whichwasacidifiedwithaqueous1MHCl(20ml), (s, 3H), 2.36 (s, 3H), 1.27 (td, J = 7.1, 2.3 Hz, 3H). Mass spectrometry: m/z:
and the product was then extracted with ethyl acetate (30 ml × 2). calcd for [C14H19O3]+ ([M+H]+), 235.13; found, 235.24.
The combined organic layers were washed with brine (10 ml), dried
over anhydrous Na2SO4, and the solvent was evaporated under reduced CDCl3): δ 15.30 (s, 1H), 8.06–7.96 (m, 2H), 7.66–7.57 (m, 1H), 7.55–7.46
Ethyl 2,4-dioxo-4-phenylbutanoate (C23.07–EA): 1H NMR (400 MHz,

(m, 2H), 7.08 (s, 1H), 4.41 (q, J = 7.1 Hz, 2H), 1.42 (t, J = 7.1 Hz, 3H). Mass Peak areas for each compound were integrated using TraceFinder
spectrometry: m/z: calcd for [C12H13O4]+ ([M+H]+), 221.08; found, 221.14. 4.1 software (Thermo Fisher Scientific).
Prodrug uptake and cleavage
Determination of prodrug stability
Escherichia coli (108 cells) were treated with different concentrations The calibration curves used to determine prodrug and drug concen-
(10–5,000 nM) of the prodrug C23.28–TPP for 30 min. The bacteria trations ranged from 50 μM to 0.012 μM with twofold serial dilutions
were washed in DPBS, lysed by freeze-thawing 5 times in liquid nitrogen (13 points in duplicate) and were generated from LC–MS quantifications
and the lysate treated with acetonitrile to a final concentration of 50%. using TraceFinder 4.1 software (Thermo Fisher Scientific). Data points
Lysates were spun down at 5,000g, passed through 0.45-μm filters and were plotted in GraphPad and respective half-lives (t1/2) were calculated
analysed by LC–MS.
using the expression t1/2 = 0.693/k where k is the rate constant. Relevant
supporting information can be found in the Source Data file.
Conversion of HMBPP to DMAPP and IPP
Bacterial viability and prodrug treatment
Escherichia coli IspH was incubated with varying concentrations
(10–5,000 nM) of TPP (control) or the IspH inhibitor C23.28 for 10 min. Escherichia coli or clinical isolates of E. aerogenes (CRE) (UCI 15),
A methyl viologen assay as described above was performed with final K. pneumoniae 1.53 (ST147+, CTX-M15+), S. enterica subsp. enterica sero-
concentrations of IspH and HMBPP of 50 nM and 1 mM, respectively. var Typhimurium (LT2 – SL7207), V. cholerae (M045), A. baumannii
At 30 min the reaction was stopped by the addition of acetonitrile to a (BC-5), A. baumannii (AB5075-UW), P. aeruginosa (PA14), P. aerugi-
final concentration of 50%. Purified HMBPP and DMAPP/IPP were used nosa (MSRN 5524), H. pylori (Hp CPY6081), B. sphaericus (CCM 2177),
as benchmarks and to obtain a dilution curve. Samples were analysed M. tuberculosis (M. tuberculosis H37Ra) and Y. pestis (KIM 10+) were
by LC–MS for the presence of HMBPP and DMAPP/IPP.
cultured to late log phase (108 cells per ml) in their respective culture
medium and quantified by measuring the optical density at 600 nm
(OD600) for 3 serial dilutions. The bacteria were spun down, resuspended
Plasma stability of prodrugs
The in vitro stabilities of the prodrugs C23.20–TPP, C23.21–TPP and in RPMI medium supplemented with 10% FBS or human serum at a
C23.28–TPP were measured in human (Sigma, P9523) mouse (Sigma, concentration of 105 cells per ml and aliquoted at 100 μl per well into a
P9275) and pig (Sigma, P2891) plasma. The lyophilized plasma was 96-well plate. Varying concentrations of candidate prodrugs (4–500 μM
reconstituted with the recommended volume of 0.05 M PBS (pH 7.4) final concentration) were added and incubated for 1–4 h (4 days for
to a concentration of 100% and prewarmed at 37ꢀ°C. The reactions M. tuberculosis) at 37ꢀ°C. Bacterial viability from each sample was tested
were initiated by the addition of the prodrugs to preheated plasma by CFU, resazurin blue (colorimetric and fluorescence) and growth
solution to yield a final concentration of 100 μM. A positive control curve assays. For proteomics and electron microscopy the bacteria
solution without the addition of plasma was also included to monitor were treated with the respective prodrugs for 8 and 24 h. The following
compound stability over the course of the experiment. The assays antibiotics were used to compare bacterial killing potency with our
were incubated at 37ꢀ°C and shaken at 200 rpm. Samples (50 μl) were prodrugs: meropenem (Sigma, 1392454), amikacin (Sigma, A0365900),
taken at 0, 15, 30, 45, 60 and 120 min and added to 200 μl acetonitrile ceftriaxone (Sigma, C0691000), cefepime (Sigma, 1097636), ciprofloxa-
to deproteinize the plasma. The samples were vortexed for 1 min and cin (Sigma, 17850), tobramycin (Sigma, T4014), ceftaroline (Bocsci,
centrifuged at 4ꢀ°C for 15 min at 20,000g. The clear supernatants were B0084-459128), kanamycin (Sigma, B5264), chloramphenicol (Sigma,
transferred to LC–MS vials for analysis.
C0378), ampicillin (Sigma, A9518), doxycycline (Sigma, D3447), gen-
tamicin (Sigma, G1264) and streptomycin (Sigma, S6501).
Liver microsome stability of prodrugs
Resazurin blue assay
The in vitro stabilities of the prodrugs C23.20–TPP, C23.21–TPP and
C23.28–TPP were measured in human (Sigma, M0317) mouse (Sigma, Control or prodrug-treated bacterial samples were treated with resa-
M9441) and monkey (Sigma, M8816) liver microsomes. A stock solu- zurin sodium salt (Sigma R7017) at a final concentration of 0.02% and
tion of the prodrug was added to a solution of 0.1 M PBS (pH 7.4) incubated for 4 h (overnight for M. tuberculosis) at 37ꢀ°C in a Biotek
containing 1 mM NADPH to a final concentration of 100 μM. This Synergy 2 plate reader. Changes in fluorescence were measured every
solution was incubated at 37ꢀ°C for 5 min at which time microsomes 20 min for 16 h (3 days for M. tuberculosis), with discontinuous shak-
were added at a final concentration of 1.0 mg ml−1, incubated at 37ꢀ°C ing, using excitation filter range 530–570 nm and emission filter range
and shaken at 200 rpm. A positive control solution without the addi- 590–620 nm. An increase in fluorescence intensity corresponds to
tion of microsomes was also included to monitor compound stability bacterial growth and is quantified by comparison with untreated
over the course of the experiment. Aliquots were removed at 0, 15, bacterial control samples. The ratio of (Tthreshold (untreated)/Tthreshold
30, 60, 90, 120 min and 10× volume of acetonitrile was added to stop (prodrug-treated)) (where T = time) was used to quantify the change in
the reaction and deproteinate the sample. Samples were centrifuged bacterial growth. To minimize inter-experimental variations, all Tthreshold
at 20,000g for 5 min at 4ꢀ°C, and the supernatant was transferred to times were corrected by subtracting the time taken for untreated con-
LC–MS vials for analysis.
trol cultures to reach minimum detectable fluorescence. At the end of
the experiment, wells were visualized for changes in colour from blue
(inviable bacteria) to pink (viable bacteria) or by measuring the fluo-
LC–MS quantification of small molecules
LC–MS analysis was performed on a Thermo Fisher Scientific Q Exactive rescence at the aforementioned excitation and emission wavelengths.
HF-X mass spectrometer equipped with a HESI II probe and coupled
Measurement of bacterial membrane integrity by SYTO 9/
propidium iodide assay
to a Thermo Fisher Scientific Vanquish Horizon UHPLC system. IPP/
DMAPP and HMBPP were analysed by hydrophilic interaction chro-
matography (HILIC) on a ZIC-pHILIC 2.1-mm i.d × 150ꢀmm column Escherichia coli cells grown to late log phase (108 cells per ml) were
(EMD Millipore). The HILIC mobile phase A was 20 mM ammonium treated with TPP (control) or DAIA prodrugs at varying concentra-
carbonate, 0.1% ammonium hydroxide, pH 9.2, and mobile phase B tions in RPMI + 10% FBS. Bacteria were spun down and washed three
was acetonitrile. Prodrug compounds were analysed by reversed phase times in Tris buffered saline (TBS) (pH 7.5). Component A (SYTO 9 dye)
(RP) chromatography on a Synergi 4 mm Polar-RP 2-mm i.d × 100ꢀmm (1.5 μl ml−1) and component B (propidium iodide) from the BacLight Live/
column (Phenomenex). The RP mobile phase A was 0.1% formic acid in Dead kit (Life Tech, L7012) were added to the bacterial samples and incu-
MilliQ water, and mobile phase B was 0.1% formic acid in acetonitrile. bated for 15 min. An aliquot was run for flow cytometry on a BD LSR II

Article
instrument (BD Biosciences). With the excitation wavelength centred at 1% dextrose for 8 or 24 h to inhibit IspH expression. At respective time
about 485 nm, the fluorescence intensities at 530 nm (green) and 630 nm points the samples were fixed in 2.5% glutaraldehyde, 2% paraformal-
(red) were measured and the data analysed using FlowJo software. TPP- dehyde at 4ꢀ°C in 100 mM cacodylate buffer (pH 7.0) containing 2 mM
or isopropanol-treated bacteria served as negative or positive controls, CaCl2 and 0.2% picric acid. Samples were briefly washed and treated
respectively, and their flow plots were used to gate the prodrug-treated for 2 h at 4ꢀ°C with 1% osmium tetroxide in 100 mM cacodylate buffer
samples. As bacteria lose their membrane integrity the green SYTO 9 (pH 7.0). After washing with distilled water 3–5 times, samples were
dye is displaced by the red propidium iodide dye. The remaining sam- dehydrated using increasing ethanol concentrations and embedded in
ples were spun down at 5,000g for 10 min, resuspended in 10 μl of TBS, Epon resin (Sigma-Aldrich). Ultrathin sections of the embedded sam-
spread on glass microscopy slides and dried. The samples were mounted ples were cut and loaded onto grids and stained further with Reynold’s
using Cytoseal 60 or Mounting Medium (Electron Microscopy Sciences). lead citrate (Sigma-Aldrich) for 3–15 min. Grids were dried overnight
Specimens were documented photographically using an 80i upright and observed using a JEOL 1010 transmission electron microscope
microscope and analysed with the NIS-Elements Basic Research software. equipped with an AMT 2k CCD camera.
Measuring respiration using a Seahorse XF Analyzer
Scanning electron microscopy
On the day before the assay, the sensor cartridges from the Seahorse Scanning electron microscope experiments were carried out at the
XFe96 FluxPaks (Agilent, 102416) were calibrated according to the CDB Microscopy Core (Perelman School of Medicine, University of
manufacturer’s instructions using pre-warmed Seahorse XF Calibrant. Pennsylvania). Bacterial samples were washed three times with 50 mM
Escherichia coli cells were grown in LB medium overnight to an OD600 Na-cacodylate buffer, fixed for 2–3 h with 2% glutaraldehyde in 50 mM
of 0.3, washed in PBS and resuspended in Seahorse XF RPMI medium, Na-cacodylate buffer (pH 7.3), spun down over 0.22-μm filter membranes
pH 7.4 (Agilent, 103576) supplemented with 1% glucose. Bacteria and dehydrated in an increasing ethanol concentration over a period of
(105, 106 or 107) were added to XF Cell Culture Microplates (Agilent, 1.5 h. Dehydration in 100% ethanol was performed three times. Dehy-
101085-004) precoated with poly-D-lysine and spun down at 2,000g for drated samples were incubated for 20 min in 50% hexamethyldisilane
10 min to attach them to the plate. The wells in the plate were divided to (HMDS; Sigma-Aldrich) in ethanol followed by 100% HMDS (refreshed
include bacteria treated with TPP (negative control) and 3 concentrations three times) and then overnight air-drying as described previously49
.
(500, 100 and 20 μM) of C23.28–TPP; 8 technical replicates were used for Samples were then mounted on stubs and sputter-coated with gold–pal-
each condition. Fresh medium (90 μl) was added to each well and 90 μl ladium. Specimens were observed and photographed using a Quanta 250
of TPP or prodrug solution was added to each injection port A. The base- FEG scanning electron microscope (FEI) at 10 kV accelerating voltage.
line oxygen consumption rate (OCR) and extracellular acidification rate
Toxicity assays in mammalian cell lines
(ECAR) were measured for 12 min, after which the TPP or prodrug solution
was injected into each sample. Readings were obtained as pmol min−1 The cytotoxicity of prodrugs to C2C12, HepG2, Raw 264.7 and Vero
(OCR) and mpH min−1 (ECAR) every 6 min for up to 90 min. The mean cells was estimated using an LDH-GloTM cytotoxic assay kit (Promega,
of the 8 technical replicates was plotted for each treatment condition J2381). Cells were grown, counted, aliquoted in 96-well plates at a cell
and changes in OCR and ECAR were compared to the control samples.
density of 105 cells per well and allowed to adhere to the bottom of the
wells for 1 day at 37ꢀ°C and 5% CO2. The cells were treated with prodrugs
at different concentrations (1–5,000 μM). Cells treated with 2% DMSO
Detection of superoxide and H2O2
Superoxide anion was measured in prodrug- and TPP-treated bacte- served as negative control whereas cells treated with 0.2% Triton X-100
ria by diluting them 1/50 into PBS containing 2 μM dihydroethidium served as positive control for cytotoxicity. Eight replicates were per-
(Sigma, D7008) just before flow cytometry (excitation, 535 nm; emis- formed for each condition. Supernatant medium samples were taken
sion, 610 nm). H2O2 production was measured in similar bacterial from each well at intervals of 24, 48 and 72 h, diluted 300-fold in PBS,
samples using the Amplex Red Hydrogen Peroxide/Peroxidase Assay added to the lactate dehydrogenase (LDH) assay reagent in a 1:1 ratio
Kit (Thermo Fisher, A22188). Fluorescence measurements were cali- (20 μl:20 μl) in white opaque 96-well plates and further incubated at
brated by comparison to calibration curves for wells containing H2O2 room temperature for 1 h in the dark. Luminescence was measured
in a final concentration ranging between 0.1 to 100 μM. Fluorescence using a Biotek Synergy 2 plate reader with integration time 1 s per
was measured using the 540/620 nm wavelength pair in a Biotek Syn- well. Cytotoxicity was calculated using the following equation: % cyto-
ergy 2 plate reader.
toxicity = 100 × (experimental LDH release − medium background)/
(maximum LDH release control − medium background).
Staining for bacterial membrane potential
Measurement of mitochondrial membrane potential
The procedure for studying the changes in membrane potential in
prodrug-treated E. coli was identical to that used for the Live/Dead To quantify the effect of IspH prodrugs on the mitochondrial membrane
assay with the following exceptions. The BacLight Bacterial Membrane potential of C2C12 myoblasts (ATCC, CRL-1772), cells were grown in
Potential Kit (Lifetech, B34950) was used in this case. Component A DMEM + 10% FBS up to 90% confluence and suspended by trypsiniza-
(10 μl) (3 mM 3,3′-diethyloxacarbocyanine (DiOC2)) was used to stain tion. Cells were washed and pelleted at 500g for 5 min and resuspended
the bacterial samples for 30 min at room temperature. TPP- or compo- in DMEM consisting of 100 nM of tetramethyl rhodamine methyl ester
nent B (carbonyl cyanide m-chlorophenyl hydrazine (CCCP))-treated (Thermo Fisher, I34361) for 30 min at 37ꢀ°C with slow shaking. Myoblasts
bacteria were used as negative or positive controls, respectively, and were pelleted down and resuspended in PBS. One million cells were
to gate prodrug-treated samples. When the membrane potential is incubated with 1, 10 or 100 μM concentrations of TPP, IspH prodrugs
intact, the DiOC2 dye forms tetramers within bacteria that fluoresce or CCCP (Invitrogen, B34950) for 10 min at room temperature. CCCP
at 630 nm (red). Loss of membrane potential leads to the formation is an ox-phos uncoupler that causes loss of mitochondrial membrane
of dimers that fluoresce at 530 nm (green). Bacteria were analysed by potential and is used as a positive control. After 10 min cells were ana-
both flow cytometry and microscopy.
lysed by flow cytometry according to the manufacturer’s instructions
and the plots were gated using negative control (unstained cells).
Transmission electron microscopy
Profiling the effect of the TPP carrier molecule on hERG channel
Bacteria (E. coli or V. cholerae) were treated with TPP or with the prodrug
C23.28–TPP in RPMI medium with 10% FBS for 0, 8 or 24 h. The ΔispH Compound profiling against hERG—to evaluate the potential car-
conditional knockdown E. coli was cultured similarly in the presence of diac liability of 6-hydroxyhexyl TPP, methyl TPP and our prodrug

C23.28–TPP—was carried out at Reaction Biology using the QPatch points was performed using Search Tool for the Retrieval of Interact-
HTX fully automated patch-clamp platform that enables the testing ing Genes/Proteins (STRING)34. Functions with at least 5 differentially
of up to 48 cells in parallel. Electrophysiological profiling was done in expressed proteins enriched at the P < 0.001 threshold were considered.
the presence of verapamil (positive control), DMSO (vehicle control)
Antibiotic-resistance assays with DAIAs
and the TPP compounds at a concentration range of 10 nM to 10 μM
(n = 3 cells per sample concentration × 6 concentrations). Exemplar For this study, K. pneumoniae and V. cholerae clinical strains mentioned
hERG traces were elicited from a holding potential of −80 mV followed in the section ‘Bacteria’ were cultured to exponential phase in RPMI
by steps from −60 to +50 mV in 10 mV increments; tail currents were medium containing 10% FBS. The bacteria were washed in DBPS and
elicited by a step to −50 mV. Response data obtained were normalized quantified by OD600. The bacteria were aliquoted into identical samples
to peak current at 0.1% DMSO. Nonlinear regression curve fits were containing 105 CFU. These aliquots were used to start fresh cultures in
used to calculate the IC50 of each compound.
RPMI medium containing 5% human serum, in the presence or absence
of 106 human PBMCs, and for each condition in the presence or absence
of the prodrug C23.28–TPP. Hygromycin and streptomycin were used
Protein isolation and western blot analysis
Bacterial samples were washed with PBS and treated with 10 mg ml−1 as control antibiotics for Vibrio and Klebsiella, respectively. After 8 h
lysozyme in 20 mM Tris-HCl, pH 8.0; 2 mM EDTA at 37ꢀ°C for 30 min. of incubation at 37ꢀ°C and 5% CO2, 50 μl of each sample was plated by
Lysates are prepared by freeze-thawing in Ripa lysis buffer (10 mM serial dilution for the measurement of CFU, and the rest of the cultures
Tris-Cl pH 8.0, 1 mM EDTA, 0.5 mM EGTA, 1% Triton X-100, 0.1% sodium were allowed to grow. After 24 h of incubation, bacteria from each
deoxycholate, 0.1% SDS, 140 mM NaCl) supplemented with protease sample were washed and quantified by OD600. Bacteria (105) from the
inhibitors at 4ꢀ°C. Whole-cell lysates (100 μg per reaction) were mixed respective samples in the first passage were used to start the next pas-
with an equal volume of 2× SDS–PAGE sample buffer supplemented with sage (cycle of selection by antibiotic) under the same conditions. For
10% 2-mercaptoethanol and heated for 5 min at 100ꢀ°C. Protein samples samples co-incubated with human PBMCs, fresh PBMCs from the same
were size-fractionated on 4–20% Tris-glycine gradient gels (Lonza) or donor were used for every passage. Bacterial growth in each passage
lab-made 12.5% Tris-glycine gels using constant voltage at room tem- was measured up to the 18th passage. Bacterial growth (CFU) in the
perature, transferred overnight onto Immuno-Blot PVDF membranes absence of antibiotics was considered as 100% growth and resistance to
(Bio-Rad, 162-0177) at 4ꢀ°C and subjected to protein blotting using the an antibiotic in each passage was defined as the percentage of bacterial
mouse anti-E.coli RNA Sigma 70 antibody (BioLegend, 663208) or rabbit growth that occurred in the presence of that antibiotic.
anti-E. coli IspH antibody (generated in this work). Both primary anti-
Flow cytometry
bodies show cross-reactivity across several bacterial species. Secondary
antibodies conjugated to horseradish peroxidase were used at a dilution Cells were washed with 2 ml of 1× PBS at 1,500 rpm for 5 min and then
of 1:10,000 (GE Healthcare, NA931V, NA934V). The immunoblots were stained with 1 ul of Zombie Yellow (BioLegend, 423103) for 20 min at
scanned using Image Quant LAS 4000. Uncropped western blots with room temperature to check the viability. The cells were stained for
molecular weight markers are shown in Supplementary Fig. 1.
cell surface markers with a combination of (where indicated) CD3-
PerCP-Cy5.5 (clone UCHT1, BD Biosciences, 560835), CD4-Alexa Fluor
700 (clone RPA-T4, BD Biosciences, 557922), CD8a-Brilliant Violet 711
Proteomics
Triplicate samples of C23.28–TPP-treated or ΔispH E. coli lysates at (clone RPA-T8, BioLegend, 301044), TCR Vγ9-FITC (clone 7A5, Invitro-
0, 8 and 24 h (a total of 18 samples) were processed. Protein samples gen, TCR2720), CD69- PE/Cy7 (clone FN50, BD Biosciences, 557745),
were concentrated (up to eightfold) by lyophilization and 25 μg from HLA-DR-Brilliant Violet 421 (clone L243, BioLegend, 307636), for 20 min
each sample was separated by SDS–PAGE for a distance of 0.5 cm into in FACS buffer (1% FBS in PBS) at room temperature. Next the cells were
the gel. The entire lanes were excised, digested with trypsin and ana- washed with PBS, fixed and permeabilized using a Fixation/Permeabili-
lysed by LC–MS/MS on a Q Exactive HF mass spectrometer using a zation Kit (BD Biosciences, 554714) for 15 min at 4ꢀ°C. After washing with
240 min LC gradient. Tandem mass spectrometry (MS/MS) spectra 1 ml of 1× permeabilization buffer, intracellular proteins were stained
were searched with full tryptic specificity against the UniProt E. coli using Perforin-Brilliant Violet 421 (clone dG9, BioLegend, 308122),
database (https://www.uniprot.org; accessed 7 December 2019) using Granulysin-Alexa Fluor 647 (clone DH2, Bio Legend, 348006), Gran-
the MaxQuant 1.6.3.3 program. The ‘match between runs’ feature was zyme A-PE/Cy7 (clone CB9, Bio Legend, 507222). Cells were washed
used to help to transfer identifications across experiments to minimize with 1× permeabilization buffer twice. The cells were resuspended in
missing values. Protein quantification was performed using razor and 300 μl of 1% paraformaldehyde fixation buffer (BioLegend, B244799)
unique peptides. False discovery rates (FDRs) for protein and peptide in PBS. Samples were run on BD LSR II (BD Biosciences) and the data
identifications were set at 1%. A total of 2,346 protein groups were analysed using FlowJo software. Cells were first gated for lymphocytes
identified, including proteins identified by a single razor and unique (forward scatter/side scatter (FSC/SSC)) then singlets (FSC-A vs FSC-H).
peptide. Label-free quantitation (LFQ) intensity was used for protein The singlets were further analysed for their uptake of the Live/Dead
quantification50. The LFQ intensity levels were log2-transformed and Aqua or Zombie Yellow stain to determine live and dead cells. Live
undetected intensities were floored to a minimum detected intensity cells were gated for CD3+ cells then gated for their identifying surface
across all proteins or a minimum across 4 samples in the case that both markers—CD4, CD8 and Vγ9 (γδ T lymphocytes)—followed by their
replicates were undetected.
respective cytotoxic markers perforin, granulysin and granzyme A
or cell surface markers of T cell activation such as HLA-DR, and CD69.
Gating strategy for every FACS plot is shown in the Source Data.
Bioinformatics analysis
Unpaired t-tests were performed to estimate the significance of dif-
ference between conditions, and false discovery rate was estimated
Validating the anti-Vδ2-TCR antibody for immunofluorescence
using a procedure described previously51. Proteins that passed the Human PBMCs from one donor were split into two aliquots; one
P < 0.05 threshold were considered significant (all passed FDR <5% sample was treated with 10 μM HMBPP and 50 ng ml−1 IL-15 to expand
threshold). A total of 525 proteins changed in both the IspH prodrug Vγ9Vδ2 T cells and the other sample was depleted of all γδ T cells using
treatment and the ΔispH conditional knockdown systems. Proteins Anti-TCRγ/δ Microbead Kit (Miltenyi, 130-050-701). HepG2 and Vero
showing more than 2-fold up- or downregulation under both condi- cells served as negative control. Cells (106 of each type) were collected
tions and at both 8- and 24-h time points were analysed using Venny52. in Eppendorf tubes and washed with 1× PBS 3 times. The cell pellet was
Enrichment analysis of proteins common to both conditions and time resuspended in PBS and fixed for 20 min by adding formaldehyde to a

Article
final concentration of 4%. Fixed cells were washed with 1× PBS, pelleted was carried out using theVirtual Ligand Screening software (https://
and embedded in 100 μl of 4% agar (Fisher Scientific, BP14232). The agar molsoft.com/vls.html) from Molsoft. Owing to the lack of a suitable
block was then treated with 70% ethanol before paraffin embedding and online repository, all docking data are available upon request as an
sectioning at the Wistar Histotechnology Facility. For immunofluores- .icb file, viewable using the free ICM browser (http://www.molsoft.
cence studies, sections were deparaffinized in xylene, then rehydrated com/icm_browser.html). LC–MS/MS spectra were searched against
in ethanol (100–95% to 80–70%) and distilled water. The endogenous the UniProt E. coli (BL21-DE3) database (https://www.uniprot.org/
peroxidase activity was eliminated by treating the sections with 0.5% proteomes/UP000002032). The proteomics data are available on
hydrogen peroxide in methanol for 10 min. The slides were washed MassIVE (https://massive.ucsd.edu/) using the accession number
under tap water for 5 min before simmering them in Tris-EDTA buffer. MSV000086359, or they can be downloaded from ftp://massive.ucsd.
The slides were washed with PBS before blocking them in 5% BSA block- edu/MSV000086359/. All reagents used or generated and all data that
ing solution for 1 h. The tissue sections were subsequently incubated support the findings of this study are available from the authors on
with primary anti Vδ2-TCR primary antibody (BioLegend, 331402) 1:50 reasonable request, see author contributions for contacts for specific
in 5% BSA overnight at 4ꢀ°C, washed next day with 1× PBS and incubated datasets. Source data are provided with this paper.
with AF647 (Invitrogen, A21236) secondary antibody 1:200 for 45 min.
41. Abagyan, R., Totrov, M. & Kuznetsov, D. ICM—a new method for protein modeling and
DAPI was added for 5 min and the sections mounted using Cytoseal 60
or Mounting Medium (Electron Microscopy Sciences). Specimens were
design - applications to docking and structure prediction from the distorted native
conformation. J. Comput. Chem. 15, 488–506 (1994).
42. Abagyan, R. & Totrov, M. Biased probability Monte Carlo conformational searches and
electrostatic calculations for peptides and proteins. J. Mol. Biol. 235, 983–1002 (1994).
43. Lam, P. C., Abagyan, R. & Totrov, M. Hybrid receptor structure/ligand-based docking and
activity prediction in ICM: development and evaluation in D3R Grand Challenge 3.
J. Comput. Aided Mol. Des. 33, 35–46 (2019).
documented photographically using a Leica TCS SP5 Scanning Confo-
cal Microscope and analysed with the NIS-Elements Basic Research
software. The validation images are shown in Extended Data Fig. 11b, c.
44. Lam, P. C., Abagyan, R. & Totrov, M. Ligand-biased ensemble receptor docking (LigBEnD):
a hybrid ligand/receptor structure-based approach. J. Comput. Aided Mol. Des. 32,
187–198 (2018).
Tissue staining and immunofluorescence
Tissues were collected and fixed in Formalde-Fresh Solution overnight
at 4ꢀ°C, washed with 1× PBS and transferred to 70% ethanol before paraf-
fin embedding and sectioning. Tissue embedding and sectioning were
performed by the Histotechnology Facility at The Wistar Institute. For
immunohistochemistry studies, tissue sections were deparaffinized in
xylene, rehydrated in ethanol (100–95% to 80–70%) and distilled water.
The endogenous peroxidase activity was quenched with 0.5% hydrogen
peroxide in methanol for 10 min. The slides were washed under tap water
for 5 min, simmered in Tris-EDTA buffer, and washed with PBS before
blocking in 5% BSA blocking solution for 1 h. The tissue sections were
subsequently incubated with anti Vδ2-TCR primary antibody (BioLegend,
331402) and anti-E. coli antibody (Abcam, ab137967) 1:50 in 5% BSA over-
night at 4ꢀ°C, washed the next day with 1× PBS and incubated with AF647
(Invitrogen, A21236) and AF488 (BioLegend, 406416) secondary antibod-
ies 1:200 for 45 min. DAPI (1:5,000) was added for 5 min and the samples
mounted using Cytoseal 60 or Mounting Medium (Electron Microscopy
45. Totrov, M. Atomic property fields: generalized 3D pharmacophoric potential for
automated ligand superposition, pharmacophore elucidation and 3D QSAR. Chem. Biol.
Drug Des. 71, 15–27 (2008).
46. Chen, Q. & Chen, J. Isolation of CD34+ cells from human fetal liver and cord blood.
Bio Protoc. 3, e991 (2013).
47. Somasundaram, R. et al. Tumor-associated B-cells induce tumor heterogeneity and
therapy resistance. Nat. Commun. 8, 607 (2017).
48. Span, I. et al. Insights into the binding of pyridines to the iron–sulfur enzyme IspH.
J. Am. Chem. Soc. 136, 7926–7932 (2014).
49. Braet, F., De Zanger, R. & Wisse, E. Drying cells for SEM, AFM and TEM by
hexamethyldisilazane: a study on hepatic endothelial cells. J. Microsc. 186, 84–87
(1997).
50. Cox, J. et al. Accurate proteome-wide label-free quantification by delayed
normalization and maximal peptide ratio extraction, termed MaxLFQ. Mol. Cell
Proteomics 13, 2513–2526 (2014).
51. Storey, J. D. & Tibshirani, R. Statistical significance for genomewide studies. Proc. Natl
Acad. Sci. USA 100, 9440–9445 (2003).
52. Oliveros, J. C. Venny v.2.1 (BioinfoGP, 2007).
Sciences). Specimens were photographed using an 80i upright micro
scope and analysed with the NIS-Elements Basic Research software.
-
Acknowledgements This research was supported by the G. Harold and Leila Y. Mathers
Charitable Foundation, Commonwealth Universal Research Enhancement Program
(CURE – Pennsylvania Department of Health) and the Wistar Science Discovery Fund (F.D.).
F.D. was supported by a Wistar Institute recruitment grant from The Pew Charitable Trusts.
R.S.S. and M.H. were funded by the Adelson Medical Research Foundation and DOD for
Hu-mice generation. We thank D. Speicher from the Proteomics Facility at the Wistar
Institute, S. Molugu from the Electron Microscopy Resource Laboratory and Y. Velich from
the Cell and Developmental Biology Microscopy core at UPenn. Support for the Wistar
Institute Proteomics and Metabolomics and Genomics Shared Resources was provided by
Cancer Center Support Grant P30 CA010815 and National Institutes of Health instrument
grant S10 OD023586. We thank M. Groll, E. Oldfield, A. Odom John and C. Morita for advice
on the bacterial isoprenoid synthesis pathway, IspH and γδ T cell fields.
Software used for data collection
NIS-Elements Basic Research (Nikon) v.4.60.00; FlowJo (FlowJo LLC)
v.10; Internal Coordinate Mechanics software (ICM) (MolSoft) v.3.7-
2a; Virtual Ligand Screening (VLS) (MolSoft) v.3.7-2a; Seahorse Wave
controller software v.2.4.2 (Agilent).
Software used for data analysis
Author contributions F.D. conceived the study and planned the experiments. M.T. set up the
atomic field property of the IspH catalytic site and performed molecular docking experiments.
K.S.S., R. Sharma, P.V. and A.S. purified the proteins, performed the biochemical activity assays,
bacterial killing experiments, mouse infection studies and contributed to the preparation of
the manuscript. P.V. performed flow cytometry and microscopy studies. K.S.S. performed the
electron microscopy studies with assistance from the UPenn electron microscopy core. A.R.G.
and H.-Y.T. ran the samples for proteomics and small-molecule studies. A.K. and R. Sharma
performed the bioinformatics and pathway analysis on proteomics data and helped to
illustrate it in a figure. J.C. performed the surface plasmon resonance studies. J.M.S. planned
the synthesis of DAIA prodrugs and P.A.N.R. synthesized them. H.C., K.M., R. Somasundaram
and M.H. provided Hu-mice. M.G. and M.E.M. performed the seahorse experiments. F.D. and
J.M.S. analysed the data. F.D. generated the figures and drafted the manuscript. J.M.S. and
P.A.N.R. provided reagents and expertise. All authors provided critical revisions to the
manuscript.
Microsoft Office 2016; Prism 7 v.7.04 (GraphPad); MaxQuant v.16.3.3;
Max Planck Institute Search Tool for the Retrieval of Interacting Genes/
Proteins (STRING) v.11; Venny v.2.1; TraceFinder v.4.1; ICM Browser
v.3.7-2a (Molsoft); Seahorse Wave analysis software (Agilent) v.2.4.2;
ChemDraw v.19.1.
Reporting summary
Further information on research design is available in the Nature
Research Reporting Summary linked to this paper.
Data availability
Competing interests The authors declare no competing interests.
Molecular docking studies were performed using the E. coli IspH struc-
ture 3KE8 from the Protein Data Bank. The atomic field property of
the IspH binding pocket was mapped using the Internal Coordinate
Mechanics (ICM) software (http://www.molsoft.com/icm_pro.html)
from Molsoft and the molecular docking of 10 million compounds
from the MolCart library (https://www.molsoft.com/molcart.html)
Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/s41586-020-
03074-x.
Correspondence and requests for materials should be addressed to J.M.S. or F.D.
Peer review information Nature thanks Herman Sintim, Ben Willcox and Gerry Wright for their
contribution to the peer review of this work.
Reprints and permissions information is available at http://www.nature.com/reprints.

Extended Data Fig. 1 | Purification of recombinant IspH proteins from
multiple microbial species and measurement of their biochemical activity
by methyl viologen assay. a, Coomassie-stained gels showing IPTG induction
of recombinant 6His-tagged P. falciparum (Pf), E. coli (Ec), P. aeruginosa (Pa)
and M. tuberculosis (MTB) IspH. b, Anti-His-tag immunoblots showing the
respective purified IspH proteins. Images in a, b are representative of 3
independent purification attempts. c, IspH uses methyl viologen (MV) as an
electron donor for the reductive dehydroxylation of HMBPP. Colourless
oxidized methyl viologen is restored to its reduced blue form by sodium
dithionite. In the absence or inhibition of IspH activity, methyl viologen stays
blue. d, e, Methyl viologen assays measuring IspH activity using different
concentrations of E. coli IspH at 10 min in the presence of 1 mM HMBPP (d) and
different concentrations of HMBPP (e) at 30 min in the presence of 50 nM E. coli
IspH. For d, e, data are mean of 3 independent experiments s.e.m. ***P < 0.001,
**P < 0.01, *P < 0.05, ns, not significant; two-tailed unpaired Student’s t-test,
relative to 0 nM IspH in d or 0 μM HMBPP in e.

Article
Extended Data Fig. 2 | In silico molecular docking with the active pocket of
E. coli IspH. a, Crystal structure of E. coli IspH (PDB: 3KE8)28 (top left) was used
to generate the atomic property field (bottom left) and mimic HMBPP
interactions in the active binding pocket (right). b, Automated virtual ligand
screening (Molsoft) identified 168 out of 9.6 million compounds on the basis of
predicted binding at the active site. c, The top 24 compounds were compared
with HMBPP visually and on the basis of their predicted number of hydrogen
bonds formed, hydrogen-bond energy, Van der Waals interaction energy and
other interactions as mentioned. In silico docking for C1–24 is shown in
Extended Data Fig. 3a.

Extended Data Fig. 3 | In silico molecular docking of compounds C1–24 and
their inhibitory activity on E. coli IspH. a, Chemical structures and in silico
docking of the top 24 candidate IspH inhibitors at the E. coli IspH active pocket
rendered by Molsoft. Structures are shown in Supplementary Fig. 2a.
b, Activity of P. aeruginosa (Pa), M. tuberculosis (MTB) and P. falciparum (Pf)
IspH pretreated with DMSO (control), C10, C17 and C23 over time. Data are
mean of 4 independent experiments s.e.m. c–e, Inhibition of E. coli IspH by
analogues of C10 (c), C17 (d) or C23 (e). Structures are shown in Supplementary
Fig. 2. For analogues with better activity than the parent compound,
***P < 0.001, **P < 0.01, *P < 0.05; two-tailed unpaired Student’s t-test, relative to
C23 (n = 8 technical replicates). Data are mean s.e.m.

Article
Extended Data Fig. 4 | See next page for caption.

Extended Data Fig. 4 | Drug binding assays, structure–activity
retention times (RT) and mass:charge (m/z) ratios. Area under the respective
peaks is measured in arbitrary units (AU) and is directly proportional to the
abundance of the molecules. f, Relative abundances of C23.28–TPP (prodrug),
TPP (carrier molecule) and C23.28 (active drug) found within E. coli treated
with different concentrations (10–5,000 nM) of C23.28–TPP (n = 3 technical
and 2 biological replicates). g, Methyl viologen assay performed by treating
1 mM HMBPP with 50 nM E. coli IspH pre-treated with 5 μM C23.28 or TPP for
30 min. Samples analysed by LC–MS to quantify relative conversion of HMBPP
(IspH substrate) to DMAPP and IPP (IspH products). Respective molecules were
identified by their respective retention times and mass:charge (m/z) ratios.
Area under the respective peaks is measured in AU and is directly proportional
to the abundance of the molecules. h, Conversion of 1 mM HMBPP (black) to
DMAPP and IPP (grey) in 30 min by 50 nM E. coli IspH in the presence of different
concentrations (10–5,000 nM) of TPP (dotted lines) or C23.28 (solid lines)
(n = 3 technical and 2 biological replicates). For f, h, data are mean of 3
independent experiments s.e.m.
relationship, testing prodrug potency with different carrier molecules
and determining prodrug cleavage and E. coli IspH inhibition by LC–MS.
a, SPR signals (resonance units (RU)) from different concentrations of HMBPP,
C23.20 and C23.21 run on E. coli IspH crosslinked NTA chip, plotted against
concentrations to calculate KD and Rmax (the amount of ligand (in RU) to be
immobilized) values (n = 3 biological and 2 technical replicates). b, Structure–
activity guided analogue design reduced the IC50 values of multiple C23
analogues compared with the parent compound. Structures are shown in
Supplementary Fig. 2. c, d, Prodrug ester forms of analogue C23.47 obtained by
linking ethanol, TPP or dimethylaminopropanol (synthetic reactions are shown
in Supplementary Fig. 3) were tested for E. coli killing by dynamic growth curves
(c) and by resazurin blue assay (d). For c, d, n = 3 biological and 8 technical
replicates. e, Escherichia coli cells treated with 5 μM C23.28–TPP for 30 min
were lysed and the lysates analysed by LC–MS to quantify the relative
abundance of C23.28–TPP (prodrug), TPP (carrier molecule) and C23.28
(active drug). Respective molecules were identified by their respective

Article
Extended Data Fig. 5 | C23 prodrugs specifically act on IspH and kill
multidrug-resistant clinical isolates of V. cholerae. a, Immunoblot shows
modulation of IspH levels in CGSC 8074 (E. coli) achieved by altering arabinose
levels in culture medium. RpoD, loading control; representative of 3 independent
experiments. b, c, Sensitivity of the CGSC 8074 strain to C23.28–TPP decreases
with increasing IspH levels, as shown by resazurin blue assay (b) and dynamic
growth curves (c). Data are mean of 3 independent experiments s.e.m.
***P < 0.001, **P < 0.01, *P < 0.05, the remainder are not significant; two-tailed
paired Student’s t-test. d–f, TPP-linked prodrug esters of C23 analogues C23.7,
C23.20, C23.21, C23.28 and C23.47 (d) were tested for ability to kill V. cholerae
(strain M045) by dynamic growth curves and resazurin blue assay (e; n = 3
biological and 8 technical replicates) or by CFU plating after 24 or 48 h
treatment (f; n = 3 biological replicates with 3 serial dilutions). The MIC90 values
for prodrug analogues tested on drug-resistant clinical isolates of different
pathogenic bacteria are shown in Extended Data Fig. 8a. Data are mean of
3 independent experiments s.e.m. *P < 0.05, **P < 0.01, ***P < 0.001, the
remainder are not significant; two-tailed unpaired Student’s t-test.

Extended Data Fig. 6 | See next page for caption.

Article
Extended Data Fig. 6 | DAIA prodrugs increase oxidative stress and cause
defects in bacterial respiration, membrane integrity and cell-wall
architecture. a, b, Respiratory changes in E. coli treated with TPP or with the
indicated concentration of the DAIA prodrug C23.28–TPP were compared by
measuring OCR (for aerobic respiration) (a) and ECAR (for glycolysis) (b).
***P < 0.001; two-tailed unpaired Student’s t-test, relative to TPP-treated
control. c, d, Superoxide (solid line, 2 h after treatment; dotted line, 4 h after
treatment) (c) and hydrogen peroxide (d) levels were simultaneously measured
by dihydroethidium (DHE) and Amplex red fluorescence respectively. n = 8
biological replicates; data are mean s.e.m. e, f, Changes in E. coli membrane
integrity, upon TPP or prodrug treatment, measured by Live/Dead (SYTO 9/
propidium iodide) assay using flow cytometry (e) or fluorescence microscopy
(f). n = 3 biological replicates. Scale bar, 2 μm. g, h, Loss of E. coli membrane
potential upon treatment with TPP or prodrug measured by BacLight (DiOC2)
assay using flow cytometry (g) or fluorescence microscopy (h). n = 3 biological
replicates. Scale bar, 2 μm. i, Scanning electron micrographs (SEM; left) and
transmission electron micrographs (TEM; right) compare the morphology of
E. coli after 8 h of TPP or prodrug treatment to that of the conditional ispH
knockdown E. coli strain CGSC 8074 (ΔispH) kept for 8 h in 1% glucose medium.
Red arrows indicate membrane blebbing. j, SEM (top) and TEM (bottom)
compare the morphology of V. cholerae after 8 h of TPP or prodrug (C23.28–
TPP) treatment. In i, j, images are representative of 20 fields from 3 technical
replicates. Scale bar, 400 nm.

Extended Data Fig. 7 | See next page for caption.

Article
Extended Data Fig. 7 | DAIA prodrugs are stable in plasma and liver
microsomes, non-toxic to mammalian cells, do not disrupt mitochondrial
membrane potential in C2C12 myoblasts and do not disrupt hERG
function. a, b, Nonlinear regression curves for degradation of prodrugs
C23.28–TPP and C23.21–TPP and the appearance of the parent drugs C23.28
and C23.21 in the presence of human, pig and mouse plasma (a) or human,
monkey and mouse liver microsomes (b). Drug and prodrug concentration
measured by LC–MS and normalized on a standard curve. The half-lives (t1/2) are
calculated from respective curves. Data are mean s.e.m. of 3 independent
experiments. c, d, Cytotoxicity of prodrug analogues on HepG2, RAW264.7
and Vero cells (c) and C2C12 myoblasts (d) measured at 24, 48 and 72 h by LDH
release (n = 3 biological and 4 technical replicates). e, Effect of TPP and
prodrugs C23.28–TPP and C23.47–TPP on mitochondrial membrane potential
of C2C12 myoblasts, measured by tetramethyl rhodamine methyl ester
(TMRM) fluorescence. CCCP, positive control (n = 3 biological and
4 technical replicates). f, Toxicity of C23.28–TPP, the carrier (6-hydroxyhexyl)
triphenylphosphonium bromide and Me-TPP to hERG channel measured by
automated Q patch assay; the normalized current response is plotted using
nonlinear regression curves and the IC50 of respective compounds is
calculated. Data are mean of 3 independent experiments s.e.m.
Verapamil was used as the positive control and DMSO as the negative control.

Extended Data Fig. 8 | See next page for caption.

Article
Extended Data Fig. 8 | Treating E. coli with an IspH inhibitor prodrug
disrupts the levels of IspH and several proteins in essential bacterial
metabolic and synthesis pathways. a, Immunoblots measure relative levels of
E. coli IspH at 8 and 24 h after C23.28–TPP treatment or after conditional
knockdown in CGSC 8074 (ΔispH) grown on 1% glucose. b, Immunoblots
measure relative levels of IspH in clinical isolates of several pathogenic bacteria
at 8 and 24 h after C23.28–TPP treatment. For a, b, RpoD immunoblot serves as
loading control and the blots are representative of 3 technical replicates.
c, Unsupervised hierarchical clustering of 2,346 proteins resolved indicates
that the 3 biological replicates for each condition clustered together. A total
of 525 proteins were either up- or downregulated both after C23.28–TPP
treatment or after conditional knockdown in CGSC 8074 (ΔispH). d, Functions
or pathways that are significantly enriched 8 and 24 h after C23.28–TPP
treatment. Bars indicate the −log10(P value) with the number of proteins
identified in each category next to the respective bar. The bars are colour-
coded for the percentage of proteins in the pathway that are up- or
downregulated. e, Venn diagram comparing the overlap in downregulated
(>2-fold) proteins at 8 or 24 h after C23.28–TPP treatment or after conditional
knockdown in CGSC 8074 (ΔispH). f, Proteins important for lipid synthesis,
ribosome modification, respiration, cell division, tRNA aminoacylation, DNA/
RNA synthesis, DNA repair, amino acid (AA) synthesis and lipopolysaccharide
(LPS) cell-wall synthesis pathways are among those significantly downregulated.
Associated with Extended Data Fig. 9a. P < 0.05 and FDR < 5%. g, Venn diagram
comparing the overlap in upregulated (>2-fold) proteins at 8 or 24 h after
C23.28–TPP treatment or after conditional knockdown in CGSC 8074 (ΔispH).
h, Ribosome component proteins or proteins important for multidrug efflux
and oxidative defence pathways are among those significantly upregulated.
Associated with Extended Data Fig. 9b. P < 0.05 and FDR <5%.

Extended Data Fig. 9 | Escherichia coli metabolic pathways up- or
downregulated after IspH inhibition. a, b, Pathway analysis of 323
downregulated (a) (Extended Data Fig. 8e, f) or 60 upregulated (b)
(Extended Data Fig. 8g, h) proteins from a proteomic screen comparing ΔispH
E. coli and E. coli after C23.28–TPP treatment to untreated wild-type E. coli.

Article
Extended Data Fig. 10 | See next page for caption.

Extended Data Fig. 10 | By dual action, IspH prodrugs expand and activate
Vγ9Vδ2 T cells and reduce the emergence of antibiotic resistant bacteria.
a, Top, uninfected (UI) human PBMCs or those co-infected with E. coli analysed
for expansion of CD3+Vγ9TCR+ (γδ) T cells and compared to untreated (UT) or
TPP-, prodrug (C23.07–TPP)- or kanamycin (Kan)-treated PBMCs. Middle,
bottom, gated γδ T cell populations analysed for cytotoxic granule proteins
granulysin (GNLY) and perforin (middle) or cell surface markers of T cell
activation CD69 and HLA-DR (bottom). Representative of 4 independent
experiments (4 donors). Percentage of Vγ9+ T cells from CD3+ population and
the percentage of Vγ9+ T cells with increased expression of granulysin,
perforin, CD69 and HLA-DR were plotted in the respective graphs. Data are
mean s.e.m. **P < 0.05, ***P < 0.001, the remainder are not significant; one-
way ANOVA relative to untreated sample. b, Uninfected human PBMCs or those
co-infected with V. cholerae (top) or M. smegmatis (bottom) were analysed for
expansion of CD3+Vγ9TCR+ (γδ) T cells and compared to untreated or TPP-,
prodrug (C23.07–TPP)- or kanamycin-treated PBMCs (n = 4 biological
replicates). Percentage of Vγ9+ T cells from the CD3+ population were plotted in
the respective graphs. Data are mean s.e.m. ***P < 0.001, rest not significant,
calculated by one-way ANOVA relative to untreated sample. c, Human PBMCs
co-infected with kanamycin-resistant E. coli or V. cholerae can kill neither on
their own. Addition of C23.07–TPP kills both V. cholerae and E. coli (n = 2
biological and 3 technical replicates). Data are mean s.e.m. ***P < 0.001, ns,
not significant; unpaired Student’s t-test relative to untreated samples. d, γδ
T cell depletion from human PBMCs is verified by treating depleted (γδ−) and
undepleted human PBMCs treated with 10 μM HMBPP and 50 ng ml−1 IL-1 5.
Representative of 4 independent experiments (4 donors). Percentage of Vγ9+
T cells from the CD3+ population on different days were plotted in the
respective graphs. Data are mean s.e.m. ***P < 0.001 comparing γδ depleted
and undepleted PBMCs calculated by unpaired t-test. e, f, Multidrug-resistant
clinical isolates of Vibrio (e) and Klebsiella (f) grown for 18 serial passages in
media (RPMI+ 10% human serum) containing DAIA prodrug (C23.28–TPP) or
conventional antibiotics (hygromycin (Hyg) or streptomycin (Strep)) gradually
develop resistance when measured by CFU (top). Similar serial passages in
presence of human PBMC inhibit the development of resistance against the
DAIA prodrug but not against hygromycin or streptomycin. Passages in γδ
depleted (γδ−) PBMCs show higher antibiotic resistance against the DAIA
prodrug. (n = 3 technical replicates). Data are mean s.e.m. ***P < 0.001, NS, not
significant; unpaired Student’s t-test. g, C57BL/6 mice infected with V. cholerae
are treated with TPP or the DAIA prodrug C23.28–TPP and monitored daily
from day 2 post-infection for survival (n = 10 mice per group). h, Vibrio load in
different organs at the experimental endpoint measured as CFU mg−1 (n = 10
mice with 3 technical replicates), comparing changes in bacterial CFU in
C57BL/6b mice after C23.28–TPP treatment. Data are mean s.e.m.
***P < 0.001; unpaired Student’s t-test, relative to TPP-treated mice. i, Hu-mice
injected i.p. with HMBPP at different concentrations show dose-dependent
expansion of γδ T cells but not αβ T cells in blood taken every day for a week
(n = 2 mice per group).

Article
Extended Data Fig. 11 | γδ T cells expand in the tissues of prodrug-treated,
E. coli-infected humanized mice. a, Hu-mice infected with E. coli (green) and
treated with TPP or the prodrug C23.07–TPP are compared for expansion of
Vδ2 TCR+ T cells (red) in multiple organs at day 5 post-infection. DAPI, blue.
Scale bar, 100 μm (representative of samples tested from 5–6 Hu-mice).
b, Vδ2 antibody validated for immunofluorescence staining of formalin-fixed
human PBMCs that are γδ expanded (HMBPP + IL15) or γδ depleted. HepG2 and
Vero cells serve as negative controls. c, Anti-E. coli antibody validated for
immunofluorescence staining of formalin-fixed HepG2 cells co-infected with
E. coli BL21 strain. HepG2 without E. coli serves as a negative control.