Identifying drug targets in tissues and whole blood with thermal-shift profiling

Article abstract of DOI:10.1038/s41587-019-0388-4

Monitoring drug–target interactions with methods such as the cellular thermal-shift assay (CETSA) is well established for simple cell systems but remains challenging in vivo. Here we introduce tissue thermal proteome profiling (tissue-TPP), which measures

Full text of DOI:10.1038/s41587-019-0388-4

Letters
https://doi.org/10.1038/s41587-019-0388-4
Identifying drug targets in tissues and whole
blood with thermal-shift profiling
Jessica Perrin1,3, Thilo Werner1,3, Nils Kurzawaꢀ ꢀ2,3, Anna Rutkowska1, Dorothee D. Childs2,
Mathias Kalxdorf1, Daniel Poeckel1, Eugenia Stonehouse1, Katrin Strohmer1, Bianca Heller1,
Douglas W. Thomson1, Jana Krause1, Isabelle Becherꢀ ꢀ2, H. Christian Eberl1, Johanna Vappiani1,
Daniel C. Sevin1, Christina E. Rau1, Holger Franken1, Wolfgang Huber2, Maria Faelth-Savitski1,
Mikhail M. Savitski2*, Marcus Bantscheffꢀ ꢀ1* and Giovanna Bergaminiꢀ ꢀ1*
Monitoring drug–target interactions with methods such as and only after heating proteins are extracted with a mild deter-
the cellular thermal-shift assay (CETSA) is well established gent (Supplementary Fig. 1), resulting in better solubilization of
for simple cell systems but remains challenging in vivo. Here membrane-bound proteins. The percentage of membrane pro-
we introduce tissue thermal proteome profiling (tissue-TPP), teins detected in crude cell extracts following this protocol was
which measures binding of small-molecule drugs to proteins fourfold higher than in detergent-free extracts (PBS extracts) and
in tissue samples from drug-treated animals by detecting almost comparable to that in intact cells (Supplementary Fig. 2a
changes in protein thermal stability using quantitative mass and Supplementary Data 1). In line with previous reports, melt-
spectrometry. We report organ-specific, proteome-wide ing temperatures (Tm) measured in HepG2 cells and HepG2 crude
thermal stability maps and derive target profiles of the non- extracts showed a correlation of r=0.58, an expected value owing
covalent histone deacetylase inhibitor panobinostat in rat to differences in concentration of cellular co-factors and the altera-
liver, lung, kidney and spleen and of the B-Raf inhibitor vemu- tion of protein–protein interactions (Supplementary Fig. 2b).
rafenib in mouse testis. In addition, we devised blood-CETSA When comparing Tm values in crude extracts from rat kidney, liver,
and blood-TPP and applied it to measure target and off-target spleen and testis, high correlations were determined (r=0.87–0.93;
engagement of panobinostat and the BET family inhibitor Supplementary Fig. 2c and Supplementary Data 2).
JQ1 directly in whole blood. Blood-TPP analysis of panobino-
We then performed two-dimensional (2D)-TPP3 experiments
stat confirmed its binding to known targets and also revealed with panobinostat in crude extracts from mouse and rat liver tissues
thermal stabilization of the zinc-finger transcription factor and from the HepG2 cell line obtaining similar target profiles across
ZNF512. These methods will help to elucidate the mechanisms all extracts (Supplementary Fig. 3a,b, Supplementary Table 1 and
of drug action invivo.
Supplementary Data 3). Notably, the multipass membrane FADS1,
TPP determines thermal stability across the proteome by mea- which is localized to the membrane of the endoplasmic reticulum
suring the soluble fractions of proteins upon heating cells and and has been previously reported to be stabilized in live cells and not
lysates to a range of temperatures. It also enables the proteome-wide detected in PBS extracts8, was stabilized in crude extracts and could
assessment of drug–protein interactions by combining quantita- now be identified as a direct target of panobinostat (Supplementary
tive mass spectrometry1 with CETSA2–4. The CETSA methodology Fig. 3c,d). This data shows that crude extracts derived from animal
has recently been applied to tissues for detecting thermal stability tissues can be used to generate in vitro cross-species selectivity pro-
changes of individual drug targets upon in vivo dosing2,5; however, a files of small molecules with TPP.
strategy for the assessment of proteome-wide thermal stability and
We then applied tissue-TPP to organ sections derived from a rat
compound-selectivity profiling in vivo has been lacking. The ability in vivo study with panobinostat. Before proceeding with the analy-
to translate target and off-target engagement measurements from sis of the dosed animals, we analyzed the thermal stability of tis-
in vitro systems to in vivo studies is critical for predicting not only sue proteomes in the control animals. Three control rats were killed
efficacy but also adverse reactions6.
and tissue samples derived from liver, lung, kidney and spleen were
Here we report a TPP-based method, tissue-TPP, to assess ther- directly incubated at different temperatures ranging from 37°C to
mal stability across tissue proteomes and to study the engagement 64°C. After mechanical tissue disruption with a bead-based pro-
of targets and off-targets in organs of animals dosed with the non- cedure and addition of DNAse and a mild detergent, the soluble
covalent small-molecule histone deacetylase (HDAC) inhibitor, protein fraction was analyzed by quantitative mass spectrometry
panobinostat, taking advantage of the extensive in vitro character- resulting in a total of 11,032 proteins detected across all four organs
ization of its selectivity in biochemical and cell-based assays7,8.
(Supplementary Data 4 and Supplementary Fig. 4a). The resulting
We first determined protein targets of panobinostat in extracts proteome-widethermal-stabilityprofilesshowedsimilarTm distribu-
derived from tissues and compared this profile to those derived tions and high pairwise correlations (Fig. 1a) irrespective of protein-
from cell lines using a crude-extract protocol4. In this lysis strat- expression levels in the different organs (Supplementary Fig. 4b,c).
egy, cells are broken up mechanically in the absence of detergent Interestingly, the thermal-stability profiles in organs were also very
1Cellzome GmbH, GlaxoSmithKline, Heidelberg, Germany. 2Genome Biology Unit, European Molecular Biology Laboratory, Heidelberg, Germany.
3These authors contributed equally: Jessica Perrin, Thilo Werner, Nils Kurzawa. *e-mail: mikhail.savitski@embl.de; marcus.x.bantscheff@gsk.com;
giovanna.2.bergamini@gsk.com
NATuRE BIoTECHNoloGy | www.nature.com/naturebiotechnology

Nature BiotechNology
Letters
similar to that generated in the rat cell line RBL-2H3 (Fig. 1b,c,
We also analyzed the melting behavior of protein complexes
Supplementary Fig. 4d and Supplementary Data 2 and 4). Despite in different organs using statistical methods previously reported
the high diversity in structure and density of the different organs, for cultured and primary cells13,15,16 (Supplementary Fig. 8a,b
heating appeared to have homogeneously affected all samples, dem- and Supplementary Data 8 and 9). Across all tissues, stabilities
onstrating the general applicability of thermal proteome profiling of 57 protein complexes, either of the whole complex or of indi-
to tissues.
vidual subunits, could be compared on the basis of melting points
To carry out a deeper analysis of the proteome thermal stabil- of their subunits (Supplementary Figs. 8c and 9a). Among those,
ity across tissues, we used a non-parametric analysis of response the proteasome complex showed consistently higher thermal
curves9 (NPARC; F-statistic-based strategy) for detecting differen- stability in the 20S proteasomal core subunits than the 19S sub-
tial melting behaviors (Supplementary Data 5). Incomplete dena- unit (Supplementary Fig. 9b). Differential thermal behavior was
turation curves were included in the analysis and the number of observed for Psmd4 and Psmd9, which was recently identified as
tested proteins increased by 16–20% as compared to the number of a chaperone involved in proteasome assembly, but not a constitu-
proteins with a defined melting point (Supplementary Fig. 5a). A tive member of the assembled proteasome complex17. Interestingly,
total of 4,110 melting curves could be fitted in liver, 3,411 in kidney, the subunit Psmd5 showed simultaneous melting with the other 19S
3,703 in lung and 4,218 in spleen (Supplementary Data 6), 2,144 components exclusively in kidney, suggesting specific regulation of
of those were overlapping in all four tissues (Supplementary Fig. the 19S complex in this organ. Also, the H+-transporting V-type
5b). While the overall thermal profiles were consistent across the ATPase, which is known to be required in kidney for deacidifica-
different organs, significantly different thermal denaturation curves tion of the blood18, showed a significantly higher median melting
between any of the four organs were identified for 637 proteins point specifically in this organ (Supplementary Data 8). Detailed
(P≤0.001) which were enriched for Gene Ontology (GO) terms comparison of the individual complex subunits revealed that along
such as ‘fatty acid β-oxidation’ and ‘cellular modified amino acid with an overall increase in thermal stability of the V1 subcomplex
metabolic process’ (Supplementary Tables 2 and 3, Supplementary members, the melting point of the C subunit of V1 (v1c1) was sig-
Fig. 5c and Supplementary Data 7). Proteins involved in these nificantly increased in kidney as compared to the other tissues (Fig.
biological processes showed higher thermal stabilities and expres- 1g). This would suggest the presence of a higher population of fully
sion levels in liver and kidney as compared to lung and spleen assembled complexes in kidney as the C subunit is known to dis-
(Supplementary Figs. 5d,e and 6a), potentially reflecting the high sociate from V1 upon detachment of V1 from V0, providing a ratio-
capacity for fatty acid β-oxidation of hepatocytes and proximal nale for the higher thermal stability measured for the complex in
tubules in the kidneys10 and the key role of kidney and liver in the this organ19 (Fig. 1g).
synthesis and interorgan exchange of amino acids in the body11. Of
This data suggests that tissue-TPP, by reflecting protein post-
note, a comparative analysis of protein abundances and thermal translational modifications as well as protein–protein interactions,
stabilities showed no overall correlation between these two datasets could be potentially used to monitor physiological processes in dif-
suggesting that tissue-TPP could provide information on protein ferent organs in vivo.
states beyond levels of protein expression (Supplementary Fig. 6b).
We then analyzed tissue samples derived from three rats admin-
Several factors can explain altered thermal stability of a protein, istered intravenously with 10mgkg−1 panobinostat and killed 90
including post-translational modifications and protein–protein minutes post-dose (Fig. 2a and Supplementary Fig. 1a). Following
interactions as previously shown in cell cultures12,13; however, so far quantitative mass spectrometry analysis (Supplementary Fig. 10a
it has been unclear whether such changes could be also detected andSupplementaryData4and10), significantstabilizationofHdac1,
in vivo. When analyzing the signal transducers and activators of Hdac2 and Ttc38 was observed in all analyzed tissues as shown for
transcription (Stat) family proteins, we observed significant dif- spleen in Fig. 2b and liver, kidney and lung in Supplementary Fig.
ferences (P≤0.001) in thermal stability, in particular, lower melt- 10b. Hdac6 was stabilized in both lung and spleen, while stabiliza-
ing points in liver as compared to the other organs (Fig. 1d and tion of Fads1, could be observed exclusively in lung. The different
Supplementary Fig. 7a). As Stat signaling is also regulated by tyro- off-target profiles obtained across tissues reflected heterogeneous
sine phosphorylation and a role in proliferation has been reported levels of protein expression and of drug exposure (Supplementary
in hepatocytes14, we speculated that the observed stability changes Data 11). Hdac1, Hdac2 and Ttc38 were less stabilized in liver
may reflect, at least in part, differences in the phosphorylation state. where the lowest concentrations of free panobinostat were mea-
Indeed, in comparison to kidney, we detected a higher proportion sured (Supplementary Fig. 10c)—in agreement with previous
of phosphorylated Stat5 (pStat5) in liver (Fig. 1e), which showed in vivo studies reporting the rapid metabolism of this drug in rats20.
lower melting temperature than the total Stat5 pool (Fig. 1f and In addition, we detected stabilization of dehydrogenase–reductase
Supplementary Fig. 7b).
SDR family member 1, Dhrs1, which was not observed in any tissue
Fig. 1 | Analysis of proteome thermal stability in tissues and cell lines. a, Scatter plots and density plots of protein Tm in rat tissue pieces. In the scatter
plots, the mean Tm (nꢀ=ꢀ3) of robustly quantified proteins (blue dots) are compared pairwise in the indicated rat tissues. The correlation coefficient
represents Pearson’s ρ. Density plots illustrate the Tm values distribution over temperature. b, Heat map of protein thermal stabilities in live RBL-2H3
cells and rat spleen. The median relative abundance across replicates (RBL-2H3, nꢀ=ꢀ2; spleen samples, nꢀ=ꢀ6) at the indicated temperature is shown for
each protein as fold change relative to the lowest measured temperature (37ꢀ°C). The 1,739 proteins robustly quantified in both sample types are plotted
in the same order. c, Scatter plots of protein Tm values in RBL-2H3 cells and rat spleen. Blue dots represent the mean Tm of robustly quantified proteins
(RBL-2H3, nꢀ=ꢀ2; spleen samples, nꢀ=ꢀ3). The correlation coefficient represents Pearson’s ρ. d, Stat5b melting curves in rat kidney (red line, Tmꢀ=ꢀ46ꢀ°C) and
liver (orange line, Tmꢀ=ꢀ44ꢀ°C) samples. e, Relative signal ratio of pStat5a/b (Ser726 and Ser731) to total Stat5b in liver (orange bar) and kidney (red bar)
samples by immunoblot (mean of two samples with data points overlaid). f, Melting curve of pStat5a/b (Tmꢀ=ꢀ41ꢀ°C) and total Stat5b (Tmꢀ=ꢀ43ꢀ°C) in rat
liver samples by immunoblot (mean of two samples). g, Thermal stabilities of the V-type ATPase subunits in rat tissue samples. Top, box plots of Tm values
of the different subunits of the V1 complex across tissues (nꢀ=ꢀ6). The significance levels were tested by two-sided Welch’s t tests for each subunit in
kidney versus any other tissue; ***Pꢀ<ꢀ0.001. Kidney versus liver, Pꢀ=ꢀ3.8ꢀ×ꢀ10−10; kidney versus spleen, Pꢀ=ꢀ2.4ꢀ×ꢀ10−9; and kidney versus lung, Pꢀ=ꢀ4.2ꢀ×ꢀ10−11
.
The center line in box plots is the median, the bounds of the boxes are the interquartile range (IQR) and the whiskers correspond to the highest or lowest
respective value or if the lowest or highest value was an outlier (>1.5ꢀ×ꢀIQR from the bounds of the boxes) it is exactly 1.5ꢀ×ꢀIQR. Bottom, Tm values of the
V-type ATPase are mapped onto its schematic representation and visualized as color gradients.
NATuRE BIoTECHNoloGy | www.nature.com/naturebiotechnology

Nature BiotechNology
Letters
crude extracts or in experiments with cell lines. To further investi- 1a). In agreement with the in vivo study, we detected in rat spleen
gate compound effects in tissue, we optimized a model in which live slices treated with panobinostat stabilization of Dhrs1 (Fig. 2c,
tissue slices can be incubated ex vivo (Fig. 2a, Supplementary Fig. Supplementary Data 12), as well as metabolism of the compound
a
b
RBL-2H3
Rat spleen
Kidney
37 °C
64 °C
66.3 °C 37 °C
0.08
0.06
0.04
0.02
0
Liver
Correlation:
0.90
60
50
40
Lung
Correlation:
0.84
Correlation:
0.87
60
50
40
Spleen
c
Correlation:
0.89
Correlation:
0.86
Correlation:
0.85
Correlation:
0.93
60
50
40
60
50
40
40
50
60
40
50
60
40
50
60
40
50
60
40
50
60
Tm spleen tissue (°C)
Tm tissue pieces (°C)
d
e
f
Stat5b
Kidney
Liver
0.20
1.0
1.0
0.15
0.01
0.05
0
Total stat5
0.5
0
0.5
0
Liver
pStat5
40 45 50 55 60 65
Temperature (°C)
Liver
Kidney
40 45 50 55 60 65
Temperature (°C)
g
Kidney
Liver
Lung
Spleen
60
56
52
48
***
T
m (°C)
H+-transporting V-type ATPase
50 54 58 62
E
E
E
E
E
A
A
A
A
B
A
A
A
A
V1
A
A
B
B
B
B
G
G
G
C
G
C
G
C
D
F
D
F
D
D
D
F
F
F
C
H
H
C
H
H
H
a
V0
Kidney
Liver
Lung
Spleen
c-ring
NATuRE BIoTECHNoloGy | www.nature.com/naturebiotechnology

Nature BiotechNology
Letters
Animal treatment
Tissue preparation
O
(E)
OH
N
H
H
N
HN
Panobinostat
MS-based identification
and quantification
and data analysis
Protein extraction and
aggregates separation
Heat treatment
3 min, 42–64 °C
Spleen
Weighing
Tissue pieces
- Bead-based
disruption
- Detergent/DNase
- Centrifugation
Tissue preparation
Ex vivo incubation
m/z
O
(E )
OH
N
H
H
N
HN
Panobinostat
150 min, 37 °C
Tissue slices
Tissue punches
b
Spleen
30
Dhrs1
Hdac1
12
10
8
Hdac1
20
10
Hdac2
Ttc38
Hdac2
Pycr3
Fam136a
Ttc38
Dhrs1
Hdgf
Hdac6
6
Hdac10
Mier1
4
Zadh2
0
2
–10
0
0
10
20
30
–10
0
10
20
30
Distance score (panobinostat/vehicle)
Stabilization score rep1
Fig. 2 | Panobinostat profiling in rat tissue invivo and ex vivo. a, Tissue-TPP work flow for tissue pieces derived from in vivo dosed rats (upper panel).
Rats were dosed intravenously with panobinostat or vehicle. One and a half hours after injection, the animals were killed and lung, spleen, liver and kidney
were dissected into pieces. Samples were weighed and heat treated at temperatures from 42 to 64ꢀ°C for 3ꢀmin in buffer. After one freeze–thaw cycle,
tissue samples were disrupted using ceramic beads and incubated with detergent and DNase for 60ꢀmin at 4ꢀ°C. The soluble proteins were separated
by centrifugation, identified and analyzed by LC–MS/MS. Work flow for ex vivo tissue-TPP in rat spleen punches (lower panel). Punches generated from
precision-cut slices (thickness 500ꢀµm, diameter 4ꢀmm) from rat spleen were treated with panobinostat (50, 5, 0.5 and 0.05ꢀµM) for 150ꢀmin at 37ꢀ°C
and 5% CO2. After heat treatment, the analysis was performed as for the in vivo samples. b, Volcano plot showing protein targets stabilized in rat spleen
following in vivo administration of panobinostat (10ꢀmgꢀkg−1). Aggregated P values (−log10 transformed) and distance scores (log2 ratios aggregated across
temperatures) of proteins quantified in spleens of panobinostat-dosed (nꢀ=ꢀ3) versus vehicle-dosed (nꢀ=ꢀ3) rats are displayed. The ratio-based approach
used for the analysis included two-sided Student t tests for ratios between treatment and control groups obtained at each temperature, aggregation of
retrieved P values per protein by Brown’s method and multiple testing adjustment by the method of Benjamini–Hochberg (BH). Black dots, significantly
affected proteins after BH correction; light gray dots, proteins that were not significantly affected; previously described targets of panobinostat8 are labeled
in orange. c, Scatter plot of protein targets of panobinostat following ex vivo incubation in spleen tissue punches. Stabilization scores were displayed
for two replicate 2D-TPP experiments (rep1, x axis; rep2, y axis). The stabilization score is a measure of monotonously increasing or decreasing protein
thermal stability in response to increasing treatment concentrations (50, 5, 0.5 and, 0.05ꢀµM) across different temperatures. Highlighted proteins were
found as hits in both replicates according to the criteria described in ref. 8, previously described targets of panobinostat8 are labeled in orange.
(Supplementary Fig. 11, Supplementary Data 13–17). Overall, these
To explore the applicability of tissue-TPP to other inhibi-
findings indicate that stabilization of Dhrs1 can only be observed in tors, we incubated ex vivo the clinical drug vemurafenib21 in
metabolically active live tissue and might be due to a downstream fragments of rat testis and demonstrated stabilization of targets
cellular event since it does not result from direct binding of pano- and off-targets (Supplementary Fig. 12 and Supplementary Data
binostat. In addition to stabilization of the known targets Hdac1, 18). These data show the suitability of ex vivo tissue-TPP for
Hdac2, Hdac10 and Ttc38 (Fig. 2c and Supplementary Data 12), investigating engagement of drug targets and off-targets in ani-
Mier1, a known Hdac1/2 interactor7, was stabilized by panobino- mal organs with the potential for future applications to human
stat, further indicating the suitability of tissue-TPP for analyzing models. Importantly, in animals dosed with non-covalent small-
protein–protein interactions in organs.
molecule inhibitors, the engagement measured for the identified
NATuRE BIoTECHNoloGy | www.nature.com/naturebiotechnology

Nature BiotechNology
Letters
40,000
35,000
30,000
25,000
HDAC2
20,000
15,000
10,000
5,000
Heat treatment
PBMC extraction
O
(E )
OH
N
H
H
N
HN
12
40
55 115 15,320
- Lysis with
detergent
Panobinostat
MW (kDa)
- Filter plate
WES
RBC
MS-based identification
and quantification
and data analysis
3 min, 42–64 °C
b
Human whole blood
HDAC2
1.5
1.0
0.5
0
10
5
m/z
HDAC1
ZNF512
TTC38
HDAC2
HDAC6
pEC50
Belinostat
0
5.3
Romidepsin 6.8
Panobinostat 7.5
–5
–9
–8
–7
–6
–5
–4
–5
0
5
10
Inhibitor concentration (log M)
Stabilization score rep1
Fig. 3 | Thermal-shift assays in whole blood. a, Workflow of thermal-shift assays in whole blood. Fresh human or rat whole blood was treated ex vivo with
panobinostat for 90ꢀmin at 37ꢀ°C and directly heated in PCR plates (3ꢀmin) with subsequent separation of PBMCs. After lysis and separation of aggregates,
protein levels were measured in the soluble fraction by LC–MS/MS or WES, in blood-TPP or blood-CETSA, respectively. b, Scatter plot of protein targets of
panobinostat following ex vivo incubation in human whole blood. Stabilization scores were displayed for two replicate 2D-TPP experiments
(rep1, x axis; rep2, y axis). Highlighted proteins were found as hits in both replicates according to previously described criteria8; previously described
targets of panobinostat8 are labeled in orange. c, Dose-dependent stabilization of HDAC2 by panobinostat, belinostat and romidepsin measured by
blood-CETSA at 54ꢀ°C (nꢀ=ꢀ3 donors); each data point is the mean of technical duplicates. Stabilization of HDAC2 in response to inhibitor treatment was
measured relative to vehicle-treated control and fitted with saturation curves to determine the concentrations leading to half-maximal stabilization, pEC50
(−log10-transformed compound concentrations).
targets and off-targets by tissue-TPP correlates with drug expo-
sure in tissue.
We then investigated dose-dependent stabilization of panobi-
nostat targets and off-targets in human blood samples subjected
Next, we devised a thermal-stability-based method to mea- to temperatures from 44 to 54 °C. Mass-spectrometry analysis
sure target and off-target engagement directly in whole blood. revealed dose-dependent stabilization of the targets HDAC1,
Monitoring direct target engagement in vivo, together with effi- HDAC2 and HDAC6 and the off-target TTC38 at similar con-
cacy biomarkers, enables the optimization of compound dosing centrations observed in cell lines and tissue sections (Fig. 3b,
regimens, which is essential to maximize efficacy as well as to avoid Supplementary Fig. 14a and Supplementary Data 19). Whilst
adverse effects. While in animal studies necropsies are typically other off-targets such as PAH and FADS1/2 were not detected in
accessible for analysis, in human studies blood samples are routinely the TPP analysis of blood samples owing to the low expression
analyzed and tissues biopsies are not always available.
levels of these proteins, we observed dose-dependent stabilization
We therefore optimized a procedure to prepare soluble proteins of the zinc finger protein ZNF512. This specific effect on ZNF512
from heated whole-blood samples for analysis by quantitative mass was never observed in the cell lines and tissues previously ana-
spectrometry. This procedure, referred to as blood-TPP, includes lyzed, most likely because of low levels of protein expression. To
heat treatment of whole blood, lysis of red blood cells and TPP analy- investigate the hypothesis that the hydroxymate moiety of pano-
sis of purified peripheral blood mononuclear cells (PBMCs) (Fig. 3a). binotstat could bind the zinc ions of this transcription factor, we
We observed similar thermal denaturation across the proteome of extended the analysis to other HDAC inhibitors and used HL-60
PBMCs purified from heated human whole blood and PBMCs heated and THP-1 cell lines, which express high levels of this transcrip-
after isolation from whole blood (Supplementary Fig. 13a) and found tion factor. Panobinostat and SAHA, another hydroxamate-con-
good correlation (r=0.79) of the melting points of the 2,054 proteins taining HDAC inhibitor, but not entinostat (which contains a
common to both datasets—of a total of 7,081 proteins identified in different zinc-binding group, orthoanalide), stabilized ZNF512,
heated whole blood (Supplementary Fig. 13b). This data indicates the as well as the transcription factor ZNF148 (Supplementary Fig.
suitability of blood-TPP profiling offering the fundamental advan- 14b and Supplementary Data 20). These observations suggest
tage with respect to TPP in purified cells of not introducing any that the previously unknown interactions with zinc-finger pro-
additional dilution step before heat treatment, thus enabling a phar- teins may contribute to the differential pharmacological profiles
macologically relevant measurement of target engagement.
of these HDAC inhibitors.
NATuRE BIoTECHNoloGy | www.nature.com/naturebiotechnology

Nature BiotechNology
Letters
To explore the applicability of blood-TPP to small molecules References
1. Bantschef, M., Lemeer, S., Savitski, M. M. & Kuster, B. Quantitative mass
spectrometry in proteomics: critical review update from 2007 to the present.
Anal. Bioanal. Chem. 404, 939–965 (2012).
other than HDAC inhibitors, we tested the bromodomain and extra-
terminal (BET) family inhibitor JQ1 (ref. 22) in human whole blood
and measured stabilization of its known targets4 (Supplementary
Fig. 14c and Supplementary Data 21).
2. Molina, D. M. et al. Monitoring drug target engagement in cells and tissues
using the cellular thermal shif assay. Science 341, 84–87 (2013).
3. Savitski, M. M. et al. Tracking cancer drugs in living cells by thermal
profling of the proteome. Science 346, 1255784 (2014).
4. Savitski, M. M. et al. Multiplexed proteome dynamics profling
reveals mechanisms controlling protein homeostasis. Cell 173,
260–274 (2018).
Finally, to enable higher throughput for the blood assay we opti-
mized an antibody-based read-out, referred to as blood-CETSA.
Using these settings, we measured dose-dependent engagement of
HDAC2 by panobinostat in human blood (pEC50 7.5, –log10-trans-
formed compound concentration leading to half-maximal stabi-
lization of the target protein), as well as two additional marketed
HDAC inhibitors, belinostat and romidepsin (Fig. 3c). The calcu-
lated pEC50 values are in line with previous reports and compound
ranking23. In addition, we tested blood-CETSA on blood freshly
derived from rats. In this setup, panobinostat induced dose-depen-
dent stability changes of Hdac2, with a pEC50 of 7.1 (Supplementary
Fig. 14d), in agreement with the measurements in human blood.
These data show how blood-CETSA could be used to monitor tar-
get engagement in preclinical studies enabling a direct cross-species
translation from animals to humans. A quantitative measurement of
the levels of target engaged after in vivo administration in preclini-
cal and clinical studies could be achieved by the aid of a compound
spiked-in control sample, which would enable normalization of
the stabilization levels measured in each individual blood sample
(Supplementary Fig. 14e).
5. Ishii, T. et al. CETSA quantitatively verifes in vivo target engagement
of novel RIPK1 inhibitors in various biospecimens. Sci. Rep. 7,
13000 (2017).
6. Morgan, P. et al. Can the fow of medicines be improved? Fundamental
pharmacokinetic and pharmacological principles toward improving phase II
survival. Drug Discov. Today 17, 419–424 (2012).
7. Bantschef, M. et al. Chemoproteomics profling of HDAC inhibitors
reveals selective targeting of HDAC complexes. Nat. Biotechnol. 29,
255–265 (2011).
8. Becher, I. et al. Termal profling reveals phenylalanine hydroxylase as an
of-target of panobinostat. Nat. Chem. Biol. 12, 908–910 (2016).
9. Childs, D Non-parametric analysis of thermal proteome profles reveals novel
drug-binding proteins. Mol. Cell Proteomics 18, 2506–2515 (2019).
10. Bhargava, P. & Schnellmann, R. G. Mitochondrial energetics in the kidney.
Nat. Rev. Nephrol. 13, 629–646 (2017).
11. van de Poll, M. C. G., Soeters, P. B., Deutz, N. E. P., Fearon, K. C. H. &
Dejong, C. H. C. Renal metabolism of amino acids: its role in interorgan
amino acid exchange. Am. J. Clin. Nutr. 79, 185–197 (2004).
12. Reinhard, F. B. M. et al. Termal proteome profling monitors ligand
interactions with cellular membrane proteins. Nat. Methods 12,
1129–1131 (2015).
Our study introduces tissue-TPP technology, which enables the
hypothesis-free discovery of drug targets and off-targets in organ-
derived crude extracts, in live tissue explants and following in vivo
drug treatment. The tissue-specific profiles obtained for panobi-
13. Tan, C. S. H. et al. Termal proximity coaggregation for system-wide
profling of protein complex dynamics in cells. Science 359,
1170–1177 (2018).
nostat highlight the importance of profiling compounds in tissues 14. Gao, B., Wang, H., Lafdil, F. & Feng, D. STAT proteins—key regulators of
anti-viral responses, infammation, and tumorigenesis in the liver. J. Hepatol.
that are exposed to the drug in vivo to ultimately assess pharma-
57, 430–441 (2012).
cology as well as potential hazards associated with drug off-targets.
15. Becher, I. et al. Pervasive protein thermal stability variation during the cell
Tissue-TPP also allows monitoring of processes at a molecular level
cycle. Cell 173, 1495–1507 (2018).
directly in organs, enabling in vitro-to-in vivo translation of funda-
mental biological findings.
16. Mathieson, T. et al. Systematic analysis of protein turnover in primary cells.
Nat. Commun. 9, 689 (2018).
17. Kaneko, T. et al. Assembly pathway of the mammalian proteasome
base subcomplex is mediated by multiple specifc chaperones. Cell 137,
914–925 (2009).
The blood-TPP and blood-CETSA approaches can readily iden-
tify drug targets and off-targets, and quantitatively measure target
engagement in whole blood. Given the generally easier access to
blood samples as compared to tissue biopsies, these approaches
could be widely applied in clinical settings. In summary, we intro-
duce and validate technologies broadly applicable both for funda-
mental research as well as in preclinical and clinical settings.
18. Al-Awqati, Q. Plasticity in epithelial polarity of renal intercalated
cells: targeting of the H+-ATPase and band 3. Am. J. Physiol. 270,
C1571–C1580 (1996).
19. Kane, P. M. Disassembly and reassembly of the yeast vacuolar H+-ATPase
in vivo. J. Biol. Chem. 270, 17025–17032 (1995).
20. Assessment report Farydak, procedure No. EMA/H/C/003725/0000
(European Medicines Agency, 2015).
online content
21. Bollag, G. et al. Clinical efcacy of a RAF inhibitor needs broad target
blockade in BRAF-mutant melanoma. Nature 467, 596–599 (2010).
22. Filippakopoulos, P. et al. Selective inhibition of BET bromodomains. Nature
468, 1067–1073 (2010).
Any methods, additional references, Nature Research reporting
summaries, source data, extended data, supplementary informa-
tion, acknowledgements, peer review information; details of author
contributions and competing interests; and statements of data and
code availability are available at https://doi.org/10.1038/s41587-
019-0388-4.
23. Khan, N. et al. Determination of the class and isoform selectivity
of small-molecule histone deacetylase inhibitors. Biochem. J. 409,
581–589 (2008).
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
Received: 3 April 2019; Accepted: 6 December 2019;
Published: xx xx xxxx
© The Author(s), under exclusive licence to Springer Nature America, Inc. 2020
NATuRE BIoTECHNoloGy | www.nature.com/naturebiotechnology

Nature BiotechNology
Letters
TPP-TR experiments in human blood. After addition of protease inhibitors (1:200,
P8340) fresh human heparinized blood samples (2ml per condition) were distributed
into a PCR plate and heat treated to ten temperatures covering the range from 37°C
to 62°C for 3min. PBMCs were retrieved by density centrifugation (Leucosep tubes,
Greiner Bio-One, 163288) for 10min at 1,000g at room temperature. After retrieval
of PBMCs, contaminating erythrocytes were removed by incubating the PBMCs with
red blood cell lysis buffer (Roche).The PBMCs were washed with PBS supplemented
with protease inhibitors. PBMCs were resuspended and incubated for 1h at 4°C
after supplementation of the buffer with Igepal CA-630 (final concentration 0.8%),
benzonase (1kUml−1) and 1.5mM MgCl2 at 4°C. The lysates were then centrifuged
for 3min at 500g at 4°C. The soluble protein fraction was retrieved by centrifugation
of the supernatant through a 384-filter plate (Thermo Fisher Scientific, 10675743)
Samples were further processed for LC–MS/MS analysis.
Methods
Reagents and cell culture. Reagents and media were purchased from Sigma
unless otherwise stated. Compounds panobinostat (S1030), belinostat (S1085),
romidepsin (S3020), vemurafenib (S1267), SAHA and entinostat were sourced from
Selleckchem. JQ1 was sourced from Carbosynth (FC43273). RBL-2H3 cells and HL-
60 cells were from DSMZ and HepG2 cells and THP1 cells were from ATCC. HepG2
cells were cultured in MEM (Gibco) containing 10% FBS (Gibco) supplemented with
1mM sodium pyruvate (Gibco) and 1% non-essential amino acids (Gibco). RBL-
2H3 cells were cultured in MEM containing 10% FBS and l-glutamine (2.0mM;
Gibco). HL-60 cells were cultured in RPMI (Gibco) 1640 medium containing
10% FBS. THP1 cells were cultured in RPMI supplemented with 10% FBS, 4.5gl−1
glucose, 10mM Hepes and 1mM sodium pyruvate. For all TPP experiments, PBS
was supplemented with EDTA-free protease inhibitors (Roche, 11814389001) or
additionally supplemented with a protease inhibitor mix (P8340). Anti-pSTAT5a/b
(Ser726, Ser731) and anti-STAT5b were purchased from Invitrogen. Anti-HDAC2
was purchased from CST (5113) and from Abcam (Ab124974). Anti-SOD1 was
purchased from Novus. Anti-Dhrs1 was purchased from Sigma (HPA000599).
Detailed information on the antibodies, the experimental design and data analysis is
included in the Life Sciences Reporting Summary.
TPP-TR experiments in human PBMC. PBMCs were first purified from 45ml of blood
by density centrifugation (Leucosep tubes, Greiner Bio-One, 163288) as described
above without the red blood cell lysis step. The PBMCs were then resuspended
in 400µl of PBS supplemented with protease inhibitors and 20µl aliquots were
distributed into a PCR plate and heat treated for 3min at the respective temperature
(ten temperatures covering 37.6–66.9°C). Heat-treated samples were then rested for
3min at room temperature and lysed as described above.
Preparation of cells or tissue crude extracts. The crude cell extract was prepared
as described4. HepG2 cells were dounce homogenized with 20 strokes in PBS
supplemented with protease inhibitors. Rat livers (Heidelberg Pharma) and mouse
livers (Charles River) were disrupted using ceramic beads (OMNI Bead Ruptor 24)
in PBS supplemented with protease inhibitors. The cell or tissue suspensions were
incubated for 1h with benzonase (1kUml−1, E1014) supplemented with 1.5mM
MgCl2 (M1028). Protein concentration was determined by the Bradford assay (Bio-
Rad, 500-0006) or the bicinchoninic acid assay (VWR, 733-1404). The preparation
of PBS extract is described in ref. 8.
2D-TPP experiments. 2D-TPP in HepG2, HL-60 and THP1 cells. Experiments
were performed as described8. In brief, HepG2 cells were treated with vehicle
(DMSO) or panobinostat (5, 1, 0.14 and 0.02µM) for 90min at 37°C and 5% CO2,
and collected by centrifugation. Cells were resuspended in PBS and transferred
to 96-well PCR plates and heated for 3min to one of the 12 tested temperatures
covering the range from 42°C to 64°C. Afer cell lysis (in presence of Igepal
CA-630 (I3021) 0.8%, MgCl2 (M1028) 1.5mM and benzonase 1kUml−1) the
aggregated proteins were separated by ultracentrifugation (20min, 100,000g).
Samples were further processed for LC–MS/MS analysis.
Preparation of tissue punches for live tissue compound incubation. Fresh spleen
and testis from 10-week-old Wistar Han rats were provided by Heidelberg Pharma.
Precision-cut slices of 500µm thickness were prepared from fresh spleen using a
Vibratome (VT1200S, Leica). Tissue punches of 4mm diameter were generated
using disposable biopsy punchers (pfm medical) and kept in cold medium
(DMEM high glucose (Gibco)) during the procedure, supplemented with 10% FBS,
penicillin–streptomycin 100Uml−1 (Gibco), l-glutamine (4.5gl−1), and 10mM
HEPES (Gibco)).
Using similar conditions, THP1 cells and HL-60 cells were treated with vehicle
(DMSO), SAHA (10, 2, 0.29 and 0.04µM), entinostat (50, 10, 1.4 and 0.2µM) or
panobinostat (5, 1, 0.14 and 0.02µM).
2D-TPP in crude cell or tissue extracts. Experiments were performed as described
above with the following changes. Crude extracts were exposed to vehicle (DMSO)
or panobinostat (5, 1, 0.14 and 0.02µM) for 15min at 25°C. The samples were
heated to 12 temperatures covering the range from 42°C to 64°C.
The epididymis and envelope of the testes were removed before cutting the
testis in the absence of medium into heterogeneously sized pieces using scalpels,
on ice. Note, TPP-temperature range (TPP-TR) experiments refer to thermal-
proteome profiles or thermal-stability profiles where the temperature is varied.
2D-TPP experiments refer to experiments where the temperature and the
concentration of the test substance are varied.
2D-TPP in live rat spleen tissues. Randomized spleen punch pieces (two per well)
were distributed over 48-well plates containing 600µl of DMEM high glucose
(Gibco), supplemented with 10% FBS, 100Uml−1 penicillin–streptomycin (Gibco),
4.5gl−1 l-glutamine and 10mM HEPES (Gibco). The spleen punch pieces were
treated with panobinostat (50µM, 5µM, 0.5µM and 0.05µM) for 2.5h at 37°C
and 5% CO2. The punch pieces were then transferred to a 96-well PCR plate
containing 100µl of PBS supplemented with protease inhibitors and the respective
panobinostat concentrations. Samples were heated in parallel for 3min to one
of the 12 tested temperatures covering the range from 42°C to 64°C and then
processed as described above.
An overview of all liquid chromatography–tandem mass spectrometry (LC–
MS/MS) experiments is provided in Supplementary Data 22.
TPP-TR experiments. TPP-TR experiments in cells. TPP-TR was performed as
described3,25. In brief, RBL-2H3 and HepG2 cells were treated with vehicle (0.5%
DMSO) for 90min at 37°C and 5% CO2, then collected by trypsination, aliquoted
and heated for 3min to a temperature range of 37 to 66.3°C in PBS in a PCR plate.
Igepal CA-630 (I3021) was added to a concentration of 0.8%, MgCl2 (M1028) to
1.5mM and benzonase to 1kUml−1, and samples were incubated for 1h at 4°C.
Te aggregated proteins were separated by ultracentrifugation (20min, 100,000g).
Samples were further processed for LC–MS/MS analysis.
2D-TPP in rat testis live tissues. Fragments of testis were weighed into 96-well
round-bottom cell culture plates containing 150µl of medium. The testis pieces
were treated with vemurafenib (40, 10, 2 and 0.4µM) for 1.5h at 37°C and 5% CO2,
and was transferred to a 96-well PCR plate, centrifuged for a short period and the
remaining medium was discarded. A volume of PBS (supplemented with protease
inhibitors and the respective vemurafenib concentrations) equivalent to the volume
of testis pieces (testis density, 1gml−1 (ref. 26)) was added to each well. The samples
were processed as described above.
TPP-TR experiments in crude extracts of cells or tissues. Experiments were
performed as described4,8. Crude extracts at 2mgml−1 protein concentration were
incubated with vehicle (DMSO) for 15min at 25°C, then 12 wells of a PCR plate
were filled with 100µl (each) of each reaction. Samples were heated in parallel for
3min to ten temperatures covering the range from 37°C to 66.3°C. Heat-treated
samples were then rested for 3min at room temperature. Igepal CA-630 (I3021)
was added to a concentration of 0.8%, MgCl2 (M1028) to 1.5mM and benzonase
to 1kUml−1, and samples were incubated for 1h at 4°C. Aggregated proteins
were separated by ultracentrifugation (20min, 100,000g). Samples were further
processed for LC–MS/MS analysis.
2D-TPP in human blood. Fresh human heparinized blood was treated with vehicle
(DMSO), panobinostat (5µM, 1µM, 0.143µM and 0.02µM) or JQ1 (40, 10, 2 and
0.4µM) for 1.5h at 37°C and 5% CO2 (1.5–2ml of blood per condition, depending
on availability). After addition of protease inhibitors (1:200, #P8340), the blood
was distributed into a PCR plate and heat treated for 3min to five temperatures
covering the range of 44 to 54°C. PBMCs were retrieved by density centrifugation
for 10min at 1,000g at room temperature using SepMate-15 tubes (STEMCELL,
85420) and Lymphoprep density gradient medium (STEMCELL, 07811) combined
with RosetteSe Human Granulocyte Depletion Cocktail (STEMCELL, 15624). The
PBMCs were washed with PBS supplemented with protease inhibitors. The PBMCs
were then lysed and processed as described above.
TPP-TR experiments in fragments of rat testis. Fragements of rat testis were added
to a 96-well plate and the weight of the individual minced pieces was tracked
using a scale. A volume of PBS supplemented with protease inhibitors equivalent
to the volume of testis tissue (testis density: 1gml−1 (ref. 26)) was added to each
well. Samples were heated in parallel for 3min to ten temperatures covering
the range from 37°C to 66.3°C. Heat-treated samples were rested for 3min at
room temperature and snap-frozen in liquid nitrogen. After thawing, the tissue
pieces were disrupted by bead ruptor. Igepal CA-630 (I3021) was added to a
Tissue-TPP: in vivo rat panobinostat study. In vivo study. Te rat in vivo study,
including preparation of tissue pieces and heat treatment was performed by
OncoDesign. Twelve 10-week-old male Han Wistar rats (Crl: WI(Han), Charles
River) were distributed randomly into two groups of six rats. Group 1 was dosed
intravenously (bolus) with 10mgkg−1 panobinostat and group 2 was dosed with
vehicle (2% DMSO, 48% PEG 300 and 2% Tween 80 in water, pH 7). All rats were
killed 90min afer dosing. Te lung, liver, kidneys and spleen were collected from
concentration of 0.8%, MgCl2 (M1028) to 1.5mM and benzonase to 1kUml−1
,
and samples were incubated for 1h at 4°C. Aggregated proteins were separated by
ultracentrifugation (20min, 100,000g). Samples were further processed for LC–
MS/MS analysis.
NATuRE BIoTECHNoloGy | www.nature.com/naturebiotechnology

Nature BiotechNology
Letters
each rat and cut with scalpels into two pieces. Half of the tissues were used for the
tissue compound concentration measurement and the other half was used to for
tissue-TPP analysis.
for single spectrum-to-sequence assignments, we required this assignment to be
the best match with a minimum Mascot score of 31 and a 10× difference over the
next best assignment (using these criteria, the decoy search results indicated a <1%
false-discovery rate); (ii) for multiple spectrum-to-sequence assignments and using
the same parameters, the decoy search results indicate <0.1% FDR. Reporter ion
intensities were read from the raw data and multiplied by ion accumulation times
(measured in ms) to yield a measure proportional to the number of ions (this measure
is referred to as ion area30). Spectra matching to peptides were filtered according to the
following criteria: mascot ion score >15; signal-to-background of the precursor ion
>4; and signal-to-interference >0.5 (ref. 31). Fold changes were corrected for isotope
purity and adjusted for interference caused by simultaneous elution near isobaric
peaks as estimated by the signal-to-interference measure30. Protein quantification
values were calculated from individual spectra matching unique peptides using a sum-
based bootstrap algorithm; 95% confidence intervals were calculated for all protein
fold changes that were quantified with more than three spectra30. An automatic
outlier removal procedure was performed on proteins with at least five quantified
PSMs (peptide–spectrum matches). In brief, each PSM and respective quantification
values were sequentially excluded from the fold-change calculation; in cases where the
confidence interval was then improved by at least 30%, the given quantification values
were permanently excluded from further analyses.
Determination of tissue compound concentration. Lung, liver, kidney and spleen
samples were homogenized in deionized water. Proteins were removed by
precipitation using acetonitrile. The parent compound concentration in the
supernatant was determined using LC–MS/MS analysis.
Heat treatment and lysis. The tissues were cut with scalpels into small pieces (20mg
to 50mg per piece). Fourteen pieces per organ were transferred into PCR plates.
Each piece of organ was weighed. Thirty microliters of cold PBS supplemented
with protease inhibitors was added to each tissue before being heated for 3min to
37°C or 12 temperatures covering the range from 42°C to 64°C. The tissue pieces
were then rested for 3min at room temperature and snap-frozen in liquid nitrogen.
After thawing, three volumes of PBS supplemented with protease inhibitors was
added per volume of tissue (lung, liver, spleen and kidney tissue density ~1gml−1
(ref. 26)), and the tissue pieces were disrupted by bead ruptor. Igepal CA-630
(I3021) was added to a concentration of 0.8% and MgCl2 (M1028) to 1.5mM, and
samples were incubated for 1h with benzonase (1kUml−1) at 4°C. Aggregated
proteins were separated by ultracentrifugation (20min, 100,000g). Samples were
further processed for LC–MS/MS analysis.
Stat5a and Stat5b have very similar sequences. When both were identified, only
peptides that were unique for one of the two genes were used for quantification,
resulting in a lower number of quantified peptides for Stat5a and Stat5b.
Preparation of samples for LC–MS/MS analysis. Proteins were denatured using
a sample based on lithium or sodium dodecyl sulfate (LDS or SDS, respectively)
(NuPAGE LDS sample buffer, NP008 or 2× sample buffer (200mM Tris-HCl
(933313), 250mM Trizma base (T1503), 20% glycerol (Merck, 1.04094), 4% SDS
(Fluka, 5030) and 0.01% bromophenol blue (B8026)) and by incubating the samples
for 30min at 25°C or 5 minutes at 95°C. Proteins were partially separated by means
of SDS gel electrophoresis or, alternatively, were directly processed (see below). Gel
lanes were cut into three slices covering the entire separation range (∼2cm) and
subjected to in-gel digestion with either LysC (Wako Chemicals) for 2h and trypsin
(Promega) overnight or trypsin only for 4h. Peptide samples were labeled with
TMT10 (Thermo Fisher Scientific) reagents. The labeling reaction was performed in
40mM triethylammonium bicarbonate, pH 8.53, at 22°C and quenched with glycine.
Alternatively, proteins were digested following a modified solid-phase-enhanced
sample preparation (SP3) protocol27. In brief, proteins in 2% SDS were bound to
paramagnetic beads (SeraMag Speed beads, GE Healthcare, 45152105050250,
651521050502) in filter plates (Multiscreen, Merck-Millipore, 10675743) and the
plates were filled with 50% ethanol. Following washing of beads with four times
200µl of 70% ethanol, beads were resuspended in trypsin and LysC in 0.1mM
HEPES, pH 8.5 containing TCEP and chloracetamide and incubated at room
temperature overnight and subjected to TMT labeling. Labeled peptide extracts were
combined and samples from 2D-TPP, TPP-CCR (compound-concentration range)
and TPP-TR experiments were subjected to additional fractionation on an UltiMate
3000 (Dionex) by reverse-phase chromatography at pH 12 on a 1-mm Xbridge
column (Waters) and up to 24 fractions were collected. Depending on the analytical
depth required, 8–24 fractions were analyzed by LC–MS/MS.
Analysis of TPP-TR experiments. Experiments were analyzed as described25.
Following LC–MS/MS analysis, protein fold changes at each temperature were
computed relative to the protein abundance at 37°C. These fold changes represent
the relative amount of non-denatured protein at the corresponding temperature.
After a global normalization procedure3, melting curves were fitted to the fold
changes of each individual protein according to the following equation:
1 � plateau
f ðTÞ ¼
þ plateau
ð1Þ
a
�b
1 þ e�
ðT
Þ
where T is the temperature and a, b and plateau are constants. The Tm of the protein
under the corresponding condition is given by the temperature at which the value
of the melting curve is 0.5. Model fitting was performed by non-linear regression
using the nls function in R, as described25. Tm values were only reported if predicted
to be located inside the measured temperature range.
Median Tm values were calculated and reported if the protein was quantified
with at least two unique peptides in both replicates. Tm values were visualized in
pairwise scatter plots (Fig. 1a,c, Supplementary Figs. 1c,d,3d and 10b), if they
additionally showed a Tm difference between replicates smaller or equal to 2°C and
an average fitted plateau parameter of smaller or equal to 0.2.
Analysis of 2D-TPP experiments. Experiments were analyzed as described8.
Each 2D-TPP experiment consisted of six TMT10 experiments (except the 2D-
TPP experiments in whole blood which consisted of three TMT10 experiments),
each containing the four compound concentrations and one vehicle control (at
two consecutive temperatures). For each TMT10 experiment, the two vehicle
conditions at the two selected temperatures were used as the reference for fold-
change calculations for the two sets of four compound concentrations and the
vehicle condition at the respective temperatures. The five different fold change
values at each temperature condition were transformed to a range of between
0 and 1.
The following filtering criteria were used to consider a protein as stabilized
in response to the compound treatment: proteins had to be identified by more
than one unique peptide (qupm>1) in at least one of the experiments of the
set. Proteins were required to be stabilized by the compound treatment at the
maximum concentration by at least 50% (untransformed fold changes) as
compared to the vehicle condition, in at least two adjacent temperature conditions.
A sigmoidal curve according to the following equation was fitted to derive
pEC50 values:
Liquid chromatography–tandem mass spectrometry. Samples were dried
in vacuo and resuspended in 0.05% trifluoroacetic acid in water. Half of the
sample was injected into an Ultimate3000 nanoRLSC (Dionex) coupled to a
Q-Exactive (Thermo Fisher Scientific). Peptides were separated on custom-
made 50cm×100um (ID) reverse-phase columns (Reprosil) at 55°C. Gradient
elution was performed from 2% acetonitrile to 40% acetonitrile in 0.1% formic
acid and 3.5% DMSO over 65min or 2h. Samples were online injected into
Q-Exactive or Orbitrap Fusion Lumos mass spectrometers. The Q-Exactive was
operated in the data-dependent top 10 mode. Mass spectrometry spectra were
acquired using a resolution of 70,000 and an ion target of 3×106. Higher-energy
collisional dissociation (HCD) scans were performed with 35% NCE at a resolution
of 35,000 (at an m/z of 200) or 17,500 setting on a research-grade Q-Exactiv
Plus mass spectrometer using the phase-constrained spectrum deconvolution
method28; the ion-target setting was set to 2×105 so as to avoid coalescence29.
The instruments were operated with Tune v.2.2 or v.2.3 and Xcalibur v.2.7 or
v.3.0.63. The Orbitrap Fusion Lumos operated with a fixed cycle time of 3s. Mass
spectrometry spectra were acquired at a resolution of 60,000 and an ion target of
4×105. HCD fragmentation was performed at 38% NCE at a resolution of 30,000
and an ion target 1×105. The instrument was operated with Tune v.2.1.1565.23
and Xcalibur v.4.0.27.10. Information on which type of sample preparation (partial
gel separation or SP3 protocol) and which instrument was used for MS/MS data
acquisition can be retrieved in Supplementary Data 22.
1
Y ¼
ð2Þ
1 þ 10ðlog EC
50�xÞslope
Generation of scatter plots of 2D-TPP experiments. For generating the plots
displayed in Fig. 2c and Supplementary Figs. 9a and 10e, a stabilization score δ was
computed for each protein in each replicate with at least one measurement with
qupm >1:
8
P
c
δstab:
;
i¼0 ðriÞ≥0
>
<
P
P
c
c
Peptide and protein identification and quantification. Mascot v.2.4 (Matrix
Science) was used for protein identification, a 10p.p.m. mass tolerance for peptide
precursors and 20-mDa (HCD) mass tolerance for fragment ions was selected.
Carbamidomethylation of cysteine residues and TMT modification of lysine
residues were selected as fixed modifications. Methionine oxidation, N-terminal
acetylation of proteins and TMT modifications of peptide N termini were selected
as variable modifications. The proteins were searched using Uniprot or Swissprot
for human samples. The following criteria were used for protein identifications: (i)
δ ¼ 0; i¼0 ðriÞ≥0 and δstab: <0 or i¼0 ðriÞ<0 and δdestab: >0
ð3Þ
>
:
P
c
δdestab:
;
i¼0 ðriÞ<0
where
� �
P
c
δ
stab: ¼ i¼0log2ðriÞ � maxij¼�01 log2 rj ; δdestab
:
� �
P
c
¼
i¼0log2ðriÞ � minij¼�01 log2 rj
NATuRE BIoTECHNoloGy | www.nature.com/naturebiotechnology

Nature BiotechNology
Letters
and where ri are the relative fold changes at treatment condition i as compared to
control and c is the number of concentrations.
where ΣrTvsC is the aggregated log2 ratio, s is the s.d. of the dataset defined as
10×median s.d. of ratios between replicates and P is the minimal accepted −
log10-transformed P value set to P=0.01. Proteins that were significant after the
Benjamini–Hochberg correction33 were highlighted. The cutoff was set to 0.1. The
Benjamini–Hochberg correction was calculated with a double amount of proteins,
as two P values were calculated per temperature and protein.
The stabilization score is a measure of monotonously increasing or decreasing
protein thermal stability in response to increasing treatment concentrations across
different temperatures.
Stabilization scores from both replicates were plotted as scatter plots and
detected targets passing the criteria above were labeled.
Functional inference of tissue differences in protein melting. To compare melting
curves between organs, we developed an approach using functional inference
(D.C. et al., manuscript in preparation). The approach is based on fitting two
competing models to the data. The null model states that the temperature response
is explained by a smooth melting curve according to equation (1). The alternative
model replaces this common function by organ-specific curves. To detect proteins
with different melting curves between organs, we constructed an F statistic that
quantified improvements in goodness-of-fit of the alternative model relative to the
null model:
Analysis of in vivo rat panobinostat study. Description of TMT10 sample set.
Each tissue type (lung, liver, spleen and kidney) was analyzed in a separate set of
TMT10 experiments. Each TMT10 sample set contained fve tissue extracts from
one vehicle-treated rat and fve tissue extracts from one panobinostat-treated rat,
heated at fve diferent temperatures (37°C and four additional temperatures;
Supplementary Fig. 7a). Tree TMT10 sample sets, when combined, enabled the
generation of a full temperature gradient with 13 distinct temperatures per rat.
Preprocessing. Protein fold-changes were calculated relative to the 37°C vehicle-
treated sample by dividing their sum ion areas. Vehicle-treated samples were
normalized at each temperature as follows. First, proteins fulfilling the following
criteria were selected for the calculation of the normalization factor: (i) quantified
with at least three individual spectra matching unique peptides and at least two
unique peptides; and (ii) a fold-change ≥0.2 relative to the 37°C-treated sample.
The empirical distribution of the fold change was generated for each replicate and
temperature. A normalization factor per temperature was calculated as the mean
of the fold change at the respective distribution maximum in each replicate. All
protein fold changes of vehicle-treated samples in the respective experiments were
multiplied with the normalization factors.
The fold changes of the panobinostat-treated sample were normalized as
described for vehicle-treated samples with the following changes: a normalization
factor per temperature and replicate was calculated as the difference between the
fold change at the distribution maximum of the panobinostat- and vehicle-treated
groups. These normalization factors were then applied to all protein fold changes
of the panobinostat-treated condition.
The normalized vehicle-treated and panobinostat-treated datasets were then
combined. Protein fold changes of the panobinostat-treatment condition were
scaled to the 37°C panobinostat-treated sample per protein.
RSS0i � RSS1i dfi2
Fi ¼

ð7Þ
RSS1i
dfi1
where RSS0i is the sum of squared residuals of the null model fit to protein i, RSSi1
is the sum of squared residuals of the alternative model fit to protein i, dfi1 is the
degrees of freedom parameter of the numerator RSS0i À RSSi1 and dfi2 is the degrees
of freedom parameter of the denominator RSS1i .
The values of dfi1 and dfi2 could not be analytically derived owing to correlated
and heteroscedastic residuals. To estimate the null distribution of Fi, we generated
a null dataset by randomly permuting the tissue labels per protein without
replacement. The F statistic numerator RSS0i À RSSi1 and the denominator RSS1i
were recalculated on the basis of this null data. The distribution parameters
2
were estimated by numerically fitting a χ distribution separately to the values of
RSS0i À RSS1i (to obtain dfi1) and of RSSi1 (to obtain dfi2) in R. A scaling parameter
2
for the χ distribution was calculated from the empirical distribution of the values
of RSS0i À RSS1i (excluding values within the upper 5% quantile) in the label-
permuted dataset according to the following equation:
VarðRSS0 � RSS1Þ
σ02
¼
ð8Þ
2  meanðRSS0 � RSS1Þ
Determination of Tm estimates for TPP-TR experiments with intact tissues pieces.
Protein melting curves were fitted according to equation (1) per condition (vehicle
and panobinostat-treatment) through one two or three data points per temperature
where each data point reflected a biological replicate. A single Tm per protein and
tissue was computed as the mean of the Tm determined from the vehicle curve and
the panobinostat curve.
All values of RSS0i À RSSi1 and of RSS1i that were obtained from the label-
permuted dataset were divided by σ20 before the numerical fit of the χ distribution.
The estimated degrees of freedom were used to calculate Fi according to
equation (8), with the numerator and denominator obtained from the original
data. We computed P values by comparing each Fi to an F distribution with the
estimated degrees of freedom. All P values were adjusted for multiple testing by the
Benjamini–Hochberg correction33.
2
Statistical analysis of panobinostat-induced thermal stability effects: ratio-based
approach. Proteins quantified in at least two replicate TMT10 experiments covering
the same temperatures were included. If the average relative fold change in
treatment or control was ≥0.4 at the highest temperature, the proteins were filtered
out. If the median R2 <0.8 from the fitted melting curve in controls or treatment,
the protein was filtered out.
Clustering. Proteins identified in all tissues and with an adjusted F test P≤0.001 in
any of the cross-tissue comparisons were used for clustering. Clustering was done
using complete linkage and Euclidian distance in Tibco Spotfire v.7.0.1.24. For the
clustering in the heat map (Supplementary Fig. 4c) the median normalized area
under the curve (AUC) (AUCi −medianall tissues) per tissue was used to display the
differences of protein melting behavior in the four tissues. For the clustering in
the heat map (Supplementary Fig. 4f), the z-transformed MS1 intensities (Top3
quantification34; at 37°C) of individual proteins were used to display the relative
protein concentrations in the four tissues. GO enrichment was performed using
the online enrichment tool Gorilla35,36 using all identified proteins as a background.
For each protein and at each temperature, two parameters were calculated:
the mean abundance ratio between treatment and control, and the significance of
an abundance difference by left- and right-tailed one-sample t tests on the basis
of log2-transformed treatment to control ratios. If the protein relative fold change
at a temperature was <0.15 in the vehicle-treated and the panobinostat-treated
condition, the log2 ratio at the specific temperature was removed. All P values were
aggregated by the empirical Brown’s method32 separately for the left-tailed test and
the right-tailed test. For the covariance matrix calculation, the log2 ratios of each
replicate and each P value was used. A protein abundance ratio was used as second
effect size to evaluate if the panobinostat treatment had a significant effect. If the
P value of the previously described t test on the log2 ratios was smaller than 0.05 at
several consecutive temperatures, the average log2 ratios were summed. If several
were recognized, the one with the highest summed ratio was selected (aggregated
log2 ratio). If a protein showed no or only one P value<0.05, the average log2 ratio
at the lowest P value was reported. Aggregated ratios were normalized to a distance
score according to the following equation:
Analysis of thermal stability of proteins in complexes. Proteins were mapped
to complexes using annotations comprising the databases CORUM37,38 and
COMPLEAT39 as previously compiled40.
To assess whether differences in melting points of complex members were
less than would be expected by chance, we used a strategy described previously
for protein half-lives16, where the s.d. of melting points of complex members was
compared to those computed from a permuted version of the same list.
To compare thermal stabilities of protein complexes across tissues, median
melting points of protein complex members identified and characterized across all
tissues were determined and compared.
To evaluate co-melting of complex members with potential interaction
partners, we applied a previously suggested strategy13, where average Euclidean
distances of fold changes at respective temperatures between two proteins of
interest were compared with 10,000 random comparisons. This approach makes
it possible to calculate P values, by comparing the Euclidean distance of a given
protein pair with the distribution of randomly drawn comparisons.
ꢀ
x � x
ð4Þ
Distance score ¼
S
where x is the aggregated ratio of the respective protein, ꢀx is the mean of
aggregated ratios and S is the s.d. of aggregated ratios.
Proteins were regarded as significantly affected if they fulfilled the following
criteria:
ꢀ
ꢁ
Determination of Stat5 and pStat5 levels and Tm in rat tissues. Liver and kidneys
were dissected from untreated 10-week-old Han Wistar rats (Heidelberg Pharma),
cut into pieces of 30 to 60mg and distributed over 96-well PCR plates. A volume of
PBS supplemented with protease inhibitors equivalent to the volume of liver and
kidney pieces was added to each well. Samples were heated in parallel for 3min to
temperatures covering the range from 37°C to 56.4°C. Heat-treated samples were
1
ifðΣrTvsC >sÞ �log10ðPÞ>
þ P
ð5Þ
ð6Þ
ðΣrTvsC � sÞ
ꢀ
ꢁ
1
ifðΣrTvsC < � sÞ �log10ðPÞ>
þ P
ðs � ΣrTvsC
Þ
NATuRE BIoTECHNoloGy | www.nature.com/naturebiotechnology

Nature BiotechNology
Letters
rested for 3min at room temperature and disrupted by bead ruptor. Non-heat-
treated tissue pieces were directly disrupted by bead ruptor. Igepal CA-630 (I3021)
was added to a concentration of 0.8% and MgCl2 (M1028) to 1.5mM, and samples
were incubated for 1h with benzonase (1kUml−1) at 4°C. Aggregated proteins
were separated by ultracentrifugation (20min, 100,000g) and denatured with an
SDS-based sample buffer. Lysates were analyzed by immunoblot for pSTAT5a/b
(Ser726, Ser731) (Invitrogen, PA5-36767) and total STAT5b (Invitrogen, 135300)
signals in two biological replicates.
of log10-transformed ion intensities in all samples is available as Supplementary
Data 14, metabolite annotation details are provided as Supplementary Data 13 and
additional information on panobinostat and its metabolites relevant for this study
are provided as Supplementary Data 15.
Panobinostat metabolisation in rat liver S9 and thermal shift assay. Male
Wistar rat liver S9 (BioIVT, M00022) at a protein concentration of 2mgml−1 were
incubated for 2h at 37°C in the presence of 100µM panobinostat or DMSO and
a co-factor mix consisting of 1mM nicotinamide adenine dinucleotide phosphate
(NADP, Roche, 10128058001), 0.5mM uridine 5′-diphosphoglucuronic acid
(UDPGA, Sigma, U6751), 0.05mgml−1 adenosine 3′-phosphate 5′-phosphosulfate
(PAPS, Sigma, A1651) and 0.05mM l-gluthatione (GSH, Sigma, G4251). Wells of
a PCR plate were filled with 100μl (each) of each reaction. Samples were heated in
parallel for 3min to 37, 47.7, 51.1, 54.2, 57.2 or 60.7°C. Heat-treated samples were
then rested for 3min at room temperature. Igepal CA-630 (I3021) was added to
To derive pEC50 values, the fold change values were transformed to a range
between 0 and 1. A sigmoidal curve was fitted using GraphPad Prism 7 to derive
pEC50 values, according to equation (2).
Blood-CETSA experiments and detection of HDAC2 by Western Assay System.
Fresh rat or human heparinized blood was treated with different concentrations
of panobinostat (5, 1, 0.2, 0.05, 0.0125 and 0.0025µM), romidepsin (30, 10, 2,
0.4, 0.2, 0.08, 0.01 and 0.0032µM), belinostat (50, 25, 10, 5, 1, 0.2 and 0.04µM),
or vehicle for 90min at 37°C and 5% CO2 (1.5–1.75ml of blood per condition,
depending on availability). After the addition of protease inhibitors (1:200, P8340),
the blood was then distributed into a 96-well PCR plate and heat treated for
3min to 54°C (human) and 53°C (rat blood). PBMCs were retrieved by density
centrifugation (Leucosep tubes, Greiner Bio-One, 163288) for 10min at 1,000g at
room temperature. The PBMCs were then lysed and processed as described under
‘TPP experiments in human blood’. The denatured samples were further diluted
in Western Assay System (WES) buffer (Protein Simple, 042-195) and relative
abundance of HDAC2 (CST 5113, human blood; Abcam Ab124974, rat blood)
and SOD1 (Novus, NBP1-90186) were analyzed by capillary gel electrophoresis
(WES, Protein Simple). The integrated peak areas of HDAC2 were divided by the
integrated peak areas of SOD1 to correct for potential variation in total protein
content. The vehicle condition (DMSO) was used as the reference for fold-change
calculations for the different compound concentration conditions at the respective
temperature. pEC50 values were calculated as described above.
a concentration of 0.8%, MgCl2 (M1028) to 1.5mM and benzonase to 1kUml−1
,
and samples were incubated for 1h at 4°C. Aggregated proteins were separated by
ultracentrifugation (20min, 100,000g). Samples were further processed for LC–
MS/MS analysis.
Synthesis of panobinostat metabolites. Unless otherwise stated, all solvents
and chemicals like the building blocks were used as purchased without further
purification. Preparative high-performance liquid chromatography (HPLC) was
performed on a Waters autopurification system (2767 sample manager, 2545
binary gradient module, 2420 ELS detector, 2996 photodiode array detector, 3100
mass detector, Waters SFO, and Waters 515 HPLC pump). The purification was
done on a Xbridge BEH C18 OBD 5μm Prep Column using a water–acetonitrile
mixture (0.2% formic acid as modifier) at a flow rate of 30mlmin−1. The purity
of all test compounds was 95% or higher as determined by analytical ultra-
performance liquid chromatography was done on a Waters Acquity BEH C18
1.7µm column (2.1×50mm) at 40°C and a flow rate of 0.5mlmin−1 with a linear
gradient over 1min (5–100% acetonitrile in water, 0.2% formic acid as modifier).
The instrument used for analysis was a waters Acquity system (detectors: Acquity
SQD and Acquity PDA). Mass signals were determined using a Waters 3100 mass
detector. Alternatively for synthesis of (2S,3S,4S,5S,6S)-3,4,5-trihydroxy-6-(((E)-
3-(4-(((2-(2-methyl-1H-indol-3-yl)ethyl)amino)methyl)phenyl)acrylamindo)oxy)
tetrahydro-2H-pyran-2-carboxylic acid the purity was determined using an Agilent
1290 Infinity II and mass signals were determined by MSD 6130. The column was
Sunfire C-18 (30×2.1mm) 3.5μm and a flow rate of 1.0mlmin−1 with a linear
gradient over 3min (0–100% acetonitrile in water, 0.1% formic acid as modifier).
(1H) NMR spectra were recorded on a Bruker Avance 400 spectrometer (400MHz)
using DMSO-d6 as solvent. Chemical shifts are given in p.p.m. (δ relative to
residual solvent peak for 1H).
Nano differential scanning fluorimetry with recombinant DHRS1. His-Flag
tagged recombinant DHRS1 (amino acids 3–262) was expressed in E. coli (strain)
and purified via Ni-NTA to a purity of 90%. Thermal-shift assays were performed
using the Prometheus NT.48 (Nano Temper technologies) according to the
manufacturer’s instructions. Panobinostat or its metabolites at 100µM were added
to recombinant DHRS1 at 100µgml−1 and analyzed using a temperature range
of 20–90°C with a temperature slope of 1.0°Cmin−1. The Tm was determined as
the first derivate of the flourescence ratio of 350nm to 330nm. All experiments
were performed in duplicate. A Tm could not be determined for metabolites with
autofluorescence.
Untargeted metabolomics analysis of rat spleen slices treated ex vivo with
panobinostat. Fresh rat spleen tissue punches were prepared as described above.
Spleen punches were treated for for 1h at 37°C with different concentrations
of panobinostat or vehicle as indicated, in DMEM high glucose (Gibco),
supplemented with 10% FBS, penicillin–streptomycin 100Uml−1 (Gibco), l-
glutamine 4.5gl−1 and 10mM HEPES (Gibco). The tissue punches were rinsed with
buffer (75mM ammonium carbonate dissolved in deinoized water, pH adjusted to
7.4 with acetic acid), snap frozen in liquid nitrogen and stored at −80°C. Frozen
tissue slices were mixed with 5µlmg−1 of deionized water and disrupted by bead
ruptor as described above. Subsequently, 50µl of homogenates were transferred to
1.5-ml tubes and mixed with 1ml of a 70:30 (vol/vol) ethanol:water solution heated
to 75°C. Samples were incubated at 75°C for 2min with occasional vortexing,
placed on ice for 10min and centrifuged at 13,000g for 10min to pellet debris.
Supernatants were vacuum dried as described above and finally reconstituted in
100µl of deionized water. To enable quantification of panobinostat and its four
metabolites M37.8, M43.5, M36.9 and T27C, defined concentrations of the four
compounds were added to spleen lysate before homogenization. Calibration
curve samples were subsequently processed identically to the remaining spleen
samples to ensure data comparability. Samples were analyzed on a LC–MS
platform consisting of a Thermo Scientific Ultimate 3000 liquid chromatography
system with autosampler temperature set to 10°C coupled to a Thermo
Scientific Q-Exactive Plus Fourier transform mass spectrometer equipped with
a heated electrospray ion source and operated in negative-ionization mode. The
isocratic flow rate was 150μlmin−1 of mobile phase consisting of 60:40 (vol/vol)
isopropanol:water buffered with 1mM ammonium fluoride, pH 9 and containing
10nM taurocholic acid and 20nM homotaurine as lock masses. Mass spectra were
recorded in profile mode from m/z of 50 to 1,000 with the following instrument
settings: sheath gas, 35 a.u. (arbritary units); auxiliary gas, 10 a.u.; auxiliary gas
heater, 200°C; sweep gas, 1 a.u.; spray voltage, −3kV; capillary temperature,
250°C; S-lens RF level, 50 a.u; resolution, 70,000 at an m/z of 2,000; AGC target,
3×106 ions; maximum injection time, 120ms; acquisition duration, 60s. Spectral
data processing was performed using an automated pipeline in R as described
previously41. Detected ions were tentatively annotated as metabolites using the
HMDB database42 as reference assuming [M-H] and [M-2H] as ionization options
and the exchange of one or two 12C with the equivalent number of 13C atoms, with
the method-inherent disability to distinguish between isomers. The full dataset
4-(((2-(2-Methyl-1H-indol-3-yl)ethyl)amino)methyl)benzoic acid (M36.9).
2-(2-Methyl-1H-indol-3-yl)ethan-1-amine (50mg, 0.287mmol) was dissolved in
ethanol (0.5ml) and to this solution was added 4-formylbenzoic acid (64.6mg,
0.430mmol) followed by sodium cyanoborohydride (36.1mg, 0.574mmol). The
reaction mixture was stirred overnight at room temperature and then diluted with
5% NaHSO4 (aq.) (10ml). This was then extracted with ethyl acetate (2×10ml)
and the organic phase was dried over sodium sulfate, filtered and evaporated. The
residue was purifed by preparatory HPLC. This afforded 4-(((2-(2-methyl-1H-
indol-3-yl)ethyl)amino)methyl)benzoic acid (17mg, 0.054mmol, 18.83% yield)
as a white solid. LC–MS m/z 307.3 [M-H]− (Rt=0.72min). 1H NMR (DMSO-d6,
400MHz) δ 12.07 (br s, 1H), 7.6–7.6 (m, 3H), 7.35 (t, 2H, J=7.7Hz), 7.2–7.3 (m,
1H), 7.07 (d, 1H, J=1.3Hz), 6.85 (s,1H), 4.0–4.2 (m, 4H), 3.61 (s, 2H), 2.26 (s, 3H).
(E)-3-(4-(((2-(2-Methyl-1H-indol-3-yl)ethyl)amino)methyl)phenyl)acrylamide
(37.8) was synthesized as previously described8.
(E)-3-(4-(((2-(2-Methyl-1H-indol-3-yl)ethyl)amino)methyl)phenyl)acrylic acid
(M43.5). Methyl (E)-3-(4-(((2-(2-methyl-1H-indol-3-yl)ethyl)amino)methyl)
phenyl)acrylate (30mg, 0.086mmol) was dissolved in tetrahydrofuran (0.4ml)
and to this was added an aqueous solution of NaOH (1M, 0.1ml). The reaction
mixture was then stirred overnight and then 5% NaHSO4 (aq) was added. The
precipitate was collected by filtration and was washed with water and ethyl acetate.
This afforded (E)-3-(4-(((2-(2-methyl-1H-indol-3-yl)ethyl)amino)methyl)phenyl)
acrylic acid (9mg, 0.027mmol, 31.3 % yield) as a white solid. LC–MS m/z 333.3
[M-H]− (Rt=0.74min). 1H NMR (DMSO-d6, 400MHz) δ 10.80 (s, 1H), 7.71
(d, 2H, J=8.1Hz), 7.59 (d, 1H, J=16.1Hz), 7.47 (d, 2H, J=8.1Hz), 7.39 (d, 1H,
J=7.6Hz), 7.24 (d, 1H, J=7.9Hz), 6.9–7.0 (m, 2H), 6.56 (d, 1H, J=16.1Hz), 4.04
(s, 2H), 2.90 (s, 4H), 2.32 (s, 3H).
N-Hydroxy-3-(4(((2-(2-methyl-1H-indol-3-yl)ethyl)amino)methyl)phenyl)
propanamide (T27c). 3-(4-(((2-(2-Methyl-1H-indol-3-yl)ethyl)amino)
methyl)phenyl)propanoic acid (63mg, 0.187mmol) was dissolved in N,N-
dimethylformamide (1ml) and to this was added O-(tert-butyl)hydroxylamine
(83mg, 0.936mmol), DIPEA (0.196ml, 1.124mmol) and finally HATU (107mg,
0.281mmol). The reaction mixture was then stirred for 18h at room temperature
and then purified directly by reverse phase column chromatography eluting with
NATuRE BIoTECHNoloGy | www.nature.com/naturebiotechnology

Nature BiotechNology
Letters
acetonitrile in water (10–80%). This afforded N-(tert-butoxy)-3-(4-(((2-(2-methyl-
1H-indol-3-yl)ethyl)amino)methyl)phenyl) propanamide (19mg, 0.047mmol,
24.90% yield) as a colorless solid. LC–MS m/z 408.3 [M+H]+ (Rt=0.80min).
N-(tert-Butoxy)-3-(4-(((2-(2-methyl-1H-indol-3-yl)ethyl)amino)methyl)
phenyl)propanamide (19mg, 0.047mmol) was dissolved in dichloromethane
(0.2ml) and to this was added trifluoroactetic acid (0.3ml, 3.89mmol). The
reaction mixture was stirred at room temperature for 5h. The solvent was then
removed in vacuo and the residue was purified by preparatory HPLC. This
afforded N-hydroxy-3-(4-(((2-(2-methyl-1H-indol-3-yl)ethyl)amino)methyl)
phenyl)propanamide (9mg, 0.026mmol, 54.9 % yield) as a colorless solid. LC–MS
m/z 352.2 [M+H]+ (Rt=0.70min). 1H NMR (DMSO-d6, 400MHz) δ 10.74 (s,
1H), 10.39 (s, 1H), 8.27 (s, 1H), 7.38 (d, 1H, J=7.6Hz), 7.1–7.3 (m, 5H), 6.8–7.0
(m, 2H), 3.80 (s, 2H), 2.7–2.9 (m, 6H), 2.31 (s, 3H), 2.2–2.3 (m, 2H).
organic phase was washed with saturated brine (30ml), dried over sodium sulfate and
evaporated in vacuo to give the crude product. The residue was purified by column
chromatography eluting with ethyl acetate in petroleum ether (0–100%) to afford
(2S,3R,4S,5S,6S)-2-(((E)-3-(4-(((tert-butoxycarbonyl)(2-(2-methyl-1H-indol-3-yl)
ethyl)amino)methyl)phenyl)acrylamido)oxy)-6-(methoxycarbonyl)tetrahydro-
2H-pyran-3,4,5-triyl triacetate (1.1g, 39.2% yield) as a pale yellow solid. Mass
spectrometry m/z of 766.3 [M+H]+ (Rt=2.62min).
To a solution of (2S,3R,4S,5S,6S)-2-(((E)-3-(4-(((tert-butoxycarbonyl)
(2-(2-methyl-1H-indol-3-yl)ethyl)amino)methyl)phenyl)acrylamido)oxy)-6-
(methoxycarbonyl)tetrahydro-2H-pyran-3,4,5-triyl triacetate (900mg, 1.175mmol)
in 1,4-dioxane (10ml), stirred under nitrogen at 0°C, was added hydrochloric acid
(4.5M) in dioxane) (5ml, 22.50mmol) dropwise. The reaction mixture was stirred at
25°C for 3h and then was concentrated under reduced pressure, diluted with water
(50ml), neutralized with 15% NaHCO3 solution and extracted with ethyl acetate
(2×60ml).The organic phase was washed with saturated brine (50ml), dried over
sodium sulfate and evaporated in vacuo. The crude compound was washed with
ether (2×10ml) and dried under reduced pressure to afford (2S,3S,4S,5R,6S)-2-
(methoxycarbonyl)-6-(((E)-3-(4-(((2-(2-methyl-1H-indol-3-yl)ethyl)amino)methyl)
phenyl)acrylamido)oxy)tetrahydro-2H-pyran-3,4,5-triyl triacetate (0.6g, 58.2% yield)
as an off white solid. Mass spectrometry m/z of 665.9 [M+H]+ (Rt=1.58min).
To a solution of (2S,3S,4S,5R,6S)-2-(methoxycarbonyl)-6-(((E)-3-(4-(((2-(2-
methyl-1H-indol-3-yl)ethyl)amino)methyl)phenyl)acrylamido)oxy)tetrahydro-
2H-pyran-3,4,5-triyl triacetate (550mg, 0.826mmol) in methanol (10ml) under
nitrogen at −10°C was added sodium hydroxide (2ml, 2.000mmol) dropwise.
The reaction mixture was stirred at 0°C for 30min. The reaction mixture was
neutralized at 0°C with acetic acid and the solvent was removed in vacuo. The
residue was purified by preparatory HPLC. This afforded (2S,3S,4S,5S,6S)-3,4,5-
trihydroxy-6-(((E)-3-(4-(((2-(2-methyl-1H-indol-3-yl)ethyl)amino)methyl)
phenyl)acrylamindo)oxy)tetrahydro-2H-pyran-2-carboxylic acid (55mg,
12.4% yield) as a pale yellow solid. Mass spectrometry m/z of 526.2 [M+H]+
(Rt=1.27min). 1HNMR (DMSO-d6, 400MHz): δ 10.76 (s, 1H), 7.57–7.55 (m,
1H), 7.47–7.39 (m, 4H), 7.23 (d, J=8.0Hz, 1H), 6.99–6.90 (m, 2H), 6.48 (d,
J=15.5Hz, 1H), 5.05 (bs, 1H),4.61 (d, J=7.2Hz, 1H), 4.0 (s, 2H), 3.49
(d, J=8.3Hz, 1H), 3.24–3.21 (m, 3H), 3.17–3.13 (m, 2H), 2.95–2.83 (m, 4H),
2.32 (s, 3H).
3-(4-(((2-(2-Methyl-1H-indol-3-yl)ethyl)amino)methyl)phenyl)propanoic acid
(M44.6). 2-(2-Methyl-1H-indol-3-yl)ethan-1-amine (200mg, 1.148mmol) was
dissolved in ethanol (2mL) and to this was added 3-(4-formylphenyl)propanoic
acid (307mg, 1.722mmol), followed by sodium cyanoborohydride (144mg,
2.296mmol). The reaction mixture was then stirred at room temperature for
18h. It was then diluted with 5% NaHSO4(aq) (10mL) and extracted with ethyl
acetate (2×20ml). The combined organic phases were dried over sodium sulfate,
filtered and evaporated. The residue was purified by reverse phase column
chromatography eluting with acetonitrile in water (0 to 100 % containing 0.2%
formic acid). The product was further purified by dissolving it in ethyl acetate
(20ml) and washing with 0.5M HCl (2×20ml). The organic phase was dried over
sodium sulfate, filtered and evaporated. The residue was finally purified by reverse
phase chromatography eluting wth acetonitrile in water (10–100% containing
1% ammonia). This afforded 3-(4-(((2-(2-methyl-1H-indol-3-yl)ethyl)amino)
methyl)phenyl)propanoic acid (20mg, 0.059mmol, 5.13 % yield) as a white solid.
MS m/z 337.3 [M+H]+ (Rt=0.77min). 1HNMR (DMSO-d6, 400MHz) 1H NMR
(DMSO-d6, 400MHz) δ 10.63 (s, 1H), 7.3 (d, 1H, J=7.7), 7.20–7.25 (m, 3H), 7.15
(d, 2H, J=8.4) 6.8–6.9 (m, 1H), 6.75–6.8 (m, 1H), 3.69 (s, 2H), 2.7–2.8 (m, 4H),
2.6–2.7 (m, 2H), 2.3–2.4 (m, 2H), 2.23 (s, 3H).
3-(4-(((2-(5-Hydroxy-2-methyl-1H-indol-3-yl)ethyl)amino)methyl)phenyl)propanoic
acid (T25g). 3-(2-Aminoethyl)-2-methyl-1H-indol-5-ol (25mg, 0.131mmol) was
dissolved in ethanol (0.5ml) and to this was added 3-(4-formylphenyl)propanoic
acid (25.8mg, 0.145mmol) followed by sodium cyanoborohydride (12.39mg,
0.197mmol). The reaction mixture was then stirred at room temperature
overnight. The solvent was removed in vacuo and the residue was purified by
reverse-phase column chromatography eluting with acetonitrile in water (0–100%
containing 0.2% formic acid). This afforded 3-(4-(((2-(5-hydroxy-2-methyl-1H-
indol-3-yl)ethyl)amino)methyl)phenyl)propanoic acid, formic acid salt (5mg,
0.012mmol, 9.4% yield) as a white solid. Mass spectrometry m/z of 353.2 [M+H]+
(Rt=0.68min).1H NMR (DMSO-d6, 400MHz) δ 10.38 (s, 1H), 7.2–7.3 (m, 2H),
7.1–7.2 (m, 4H), 6.92 (d, 1H, J=8.5Hz), 6.63 (d, 1H, J=2.1Hz), 6.41 (dd, 1H,
J=2.1, 8.5Hz), 4.35 (s, 1H), 3.92 (br s, 2H), 2.7–2.8 (m, 6H), 2.4–2.5 (m, 2H), 2.17
(s, 3H).
Density plots. Densityplots shown were generated using R and the ggplot2
package.
Box plots. Box plots shown were generated using R and the ggplot2 package.
The lower and the upper hinges of the boxes correspond to the 25th and 75th
percentiles, the bar in the box shows the median. The upper whisker extends from
the upper hinge to the largest value (up to a maximum of 1.5×IQR). Similarly, the
lower whisker extends from the lower hinge to at most 1.5×IQR. Points outside of
the whiskers were plotted individually as outliers.
Group tests were performed using a nonparametric Wilcoxon test and
statistical significance is indicated by: *P≤0.05, **P≤0.01 and ***P≤0.001 if not
stated otherwise.
(2S,3S,4S,5S,6S)-3,4,5-Trihydroxy-6-(((E)-3-(4-(((2-(2-methyl-1H-indol-3-yl)
ethyl)amino)methyl)phenyl)acrylamindo)oxy)tetrahydro-2H-pyran-2-carboxylic
acid (M34.4). To a solution of methyl (E)-3-(4-(((2-(2-methyl-1H-indol-3-yl)
ethyl)amino)methyl)phenyl)acrylate (4g, 11.48mmol) in 1,4-dioxane (50ml) and
water (5ml) at 0°C was added triethylamine (4.80ml, 34.4mmol), followed by
Boc-anhydride (8.00ml, 34.4mmol). The reaction mixture was stirred at 25°C for
48h and then the solvent was evaporated in vacuo. The residue was partitioned
between ethyl acetate (100ml) and water (100ml). The organic phase was washed
with saturated brine (50ml), dried over sodium sulfate and evaporated in vacuo to
afford (E)-3-(4-(((tert-butoxycarbonyl)(2-(2-methyl-1H-indol-3-yl)ethyl)amino)
methyl)phenyl)acrylate (5.0g, 92% yield) as a brown oil. Mass spectrometry m/z of
393.2 [M(-tBu)+H]+ (Rt=2.82min).
Ethics statement. Human blood was sourced ethically with written informed
consent and its research use was in accordance with the terms of the informed
consent under an IRB/EC approved protocol (Ethics Commission at the
Landesärztekammer Baden-Wuerttemberg, Germany). The animal experiments
were ethically reviewed by the GSK Office for Animal Welfare, Ethics and Strategy
and carried out in accordance with European Directive 2010/63/EEC and the GSK
Policy on the Care, Welfare and Treatment of Animals.
Reporting Summary. Further information on research design is available in the
Nature Research Reporting Summary linked to this article.
Data availability
A solution of methyl (E)-3-(4-(((tert-butoxycarbonyl)(2-(2-methyl-1H-indol-
3-yl)ethyl)amino)methyl)phenyl)acrylate (5g, 11.15mmol) in tetrahydrofuran
(50ml) and methanol (5ml) was cooled to 0°C and a solution of lithium hydroxide
(0.534g, 22.29mmol) in water (5mL) was added. The reaction mixture was stirred
at room temperature for 16h then diluted with water (150ml) and cooled to 0°C.
The solution was neutralized with 1.5M HCl solution and extracted with ethyl
acetate (2×150ml). The combined organic layers were washed with brine (100ml)
and dried over anhydrous sodium sulfate, filtered and concentrated under reduced
the pressure. The residue was washed with ether (2×50ml) and this afforded E)-
3-(4-(((tert-butoxycarbonyl)(2-(2-methyl-1H-indol-3-yl)ethyl)amino) methyl)
phenyl) acrylic acid (3.9g, 79% yield) as a white solid. Mass spectrometry m/z of
435.1 [M+H]+ (Rt=2.49min).
All data is available in the main text or the supplementary materials. The mass
spectrometry proteomics data have been deposited to the ProteomeXchange
Consortium via the PRIDE24 partner repository with the dataset identifiers
PXD015458, PXD015427, PXD015397, PXD015373 and PXD016277.
Code availability
The code used for this study will be made available by the corresponding authors
upon reasonable request.
References
To a solution of (E)-3-(4-(((tert-butoxycarbonyl)(2-(2-methyl-1H-indol-3-yl)
ethyl)amino) methyl)phenyl) acrylic acid (1.368g, 3.15mmol) in DMF (15ml), stirred
under nitrogen at 0°C, were added HATU (1.437g, 3.78mmol), DIPEA (1.100ml,
6.30mmol) and a solution of (2S,3R,4S,5S,6S)-2-(aminooxy)-6-(methoxycarbonyl)
tetrahydro-2H-pyran-3,4,5-triyl triacetate (1.1g, 3.15mmol) in DMF (5ml) dropwise.
The reaction mixture was stirred at 25°C for 5h and then concentrated in vacuo
and the residue was partitioned between ethyl acetate (80ml) and water (60ml). The
24. Perez-Riverol, Y. et al. Te PRIDE database and related tools and resources in
2019: improving support for quantifcation data. Nucleic Acids Res. 47,
D442–D450 (2019).
25. Franken, H. et al. Termal proteome profling for unbiased identifcation of
direct and indirect drug targets using multiplexed quantitative mass
spectrometry. Nat. Protoc. 10, 1567–1593 (2015).
NATuRE BIoTECHNoloGy | www.nature.com/naturebiotechnology

Nature BiotechNology
Letters
26. Density and mass of each organ/tissue Bionumbers http://kirschner.med.
harvard.edu/fles/bionumbers/Density%20and%20mass%20of%20each%20
organ-tissue.pdf (2014).
40. Ori, A. et al. Spatiotemporal variation of mammalian protein complex
stoichiometries. Genome Biol. 17, 47 (2016).
41. Fuhrer, T., Heer, D., Begemann, B. & Zamboni, N. High-throughput, accurate
mass metabolome profling of cellular extracts by fow injection-time-of-fight
mass spectrometry. Anal. Chem. 83, 7074–7080 (2011).
27. Moggridge, S., Sorensen, P. H., Morin, G. B. & Hughes, C. S. Extending the
compatibility of the SP3 paramagnetic bead processing approach for
proteomics. J. Proteome Res. 17, 1730–1740 (2018).
42. Wishart, D. S. et al. HMDB 4.0: the human metabolome database for 2018.
Nucleic Acids Res. 46, D608–D617 (2018).
28. Kelstrup, C. D. et al. Limits for resolving isobaric tandem mass tag reporter
ions using phase-constrained spectrum deconvolution. J. Proteome Res. 17,
4008–4016 (2018).
Acknowledgements
29. Werner, T. et al. Ion coalescence of neutron encoded TMT 10-plex reporter
ions. Anal. Chem. 86, 3594–3601 (2014).
The authors thank J. Stuhlfauth, N. Garcia-Altrieth, L. Vitali, K. Beß and B. Dlugosch for
the supporting cell culture and lysate production; M. Bösche, T. Rudi, M. Klös-Hudak, K.
Kammerer and M. Steidel for assistance with mass spectrometry; A. Wolf for supporting
the DSF assay; S. Gade for support during data uploading; C. Schofield, T. Walker, G.
Drewes, F. Reinhard, P. Grandi and P. Wier for scientific discussions; and E. Nicodeme
and F. Blandel for excellent technical support during the animal study.
30. Savitski, M. M. et al. Measuring and managing ratio compression
for accurate iTRAQ/TMT quantifcation. J. Proteome. Res. 12,
3586–3598 (2013).
31. Savitski, M. M. et al. Targeted data acquisition for improved reproducibility
and robustness of proteomic mass spectrometry assays. J. Am. Soc. Mass.
Spectrom. 21, 1668–1679 (2010).
Author contributions
32. Poole, W., Gibbs, D. L., Shmulevich, I., Bernard, B. & Knijnenburg, T. A.
Combining dependent P-values with an empirical adaptation of Brown’s
method. Bioinformatics 32, i430–i436 (2016).
J.P., T.W., C.E.R., E.S., K.S., J.K., B.H., I.B. and D.P. designed and performed experiments
and analyzed TPP data; A.R., K.S., B.H. and T.W. designed and performed experiments
and analyzed the data for whole blood; D.W.T. synthesized compounds; D.C.S. and
J.V. performed and analyzed the metabolomics experiments; H.C.E. analyzed the DSF
experiment; N.K., D.D.C., M.K., H.F. and M.F.-S. performed the statistical analysis; W.H.
gave scientific and data analysis advice; G.B., M.B. and M.M.S. designed experiments,
analyzed data and wrote the manuscript; and G.B. supervised the work.
33. Benjamini, Y. & Hochberg, Y. Controlling the false discovery rate: a practical
and powerful approach to multiple testing. J. R. Stat. Soc. Ser. B Methodol 57,
289–300 (1995).
34. Silva, J. C., Gorenstein, M. V., Li, G.-Z., Vissers, J. P. C. & Geromanos, S. J.
Absolute quantifcation of proteins by LCMSE: a virtue of parallel MS
acquisition. Mol. Cell. Proteomics 5, 144–156 (2006).
35. Eden, E., Navon, R., Steinfeld, I., Lipson, D. & Yakhini, Z. GOrilla: a tool for
discovery and visualization of enriched GO terms in ranked gene lists. BMC
Bioinformatics 10, 48 (2009).
Competing interests
J.P., T.W., A.R., D.P., E.S., K.S., B.H., D.W.T., J.K., J.V., D.C.S., H.C.E., H.F., M.F.S., M.B.
and G.B. are GSK employees. M.M.S. is a GSK shareholder.
36. Eden, E., Lipson, D., Yogev, S. & Yakhini, Z. Discovering motifs in ranked
lists of DNA sequences. PLoS Comput. Biol. 3, e39 (2007).
37. Ruepp, A. et al. CORUM: the comprehensive resource of mammalian protein
complexes. Nucleic Acids Res. 36, D646–D650 (2008).
Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/
s41587-019-0388-4.
38. Ruepp, A. et al. CORUM: the comprehensive resource of mammalian protein
complexes—2009. Nucleic Acids Res. 38, D497–D501 (2010).
39. Vinayagam, A. et al. Protein complex-based analysis framework for
high-throughput data sets. Sci. Signal. 6, rs5 (2013).
Correspondence and requests for materials should be addressed to M.M.S., M.B. or G.B.
Reprints and permissions information is available at www.nature.com/reprints.
NATuRE BIoTECHNoloGy | www.nature.com/naturebiotechnology