Articles

Image-based ex-vivo drug screening for patients with

aggressive haematological malignancies: interim results
from a single-arm, open-label, pilot study
Berend Snijder*, Gregory I Vladimer*, Nikolaus Krall, Katsuhiro Miura, Ann-Sofie Schmolke, Christoph Kornauth, Oscar Lopez de la Fuente,
Hye-Soo Choi, Emiel van der Kouwe, Sinan Gültekin, Lukas Kazianka, Johannes W Bigenzahn, Gregor Hoermann, Nicole Prutsch, Olaf Merkel,
Anna Ringler, Monika Sabler, Georg Jeryczynski, Marius E Mayerhoefer, Ingrid Simonitsch-Klupp, Katharina Ocko, Franz Felberbauer‡,
Leonhard Müllauer, Gerald W Prager, Belgin Korkmaz, Lukas Kenner, Wolfgang R Sperr, Robert Kralovics, Heinz Gisslinger, Peter Valent,
Stefan Kubicek, Ulrich Jäger, Philipp B Staber†, Giulio Superti-Furga†

Summary
Background Patients with refractory or relapsed haematological malignancies have few treatment options and short Lancet Haematol 2017;
survival times. Identification of effective therapies with genomic-based precision medicine is hampered by 4: e595–606

intratumour heterogeneity and incomplete understanding of the contribution of various mutations within specific Published Online
November 15, 2017
cancer phenotypes. Ex-vivo drug-response profiling in patient biopsies might aid effective treatment identification;
http://dx.doi.org/10.1016/
however, proof of its clinical utility is limited. S2352-3026(17)30208-9
See Comment page e567
Methods We investigated the feasibility and clinical impact of multiparametric, single-cell, drug-response profiling in *Contributed equally as first
patient biopsies by immunofluorescence, automated microscopy, and image analysis, an approach we call authors
pharmacoscopy. First, the ability of pharmacoscopy to separate responders from non-responders was evaluated †Contributed equally as last
retrospectively for a cohort of 20 newly diagnosed and previously untreated patients with acute myeloid leukaemia. authors
Next, 48 patients with aggressive haematological malignancies were prospectively evaluated for pharmacoscopy-guided CeMM Research Center for
treatment, of whom 17 could receive the treatment. The primary endpoint was progression-free survival in Molecular Medicine, Vienna,
Austria (Prof B Snijder PhD,
pharmacoscopy-treated patients, as compared with their own progression-free survival for the most recent regimen on
G I Vladimer PhD, N Krall PhD,
which they had progressive disease. This trial is ongoing and registered with ClinicalTrials.gov, number NCT03096821. O Lopez de la Fuente BA,
J W Bigenzahn MD, M Sabler MSc,
Findings Pharmacoscopy retrospectively predicted the clinical response of 20 acute myeloid leukaemia patients to R Kralovics PhD, A Ringler MSc,
S Kubicek PhD,
initial therapy with 88·1% accuracy. In this interim analysis, 15 (88%) of 17 patients receiving pharmacoscopy-guided Prof G Superti-Furga PhD);
treatment had an overall response compared with four (24%) of 17 patients with their most recent regimen (odds Department of Biology,
ratio 24·38 [95% CI 3·99–125·4], p=0·0013). 12 (71%) of 17 patients had a progression-free survival ratio of 1·3 or Institute of Molecular Systems
higher, and median progression-free survival increased by four times, from 5·7 (95% CI 4·1–12·1) weeks to Biology, ETH Zurich, Zurich,
Switzerland (Prof B Snijder);
22·6 (7·4–34·0) weeks (hazard ratio 3·14 [95% CI 1·37–7·22], p=0·0075). Allcyte, Vienna, Austria
(G I Vladimer); Department of
Interpretation Routine clinical integration of pharmacoscopy for treatment selection is technically feasible, and led to Internal Medicine I, Division of
improved treatment of patients with aggressive refractory haematological malignancies in an initial patient cohort, Hematology and
Hemostaseology (K Miura MD,
warranting further investigation. A-S Schmolke PhD, H-S Choi PhD,
E van der Kouwe PharmD,
Funding Austrian Academy of Sciences; European Research Council; Austrian Science Fund; Austrian Federal S Gültekin PhD, L Kazianka MD,
Ministry of Science, Research and Economy; National Foundation for Research, Technology and Development; G Jeryczynski MD,
Prof H Gisslinger MD,
Anniversary Fund of the Austrian National Bank; MPN Research Foundation; European Molecular Biology Prof U Jäger MD,
Organization; and Swiss National Science Foundation. Prof P B Staber MD,
Prof W R Sperr MD,
Copyright © The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license. Prof P Valent MD), Clinical
Institute of Pathology
(C Kornauth MD, N Prutsch MSc,
Introduction 30% longer progression-free survival with treatment Prof O Merkel PhD,
Genetic studies have identified several genomic selected on the basis of genetic profiling than they did Prof I Simonitsch-Klupp MD,
Prof L Müllauer MD,
alterations associated with the development of haema­ with their previous treatment. However, the SHIVA
Prof L Kenner MD), Department
tological malignancies. However, barriers remain in fully study,5 one of the first randomised trials of genomic- of Laboratory Medicine
translating this genomic information into direct clinical based precision medicine, did not show a benefit in (Prof G Hoermann MD),
benefit for patients. Current efforts to introduce per­ progression-free survival for patients assigned to genome- Department of Biomedical
Imaging and Image-guided
sonalised medicine in patients with cancer, which focus based targeted treatment compared with treatment
Therapy
on genetic and molecular patient stratification, have according to physician’s choice in heavily pretreated (Prof M E Mayerhofer PhD),
produced varying results.1–5 In a pioneering study,4 which patients with cancer. Genome-based therapy decisions Division of General Surgery,
used each patient as their own control, 27% of patients are limited by our incomplete understanding of the Department of Surgery
(F Felberbauer MD,
with recurrent metastatic cancer of any kind had a relationship between cancer phenotype and genotype,

www.thelancet.com/haematology Vol 4 December 2017 e595

 Articles

B Korkmaz MD), Department of

Internal Medicine I, Division of Research in context
Oncology, Comprehensive
Cancer Center Evidence before this study Added value of this study
(Prof G W Prager MD), Ludwig We did a systematic search of PubMed using the search terms To our knowledge, our study is the first prospective study
Boltzmann Cluster Oncology (“functional screening” [Title/Abstract] OR “chemosensitivity showing feasibility and efficacy of ex-vivo drug-response
(Prof P Valent), Center for
Physiology and Pharmacology
test” [Title/Abstract] OR “chemoresistance test” [Title/Abstract] profiling to guide personalised treatment selection across large
(Prof G Superti-Furga), Medical OR “drug profiling” [Title/Abstract] OR “drug response” panels of possible treatments for patients suffering aggressive
University of Vienna, Vienna, [Title/Abstract]) AND (“leukemia” [Title/Abstract] OR haematological malignancies. We do so with a new
Austria; Christian Doppler “leukaemia” [Title/Abstract] OR “lymphoma” [Title/Abstract] image-based, drug-response profiling technique that we call
Laboratory for Chemical
Epigenetics and Anti-
OR “myeloma” [Title/Abstract] OR “hematologic” pharmacoscopy, which uses high-throughput, automated,
Infectives, Vienna, Austria [Title/Abstract]). We did not restrict the search by date, confocal microscopy; immunofluorescence; and single-cell
(A Ringler, Prof S Kubicek); language, or article type. We did one search before initiating image analysis.
Pharmacy Department, Vienna this study on Sept 1, 2015, and we repeated this search on
General Hospital, Vienna, Implications of all the available evidence
Austria (K Ocko MSc); Ludwig
Oct 13, 2017. Several studies have shown the potential for
Our interim study results indicate that adapting treatment
Boltzmann Institute for Cancer retrospective patient stratification based on a variety of ex-vivo
regimens of patients with aggressive haematological
Research, Vienna, Austria drug-response profiling techniques; however, no reports were
(Prof L Kenner); Unit of malignancies to pharmacoscopy is feasible, safe, and effective.
found of studies in which patient treatment for haematological
Laboratory Animal Pathology, More patients whose treatment protocols were selected by the
University of Veterinary malignancies were adapted to ex-vivo drug-response profiling
haematological tumour board based on pharmacoscopy results
Medicine, Vienna, Austria across large panels of drugs. We also searched ClinicalTrials.gov
had an overall response and had longer progression-free
(Prof L Kenner); and Ludwig for published clinical trials using the search terms described
Boltzmann Institute for Cancer survival with pharmacoscopy than their previous treatment.
above. This search retrieved only one other clinical trial
Research, Vienna, Austria Further studies with randomised trial designs and larger patient
(Prof W R Sperr) (currently recruiting patients and with feasibility as endpoint),
cohorts than our study are justified to further elucidate the
‡ Dr Felberbauer died in
in which patient treatment for haematological malignancies is
clinical impact of our novel, image-based, ex-vivo
October, 2017 being adapted to the drug-response profiles of primary biopsies
drug-response profiling platform.
Correspondence to: across at least 100 different drugs tested.
Prof Giulio Superti-Furga, CeMM
Research Center for Molecular
Medicine of the Austrian and the complex genetics underlying cancer are the result called pharmacoscopy.30 Pharmacoscopy enables tumour-
Academy of Sciences, 1090
Vienna, Austria
of dynamic microevolutionary processes.6 For instance, cell specific quantification of biological parameters of
gsuperti@cemm.at whereas several studies have linked cytogenetic and millions of adherent and non-adherent individual cells
molecular abnormalities with distinct clinical outcomes with high sample efficiency, minimal sample mani­
in acute myeloid leukaemia,7 accurate prediction of pulation, extensive automation, and fast turn-around
treatment response of individual patients with acute times. We thus aimed to evaluate the feasibility of
myeloid leukaemia to induction therapy remains integrating pharmacoscopy into the clinic, and to assess
challenging.8–12 Furthermore, patients with aggressive clinical response in patients who received a treatment
haematological malignancies, who have failed at least two according to pharmacoscopy results as an individual
lines of therapy, are often without further standard healing attempt.
treatment options and have a poor prognosis.13 These
patients will usually receive either best available therapy, Methods
supportive care, or will be enrolled in clinical trials. Study design and participants
Therefore, dynamic approaches that measure drug For this single-arm, open-label, pilot study, we collected
responses in cancer cells derived from patient biopsies samples and clinical data from patients with late-stage
might complement such static genetic measurements. haematological malignancies. Patients were eligible for
For example, ex-vivo chemosensitivity tests have been inclusion if no further standard treatments or clinical trials
done in samples from patients with chronic or acute were available for the patient; the patient had undergone at
leukaemia and multiple myeloma,14–21 in breast cancer- least two lines of previous therapy; the patient gave written
derived stable cell lines,22 in patient-derived xenografts in informed consent; cancer cell-containing samples could be
mice,23,24 and in gut stem-cell-derived organoids.25,26 These biopsied after written informed consent; the clinical
pioneering functional assays have provided proof of decision was made by a board consisting of haematologists,
concept by showing that ex-vivo responses might match pathologists, pharmacists, and molecular biologists; and
clinical response; however, these studies have not been candidate treatments identified by pharmacoscopy were
integrated into clinical routine because of practical clinically available and considered safe given the patient’s
limitations and scarce proof of clinical benefit.27–29 health condition. Pharmacoscopy-guided therapy was
Here, we investigate the clinical impact of a newly provided to individual late-stage patients as an individual
developed technology platform that combines multi­ healing attempt, in accordance with European Union and
parametric immunofluorescence with high-throughput Austrian named-patient use legislation. Ethical approval
automated microscopy and single-cell image analysis, was granted by the Ethics Commission of the Medical

e596 www.thelancet.com/haematology Vol 4 December 2017

 Articles

University of Vienna (Ethik Kommission 1830/2015, concentration combinations in four technical repeats. To
2008/2015, 1895/2015). determine the ex-vivo drug-induced cytotoxicity, we
quantified the number of non-fragmented nuclei in each
Procedures image after drug treatment. Drug-induced cell death based
Mononuclear cells from bone marrow aspirates, peripheral on nuclear morphology was measured after overnight
blood, pleural effusion, ascites, or excised lymph node drug incubation, and subpopulation specificity was
samples were purified using Ficoll density gradient (bone assessed on the cells that stained positive for CD34 or KIT
marrow, peripheral blood, pleural effusion, ascites; Axis- (CD117). Both markers are commonly present on leu­
Shield, Oslo, Norway) or homogenised and filtered kaemic blasts in acute myeloid leukaemia.34 Because all
through a 70-μm mesh filter (lymph tissue). The resulting patient samples contained a combination of blast cells and
single-cell suspensions of mononuclear cells were seeded non-malignant cells, we calculated the RBF of drug-treated
in 384-well imaging plates containing small compound cells to compare on target drug-induced cytotoxicity with
libraries which were incubated overnight (18 h at 37°C and that of drug-induced cytotoxicity in blast-marker-negative
5% CO2). For most patients, libraries included 139 different cells (figure 1A). The RBF is thus defined as the fraction of
drugs in two concentrations and five technical replicates viable blasts surviving drug treatment relative to the
total (appendix p 9). Bone marrow samples for the average fraction of viable blasts observed in negative See Online for appendix
retrospective acute myeloid leukaemia study came from control samples. Drug sensitivity per patient was integrated
frozen biopsies, whereas biopsies used for the prospective over the drug matrix by averaging the number of RBF
study were all freshly acquired and not stored frozen. datapoints above (scored with +1) or below (–1) the
A comparison of pharmacoscopy results from fresh and hyperplane that best separated responders from non-
frozen material from the same biopsy showed good responders (figure 1G), weighted by the area under the
consistency in results (appendix p 2). Immunofluorescence receiver-operating characteristic curve (AUROC) of each
staining, imaging by automated microscopy (Opera cor­responding concentration point in the drug matrix.
Phenix; Perkin Elmer, Waltham, MA, USA), image ana­ We also compared clinical response with cytogenetic and
lysis (CellProfiler; Broad Institute of Harvard and molecular risk classification.
Massachusetts Institute of Technology, Boston, MA, USA), For the prospective study, eligible patients were assessed
and data analysis (Matlab; versions R2015a, R2015b, by the board (PBS, UJ, GIV, KM, CK, GH, IS-K, KO, WRS)
R2016a, R2016b, R2017a, R2017b) were done as described and those who met inclusion criteria were tested by
pre­viously.30 The antibodies used to identify the target blast pharmacoscopy as outlined above. The markers used to
population were selected based on clinical pathology identify the blast populations were selected individually
antigen expression assessment reports, and included CD3 for each patient based on their disease indication and
(HIT3a), CD19 (HIB19), CD20 (2H7), CD79a (HM47), clinical diagnostics. For the prospective study, on-target
CD34 (4H11), CD117 (104ED2), and CD138 (DL-101; cytotoxicity was identified by calculating the RBF as in the
eBiosciences [Thermo Fisher, Waltham, MA, USA]). acute myeloid leukaemia retrospective analysis, in which
Relative blast fractions (RBFs) were calculated as the blast, in this context, now referred to any cancer cell. Thus,
fraction of marker-positive viable cells after drug treatment top-scoring drugs achieved the most specific reduction of
divided by the average fraction of marker-positive viable the tumour-cell-enriched cell fraction ex vivo, while
cells measured in dimethyl sulfoxide (DMSO)-containing causing minimal cytotoxicity to the marker-negative
control wells. For hierarchical clustering of ex-vivo drug healthy cells also present in the sample. The board then
responses, all RBF values per patient and marker assessed the results, taking into account an individual
combination were first averaged over technical replicates patient’s previous treatment outcomes to recommend the
and concentrations per drug, and subsequently normalised next treatment regimen. Patients assessed by the board,
into pharmacoscopy scores via (1 – RBF)/max (1 – RBF). but who had further standard treatment options, were
Thus, a pharmacoscopy score of 1 represents the strongest used as an observational cohort. Integration of data from
on target ex-vivo response, a pharmacoscopy score of 0 both patient groups allowed us to test whether chemo­
indicates no ex-vivo effect, and negative pharmacoscopy resistance measured by pharmacoscopy (eg, ex-vivo
scores indicate ex-vivo chemoresistance. survival of blast cells coinciding with death of non-
To explore whether pharmacoscopy is predictive of malignant cells) is predictive of poor clinical response. To
clinical response, we designed a retrospective study using gain an overview of the complete dataset, we first set out
samples from patients with acute myeloid leukaemia to cluster the drug-response profiles. For this purpose,
collected before receiving standard first-line remission RBF values were normalised to pharma­coscopy scores;
induction therapy (figure 1A). Roughly 60% of patients negative values indicate drug resistance (blast survival and
typically respond with complete remission to induction non-malignant-cell death), and positive values indicate on-
therapy,31 which consists of cytarabine combined with target chemosensitivity (blast death and non-malignant-
daunorubicin and etoposide.32,33 Each patient sample cell survival; appendix p 5).
was screened through a drug combination matrix of all To account for the complicating fact that for most
three first-line drugs, consisting of 125 unique drug treatment regimens comprised of multiple drugs, ex-vivo

www.thelancet.com/haematology Vol 4 December 2017 e597

 Articles

A B
8000 Correlation = 0·99

cell number (caspase 3)

Total non-apoptotic
Non-responders A
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

(ten patients)
B
C
D
E
F
G

4000
H
I
J
K
L
M

Complete remission
N
O
P

(ten patients) DAPI

CD34 + CD117+ On target Inactive Toxic
Brightfield 0
0 4000 8000
Biobanked bone marrow Cytarabine, etoposide, and daunorubicin Automated microscopy Quantify cell death based Total number of
aspirates from 20 patients with 125 drug combinations in and image analysis on nuclear morphology non-fragmented nuclei
acute myeloid leukaemia quadruplicate 18 h incubation. 15 360 images per patient
taken at time of diagnosis 40 × 384-well plates

C D
p<0·0001
0 p<0·0001
Non-responders

1
(% blasts relative to DMSO)

1·2 3 1·0
Relative blast fraction

Relative blast fraction

10 Non-responders
Cytarabine (μM)
1·0
20
0·8 0·8
0·6 0
p<0·0001
1
Complete
remission

0·4 3 0·6 Complete remission

10
20
Etoposide (μM) 0 1 3 10 20 0 1 3 10 20 0 1 3 10 20 0 1 3 10 20 0 1 3 10 20 0·4
Daunorubicin (μM) 0 0·1 1 3 10 0 0·1 1 3 10

E F
0
Non-responders

Total cell number (relative to DMSO)

1 1·0
1.0 3
(relative to DMSO)
Total cell number

Cytarabine (μM)

10 0·8
0.8
20
0.6 p<0·0001
0 0·6
0.4 Non-responders
1
Complete
remission

0.2
3 0·4
Complete remission
10
20
0·2
Etoposide (μM) 0 1 3 10 20 0 1 3 10 20 0 1 3 10 20 0 1 3 10 20 0 1 3 10 20
Daunorubicin (μM) 0 0·1 1 3 10 0 0·1 1 3 10
Daunorubicin (μM)

G H I J
p<0·0001 90 AUROC (relative blast fraction) = 0·97
Cross validation accuracy (%)

1·0
Integrated relative blast fraction
Relative blast fraction

0·5
80
True positive rate

1·2 0·91 0·86

score per patient

1·0 70
0·8 0·5
60 0·50
0·6 0
0·4 5
4 50
0 3
5 10 15 2 40 0
20 25 1 Random Total cell Total Relative 0 0·5 1·0
–0·5 number blasts blast fraction False positive rate
Non-responders
Optimal separating plane Drug space Non- Complete
Complete remission responders remission Random Total cell number Total blasts Relative blast fraction

Figure 1: Pharmacoscopy and response to first-line acute myeloid leukaemia therapy

(A) Schematic overview of the retrospective analysis using biobanked bone marrow samples of 20 patients with acute myeloid leukaemia taken before first-line treatment. (B) Comparison
of two cell-death readouts: count of cells without activated caspase-3 and count of non-fragmented nuclei. Dots represent values from individual wells, for drug-screen results aggregated from
three different patient samples. (C) Heatmap of the DMSO-relative fraction of CD34+ CD117+ blasts (C), and total cell number relative to DMSO (E), averaged for the non-responders and complete
remission patient groups. Comparison of the DMSO-relative fraction of CD34+ KIT+ blasts (D) and total cell number (F) for the non-responders and complete remission groups, plotted as function of
increasing concentrations of daunorubicin. Data are mean (SE) of patients. p values are for one-sided t test testing for reductions in complete remission group compared to non-responders.
(G) Surface plot indicating the ideal separating hyperplane between the complete remission and non-responders groups. The drug space of 25 columns and five rows represents the same drug space
as shown in (C) and (E). (H) Integrated RBF response scores per patient. Boxplots show distributions, dots are values for individual patients; crosses indicate datapoints that do not fall between the
whiskers. (I) Cross-validation accuracies for integrated response scores for different cellular drug response readouts. (J) Average ROC curves over all cross-validation runs for different cell death
readouts. AUROC values are indicated. (C–F) Assays done in technical quadruplicates for each of the 20 patient bone marrow samples. (I–J) Averaged classification results over 2025 cross-validation
runs. AUROC=average area under ROC curve. DMSO=dimethyl sulfoxide. ROC=receiver operating characteristics.

e598 www.thelancet.com/haematology Vol 4 December 2017

 Articles

testing was in fact done with single drug treatments, and R2016b, R2017a, R2017b), and Microsoft Excel (version
that multiple and varying number of blast markers were 2016).
measured in different patients depending on their clinical The trial was registered at the ClinicalTrials.gov trial
diagnostic results, we summed the relevant pharma­ registry, number NCT03096821.
coscopy values over all drugs and markers per patient,
resulting in an integrated pharmacoscopy (i-PCY) score. Role of the funding source
We quantified overall response as 1=progressive disease, The funders of the study had no role in study design, data
2=stable disease, 3=partial response, 4=complete remis­ collection, data analysis, data interpretation, or writing of
sion, and determined the correlation with i-PCY. the report. PBS and UJ had access to patient annotated
clinical data, BS, GIV, and GS-F had access to the
Outcomes anonymised patient clinical data and correlated drug
The primary outcome measure was the proportion of responses. The corresponding author had full access to all
patients achieving progression-free survival, and the the anonymised results and final responsibility for the
secondary outcome measure was the proportion of patients decision to submit for publication.
with an overall response (achieving either a complete
remission or partial response). Progression-free survival Results
was calculated as the time from the first day of treatment The retrospective acute myeloid leukaemia study to
to the date of the first reported disease progression or determine whether pharmacocopy is predictive of clinical
relapse, initiation of a new (unplanned) anticancer treat­ response used 20 biobanked bone marrow samples;
ment, or death as a result of any cause. Overall response ten samples from patients achieving stable complete
was defined by achieving either complete remission or remission to induction therapy, and ten from non-
a partial response, defined by standard response definition responders to induction therapy.32,33 The correlation
guidelines.35,36 For patients with lymphoma, responses coefficient (r) when examining the number of non-
were classified as complete remission, partial response, fragmented nuclei in each image after drug treatment was
stable disease, or progressive disease according to the 0·99 with immunofluorescence against activated caspase-3
criteria proposed by the international working group on as a measure of cell death over samples from three patients,
malignant lymphoma.35 For patients with leukaemia, confirming the nuclear morphology readout (figure 1B).
responses were assessed following the response criteria Bone marrow immunohistochemistry from clinical
defined by the recommen­ dations of the European diagnostics confirmed the presence of CD34 and CD117 on
LeukemiaNet.36 All patients that were included in the blasts of all 20 patients. The 20 patients represented both
prospective trial had uniform follow-up intervals of sexes and diverse ages, had diverse genetic lesions and
4 weeks. karyotypes, and blast fractions at time of sampling ranging
from 30% to over 90% (appendix p 6). The clinical response
Statistical analysis to treatment in our retrospective cohort of patients with
The treatment was deemed to be of clinical benefit for the acute myeloid leukaemia only partially followed the
individual patient who has a progression-free survival ratio cytogenetic and molecular risk classification (appendix
(progression-free survival on pharmacoscopy-guided pp 3, 6), with, for instance, all four patient who had a FLT3-
therapy/progression-free survival on prior therapy) of 1·3 ITD mutation in the non-responders group and both
or higher. In such cases, we rejected the null hypothesis, inv(16) patients in the complete remission group.
defined as 15% or fewer patients having a progression-free The RBF was significantly different between complete
survival ratio of 1·3 or higher. Thus, the individual patient remission and non-responders groups, with significantly
was their own control. Comparisons of the overall response stronger on-target effects observed with ex-vivo dauno­
to previous treatment and pharmacoscopy-guided treat­ rubicin treatment for the complete remission patient
ments were calculated using a one-sided McNemar’s test cohort (p<0·0001; figures 1C, 1D, appendix pp 3, 7).
for paired binomial data with continuity correction. The Conversely, population-averaged cytotoxicity measure­
odds ratio (OR) could not directly be calculated as one of ments (total cell death) did not correctly stratify patients
the discordant values (those patients who did respond to based on their clinical response (figures 1E, 1F), indicating
the most recent treatment, but who did not respond to the need for the relative drug sensitivity measurements. As
pharmacoscopy-guided treatment) was equal to zero. We expected, the integrated response score for drug sensitivity
therefore calculated the overall response-associated OR revealed good separation of responders and non-
using the standard calculation for contingency tables. responders (figure 1H, appendix p 3). One particularly
Significance testing for progression-free survival dif­ strong outlier was observed, complete remission in
ferences was done using the log-rank (Mantel-Cox) test. All patient 10 for whom no ex-vivo response was measured.
correlations are Pearson correlation coefficients. All other This discrepancy could not be attributed to differences in
p values are two-tailed t tests, unless stated otherwise. clinical parameters nor technical issues. Cross-validation
Statistical analyses were done in GraphPad Prism by leaving out and reclassifying every possible combination
(version 7), Matlab (versions R2015a, R2015b, R2016a, of two patient samples, and calculation of the ideal

www.thelancet.com/haematology Vol 4 December 2017 e599

 Articles

hyperplane based on the remaining 18 samples, revealed Consistently, we observed reduced classification power for
an average classification accuracy of 88·1% for the RBF this cohort with population-averaged readouts: overall cell
(figure 1I), and an average AUROC of 0·97 (figure 1J). death, quantified by the total cell number, led to a

A B
Insufficient material (n=5) Drug not available (n=6) 100 Correlation = 0·92 CD117
Worsening condition (n=3) Worsening condition (n=8) CD138
Lost to follow-up (n=1) Lost to follow-up (n=4) CD14

Diagnostic pathology
CD19

(flow cytometry)
Lost Lost
CD3
(n=9) (n=18) CD30
Yes Pharmacoscopy-guided
50 CD34
A
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

treatment (n=17) CD34 CD10

Informed Inclusion
B
C

Tumour board CD56

E

F
G

consent criteria met?

I

CD7
J
K

Physician’s choice of
N

No
O

CD79a
P

Biopsy Pharmacoscopy treatment (n=13)

Patient (n=57) (n=48) 0
0 50 100
Test >100 clinically 1 Two or more previous treatment lines Pharmacoscopy
approved drugs 2 No further standard treatment available (automated microscopy)

C Patient 2, diffuse large B-cell lymphoma, CD20+ cells G Before After

1·2 NS p<0·05 Drug name Rank RBF p value

Ibrutinib
Relative blast fraction

1·0 Cisplatin 1 0·504 <0·0001

Ibrutinib 2 0·661 0·00048
0·8 Ixazomib 3 0·797 0·0092
Oxaliplatin 4 0·807 0·0067
0·6 Vinblastine sulfate 5 0·829 0·017
EPZ-5676 6 0·836 0·016
0·4
104 drugs

D Patient 3, precursor B-cell lymphoblastic lymphoma, CD19+ cells H

1·2 Drug name Rank RBF p value

Bortezomib
Relative blast fraction

6-mercaptopurine
1·0 Anastrozole 1 0·210 <0·0001
0·8 Cisatracurium besylate 2 0·284 <0·0001
Raltegravir 3 0·366 <0·0001
0·6 Bortezomib 8 0·498 0·00015
0·4 6-mercaptopurine 13 0·580 0·00095
Tamoxifen 14 0·583 0·0023
0·2
266 drugs

E Patient 5, diffuse large B-cell lymphoma, CD79a+ cells I

1·2 Dexamethasone Drug name Rank RBF p value

Bortezomib
Cladribine
Relative blast fraction

1·0 Bortezomib 1 0·589 <0·0001

Carfilzomib 2 0·669 <0·0001
0·8 Vorinostat 3 0·675 <0·0001
Romidepsin 4 0·712 <0·0001
0·6 Cladribine 5 0·727 0·00029
Dexamethasone 15 0·866 0·050
0·4
139 drugs

F Patient 9, primary mediastinal large B-cell lymphoma, CD20+ CD79a+ cells J

1·2 Drug name Rank RBF p value

Cladribine
Relative blast fraction

Teniposide 1 0·870 0·015

Vandetanib 2 0·919 0·021
1·0 Vorinostat 3 0·920 0·127
Paclitaxel 4 0·925 0·098
Carboplatin 5 0·925 0·027
Cladribine 6 0·931 0·025
0·8
139 drugs

Figure 2: Study outline and representative responses to pharmacoscopy-guided treatments

(A) Outline of the workflow of this study, including patient numbers (n) and reasons for patient dropping out of the study. (B) Comparison of percentage of marker-positive cells identified by flow
cytometry-based diagnostic pathology in the clinic and by automated microscopy and single-cell image analysis in patients for which both datasets were available. (C–F) RBFs for all screened drugs for
selected case studies of patients, with corresponding patient numbers, diagnoses, and used blast marker indicated. Bar graphs show drugs ranked by average RBF, indicating significant (dark grey) and
non-significant (light grey) RBFs; values of RBF>1·2 are capped at 1·2. Horizontal line at value 1 indicates DMSO-control levels. Table showing top-hit drug names, ranks, RBF, and p values (two-tailed t test
against controls); yellow highlights indicate selected drugs provided during pharmacoscopy-guided treatment. (G–J) PET-CT or PET-MRT scans corresponding to the patients described in (A–D), at the time
of last relapse (before) and after pharmacoscopy-guided treatments (after). White outlined boxes indicate tumour foci. NS=not significant. RBF=relative blast fractions.

e600 www.thelancet.com/haematology Vol 4 December 2017

 Articles

Diagnosis Age Previous Sample type Clinical Cell markers Pharmacoscopy-guided Overall Progression- Ongoing
(years) treatment diagnostic used treatment response free survival response
lines mutations (weeks)
1 B-cell acute lymphoblastic 23 5 Peripheral NRAS, CDKN2A CD10, CD34 Bortezomib Partial response 5·3 No
leukaemia blood
2 Diffuse large B-cell 69 7 Dissociated MYD88, CDKN2A CD20 Ibrutinib Complete 42·0 No
lymphoma lymph node remission
3 Precursor B-cell 51 3 Pleural Not determined CD19, CD20 Obinutuzumab, Partial response 12·9 No
lymphoblastic lymphoma effusion 6-mercaptopurine, bortezomib
4 Peripheral T-cell lymphoma 56 4 Bone marrow TP53 CD3 Ixazomib, lenalidomide, Complete 22·6 No
dexamethasone remission
5 Diffuse large B-cell 29 2 Dissociated No alterations CD79a Bortezomib, cladribine, Complete 34·0 Yes
lymphoma lymph node detected dexamethasone remission
6 B-cell acute lymphoblastic 29 2 Peripheral FLT3, KRAS CD20, CD34 Bortezomib, azacitidine Complete 37·1 Yes
leukaemia blood remission
7 Diffuse large B-cell 60 5 Dissociated MYD88 CD19, CD20 Imatinib, ibrutinib, Stable disease 37·3 Yes
lymphoma lymph node lenalidomide, obinutuzumab;
fludarabine,
cyclophosphamide*
8 Acute myeloid leukaemia 72 2 Peripheral NRAS CD34, CD117 Azacitidine Complete 22·4 No
blood remission
9 Primary mediastinal large 27 6 Dissociated No alterations CD20, CD30, Brentuximab vedotin, Complete 34·7 Yes
B-cell lymphoma lymph node detected CD79a cladribine remission
10 T-cell lymphoblastic 31 4 Peripheral PIK3CA, FBXW7, CD3 Bortezomib, cyclophosphamide, Partial response 4·1 No
lymphoma blood NOTCH1 dexamethasone
11 Acute myeloid leukaemia 72 3 Peripheral NPM1, KRAS CD34, CD117 Decitabine Partial response 8·4 No
blood
12 Diffuse large B-cell 67 3 Lymph node MYC CD20, CD79a Ibrutinib Complete 21·9 Yes
lymphoma remission
13 Follicular lymphoma grade 63 3 Skin biopsy TP53 CD19, CD20, Bortezomib, cladribine, Complete 19·3 Yes
3A CD79a dexamathasone remission
14 T-cell prolymphocytic 40 2 Peripheral No alterations CD3 Venetoclax Partial response 13·9 No
leukaemia blood detected
15 Acute myeloid leukaemia 76 4 Bone marrow No alterations CD34, CD117 Azacitidine Partial response 3·6 Yes
detected
16 Diffuse large B-cell 53 3 Dissociated TP53 CD19, CD79a Pixantrone, idelalisib, Partial response 7·4 Yes
lymphoma lymph node obinotuzumab
17 Diffuse large B-cell 50 3 Bone marrow TP53 CD19, CD20, Azacitidine, panobinostat, Stable disease 3·3 No
lymphoma CD79a atorvastatin

Data are provided for each patient (number 1–17). Patients 5, 6, and 9 could proceed to allogeneic stem-cell transplantation, patient 7 received CART-19 transfusion. *Administered sequentially.

Table: Characteristics, treatments, and clinical responses of the 17 patients receiving pharmacoscopy-guided treatment

classification accuracy of 68·5% (AUROC 0·86), and cell (IQR 4·5–8·7). The characteristics of the 17 patients who
death of marker-positive cells, quantified as the total blasts, had pharmacoscopy-guided treatment are listed in the
led to a classification accuracy of 78·1% (AUROC 0·91; table. The 13 patients reviewed by the board that received
figures 1I, 1J). treatments not guided by pharmacoscopy served as an
In the prospective study of the 57 patients with observational cohort (appendix p 8). A comparison be­
aggressive haematological malignancies, nine patients tween the percentage of marker-positive cells measured
were not assessed by the board for reasons given in from the same biopsies by clinical diagnostics-based
figure 2A. Of 48 patients who were assessed by the board, flow cytometry, the current gold standard, and by pharma­
18 were not included and 13 still had further treatment coscopy revealed strong consistency between the
options, leaving 17 patients who met the inclusion criteria two methods (r=0·92, p<0·0001; figure 2B).
to receive pharmacoscopy-guided treatment (figure 2A). Pharmacoscopy-guided treatment regimens resulted in
For these 17 patients, pharmacoscopy was always done on encouraging partial and complete remissions (table),
the same day as the biopsy procedure, and median time to documented as indicated by the type of malignancy, for
report pharmacoscopy results back to clinicians was 5 days example by PET-CT or PET-MRI for lymphoma. To visualise
(IQR 2–8). The trial started on Sept 1, 2015, the censoring the workflow, we present data for the first four patients who
date for the interim analysis for all patients was Nov 11, had partial or complete response, as documented by PET-
2016, and the median follow-up time was 7·6 months CT or PET-MRT. Patient 2, a 69-year-old man with diffuse

www.thelancet.com/haematology Vol 4 December 2017 e601

 Articles

however, the Bruton’s tyrosine kinase inhibitor ibrutinib

A
p=0·0013 had the second strongest ex-vivo efficacy (RBF 0·61,
18 (4/17) (15/17) Response p=0·00048; figure 2C). PET-CT imaging on day 49 of
Complete
16
remission ibrutinib treatment confirmed a complete remission for
14 Partial the patient (figure 2G). Good clinical responses have been
12 response reported for a small subset of patients with diffuse large
B-cell lymphoma carrying MyD88 mutations.37 Subsequent
Patients

10
No response
8
Stable
sequencing confirmed that patient 2 had a MyD88 mutation
6 disease (table). Patient 3, a 51-year-old women with precursor B-cell
4 Progressive lymphoblastic lymphoma, had three lines of previous treat­
disease
2 ment, and was progressive after immuno­therapy with the
0 bispecific CD3–CD19 antibody blinatumomab (table). Cells
Most recent Pharmacoscopy-
regimen guided treatment isolated from a pleural effusion were tested by pharma­
coscopy against a panel of 266 compounds in duplicate,
B which revealed significant ex-vivo sensitivity to the
100 Pharmacoscopy-guided treatment
Most recent regimen proteasome inhibitor bortezomib (RBF 0·498, p=0·00015)
Censored and the thiopurine 6-mercaptopurine (RBF 0·580,
80
Progression-free survival (%)

p=0·00095; figure 2D). 6-mercaptopurine and bortezomib

60
were combined with anti-CD20 obinutuzumab. After
p=0·0075 28 days PET-CT confirmed a partial response (figure 2H).
40 For patient 5, cells from an excised lymph node were tested
for 139 drugs (figure 2E). The patient achieved a complete
20 remission (figure 2I) to a combination of the single
strongest ex-vivo acting drug bortezomib (RBF 0·589,
0 p<0·0001), with cladribine ranked fifth (RBF 0·727;
0 10 20 30 40 50 60
Weeks
p=0·00029) and dexamethasone ranked 15th (RBF 0·866;
Number at risk p=0·050; figure 2E). And after ex-vivo sensitivity to
Pharmacoscopy 17 12 9 6 2 0 0
Most recent 17 7 4 3 3 0 0 cladribine was measured for patient 9 (figure 2F),
regimen complete remission was observed with cladribine treat­
ment in combination with CD30-targeted immuno­therapy
C
60 Most recent regimen
brentuximab vedotin (figure 2J). Data for patient 7, one of
Pharmacoscopy-guided treatment the two patients who did not respond to pharmacoscopy-
Progression-free survival (weeks)

guided treatment, is shown and further discussed in the

appendix (p 4).
40
*
* *
*
Overall response and progression-free survival of
pharmacoscopy were compared with overall response and
progression-free survival for the most recent regimen on
20 *
* which the patient had progressed. Four (24%) of 17 patients
achieved an overall response with the most recent regimen
*
* compared with 15 (88%) of 17 patients who achieved an
0 overall response with pharmacoscopy-guided treatment
17 patients from the pharmacoscopy-guided cohort
(odds ratio 24·38 [95% CI 3·99–125·4], p=0·0013;
Figure 3: Overall response and progression-free survival with figure 3A). Five (38%) of 13 patients receiving standard
pharmacoscopy-guided treatment salvage treatment based on physician’s choice achieved
(A) Comparison of overall response with the most recent regimen and an overall response (appendix p 8). Notably, none of the
of pharmacoscopy-guided treatments for 17 patients with aggressive
haematological malignancies. p value was calculated by McNemar’s test for 17 patients receiving pharmacoscopy-guided treatments
paired binomial data. (B) Kaplan-Meier plot showing progression-free survival had progressive disease as best overall response, whereas
with the most recent regimen and pharmacoscopy-guided treatments for seven patients had progressive disease in response to their
17 patients. (C) Progression-free survival with most recent regimen or most recent regimen (figure 3A). Furthermore, pharma­
pharmacoscopy-guided treatment per patient. *Ongoing response at time of
analysis. coscopy-guided treatments also led to a significantly
improved median progression-free survival (22·6 weeks
large B-cell lymphoma, relapsed after seven lines of [95% CI 7·4–34·0]) compared with a median of
previous treatment. Lymphoma cells of the sample were 5·7 weeks (4·1–12·1) in the same patients with the most
resistant to most of the 104 drugs tested, and only recent regimen (hazard ratio 3·14 [95% CI 1·37–7·22],
six compounds had significant on-target effects ex vivo p=0·0075; figure 3B). 12 (71%) of 17 patients had a
(figure 2C). Cisplatin and oxaliplatin were not regarded as progression-free survival ratio of 1·3 or higher (figure 3C).
feasible given the patient’s history, age, and comorbidities; The null hypothesis was therefore rejected. Notably,

e602 www.thelancet.com/haematology Vol 4 December 2017

 Articles

A B
p=0·0019
Patient 16 Patient 1 Patient 4 Patient 13
p=0·0031 Partial response Partial response Complete remission Complete remission
0·3 p=0·011 Diffuse large B-cell acute Peripheral T-cell Lymphoma
p=0·0081 B-cell lymphoma lymphoblastic lymphoma i-PCY=2·4
Average pharmacoscopy score

i-PCY=2·9 lymphoma i-PCY=1·6

Dexamethasone
0·2 i-PCY=2·2

Bortezomib
Methotrexate

Dexamethasone

Cladribine
Bortezomib

Lenalidomide
Pixantrone

Cytarabine
Ifosfamide
Vincristine

Etoposide
0·1

Idelalisib

Ixazomib
CD20
0 CD19 CD34 CD19
CD79a CD10 CD3 CD79a
–0·1

3
p=0·003
–0·2
2
Progressive Stable Partial Complete
disease disease response remission 1
Best overall response
0
i-PCY score

C −1 Accuracy=92%

−2
AUC=0·84
0·4 p<0·0001
−3
Average pharmacoscopy score

0·2 −4

−5 Correlation=0·49
0 p=0·0065
−6
p<0·0001

–0·2 Pro- Stable Partial Complete Progressive Partial response +

p=0·0079 gressive disease response remission disease complete remission
disease n=4 n=8 n=11 n=6 n=19
–0·4 n=6
Previous treatment Latest treatment Pharmacoscopy-cohort patient
Reference cohort patient
D Average i-PCY score
Correlation=0·44
Drug resistance (% pharmacoscopy score less than –0·1)

Deceased
25 p=0·016

CD79a CD79a CD20 CD79a

CD20 CD19 CD19 CD20
20 CD19
Vincristine
6-mercaptopurine
Cyclophosphamide
Cytarabine
Daunorubicin
Dexamethasone
Etoposide
Methotrexate
Vincristine

6-mercaptopurine
Cyclophosphamide
Cytarabine
Daunorubicin
Dexamethasone
Etoposide
Methotrexate

Ibrutinib
Imatinib
Lenalidomide

5-azacytidine
Atorvastatin
Panobinostat

15

10
Patient 37 Patient 33 Patient 7 Patient 17
Progressive disease Progressive disease Stable disease Stable disease
5 Mantle cell lymphoma Diffuse large B-cell Diffuse large Diffuse large
i-PCY=−6·2 lymphoma B-cell B-cell
0 1 2 3 4 5 6 7 i-PCY=−1·9 lymphoma lymphoma
Number of previous treatment lines i-PCY=−1·1 i-PCY=1·3

Figure 4: Pharmacoscopy and therapeutic response

(A) Average pharmacoscopy scores in all patients per best overall response reveal negative scores associated with progressive disease. (B) i-PCY score per patient by
best overall response in 29 patients. Individual dots correspond to individual patients. Bars show average i-PCY scores by overall response. Box and whisker plots
show i-PCY scores for progressive disease and partial response and complete remission responses. Inset shows the corresponding ROC curve. Coloured boxes show
pharmacoscopy data for all markers and drugs for selected patient; heat map colours range from dark red (iPCY<–1) to white (iPCY=0) to dark blue (PCY>1), see also
the legend in the appendix (p 5). i-PCY=integrated pharmacoscopy. (C) Average pharmacoscopy scores. p values directly above bars indicate significant deviation
from 0; p values of pairwise comparisons are indicated by corresponding connecting lines (A, C). (D) Box and whisker plots of the tested drugs to which ex-vivo
resistance is observed by pharmacoscopy per number of previous treatment lines in 29 patients. Individual patient values are plotted as black dots next to the
boxplots. ROC=receiver operating characteristics. Crosses indicate datapoints that do not fall between the whiskers.

eight (47%) of 17 patients that received pharma­coscopy- transfusion. Five (29%) of 17 patients received treatment
guided treatment regimens still had ongoing responses at regimens that included immunotherapy as part of their
the time of analysis (figure 3C), including patients 5, 6, pharmacoscopy-guided treatment, potentially confounding
and 9, who could proceed to allogenic stem-cell our interpretation of the results. We therefore re-tested
transplantation, and patient 7, who proceeded to CART-19 clinical response after exclusion of these five patients,

www.thelancet.com/haematology Vol 4 December 2017 e603

 Articles

which showed that both overall response (p=0·0002) of less than –0·2, less than –0·3, or less than –0·4. Patients
and progression-free survival (p=0·025) remained signif­ who had received no or only one previous treatment line
icantly improved for pharma­ coscopy-guided treatments showed ex-vivo chemoresistance (a pharmacoscopy score
compared with the most recent regimen. This reanalysis less than –0·1) to 9% of tested drugs, whereas patients
allowed us to exclude the possibility that the addition of that had received five or more previous treatment lines
antibody-based immuno­ therapies affected our interpre­ showed ex-vivo chemoresistance to 17% of tested drugs
tation of the results. Taken together, pharmacoscopy- (p=0·016; figure 4D).
guided treatment regimens de­ monstrated strongly im­ Patient outcomes correlated positively with the integrated
proved clinical responses and survival benefit in an initial pharmacoscopy scores (r=0·49, p=0·0065; figure 4B).
cohort of 17 late-stage patients with aggressive relapsed 18 (94%) of 19 responding patients (partial response and
and refractory haematological malignancies. complete remission) had i-PCY scores between 0 to 3,
To test whether chemoresistance measured by pharma­ whereas four (67%) of six patients with progressive disease,
coscopy is predictive of poor clinical response we did all of whom did not receive pharmacoscopy-guided
a cluster analysis including the 17 patients receiving treatments, had i-PCY scores in the negative range
pharma­coscopy-guided treatment with the 12 observation between –0·75 and –7. Patient treatments associated with
cohort patients whose subsequent treatments were also high i-PCY scores combined drugs acting on-target on all
tested ex-vivo before treatment initiation. Hierarchical tested blast markers, or combined neutral, ex-vivo acting
clustering of the pharmacoscopy response profiles per drugs with ex-vivo on-target acting drugs. Conversely,
patient and blast-markers as determined by clinical patients res­ ponding with progressive disease as best
diagnostics, overlaid with the best overall response overall response had treatments including drugs to which
corresponding to drug and patient pairs, revealed strong, ex-vivo chemoresistance was measured. One of two
extensive patient-to-patient vari­ability in both the number non-responding (stable disease) patients receiving pharma­
and identity of drugs to which either chemo­resistance or coscopy-guided treatments had an i-PCY score of below –1,
chemosensitivity was measured (appendix p 5). Similar indicating that the pharmacoscopy test did not strongly
indications displayed remarkable heterogeneity in support the final personalised treatment regimen for this
response profiles, indicating an absence of character­ non-responding patient, due to ex-vivo discordance
istic ex-vivo responses for the tested indications in this depending on the used blast markers (figure 4B). Overall,
partially heavily pretreated cohort. Hierarchical clustering the i-PCY score separated progressive disease from
repeatedly grouped drug classes with the same mode of patients who had achieved a partial response and complete
action, including immuno­modulatory drugs (thalidomide, remission with a classification accuracy of 92% and an
lenal­idomide, and pomalidomide), anthra­cycline chemo­ AUC of 0·84 (figure 4B).
therapies (daunorubicin, doxorubicin, and valru­ bicin),
and histone deacetylase inhibitors (belinostat, pano­ Discussion
binostat, and vorinostat). The clustering further high­ This single-centre study shows technical feasibility of
lighted the diversity of treatments given to the patients. integrating automated microscopy-based, ex-vivo drug-
30 unique drugs, distributed across the clus­tering, were response profiling for patients with aggressive haemato­
tested by pharmacoscopy and subsequently administered logical malignancies into clinical practice. The test-guided
to patients, enabling robust pan-treatment statistical treatment regimens led to significantly longer progression-
analysis (figure 4, appendix p 5). free survival and improved overall response in patients
Further analyses demonstrated the association between with various haematological malignancies compared with
ex-vivo chemoresistance and poor clinical outcome. First, their most recent regimens, warranting further disease-
plotting the average pharmacoscopy scores over all specific clinical studies that include larger patient cohorts
markers and drugs in relation to associated overall and randomised control groups.38 Although the trial did
response to those drugs showed that treatments leading to not include a randomised control group and had a
progressive disease were associated with negative pharma­ relatively small cohort size of 17 patients, our results
coscopy scores, whereas treatments leading to partial suggest that a wide array of working chemotherapeutics
response or complete remission resulted in significantly and targeted inhibitors already exist, which, in principle,
positive pharma­coscopy scores (figure 4A). Second, the are capable of breaking drug resistance even in
treatments to which the patient had relapsed before multirefractory cancers, if the right drugs are selected at
pharmacoscopy testing had on average negative the right time for each individual patient. We found that an
pharmacoscopy scores (p=0·0079; figure 4C). Third, the integrative combination of chemosensitivity of the
percentage of tested drugs to which ex-vivo resistance was leukaemic blasts and chemoresistance of the marker-
measured (at pharmacoscopy score less than –0·1) negative, non-malignant cells predicted clinical response
increased with the number of previous treatment rounds to first-line acute myeloid leukaemia treatment with the
of each of the 29 patients (r=0·44; p=0·016; figure 4D). highest accuracy. Furthermore, the same readout guided
Similar significantly positive correlations were found selection of treatments associated with favourable clinical
when defining chemoresistance as pharmacoscopy scores responses, and predicted both good as well as poor clinical

e604 www.thelancet.com/haematology Vol 4 December 2017

 Articles

responses. The positive relation observed between the selection of clinically beneficial combination treatments
number of previous treatment lines and ex-vivo drug suggests that not every drug combination needs to be
resistance is intuitive, and might reflect acquired drug tested in combination ex vivo, thus allowing for larger drug
resistance as well as refractory disease being more resistant panels to be tested; future studies are needed to further
from disease onset. refine optimal, ex-vivo drug-testing regimens.
Our investigation was designed as a prospective, non- Comprehensive drug response profiles of individual
randomised study in which every patient acted as their people, as generated here, represent the outcome of
own control. This approach allowed us to assess the overall interplay between various molecular parameters of the
effect across heterogeneous diseases and treatment responding cells, including not only the genetic, proteomic,
regimens; however, the absence of random­isation could and metabolic state of the cells, but also the direct and
have led to bias.6,38 Future randomised trials testing indirect molecular interactions with other cells.30 We
pharmacoscopy-guided therapies versus physician’s choice therefore hypothesise that such comprehensive drug-
are therefore warranted, and should focus on individual response profiles can offer novel functional insight into
disease entities. the underlying health status of an individual, with
Not all patients in our study had correlation between potentially wide-ranging implications in preventive and
pharmacoscopy results and outcome, in particular one participatory medicine. Given the fast throughput of the
outlier patient (patient 10 in the retrospective acute myeloid method, both experimentally and analytically, future
leukaemia study). Identifying the causes for such outliers studies can include higher patient numbers that will be of
will thus require repetition with larger cohort sizes and great interest to investigate the translatability of pharma­
integration with systematic molecular data. In our coscopy further.
comparison of con­ ventional genetics with response Pharmacoscopy provided useful treatment-guidance in
in the retrospective acute myeloid leukaemia cohort, our an initial late-stage patient cohort, warranting further
results matched those in previous studies.39 investigation in larger and indication-specific clinical trials.
A benefit of pharmacoscopy resides in the analytical It is likely that the approach will synergise well with
power derived from monitoring with computer-aided molecular profiling techniques such as genomics and
precision millions of individual single-cell drug responses, proteomics for personalised treatment identi­fication. Such
which combined with the ability to discriminate cell types studies could lead to improved patient treatment and be a
allows us to score specific rather than general and averaged useful route to mechanistic elucidation of clinically
cytotoxic effects. Pharmacoscopy will likely be instructive relevant genotype-to-phenotype relationships.
for the personalised identification of clinically effective Contributors
therapies for other malignancies beyond those tested here. BS, GIV, NK, CK, MS, OLdlF, EvdK, SG, LKa, JWB, NP, AR, LKe, PV, and
The selection of personalised therapy by pharmacoscopy SK did the experiments. BS, GIV, NK, KM, A-SS, OLdlF, UJ, PBS, and
GS-F collected and analysed data. KM, A-SS, CK, OLdlF, H-SC, JWB, GH,
benefits from the ability to measure hundreds to thousands NP, OM, AR, GJ, MEM, IS-K, KO, FF, LM, GWP, BK, LKe, WRS, RK,
of drug exposures using small patient samples, in which HG, PV, SK, UJ, PBS, and GS-F provided support and reagents. KM,
each ex-vivo treatment includes healthy cell controls from A-SS, CK, EvdK, SG, LKa, GH, NP, GJ, IS-K, KO, WRS, HG, PV, UJ,
the same patient sample. Pharmacoscopy detects cancer and PBS diagnosed patients. GJ, IS-K, WRS, HG, PV, UJ, and PBS were
involved in patient care. BS, GIV, NK, UJ, PBS, and GS-F wrote the
cells with fluorescently labelled antibodies against clinically manuscript. All authors contributed to and approved the final manuscript.
used diagnostic markers, which means the test synergises
Declaration of interests
with, and uses similar antibodies as, clinical flow BS is a shareholder of Allcyte, has a patent WO2016046346 licensed to
cytometry-based diagnostics. Both microscopy and flow Allcyte, and is a scientific cofounder of Allcyte. GIV is a shareholder of
cytometry or optometry share the limitations of detection Allcyte, has a patent WO2016046346 licensed to Allcyte, and is a scientific
of cancer cells by antibody-based immuno­fluorescence, cofounder of Allcyte. NK is a shareholder of Allcyte, has a patent
WO2016046346 licensed to Allcyte, and is a scientific cofounder of Allcyte.
whereas pharmacoscopy allows for reduced sample KM reports personal fees from Chugai, Kyowahakko Kirin, outside the
processing and increased throughput and automation. submitted work. GH reports grants and personal fees from Gilead; and
Our results show that single-cell detection of blast markers personal fees from Novartis, Amgen, and Ariad, outside the submitted
by pharma­coscopy enables a clinically useful comparison work. KO reports personal fees from Novartis, outside the submitted
work. RK reports personal fees from and has been an advisory board
of on-target and off-target cytotoxicity, while the minimal member for AOP Orphan, PharmaEssentia, and Qiagen; and has received
ex-vivo culturing of cells, and compatibility with clinical personal fees from Novartis, outside the submitted work. UJ reports
diagnostic markers, ensure fast and relevant feedback. personal fees from Janssen Cilag, during the conduct of the study; grants
and personal fees from Roche, Celgene, Gilead, Novartis, and True North
Specifically, the platform allowed us to test 768 conditions
Therapeutics; and personal fees from Amgen, Takeda, AbbVie, Infinity,
for almost all of the 17 patient samples, returning results to outside the submitted work. GS-F reports grants from ERC Proof of
the clinic within 5 days of receiving a sample. A crucial Concept, during the conduct of the study; and is a shareholder of Allcyte,
trade-off nonetheless remains between the number of has a patent WO2016046346 licensed to Allcyte, and is a scientific
cofounder of Allcyte. All other authors declare no competing interests.
different drugs, technical replicates, concentration ranges,
timepoints, and drug combinations that can be tested from Acknowledgments
We thank the patients and their families for their trust in taking part in
one biopsy. In that regard, the observation made in this
this study, and Manuele Rebsamen, Giorgia Jurisic, Katrina Vanura, and
study that ex-vivo testing of single treatments can aid

www.thelancet.com/haematology Vol 4 December 2017 e605

 Articles

Bernadette Hilgarth for assistance and critical reading of the manuscript. 17 Montero J, Sarosiek KA, DeAngelo JD, et al. Drug-induced death
The study was academically funded. CeMM is supported by the Austrian signaling strategy rapidly predicts cancer response to chemotherapy.
Academy of Sciences, and we gratefully acknowledge funding from the Cell 2015; 160: 977–89.
European Research Council Proof of Concept grant to GS-F, the Austrian 18 Frismantas V, Dobay MP, Rinaldi A, et al. Ex vivo drug response
Science Fund (FWF; grants F4704-B20 to PV, F4711-B20 to GS-F, and profiling detects recurrent sensitivity patterns in drug-resistant
P27132-B20 to PBS), the Anniversary Fund of the Austrian National Bank acute lymphoblastic leukemia. Blood 2017; 129: e26–37.
P-15936 to PBS, the Austrian Federal Ministry of Science, Research and 19 Kurtz SE, Eide CA, Kaempf A, et al. Molecularly targeted drug
Economy to SK, the National Foundation for Research, Technology and combinations demonstrate selective effectiveness for myeloid- and
Development to SK, the MPN Research Foundation to RK, the Swiss lymphoid-derived hematologic malignancies.
Proc Natl Acad Sci USA 2017; 114: e7554–63.
National Science Foundation (grants P300P3_147897 and PP00P3_163961
to BS, and P2EZP3_159114 to NK), and European Molecular Biology 20 Pemovska T, Kontro M, Yadav B, et al. Individualized systems
medicine strategy to tailor treatments for patients with
Organization long-term fellowship (number 1543-2012 to GIV).
chemorefractory acute myeloid leukemia. Cancer Discov 2013;
References 3: 1416–29.
1 Khwaja A, Bjorkholm M, Gale RE, et al. Acute myeloid leukaemia. 21 Majumder MM, Silvennoinen R, Anttila P, et al. Identification of
Nat Rev Dis Primers 2016; 2: 16010. precision treatment strategies for relapsed/refractory multiple
2 Hood L, Friend SH. Predictive, personalized, preventive, myeloma by functional drug sensitivity testing. Oncotarget 2017;
participatory (P4) cancer medicine. Nat Rev Clin Oncol 2011; 8: 184–87. 8: 56338–50.
3 Tsimberidou AM, Wen S, Hong DS, et al. Personalized medicine 22 Yu M, Bardia A, Aceto N, et al. Cancer therapy. Ex vivo culture of
for patients with advanced cancer in the phase I program at MD circulating breast tumor cells for individualized testing of drug
Anderson: validation and landmark analyses. Clin Cancer Res 2014; susceptibility. Science 2014; 345: 216–20.
20: 4827–36. 23 Hidalgo M, Amant F, Biankin AV, et al. Patient-derived xenograft
4 Von Hoff DD, Stephenson JJ, Jr, Rosen P, et al. Pilot study using models: an emerging platform for translational cancer research.
molecular profiling of patients’ tumors to find potential targets and Cancer Discov 2014; 4: 998–1013.
select treatments for their refractory cancers. J Clin Oncol 2010; 24 Roife D, Dai B, Kang Y, et al. Ex vivo testing of patient-derived
28: 4877–83. xenografts mirrors the clinical outcome of patients with pancreatic
5 Le Tourneau C, Delord JP, Goncalves A, et al. Molecularly targeted ductal adenocarcinoma. Clin Cancer Res 2016; 22: 6021–30.
therapy based on tumour molecular profiling versus conventional 25 Gao D, Vela I, Sboner A, et al. Organoid cultures derived from
therapy for advanced cancer (SHIVA): a multicentre, open-label, patients with advanced prostate cancer. Cell 2014; 159: 176–87.
proof-of-concept, randomised, controlled phase 2 trial. 26 van de Wetering M, Francies HE, Francis JM, et al. Prospective
Lancet Oncol 2015; 16: 1324–34. derivation of a living organoid biobank of colorectal cancer patients.
6 Friedman AA, Letai A, Fisher DE, Flaherty KT. Precision medicine Cell 2015; 161: 933–45.
for cancer with next-generation functional diagnostics. 27 Selby P, Buick RN, Tannock I. A critical appraisal of the “human
Nat Rev Cancer 2015; 15: 747–56. tumor stem-cell assay”. N Engl J Med 1983; 308: 129–34.
7 Hou HA, Lin CC, Chou WC, et al. Integration of cytogenetic and 28 Von Hoff DD. “Send this patient’s tumor for culture and
molecular alterations in risk stratification of 318 patients with sensitivity”. N Engl J Med 1983; 308: 154–55.
de novo non-M3 acute myeloid leukemia. Leukemia 2014; 28: 50–58. 29 Samson DJ, Seidenfeld J, Ziegler K, Aronson N. Chemotherapy
8 Byrd JC, Mrozek K, Dodge RK, et al. Pretreatment cytogenetic sensitivity and resistance assays: a systematic review. J Clin Oncol
abnormalities are predictive of induction success, cumulative 2004; 22: 3618–30.
incidence of relapse, and overall survival in adult patients with de 30 Vladimer GI, Snijder B, Krall N, et al. Global survey of the
novo acute myeloid leukemia: results from Cancer and Leukemia immunomodulatory potential of common drugs. Nat Chem Biol
Group B (CALGB 8461). Blood 2002; 100: 4325–36. 2017; 13: 681–90.
9 Schlenk RF, Dohner K, Krauter J, et al. Mutations and treatment 31 Walter RB, Kantarjian HM, Huang X, et al. Effect of complete
outcome in cytogenetically normal acute myeloid leukemia. remission and responses less than complete remission on survival
N Engl J Med 2008; 358: 1909–18. in acute myeloid leukemia: a combined Eastern Cooperative
10 Pastore F, Dufour A, Benthaus T, et al. Combined molecular Oncology Group, Southwest Oncology Group, and M D Anderson
and clinical prognostic index for relapse and survival in Cancer Center Study. J Clin Oncol 2010; 28: 1766–71.
cytogenetically normal acute myeloid leukemia. 32 Burnett A, Wetzler M, Lowenberg B. Therapeutic advances in acute
J Clin Oncol 2014; 32: 1586–94. myeloid leukemia. J Clin Oncol 2011; 29: 487–94.
11 Walter RB, Othus M, Paietta EM, et al. Effect of genetic profiling on 33 Dombret H, Gardin C. An update of current treatments for adult
prediction of therapeutic resistance and survival in adult acute acute myeloid leukemia. Blood 2016; 127: 53–61.
myeloid leukemia. Leukemia 2015; 29: 2104–07.
34 Wells SJ, Bray RA, Stempora LL, Farhi DC. CD117/CD34 expression
12 Zuber J, Radtke I, Pardee TS, et al. Mouse models of human AML in leukemic blasts. Am J Clin Pathol 1996; 106: 192–95.
accurately predict chemotherapy response. Genes Dev 2009;
23: 877–89. 35 Cheson BD, Pfistner B, Juweid ME, et al. Revised response criteria
for malignant lymphoma. J Clin Oncol 2007; 25: 579–86.
13 Pfreundschuh M, Trümper L, Osterborg A, et al.
CHOP-like chemotherapy plus rituximab versus CHOP-like 36 Dohner H, Estey EH, Amadori S, et al. Diagnosis and management
chemotherapy alone in young patients with good-prognosis diffuse of acute myeloid leukemia in adults: recommendations from an
large-B-cell lymphoma: a randomised controlled trial by the international expert panel, on behalf of the European LeukemiaNet.
MabThera International Trial (MInT) Group. Blood 2010; 115: 453–74.
Lancet Oncol 2006; 7: 379–91. 37 Wilson WH, Young RM, Schmitz R, et al. Targeting B cell receptor
14 Irish JM, Hovland R, Krutzik PO, et al. Single cell profiling of signaling with ibrutinib in diffuse large B cell lymphoma.
potentiated phospho-protein networks in cancer cells. Cell 2004; Nat Med 2015; 21: 922–26.
118: 217–28. 38 Le Tourneau C, Kamal M, Tsimberidou AM, et al. Treatment
15 Kornblau SM, Minden MD, Rosen DB, et al. Dynamic single-cell algorithms based on tumor molecular profiling: the essence of
network profiles in acute myelogenous leukemia are associated precision medicine trials. J Natl Cancer Inst 2016; 108: djv362.
with patient response to standard induction therapy. 39 Estey EH. Acute myeloid leukemia: 2012 update on diagnosis,
Clin Cancer Res 2010; 16: 3721–33. risk stratification, and management. Am J Hematol 2012;
16 Lacayo NJ, Alonzo TA, Gayko U, et al. Development and validation 87: 89–99.
of a single-cell network profiling assay-based classifier to predict
response to induction therapy in paediatric patients with de novo
acute myeloid leukaemia: a report from the Children’s Oncology
Group. Br J Haematol 2013; 162: 250–62.

e606 www.thelancet.com/haematology Vol 4 December 2017