Quantitative proteomics and phosphoproteomics onserial tumor biopsies from a sorafenib-treatedHCC patientEva Dazerta, Marco Colombia, Tujana Boldanovab, Suzette Moesa, David Adametzc, Luca Quagliatad, Volker Rothc,Luigi Terraccianod, Markus H. Heimb, Paul Jenoea, and Michael N. Halla,1

aBiozentrum, University of Basel, 4056 Basel, Switzerland; bDepartment of Biomedicine, University Hospital Basel, 4031 Basel, Switzerland; cDepartment ofMathematics and Computer Science, University of Basel, 4051 Basel, Switzerland; and dMolecular Pathology, University Hospital Basel, 4003 Basel, Switzerland

Contributed by Michael N. Hall, December 9, 2015 (sent for review September 18, 2015; reviewed by Bernd Bodenmiller and Pierre-Alain Clavien)

Compensatory signaling pathways in tumors confer resistanceto targeted therapy, but the pathways and their mechanisms ofactivation remain largely unknown. We describe a procedure forquantitative proteomics and phosphoproteomics on snap-frozenbiopsies of hepatocellular carcinoma (HCC) and matched nontumorliver tissue. We applied this procedure to monitor signaling pathwaysin serial biopsies taken from an HCC patient before and duringtreatment with the multikinase inhibitor sorafenib. At diagnosis, thepatient had an advanced HCC. At the time of the second biopsy,abdominal imaging revealed progressive disease despite sorafenibtreatment. Sorafenib was confirmed to inhibit MAPK signaling in thetumor, as measured by reduced ribosomal protein S6 kinase phos-phorylation. Hierarchical clustering and enrichment analysis revealedpathways broadly implicated in tumor progression and resistance,such as epithelial-to-mesenchymal transition and cell adhesion path-ways. Thus, we describe a protocol for quantitative analysis ofoncogenic pathways in HCC biopsies and obtained first insights intothe effect of sorafenib in vivo. This protocol will allow elucidation ofmechanisms of resistance and enable precision medicine.

liver | evasive resistance | signal transduction | MAPK | EMT

Hepatocellular carcinoma (HCC) is a global health concernwith an estimated 750,000 new cases per year (1). In more

than 80% of cases, HCC arises in a setting of liver cirrhosismainly of alcoholic or viral origin (2). The prognosis for HCCpatients is poor, with less than 30% qualifying for curativetreatments such as tumor resection or liver transplantation (2).Median survival time of patients that cannot be treated surgicallyis less than 1 y. Sorafenib is the only approved targeted therapyfor HCC, prolonging median patient survival by ∼3 mo (3).Sorafenib is a multikinase inhibitor of Raf (B and C), vascularendothelial growth factor receptor (VEGFR), and platelet-derived growth factor receptor (PDGFR) (4), which presumablyinhibits not only tumor cells but also endothelial cells responsiblefor tumor vascularization.Resistance to a targeted cancer drug can be intrinsic or adaptive

(5). Sorafenib is largely cytostatic (6), suggesting that intrinsic re-sistance is more common in tumors, although some reports describetumor shrinkage upon sorafenib treatment (7). Studies involvingHCC cell lines or immunohistochemical staining of tumor sectionsrevealed that sorafenib resistance correlates with the up-regulationof several signaling pathways, including the mammalian target ofrapamycin (mTOR) pathway as assayed by S6 S235/236 (8) and AktS473 phosphorylation (9). Other potential resistance mechanismsinvolve epithelial-to-mesenchymal transition (EMT) and autophagy(10, 11). However, the molecular mechanisms of sorafenib re-sistance in patients are largely unknown. Understanding the path-ways that confer intrinsic or adaptive resistance would allowprecision medicine and increase treatment efficacy.Proteomic analysis allows the identification of drug targets for

cancer treatment and biomarkers for cancer classification orrecurrence. In particular, MS is a powerful tool for resolving the

complexity of cancer signaling pathways. With regard to HCC,qualitative proteomics has been performed on resected tumormaterial (12), laser-capture microdissected material from tissuesections (13, 14), and primary hepatocytes or serum derived frompatients (15, 16). These studies (17, 18) identified HCC bio-markers such as glutamine synthetase and heat shock protein 70(Hsp70) that are currently in use for diagnosis (19, 20). Quan-titative proteomics has been performed on HCC resected tissueand serum (21, 22). Recently, proteomics has been performed ontumor biopsies of renal cell carcinoma patients (23). Severalstudies also have described phosphoproteomic analyses ofresected HCC or other cancer material (24–26), in some casesquantifying up to 8,000 phosphorylated sites (hereafter referredto as “phosphosites”) starting with 2 mg of protein (18, 27–30).However, to our knowledge, quantitative proteomics and phos-phoproteomics, hereafter collectively referred to as “(phospho)pro-teomics,” have yet to be performed on tumor biopsies, possiblybecause biopsy material is nonrenewable and typically providesonly a very small amount of protein. Importantly, quantitative(phospho)proteomics on serial biopsies taken before and duringtreatment has not been described. We note that although a biopsyprocedure generates less material than a resection, it has the im-portant advantage of capturing normally dynamic properties of atumor, such as the phosphorylation status of signaling pathways.Biopsies are immediately snap-frozen upon removal from the pa-tient and, unlike resected tissue, are obtained without causing is-chemia or hypoglycemia in the collected tissue. Needle biopsies aretaken routinely to diagnose and stage the disease. Another impor-tant consideration is a method to perform quantitative (phospho)pro-teomics, such as super-SILAC (“SILAC” is an acronym for

Significance

Elucidation of evasive resistance to targeted therapies is a majorchallenge in cancer research. As a proof-of-concept study, wedescribe quantitative proteomics and phosphoproteomics on serialtumor biopsies from a sorafenib-treated hepatocellular carcinoma(HCC) patient. This approach reveals signaling pathway activity in atumor and how it evades therapy. The described method willallow precision medicine based on phenotypic data. In particular,application of the method to a patient cohort will potentiallyidentify new biomarkers, drug targets, and signaling pathwaysthat mediate evasive resistance.

Author contributions: E.D., M.H.H., P.J., and M.N.H. designed research; E.D., T.B., S.M.,L.Q., and P.J. performed research; T.B., L.T., M.H.H., and P.J. contributed new reagents/analytic tools; E.D., M.C., D.A., V.R., L.T., M.H.H., P.J., and M.N.H. analyzed data; and E.D.,M.H.H., P.J., and M.N.H. wrote the paper.

Reviewers: B.B., University of Zurich; and P.-A.C., University Hospital of Zurich.

The authors declare no conflict of interest.1To whom correspondence should be addressed. Email: m.hall@unibas.ch.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1523434113/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1523434113 PNAS | February 2, 2016 | vol. 113 | no. 5 | 1381–1386

MED

ICALSC

IENCE

S

Dow

nloa

ded

by g

uest

on

June

16,

202

0

“stable isotope labeling of amino acids in cell culture”), thatallows direct comparison of biopsies obtained at different timesor from different patients (31).We describe quantitative (phospho)proteomic analyses of

needle biopsies of HCC and matched nontumor tissue from ahuman patient. These analyses provide a global snapshot ofsignaling pathways in the biopsy material. Analyzing serial bi-opsies taken from a patient before and during therapy, wemeasured differences in signaling pathways between tumor andmatched nontumor control tissue and the changes in these sig-naling pathways upon sorafenib treatment. Our findings provideinsight into mechanisms of tumor progression and resistance tocancer therapy.

ResultsExperimental Setup for Quantitative (Phospho)Proteomics on Biopsies.To monitor signaling pathways in HCC, we established an experi-mental protocol to perform (phospho)proteomics on tumor bi-opsies (Fig. 1A). The mean wet weight of eight examined biopsies,including biopsies used in pilot experiments, was 21 mg, fromwhich, on average, ∼0.9 mg of total protein was extracted forproteome and phosphoproteome analyses (Fig. 1B). After test-ing many permutations (Table S1), urea-based protein extractioncombined with strong cation exchange chromatography (SCX)and titanium dioxide (TiO2) phosphopeptide enrichment (32)was implemented as the most effective protocol (Fig. 1C).To perform quantitative (phospho)proteomics, we generated a

spike-in standard consisting of a mixture of cell extracts of fiveHCC cell lines metabolically labeled with heavy isotopes (13C- and15N-arginine and lysine), a method commonly known as super-SILAC (31). The spike-in standard was validated by measuring a1:1 mix of the labeled standard with corresponding unlabeledextracts (Fig. S1 A and B). Over 90% of all heavy-to-light (H/L)protein ratios were <1.5 (Fig. S1C), confirming the high labeling

efficiency and thereby the validity of the spike-in standard. From1 mg of spike-in standard, we could identify ∼4,500 proteins, andnearly all (80%) proteins could be quantified in two of three MSmeasurements (Fig. S1D).To determine the sensitivity of our experimental protocol (Fig.

1C) on biopsy material, we measured the proteome and phos-phoproteome (Datasets S1 and S2) of a liver biopsy from a pa-tient with hereditary hemochromatosis. Approximately 1.4 mg oftotal protein was extracted from the biopsy, from which ∼5,200proteins were identified (Fig. 1D); 98% of those proteins couldbe quantified. Approximately 8,500 phosphosites, correspondingto about 3,000 phosphoproteins, were identified (Fig. 1E); 92%of those phosphoproteins could be quantified. In total, com-bining the proteome and phosphoproteome, we were able toquantify ∼6,000 proteins, of which ∼1,800 were found in bothdatasets (Fig. S1E). The high percentage of quantified proteinsreflects good coverage of the liver biopsy by the spike-in standard(Fig. S1F). Thus, the spike-in standard was suitable for quanti-tative (phospho)proteomics on liver biopsies.Next, we sought to determine the coverage of signaling path-

ways in our quantified (phospho)proteome. We quantified 234kinases and several components of prominent signaling pathwaysinvolved in HCC, as well as important HCC biomarkers, in the(phospho)proteome (Table S2). Thus, our overall experimentalprotocol allows monitoring of cancer-related signaling pathwaysin liver biopsies.

Quantitative Pre- and On-Therapy (Phospho)Proteomics on Biopsiesfrom a Sorafenib-Treated HCC Patient. Next, we examined if ourexperimental protocol could quantify signaling changes in an HCCpatient upon sorafenib treatment, in particular in ontreatment bi-opsies compared with pretreatment biopsies. For this proof-of-principle study, a 62-y-old male patient with alcoholic liver cirrhosisand an advanced HCC was asked to donate additional biopsies forresearch purposes (Fig. 2A). Biopsies of the primary tumor andcontrol nontumor tissue were taken from the patient 7 wk beforesorafenib treatment and then again 14.5 wk later, after 7.5 wk ofsorafenib treatment. Shortly thereafter, treatment was stopped,and the patient died 2 mo later. HCC was confirmed by histo-pathological examination. The pre- and ontreatment tumor bi-opsies were 50% and 90% HCC, respectively (Fig. 2B). Bloodlevels of the HCC biomarker α-fetoprotein (AFP) (Fig. 2A) andtumor size increased throughout the observation period. A follow-up imaging study after 8 wk of sorafenib treatment documented aportal vein occlusion (probable tumor invasion of the portal vein)and ascites, suggesting progressive disease. As a consequence,sorafenib treatment was stopped.

Proteome P-peptide enrichment

(TiO2)

Phospho-proteome

Mix 1:1

HCC patient biopsy

(CTRL/TU)

HCC super-SILAC

standard

B 1 5

~10 fractions

Reduce/alkylate Digest

SCX

A

Biopsywet

weight

Reco-vered

protein

mg

D

# of

pro

tein

s x

103

Identified Quantified

E

# of

p-s

ites/

p-pr

otei

ns x

103

P-sites P-proteins0

2

4

6

8

10

Identified Quantified

Biopsy 1CTRL/TU

Sorafenib

Biopsy 2CTRL/TU

(Phospho)Proteome

(Phospho)Proteome

C

05101520253035

0.87

0123456

...

Fig. 1. Experimental setup for quantitative (phospho)proteomics on bi-opsies. (A) Schematic overview of proposed study. (B) Mean wet weight andrecovered protein of eight liver biopsies. (C) Workflow for quantitative(phospho)proteomics. Five different human HCC cell lines (numbered 1–5)were used to prepare the super-SILAC spike-in standard. (D and E) Proteins(D) and phosphosites and phosphoproteins (E) identified and quantified inbiopsy of hemochromatosis patient. CTRL, control; TU, tumor.

A

TUC

TRL

TU

100X 200X

CTR

L

Bpre sorafenib

on sorafenib

B B

0

40

80

AFP

IU /

ml

Sorafenib

CTRL

TU

8 12 16

CTRL

TU

pre sorafenib on sorafenib

1 2 3

Weeks 4 20 310

PVO

Fig. 2. Clinical data of the sorafenib-treated HCC patient. (A) Clinical historyof the sorafenib-treated HCC patient. The green bar indicates sorafenibtreatment. PVO, portal vein occlusion. (B) H&E staining of HCC patient bi-opsies. (Scale bars: 100× magnification, 200 μm; 200× magnification, 100 μm.)

1382 | www.pnas.org/cgi/doi/10.1073/pnas.1523434113 Dazert et al.

Dow

nloa

ded

by g

uest

on

June

16,

202

0

The wet weights of the pre-sorafenib control, pre-sorafenibtumor, on-sorafenib control and on-sorafenib tumor biopsieswere 40, 11.2, 30.4, and 31.9 mg, which yielded 1.4, 0.17, 1.12,and 0.52 mg of protein, respectively. Thus, the tumor biopsieswere smaller and yielded less protein than the control nontumorbiopsies (Fig. 3A), probably because of tumor fibrosis thathampered manipulation.We analyzed the (phospho)proteome of the tumor and non-

tumor control biopsies obtained before and during sorafenibtreatment (Datasets S3 and S4). In each control biopsy, about5,000 proteins were identified, of which about 4,000 (∼80%) couldbe quantified. In agreement with the lower protein yield, thenumber of proteins identified in the tumor biopsies ranged from3,000–4,000, of which about 80% could be quantified (Fig. 3A).Approximately 3,800 phosphorylated sites in about 1,500 phos-phoproteins were detected in each control biopsy, and 1,000–2,000 phosphosites in ∼500–1,000 phosphoproteins were de-tected in the tumor biopsies (Fig. 3B). The lower number ofdetected phosphoproteins in the tumor biopsies was again theresult of lower protein yields. As expected, the (phospho)pro-teomes of the tumor and control biopsies largely overlapped(Fig. S2 A and B). Finally, combining data from the proteomeand phosphoproteome, we quantified 4,379 proteins in, for ex-ample, the pre-sorafenib control biopsy in two of three mea-surements (Fig. S2C). Of those, 3,365 proteins were measuredonly in the proteome, 332 proteins were found exclusively in thephosphoproteome, and 682 were in both datasets. These resultsare comparable to the results obtained with the control biopsyfrom the hemochromatosis patient analyzed previously (Fig.S1E). The slightly lower number of phosphorylated pro-teins detected in the biopsies from the HCC patient could beexplained by reduced tissue quality.As a proof of principle, we analyzed the expression of the

three most common HCC biomarkers, glutamine synthetase,glypican3, and Hsp70 (33). As shown in Fig. S2 D and E, glu-tamine synthetase expression was higher in the tumor than innontumor tissue and increased further upon sorafenib treatment.Glypican3 and Hsp70 were higher in the tumor but only uponsorafenib treatment. This finding suggests that the tumor pro-gressed during sorafenib treatment, in agreement with the ob-served increase in serum AFP levels and tumor size.How did quantified protein and phosphorylation levels change

significantly between biopsies? As shown in the volcano plots ofFig. S3A, 140 proteins were significantly up-regulated and 142were down-regulated in the pretreatment tumor comparedwith the matched nontumor tissue, whereas 290 proteins were

up-regulated and 177 were down-regulated in the treated tumorcompared with its matched control. With regard to phosphory-lation levels (Fig. S3B), 82 phosphosites were up-regulated and 8were down-regulated in the pretreatment tumor compared withmatched nontumor tissue, and 70 phosphosites were up-regu-lated and 19 were down-regulated in the treated tumor com-pared with matched nontumor tissue. Thus, our quantitativeanalysis of only ∼1 mg of protein recovered from biopsies pro-vided good coverage of the (phospho)proteome and detectedsignificant changes in protein expression and phosphorylation.Importantly, even from a small biopsy containing much lessprotein (0.17 mg), about 1,010 phosphosites could be detected,of which 424 could be reliably quantified.

Expression and Phosphorylation Clusters in Response to Sorafenib.We performed unsupervised hierarchical clustering to identifygroups of proteins and phosphosites that display similar ex-pression patterns across all four biopsies, i.e., the two nontumorcontrol biopsies and the two tumor biopsies, each minus and plussorafenib. The heat maps in Fig. 4 display proteins and phos-phosites that were quantified in at least two of three measurementsper biopsy and that passed an ANOVA-based multisample test forstatistically significant up- or down-regulation, with a false discoveryrate (FDR) of 2% or 5%. Column-wise clustering of triplicatemeasurements of a given biopsy demonstrated that interbiopsybiological variation was greater than intrabiopsy experimentalvariation. Row-wise clustering of proteins and phosphosites(hereafter collectively referred to as “factors”) revealed fiveprominent groups based on tissue distribution and response tosorafenib (Fig. 4 and Datasets S5 and S6). Group 1 containsfactors that are decreased in the tumor biopsies relative tonontumor biopsies. These factors could be tumor suppressorswhose absence in the tumor conferred intrinsic resistance. Group2 includes factors that are increased only in the tumor and onlybefore sorafenib treatment. These factors could be oncogenicsorafenib targets. Importantly, phosphorylation of ribosomalprotein S6 kinase (Rsk-2)-S369, a site phosphorylated by theMAPK Erk, is in this group (see below). Group 3 consists offactors that are down-regulated specifically in the tumor before

A BIdentified proteins Quantified proteins mg protein recovered

0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6

0

1

2

3

4

5

6

mg protein recovered

CTRL-

TU-

CTRL+

TU+Sfb

CTRL-

TU-

CTRL+

TU+

Identified p-sitesQuantified p-sitesIdentified p-proteins Quantified p-proteins

Sfb

0

1

2

3

4

# of

pro

tein

s x

103

# of

p-s

ites/

p-pr

otei

ns x

103

from biopsy

Fig. 3. Statistics of quantitative (phospho)proteomics on biopsies from asorafenib-treated HCC patient before and during therapy. Graphs show thenumber of proteins (A) and phosphosites and phosphoproteins (B) identifiedand quantified in two of three measurements per biopsy.

3

CTR

L

1

A

Intrinsic res. factors

Intrinsic res. factors

Adaptive res. factors

Sfb targets

Sfb - + - +

CTR

L

TU

TU

2 31 2 31 2 31 2

1

2

3

4

5

B

- + +-

CTR

L

CTR

L

TU

TU

31 2 31 2 31 2 31 2

2

3

4

5

Sfb targets

Sfb

Fig. 4. Expression and phosphorylation clusters in response to sorafenib.Unsupervised hierarchical clustering using Euclidian distance of the pro-teome (A) and the phosphoproteome (B) data from at least two of threemeasurements per biopsy. Numbers correspond to groups in A.

Dazert et al. PNAS | February 2, 2016 | vol. 113 | no. 5 | 1383

MED

ICALSC

IENCE

S

Dow

nloa

ded

by g

uest

on

June

16,

202

0

treatment. These factors behave as expected for a tumor sup-pressor whose presence is restored upon sorafenib treatment.Group 4 contains factors that are up-regulated specifically inthe tumor upon sorafenib treatment. These are factors possiblyinvolved in tumor progression and adaptive resistance. Membersof group 4 could be targeted by a second drug in combination withsorafenib to achieve more effective treatment. Group 5 containsfactors that are up-regulated specifically in the tumor biopsies,with and without sorafenib. These could be oncogenic, intrinsicresistance factors. The clinical significance of the factors in theabove groups requires further characterization and confirmation.We performed pathway enrichment analysis on the proteins

that could be adaptive resistance factors, i.e., the proteins ofgroup 4 (Fig. S3C). Using four different over-representationanalysis methods, we found several annotation terms that wereidentified more than once, e.g., lysosome, glycolysis and pyruvatemetabolism, apoptosis, and the c-myc pathway. Interestingly,these terms are consistent with the notion that sorafenib inhibitsvascularization, thus limiting oxygen and nutrient availability inthe tumor.

Changes in MAPK Pathway Activity upon Sorafenib Treatment. Wenext investigated the effect of sorafenib on its known targetin vivo. To this end, we examined phosphorylation changes in theRaf–MAPK/ERK kinase–ERK (Raf–Mek–Erk) MAPK pathway(34–37) and downstream substrates in the tumor in response tosorafenib (Fig. 5). In particular, we detected and quantifiedphosphorylation of C-Raf-S621, B-Raf-S729, Rsk-2-S369, eIF4B(eukaryotic translation initiation factor 4B)-S422, Filamin-A-S2152, raptor-S863 (mTORC1 component), S6-S236, and -S240,and CAD (carbamoyl-phosphate synthetase 2, aspartate trans-carbamylase, and dihydroorotase)-S1859. Phosphorylation of thesites in C-Raf and Rsk-2 decreased in the tumor upon sorafenibtreatment, albeit significantly only for Rsk-2. This effect, whichwas even more pronounced when normalized to matched non-tumor tissue, indicates that sorafenib treatment inhibited Raf–Erk signaling. Phosphorylation of the pyrimidine biosynthesisprotein CAD also decreased in the tumor upon treatment butstill remained at a high level. Conversely, phosphorylation ofFilamin-A-S2152 and S6-S240 increased in the tumor uponsorafenib treatment. The Filamin-A site appeared as a potentialadaptive resistance factor, as revealed by hierarchical clusteringof the phosphoproteome (Fig. 4B). The high level of eIF4Bphosphorylation in the tumor both before and during sorafenibtreatment suggests that eIF4B may be an intrinsic resistancefactor (see above). Finally, B-Raf-S729, raptor-S863, and S6-S236phosphorylation did not change significantly upon sorafenibtreatment. The quantitative decrease in C-Raf and Rsk-2 phos-phorylation upon sorafenib treatment suggests that sorafenibindeed inhibited Raf–Erk signaling in the tumor. Importantly, toour knowledge, this is the first demonstration that sorafenib in-hibits MAPK signaling in a cancer patient. However, the elevatedlevels of eIF4B, Filamin-A, S6-S240, and CAD phosphorylationsuggest that sorafenib treatment failed to inhibit signaling distal toRsk-2. This lack of sorafenib responsiveness may suggest that analternative pathway compensates for the block in Raf–Erk signal-ing. Further analysis, including a cohort of patients, may revealthe identity of the compensatory pathway and allow better-informed treatment.

Pathway Enrichment Analysis Reveals Potential Mechanisms ofSorafenib Resistance. To obtain insight into global changes oc-curring in the biopsied tissues, two different pathway enrichmentanalyses were performed using MetaCore (Fig. S4). In the firstanalysis (Fig. S4A and Dataset S7), matched pretreatment tumorand nontumor biopsies were compared directly. The proteomeshowed a strong enrichment of metabolic pathways, in particularproteins involved in androgen metabolism. Enzymes with a role

in the “backdoor” pathway of androgen activation [AKR1C1–3(Aldo-keto reductase C1–C3)] (38) were up-regulated in the tumorbefore treatment and increased even further in the tumor follow-ing treatment (Fig. S5A and Dataset S3). The enzyme AKR1C3has been described to increase aggressiveness in prostate cancer(39). Increased expression of AKR1C3 together with the ob-served expression of 17β-HSD6, which catalyzes the final step ofdihydrotestosterone synthesis (40), and decreased expressionof 17β-HSD2, which inactivates dihydrotestosterone (41), sug-gests enhanced local production of active androgens and possiblyandrogen-dependent cancer cell proliferation. We also observeddecreased expression of enzymes involved in degradation of such

A RTK

MNK

MEK

eIF4E

Grb2 RasC-RafSOS

Translation

S621

AKT

FLN-A S2152

eIF4B S422

S6mTOR

S236S240

Sorafenib

PEA15S863

raptor

TSC

B-Raf S729 AMPK

Cytoskeleton

4EBPPKA S83

S116

AKT

S6KCAD S1859

Cell growth

P-s

ite c

hang

es (n

orm

. cha

nge

upo

n tre

atm

ent,

log 2 f

old)

Set to 0: CTRL pre sorafenib TU pre sorafenib CTRL on sorafenib TU on sorafenib

P-s

ite c

hang

es (n

orm

. to

CTR

L pr

e S

fb, l

og2 r

atio

H/L

)

B

C

C-Raf

S621 B-R

af

S729 Rsk

-2

S369 S6

S236 S6

S240 rap

tor

S863 CAD

S1859

eIF

4B

S422 FLN

-A

S2152

* *

-2

-1

0

1

-3 -2 -1 0 1 2 3 4

S

S

S

* * *

S S

S369RSK-2

ERK

Shc

Transcription

Fig. 5. Changes in MAPK pathway (34–37) activity upon sorafenib treat-ment. (A) Quantitative data for proteins and phosphosites are color coded:gray, protein quantified in proteome; white, protein not quantified inproteome; blue, phosphosite down-regulated in tumor upon treatment;black, phosphosite unchanged; red, phosphosite up-regulated in tumorupon treatment; red cross, inhibitory phosphosite; black star, phosphositeconstitutively high in tumor. (B) Quantitative results (−log2-transformedmean ± SD of H/L ratios of each biopsy normalized to control pre-sorafenibbiopsy) of prominent signaling nodes from A. S, single ratio measurementonly; *, significant in ANOVA-based two-sample test comparing the bi-opsies, FDR 5% (quantified in at least two of three measurements perbiopsy). (C ) Normalized change upon treatment (see Fig. S2D): −log2

[(TU/CTRL on-sorafenib)/(TU/CTRL pre-sorafenib)] of prominent signalingnodes from A. *, significant in ANOVA-based multisample test over all fourbiopsies, FDR 5% (quantified in at least two of three measurements perbiopsy).

1384 | www.pnas.org/cgi/doi/10.1073/pnas.1523434113 Dazert et al.

Dow

nloa

ded

by g

uest

on

June

16,

202

0

hormones (UGT2B15, UGT2B17, UGT1A4) in the tumor com-pared with control tissue, likely indicating a loss of liver-specificcharacteristics (42). Interestingly, elevated levels of active andro-gens represent a well-described risk factor for HCC development,accounting for the higher incidence of HCC among males (43).Finally, the proteome also was enriched in immune responsepathways, possibly indicative of inflammation in the tumor. Thephosphoproteome exhibited enrichment in pathways involved incell adhesion, translation, and insulin signaling, all pathwayscommonly implicated in cancer.In the second analysis (Fig. S4B and Dataset S8), we com-

pared tumor tissue obtained before and during sorafenib treat-ment. In this case, changes were normalized to the matchednontumor control tissues (Fig. S2D). This analysis identifieschanges that are sorafenib-induced or changes caused by tumorprogression; it is not possible to distinguish these two scenarios,because untreated biopsies from the second time point arenot available. The proteome displayed a strong enrichment ofpathways involved in central carbon metabolism, suggesting anincrease in glycolysis and a decrease in Krebs cycle componentsin the sorafenib-treated tumor. The increased glycolytic enzymeswere aldolase, enolase, and pyruvate kinase; the decreased Krebscycle enzymes were isocitrate dehydrogenase, α-ketoglutaratedehydrogenase, and succinate dehydrogenase (Dataset S3). Thisfinding is in agreement with the recent observation that in-creased glycolysis correlates with sorafenib resistance (44) andthat sorafenib inhibits vascularization. Furthermore, both theproteome and phosphoproteome revealed enrichment of path-ways involved in cell proliferation, including the MAPK pathway,as described above (Fig. 5).Both MetaCore analyses revealed significant enrichment of cell

adhesion pathways (Fig. 6 and Fig. S5B). Interestingly, vimentinand the extracellular matrix (ECM) proteins fibronectin andvitronectin were up-regulated, and desmoplakin, plakoglobin,the junctional adhesion molecules JAM-1/2 and zona occlu-dens-1 (ZO-1), the putative tumor suppressor p120 catenin, andthe cytoskeleton regulator α-catenin were down-regulated inthe tumor upon treatment. Collectively, these changes suggest

that EMT occurred in the tumor, during tumor progression and/orin response to sorafenib (45, 46). Importantly, EMT correlates withsorafenib resistance and tumor invasiveness (10, 47), whereas theECM plays a role in drug resistance in a process called “cellularadhesion molecule-dependent drug resistance” (CAM-DR) (48,49). Both processes may have contributed to increased tumor in-vasiveness and the failure of sorafenib treatment in the patient. Inagreement with this notion, a portal vein occlusion (probable tumorinvasion of the portal vein), a finding interpreted as progressivedisease, was observed in the patient after 8 wk of sorafenib treatment.

DiscussionWe describe a protocol for quantitative (phospho)proteomicanalysis of needle biopsies. We used this protocol to examinecontrol and tumor biopsies obtained from an HCC patient be-fore and during sorafenib treatment. The small, needle biopsiesyielded 0.17–1.4 mg of protein from which we were able toquantify 2,000–4,000 proteins and 500–2,100 phosphosites. Toour knowledge, this is the first quantitative phosphoproteomicanalysis of serial biopsies from a cancer patient. Phosphorylationin the Raf–Erk–Rsk pathway was reduced in the sorafenib-treated tumor, indicating that the drug was indeed effective ininhibiting its target kinase, Raf. Interestingly, phosphorylation ofdownstream targets was not similarly inhibited, suggesting that acompensatory pathway(s) may have been active in the tumor,possibly accounting for the treatment failure observed in thepatient. Pathway enrichment analysis suggests EMT and CAM-DR as processes potentially involved in tumor progression andtreatment failure. The protocol described here could eventuallybe translated to the clinic to provide precision therapy and thusmore effective treatment of patients.Studies analyzing phosphorylation in biopsies thus far have

used antibody-based immunohistochemistry, immunoblotting, orreverse-phase protein arrays (8). In contrast, our protocol allowsquantitative analysis of multiple pathways in parallel and eventhe detection of new phosphorylation sites. A critical factor is thequality of the biopsy material. The coaxial needle biopsy tech-nique used allows sampling multiple times at the same site on thetumor. It is essential that one of the biopsy samples is evaluatedby a pathologist for diagnosis and grading of the HCC. Impor-tantly, the pathologist also provides information on the relativeamounts of vital tumor cells, stromal cells, fibrosis, and sclerosisin the biopsy, and the amount of nonmalignant liver tissue. Anaccurate pathological evaluation is even more important for bi-opsies of infiltrative type HCC that often do not have discreteboundaries in imaging studies. Based on the results of thepathological assessment, one can determine whether HCC bi-opsies obtained at different times during disease progression canbe compared. For serial biopsies before and during treatment,we sampled the same tumor nodule. However, tumor shape canchange over time. Given this limitation in the serial biopsyprocedure, it is imperative to use robust statistics for the analysisof (phospho)proteomic data. Only changes of more than twofoldor changes rated as significant by ANOVA-based tests are takeninto consideration. Importantly, we did not observe a bias towardup- or down-regulated proteins that could be correlated with thepercentage of tumor tissue present in the biopsy.We obtained evidence for EMT and up-regulation of ECM

proteins in the tumor upon treatment. This finding suggests thatthe tumor progressed toward a more aggressive and invasivephenotype. Indeed, the patient displayed portal vein occlusion(probably caused by tumor invasion of the portal vein), sug-gesting progressive disease. Furthermore, EMT and the up-regulation of ECM proteins could be one of the potential com-pensatory pathways driving progression of HCC upon sorafenibtreatment, as observed in the patient. Data from a cohort ofpatients will provide a more complete picture of how HCCprogresses and the mechanisms of sorafenib resistance. The

Receptoractivationunspecified

Ligand

VE-cadherin

N-cadherin

Connexin43/(40)/(37)

GAP JUNCTION ADHERENS JUNCTION

JAM-1/2

Claudin1/5/(3)/(14)

Occludin

TIG

HT

JUN

CTI

ON

β-cat.

α-cat.

Plakoph.4

ZO-1

ZO-2

Cing.

Afadin

α-act.

Actin γ1

Desmopl.

Vimentin

BAIAP1

Plakoglobin

Rap1 regulationpathway

p120-cat.

CYTOSKELETONREMODELING

P P

P P PP

Fig. 6. Pathway enrichment analysis indicates EMT upon sorafenib treat-ment. Pathway map of junctional mechanism of cell adhesion generatedusing MetaCore (Fig. S4). The left half of the protein icon and the leftphosphate group represent tumor vs. control pre-sorafenib; the right half ofthe protein icon and the right phosphate group indicate normalized changeupon treatment (see Fig. S2D); white, not detected; gray, not significantlychanged; blue, significantly reduced more than 0.75-fold (log2); red, signif-icantly increased more than 0.75-fold (log2).

Dazert et al. PNAS | February 2, 2016 | vol. 113 | no. 5 | 1385

MED

ICALSC

IENCE

S

Dow

nloa

ded

by g

uest

on

June

16,

202

0

described protocol could facilitate the search for new therapeuticoptions for HCC and also could be applied to other cancerstreated with other drugs to further the understanding of carci-nogenesis and cancer drug resistance in general. Another futureperspective is to identify predictive molecular patterns thatwould aid in patient stratification.

Materials and MethodsStudy Design and Patient Biopsy Selection. Patients were recruited in the Clinicfor Gastroenterology and Hepatology of the University Hospital Basel,Switzerland. They gave written informed consent to donate liver biopsyspecimens for research. The pilot experiment biopsies and the hemochro-matosis patient biopsy were collected to build up a biobank; the HCC patient

biopsies were collected as part of a study protocol, number EK 152/10. Bothprotocols were approved by the Ethics Committee of the Canton Basel,Switzerland. The biopsies were performed under ultrasound guidance using acoaxial needle technique allowing repetitive sampling from the same part ofa focal lesion with a full-core biopsy instrument (BioPince; Angitech). Di-agnosis of HCC was made by histopathological analysis.

Other Methods. For further information on materials and methods, see SIMaterials and Methods.

ACKNOWLEDGMENTS. We acknowledge support from the Goldschmidtand Jacobson Foundation (E.D.), the European Research Council (Mecha-nisms of Evasive Resistance in Cancer), and the Swiss National ScienceFoundation (M.N.H.).

1. Yang JD, Roberts LR (2010) Hepatocellular carcinoma: A global view. Nat RevGastroenterol Hepatol 7(8):448–458.

2. Mazzanti R, Gramantieri L, Bolondi L (2008) Hepatocellular carcinoma: Epidemiologyand clinical aspects. Mol Aspects Med 29(1-2):130–143.

3. Llovet JM, et al.; SHARP Investigators Study Group (2008) Sorafenib in advanced he-patocellular carcinoma. N Engl J Med 359(4):378–390.

4. Wilhelm SM, et al. (2004) BAY 43-9006 exhibits broad spectrum oral antitumor activityand targets the RAF/MEK/ERK pathway and receptor tyrosine kinases involved intumor progression and angiogenesis. Cancer Res 64(19):7099–7109.

5. Bergers G, Hanahan D (2008) Modes of resistance to anti-angiogenic therapy. Nat RevCancer 8(8):592–603.

6. Carlo-Stella C, et al. (2013) Sorafenib inhibits lymphoma xenografts by targetingMAPK/ERK and AKT pathways in tumor and vascular cells. PLoS One 8(4):e61603.

7. Escudier B, et al.; TARGET Study Group (2007) Sorafenib in advanced clear-cell renal-cell carcinoma. N Engl J Med 356(2):125–134.

8. Masuda M, et al. (2014) Alternative mammalian target of rapamycin (mTOR) signalactivation in sorafenib-resistant hepatocellular carcinoma cells revealed by array-based pathway profiling. Mol Cell Proteomics 13(6):1429–1438.

9. Chen KF, et al. (2011) Activation of phosphatidylinositol 3-kinase/Akt signalingpathway mediates acquired resistance to sorafenib in hepatocellular carcinoma cells.J Pharmacol Exp Ther 337(1):155–161.

10. van Malenstein H, et al. (2013) Long-term exposure to sorafenib of liver cancer cellsinduces resistance with epithelial-to-mesenchymal transition, increased invasion andrisk of rebound growth. Cancer Lett 329(1):74–83.

11. Zhai B, et al. (2014) Inhibition of Akt reverses the acquired resistance to sorafenib byswitching protective autophagy to autophagic cell death in hepatocellular carcinoma.Mol Cancer Ther 13(6):1589–1598.

12. Li N, et al. (2009) Proteomic analysis of differentially expressed proteins in hepatitisB virus-related hepatocellular carcinoma tissues. J Exp Clin Cancer Res 28:122–132.

13. Li C, et al. (2004) Accurate qualitative and quantitative proteomic analysis of clinicalhepatocellular carcinoma using laser capture microdissection coupled with isotope-coded affinity tag and two-dimensional liquid chromatography mass spectrometry.Mol Cell Proteomics 3(4):399–409.

14. Ai J, et al. (2006) Proteome analysis of hepatocellular carcinoma by laser capturemicrodissection. Proteomics 6(2):538–546.

15. Bao H, et al. (2009) Quantitative proteomic analysis of a paired human liver healthyversus carcinoma cell lines with the same genetic background to identify potentialhepatocellular carcinoma markers. Proteomics Clin Appl 3(6):705–719.

16. Baker ES, et al. (2014) Advancing the high throughput identification of liver fibrosisprotein signatures using multiplexed ion mobility spectrometry. Mol Cell Proteomics13(4):1119–1127.

17. Sun W, et al. (2007) Proteome analysis of hepatocellular carcinoma by two-dimensional difference gel electrophoresis: Novel protein markers in hepatocellularcarcinoma tissues. Mol Cell Proteomics 6(10):1798–1808.

18. Xu B, et al. (2014) Large-scale proteome quantification of hepatocellular carcinomatissues by a three-dimensional liquid chromatography strategy integrated withsample preparation. J Proteome Res 13(8):3645–3654.

19. Luk JM, et al. (2006) Proteomic profiling of hepatocellular carcinoma in Chinese co-hort reveals heat-shock proteins (Hsp27, Hsp70, GRP78) up-regulation and their as-sociated prognostic values. Proteomics 6(3):1049–1057.

20. Long J, et al. (2010) Glutamine synthetase as an early marker for hepatocellularcarcinoma based on proteomic analysis of resected small hepatocellular carcinomas.Hepatobiliary Pancreat Dis Int 9(3):296–305.

21. Liu Y, et al. (2014) The role of von Willebrand factor as a biomarker of tumor de-velopment in hepatitis B virus-associated human hepatocellular carcinoma: A quan-titative proteomic based study. J Proteomics 106:99–112.

22. Megger DA, et al. (2013) Proteomic differences between hepatocellular carcinomaand nontumorous liver tissue investigated by a combined gel-based and label-freequantitative proteomics study. Mol Cell Proteomics 12(7):2006–2020.

23. Guo T, et al. (2015) Rapid mass spectrometric conversion of tissue biopsy samples intopermanent quantitative digital proteome maps. Nat Med 21(4):407–413.

24. Monetti M, Nagaraj N, Sharma K, Mann M (2011) Large-scale phosphosite quantifi-cation in tissues by a spike-in SILAC method. Nat Methods 8(8):655–658.

25. Zanivan S, et al. (2013) In vivo SILAC-based proteomics reveals phosphoproteomechanges during mouse skin carcinogenesis. Cell Reports 3(2):552–566.

26. Schweppe DK, Rigas JR, Gerber SA (2013) Quantitative phosphoproteomic profiling ofhuman non-small cell lung cancer tumors. J Proteomics 91:286–296.

27. Song C, et al. (2011) Improvement of the quantification accuracy and throughput forphosphoproteome analysis by a pseudo triplex stable isotope dimethyl labeling ap-proach. Anal Chem 83(20):7755–7762.

28. Narumi R, et al. (2012) A strategy for large-scale phosphoproteomics and SRM-basedvalidation of human breast cancer tissue samples. J Proteome Res 11(11):5311–5322.

29. Zanivan S, et al. (2008) Solid tumor proteome and phosphoproteome analysis by highresolution mass spectrometry. J Proteome Res 7(12):5314–5326.

30. Harsha HC, Pandey A (2010) Phosphoproteomics in cancer. Mol Oncol 4(6):482–495.31. Geiger T, Cox J, Ostasiewicz P, Wisniewski JR, Mann M (2010) Super-SILAC mix for

quantitative proteomics of human tumor tissue. Nat Methods 7(5):383–385.32. Kettenbach AN, Gerber SA (2011) Rapid and reproducible single-stage phosphopep-

tide enrichment of complex peptide mixtures: Application to general and phospho-tyrosine-specific phosphoproteomics experiments. Anal Chem 83(20):7635–7644.

33. Di Tommaso L, et al. (2009) The application of markers (HSP70 GPC3 and GS) in liverbiopsies is useful for detection of hepatocellular carcinoma. J Hepatol 50(4):746–754.

34. Matallanas D, et al. (2011) Raf family kinases: Old dogs have learned new tricks. GenesCancer 2(3):232–260.

35. McCubrey JA, et al. (2007) Roles of the Raf/MEK/ERK pathway in cell growth, ma-lignant transformation and drug resistance. Biochim Biophys Acta 1773(8):1263–1284.

36. Mendoza MC, Er EE, Blenis J (2011) The Ras-ERK and PI3K-mTOR pathways: Cross-talkand compensation. Trends Biochem Sci 36(6):320–328.

37. Hauge C, Frodin M (2006) RSK and MSK in MAP kinase signaling. J Cell Sci 119(Pt 15):3021–3023.

38. Fiandalo MV, Wilton J, Mohler JL (2014) Roles for the backdoor pathway of androgenmetabolism in prostate cancer response to castration and drug treatment. Int J BiolSci 10(6):596–601.

39. Dozmorov MG, et al. (2010) Elevated AKR1C3 expression promotes prostate cancercell survival and prostate cell-mediated endothelial cell tube formation: Implicationsfor prostate cancer progression. BMC Cancer 10:672–688.

40. Ishizaki F, et al. (2013) Androgen deprivation promotes intratumoral synthesisof dihydrotestosterone from androgen metabolites in prostate cancer. Sci Rep3:1528–1537.

41. Lévesque É, et al. (2013) Molecular markers in key steroidogenic pathways, circulatingsteroid levels, and prostate cancer progression. Clin Cancer Res 19(3):699–709.

42. Oda S, Fukami T, Yokoi T, Nakajima M (2015) A comprehensive review of UDP-glucuronosyltransferase and esterases for drug development. Drug MetabPharmacokinet 30(1):30–51.

43. Yeh SH, Chen PJ (2010) Gender disparity of hepatocellular carcinoma: The roles of sexhormones. Oncology 78(Suppl 1):172–179.

44. Shen YC, et al. (2013) Activating oxidative phosphorylation by a pyruvate de-hydrogenase kinase inhibitor overcomes sorafenib resistance of hepatocellular car-cinoma. Br J Cancer 108(1):72–81.

45. Lamouille S, Xu J, Derynck R (2014) Molecular mechanisms of epithelial-mesenchymaltransition. Nat Rev Mol Cell Biol 15(3):178–196.

46. Zeisberg M, Neilson EG (2009) Biomarkers for epithelial-mesenchymal transitions.J Clin Invest 119(6):1429–1437.

47. Tsai JH, Yang J (2013) Epithelial-mesenchymal plasticity in carcinoma metastasis.Genes Dev 27(20):2192–2206.

48. Miyamoto H, et al. (2004) Tumor-stroma interaction of human pancreatic cancer:Acquired resistance to anticancer drugs and proliferation regulation is dependent onextracellular matrix proteins. Pancreas 28(1):38–44.

49. Trédan O, Galmarini CM, Patel K, Tannock IF (2007) Drug resistance and the solidtumor microenvironment. J Natl Cancer Inst 99(19):1441–1454.

50. Dephoure N, Gygi SP (2011) A solid phase extraction-based platform for rapidphosphoproteomic analysis. Methods 54(4):379–386.

51. Cox J, Mann M (2008) MaxQuant enables high peptide identification rates, in-dividualized p.p.b.-range mass accuracies and proteome-wide protein quantification.Nat Biotechnol 26(12):1367–1372.

52. Cox J, Mann M (2012) 1D and 2D annotation enrichment: A statistical method in-tegrating quantitative proteomics with complementary high-throughput data. BMCBioinformatics 13(Suppl 16):S12–S23.

53. Zhang B, Kirov S, Snoddy J (2005) WebGestalt: An integrated system for exploringgene sets in various biological contexts. Nucleic Acids Res 33(Web Server issue):W741–748.

54. Huang W, Sherman BT, Lempicki RA (2009) Systematic and integrative analysis oflarge gene lists using DAVID bioinformatics resources. Nat Protoc 4(1):44–57.

1386 | www.pnas.org/cgi/doi/10.1073/pnas.1523434113 Dazert et al.

Dow

nloa

ded

by g

uest

on

June

16,

202

0