Cancer Cell, Volume 39

Supplemental information

Three subtypes of lung cancer fibroblasts define

distinct therapeutic paradigms

Haichuan Hu, Zofia Piotrowska, Patricia J. Hare, Huidong Chen, Hillary E. Mulvey, AislinnMayfield, Sundus Noeen, Krystina Kattermann,MaxGreenberg, AugustWilliams, AmandaK. Riley, Jarad J. Wilson, Ying-Qing Mao, Ruo-Pan Huang, Mandeep K. Banwait, JeffreyHo, Giovanna S. Crowther, Lida P. Hariri, Rebecca S. Heist, David P. Kodack, LucaPinello, Alice T. Shaw, Mari Mino-Kenudson, Aaron N. Hata, Lecia V. Sequist, Cyril H.Benes, Matthew J. Niederst, and Jeffrey A. Engelman

Figure S1. Establishment of patient-derived fibroblasts from tumor biopsies. Related to Figure 1. A. Biopsies from an EGFR+ lung cancer patient shows abundant tumor stroma is present at different timepoints and in multiple metastasis sites. Desmoplastic areas (black arrows) constitute the microenvironment surrounding cancer cells (white arrows) in the primary lung lesion and metastatic lesions. B. Schematic of the PDF library establishment, experimental use and data integration. C. The workflow of PDF model establishment process and validation.

A B

PDF library

In silico validation(TCGA, GTex…)

Biomarker validation Pharmaceutical validation

Tumor secretome In situ validation

Candidatebiomarkers

Resistance phenotypescreening

PDF secretome profiling

Pharmaceuticalcandidate(s)

Clinicaltrials

Freshbiopsy

Biopsy(FFPE)

Earlyprimaryculture

Super-natant

Drugtargeting in vitro

Drugtargeting in vivo

A 62 years-old male patient was diagnosed with stage IV EGFR mutant (ex19 del) NSCLC with multiple metastases.

Lung primary tumor

Epidural metastasis

Vertebral metastasis

Adrenalgland

Brain

AfatinibO

simertinib

Osi+Chem

o

Deceased

Biopsy early culture

Stromal overtakes the culture Passage and expand

(natural selection) Pure PDF, and immortalizationOther techniques:

- Differential trypsinization- Miltenyi MACS- Clone ring selection

Cancer cells grow well with minimum stroma

Both cancer and stromal cells grow

Feeder based culture is almost always used with other conditions only attempted for larger tissue samples

Multiple passages may be needed for fibroblasts to be isolated

Early culture Purify / separate fibroblast Finished PDF

Overall PDF take-rate is ~80% from core biopsies and autopsies

76.27% Feeder based

23.73% Non-feeder based

Early culture condition

0 5 10 150

10

20

Passages

Pro

porti

on

85% ≤ 5 passages

Passages before immortalization

C

77.97% Natural selection

22.03% Other techniques

Fibroblast separation

Figure S2. Validation of PDFs and their expression of CAF markers. Related to Figure 1. A. An example of EGFR genotyping analysis comparing PDF and the corresponding cancer cells. B. The mRNA level of 10 CAF marker genes and the epithelial marker gene KRT18 assessed by q-PCR in PDFs

0

4

8 ✱ ✱

024

ACTA2

COL1A2

DDR2 DE

SFAP

ITGA1

P4HA3

PDGFRA

S100A4 VI

M0

5

10 ✱ ✱ ✱

Laughney 2020

Kim 2020

Maynard 2020

Norm

. exp

ress

ion

EGFR CAFnon-EGFR CAF

30 60 900

4

8

**

30 60 900

4

8

***

30 60 90-2

2

6

ns

30 60 90-15

-5

5

ns

30 60 90-2

3

8

ns

30 60 900.0

12.5

25.0

ns

30 60 90-5

5

15

ns

30 60 90-2

2

6

ns

30 60 90-5.0

2.5

10.0

ns

30 60 900.0

7.5

15.0

ns

0

2

4

6

8ns ns

ns

0

2

4

6

8ns ns

ns

012345

ns nsns

ACTA2

DDR2

DES

FAP

ITGA1

PDGFRA

S100A4

VIM

P4HA3

AgeBiopsy

siteSmoking

statusLast

treatmentTumor

oncogene

-15

-10

-5

0

ns nsns

0

2

4

6

8ns ns*

0

10

20

30ns ns

ns

-2

0

2

4

6ns ns

ns

012345

ns * *

-6-4-20246 * ****

0

5

10

15ns ns

ns

0

2

4

6 ns

0

2

4

6*

012345 ns

-10-8-6-4-20

ns

0

2

4

6 ns

0

5

10

15

20 ns

-2

-1

0

1

2 ns

0

1

2

3

4 ns

012345 ns

0

5

10

15 ns

02468

10 ns

0

5

10

15

20 ns

0

5

10

15*

-15

-10

-5

0

ns

0

5

10

15 ns

0

5

10

15

20 *

-15

-10

-5

0

5 ns

0

5

10

15

20 ns

-40-30-20-10

010 ns

05

10152025 ns

0

2

4

6

8 ns

0

2

4

6

8 ns

012345 ns

-15

-10

-5

0

ns

02468

10 ns

05

10152025 ns

012345 ns

0

2

4

6 ns

012345 ns

0

5

10

15 ns

Year-old Lung

Others

Never

Smoker TKI

Chemo

EGFROthe

rs

COL1A2

RN

A ex

pres

sion

(L2F

C)

Liver

ACTA2

COL1A2

DDR2DES

FAP

ITGA1

P4HA3

PDGFRA

S100A

4VIM

KRT18-20-10

0102030

RN

A (L

2FC

)

PDFs (n=26) Cancer (n=10)

Fibroblast markers

****

Epithelialmarker

**** **** **** **** ** **** **** ****ns ****Cancer cell(EGFR 19del)

PDF (WT)

Marker

Parental

hTERT

A B C

-8 0 8-8

0

8

r = -0.0856-8 0 8

-8

0

8 r = 0.5871

-8 0 8-8

0

8 r = 0.9284

Par

enta

l PD

F

Non-matchedhTERT PDFs

MatchedhTERT PDF

Normalized fibroblast marker genes’ expression

Correlation0 0.7 1

hTERT Immortalized PDFs (#1 – 12)

Pare

ntal

PDF

s (#

1 –

12)

Matched pairs (immortalized / parental PDFs)

#1 #2 #3 #4 #5

0123401234

#1 #2 #3 #4 #5

0123401234

Expressionscore

Biop

sy

#1 #2 #3 #4 #5

20µm

20µm

PDF

Biop

syPD

F

ACTA2

S100A4

D E

F

abc de

ParallelPDFsMetastasis

row max row min

Relative expression

ACTA2b e a c d

COL1A2a e b d c

DDR2e b a d c

DESb e a d c

FAPe d b a c

ITGA1e a c d b

P4HA3c b a d e

PDGFRAe d b a c

S100A4a b e d c

VIMb a c e d

5 parallel PDFs and 25 other PDFs

H

G

(n=26) and cancer cell lines (n=10). Means with 95%CI are shown. Two-tailed t test is used. C. Example of the results of TRAPeze assay showing the telomerase activity in a hTERT transduced PDF and its parental PDF. D. Fibroblast marker genes (n=10) signature assessed by qRT-PCR in 12 parental PDFs and compared to their hTERT immortalized derivatives. E. Expressions of fibroblast markers ACTA2 (αSMA) and S100A4 (FSP1) in EGFR+ lung cancer sample and PDFs developed from the corresponding biopsies. In situ staining is performed by RNAscope, and the result is scored based on manufacture’s semi-quantitative criteria. F. Breakdown of PDFs’ expression of CAF markers (n=10) based on patients’ clinical features. Patients’ clinical features are shown in x-axis and RNA expression (qRT-PCR, L2FC) is shown in y-axis. Mean value with 95%CI is shown (two-tailed Spearman’s r for age correlation analysis and t-test for group comparison). G. Expression of CAF markers in lung CAF single cells based on the oncogene background of NSCLC. Mean value of CAFs in each tumor sample is first calculated, and bars show the average of all samples in each group. Two-sided t-test is used. H. Five liver metastasis samples obtained from a single patient were analyzed for the expression of CAF marker genes using qRT-PCR. * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001.

Figure S3: PDFs captured the overall molecular repertoire of NSCLC CAFs. Related to Figure 1. A. Uniform Manifold Approximation and Projection (UMAP) analysis of 1,465 individual fibroblasts in NSCLC resection samples from Lambrechts et al. (2018) reveals 8 molecular classes. The expression of representative marker genes is shown by each UMAP class. Color scale is based on the raw counts. B. The top 10 markers per UMAP class. UMAP-4 corresponds to poor quality single cells and is excluded from further analysis. C. PDFs mapped to each UMAP class consistently expressed the corresponding marker genes defined by CAF single cells. D. Unsupervised clustering shows the signature proximity between PDF molecular classes (right) and that between CAF single cells (left). Each row is the average of top marker genes per UMAP class, each column is a unique PDF or CAF single cell.

B

D

C1 2 3 4* 5 6 7 8

IGFBP7 EEF1A1 CTHRC1 TMSB4X C3 GPX3 MALAT1 HLA-DRANDUFA4L2 FTH1 COL3A1 RPLP1 CCDC80 MFAP4 CCL5 TYROBPRGS5 RPL17 VCAN RPS14 SERPINF1 A2M CXCR4 HLA-DRB1COX4I2 GABARAP COL1A1 RPS12 DCN CFD BTG1 FCER1GPPP1R14A GAPDH BGN SERF2 JUN RGCC TMSB4X HLA-DPA1CALD1 NBEAL1 COL1A2 LGALS1 NNMT TIMP3 B2M CD74PDGFRB MIF LUM RPS18 PTGDS SEPP1 CD52 LYZHIGD1B RPL3 POSTN RPL34 C7 CYR61 ARHGDIB SRGNCOL4A1 LDHA AEBP1 RPLP2 RARRES1 ZFP36 CCL4 LAPTM5TINAGL1 HNRNPA1 SULF1 RPS19 IGF1 PTGDS HLA-E HLA-DPB1LHFP RPS2 COL5A2 TMSB10 C1S DUSP1 PTPRC AIF1* UMAP-4 is corresponding to poor quality single cells, and is excluded from further analysis

1 2 3 5 6 7 8 1 2 3 5 6 7 8IGFBP7

NDUFA4L2EEF1A1RPL17

CTHRC1COL3A1

C3SERPINF1

GPX3MFAP4MALAT1CCL5

HLA-DRAHLA-DRB1

UMAP subgroup

Single cells(positive cell %)

PDFs(L2FC)

row min row max

Markers of1235678

row min row max

PDFsCAF single cells

UMAP-1UMAP-2UMAP-3UMAP-5UMAP-6UMAP-7UMAP-8

UMAP-1UMAP-2UMAP-3UMAP-5UMAP-6UMAP-7UMAP-8

Expressionrow min row max

(L2FC)row min row max

A

Figure S4: HGF partially defines PDFs functional heterogeneity in conferring EGFRi resistance. Related to Figure 3. A. Comprehensive secretome profiling was carried out in 12 PDFs, with the heatmap showing the profile of secretion of 272 proteins detectable in at least one PDF. B. The list of secreted proteins strongly correlated with PDF rescue. C. The heterogeneous HGF mRNA expression in PDFs. Cancer cells’ expression is shown for comparison. D. RNAscope staining of HGF in stroma cells in EGFR+ lung cancer (n=11). HGF expression is measured by staining quantification (dots per 100 stromal cells, see Methods). E. Signaling changes induced by recombinant HGF (0, 0.625, 2.5, 10, and 40ng/mL) added to EGFR-driven NSCLC models in the absence of presence of the MET inhibitor INC280 (200nM). MET phosphorylation and downstream pathway activation measured by western blotting with the indicated antibodies. F. Schematics illustrating the bypass signaling mediated by alternative receptor kinase activation such as HGF-MET. G. Comparing PDFs with higher than medium HGF secretion with PDFs with lower than

A2M

ACE

Activin A

Activin RIB

ADAM23

ADAM9

Adiponectin

Adipsin

aFGF

AFP

Aggrecan

ALCAM

AMIGO

ANG-1

Angiogenin

ANGPTL3

ANGPTL4

ApoA1

ApoC1

ApoC2

ApoC3

ApoE

ApoH

APRIL

AR Artemin

b-NGF

B7-H1

B7-H2

B7-H3

BCAM

BDNF

bFGF

bIG-H3

BLAME

BMP-4

BMP-5

BMP-7

BMPR-II

Brevican

C5a

CA125

CA9

Cadherin-11

Cadherin-4

Carbonic Anhydrase XII

Cardiotrophin-1

Cathepsin B

Cathepsin L

Cathepsin S

CD155

CD99

CEA

CEACAM-5

Chemerin

CHI3L1

Common beta Chain

Contactin-1

Contactin-2

CRIM1

CRP

CRTAC1

CTLA4

DcR3

Desmoglein-1

Desmoglein-3

DKK-1

DNAM-1

DPPII

DPPIV

ECM-1

EG-VEGF

EGF

EMMPRIN

EpCAM

EphA1

EphB6

Ephrin-B3

ErbB2

ErbB4

ESAM

FABP2

FAP

Fas

FCAR

Ferritin

FGF-12

FGF-17

FGF-20

FGF-21

FGF-4

FGF-5

FGF-7

FGF-9

FLRG

Flt-3L

Follistatin

Follistatin-like 1

FOLR2

FSH

Galectin-1

Galectin-3

Galectin-7

Galectin-8

Gas 1

Gas6

GDF-15

GDF-8

GDNF

GH GHR

Glypican 1

Glypican 5

gp130

GROa

HB-EGF

hCGb

HGF

HGF R

HVEM

ICAM-2

IgA

IgD

IGF-1

IGF-1R

IGF-2

IGFBP-1

IGFBP-2

IGFBP-3

IGFBP-4

IGFBP-5

IGFBP-6

IgG1

IgG2

IgG3

IgG4

IL-2

IL-5

IL-6

IL-7

IL-8

Insulin

Insulin R

Integrin alpha 5

JAM-A

Kallikrein 7

Kirrel3

LAMP1

LAP(TGFb1)

Legumain

Leptin

Leptin R

LIF R alpha

LIMPII

Lipocalin-1

Lipocalin-2

LOX-1

LRIG3

LRP-6

LTbR

Lumican

MCSF R

MDM2

Mesothelin

MFRP

MICA

Midkine

MMP-1

MMP-10

MMP-13

MMP-2

MMP-3

MMP-7

MMP-8

MMP-9

NCAM-1

Nectin-1

Nectin-3

Nectin-4

Neuropilin-2

Neurturin

NGF R

Nidogen-1

Nidogen-2

Nogo Receptor

Notch-3

NOV

NRG1-b1

NSE

NT-3

NT-4

Olfactomedin-2

OPG

OSM

Osteoactivin

OX40 Ligand

P-Cadherin

p53

PAI-1

PD-ECGF

PDGF-AA

PDGF-AB

PDGF-CC

Pentraxin 3

Periostin

Persephin

PIGF

Procalcitonin

Progranulin

Prolactin R

RELT

Renin

Ret

RGM-B

ROBO3

ROBO4

S100A8

SAA

SCF

SCF R

Semaphorin 6B

Serpin F1

sFRP-3

Shh-N

Siglec-10

Siglec-7

Siglec-9

SREC-I

SREC-II

Syndecan-3

Syndecan-4

TACE

TF TfR

TGFa

TGFb RIII

TGFb1

TGFb2

TGFb3

Thrombomodulin

Thrombospondin-2

Thrombospondin-5

Thyroglobulin

Tie-1

Tie-2

TIMP-4

TLR1

TLR3

TPO

TPP1

TRAIL

TRAIL R4

Transferrin

TREM-2

TrkC

TROY

TSH

TWEAK

ULBP-1

ULBP-2

uPAR

Uromodulin

VCAM-1

VEGF

VEGF R1

VEGF R2

VEGF R3

VEGF-C

VEGF-D

Vitronectin

vWF

XIAP

PDFs

Secreted proteins

Relative expression

-4 0 4

-16-808

HG

F R

NA

(L2F

C)

Cell culturesPDFs Cancer

Sample ranks

0 40 0 40 0 40 0 40

0

200

0

200

MG

H134

MG

H707

HGF (ng/mL)

INC

280

(nM

)

p-MET p-ERK1/2 p-S6 Β-Actin

Top 30 PDF rescue correlates (ranked by correlation)

A B

C D

0

50

100

MGH154

MGH708

MGH134

H4006

MGH707

HCC827

MGH164

MGH805

MGH119PC9

MGH121

H1975R

esis

tanc

e (%

)

1.Fas2.HGF3.SREC-II4.Follistatin5.FGF76.IGFBP-27.PTX38.Angiogenin9.TGFa10.Nidogen-2

11.GROa12.AR13.DKK-114.BDNF15.GDF-1516.SCF17.TrkC18.b-NGF19.NOV20.EG-VEGF

21.ANG-122.vWF23.NT-324.VEGF-D25.Syndecan-426.Neurturin27.VEGF R328.Periostin29.aFGF30.LAP

HGF

downstream (MAPK, PI3K-AKT)

Mutant EGFRMET

EGFR TKI

20µmHGF KRT18

E F

HGFhigh

HGFlow

0

35

70 ****

Ave

. res

ista

nce

(%)

0

100

200

Patients

HGF in stromal cells

Nor

m. e

xpre

ssio

n

G H

Biopsy tissue

EGFRiEGFRi + METi

medium HGF secretion, assessed by ELISA, among 38 EGFR+ PDFs in rescuing cancer cells from EGFRi effect. H. The effect of METi in countering PDF-driven resistance to EGFRi in EGFR+ cancer models. Each bar is the average effect of 38 EGFR+ PDFs, with error bar showing 95%CI.

Figure S5: PDFs-driven activation of MET and FGFR pathways in cancer cells conferred resistance against EGFRi. Related to Figure 3 and Figure 4.

EGFR i + + +MET i +FGFR i +

C

G H

EGFRmut

cancer

PDFHGF-independentresistance

Restoresensitivity

Compound X+EGFR TKI MET TKI++ Pathway

screen No PDF PDF190 PDF731

MG

H134

MG

H707

Viab

le c

ells

to c

ontro

l (%

)

BGJ398 (nM)

BGJ398 alone BGJ398 with AZD9291+INC280

BL 10 100100010000

0

50

100

BL 10 100100010000

0

50

100

BL 10 100100010000

0

50

100

BL 10 100100010000

0

50

100

BL 10 100100010000

0

50

100

BL 10 100100010000

0

50

100

0 10 20 30 40 500

2×1054×1056×1058×1051×106

Days post injection

Clu

c R

eado

ut Cancer+No fibroblastCancer+10M fibroblastsCancer+5M fibroblasts

F

D

PDF800-2 PDF190-1 PDF808-1 PDF719-10

40

80

PDF model

For the same cancer model (MGH707)

MGH121 MGH119 MGH707 MGH1340

50

100

150

Cancer model

For the same PDF model (PDF808)Mainly

HGF-METMainly

FGF-FGFRHGF-METFGF-FGFR

Norescue

Mainly HGF-MET

MainlyFGF-FGFR

HGF-METFGF-FGFR

Norescue

E

No drug

EGFRi

EGFRi

+ METi

Ctrl HGF PDF190

Hoechst

PDF731A B

CTRLHGF

PDF190

PDF731

0

50

100

MGH134

CTRLHGF

PDF190

PDF731

0

15

304080

MGH707

EGFRiEGFRi + METi

Res

ista

nce

(%)

Res

ista

nce

(%)

0 50 1000

50

100rs = 0.9759p < 0.0001

Diagonal

Overall resistance %(rescue to EGFRi)

Addi

tive

effe

ct %

(MET

+ F

GFR

dep

ende

nt) CCD19-Lu

EGFR i + + + +MET i + +FGFR i + +

+ + + ++ +

+ +

Res

ista

nce

(%)

A-B. Representative images (A) and quantification (B) of HGFhigh PDF (PDF190) and HGFlow PDF (PDF731) rescuing cancer cell lines MGH134 (A and B) and MGH707 (B) from EGFRi alone or EGFRi in combination with METi. Values are the averages of 4 replicates with error bars representing 95%CI. C. Efficacy of indicated compounds to negate HGF-independent resistance across 2 PDFs and 2 cancer cells. The relative efficacy (IC50 shift) of each compound is measured with the compound’s IC50 in the presence of EGFR and MET dual inhibition subtracted from the IC50 when the testing compound is used alone. D. Dose response curves for the effect of BGJ398 alone or in the presence of EGFRi + METi on cancer cells’ viability. E. Examples showing different modalities of PDF-mediated rescuing activities of different PDFs on the same cancer cell line (left) and the same PDF on different cancer cell lines (right). F. The PDFs' overall rescue activity against EGFRi compared with the additive effects of FGFR-mediated rescue against EGFRi plus METi and MET-mediated rescues against EGFRi plus FGFRi. Each dot represents a PDF’s effect. G. Lack of persistence of human fibroblast in nude mice. Fibroblasts (CCD19-Lu) were engineered to secrete luciferase (Cypridina Luciferase, Cluc) and co-injected with MGH707 cancer cells subcutaneously in nude mice. Seven days after injection, tail vein blood was serially collected to measure the Cluc level. H. Conditioned media from CCD19-Lu prominently rescues 12 EGFR-driven cancer cell lines, including MGH707, through activating both MET and FGFR in vitro. Mean value and 95%CI are shown.

Figure S6: The specific FGF7-FGFR2 signaling axis between CAFs and cancer cells is a prevalent mechanism mediating EGFRi resistance in NSCLC. Related to Figure 5. A. Each PDF’s average resistance effect to ALKi (lorlatinib) assessed in 4 ALK+ cancer models is plotted against their resistance effect to EGFRi (osimertinib) assessed in 12 EGFR+ cancer models. Each dot

FGF1FGF2FGF5FGF7FGF9

-0.3

0.0

0.3

A

0

40

80

Ave

FGFR effectMET effect

-40

0

40

60 fibroblasts

MET effect- FGFR effect

Subtype I Subtype II Subtype III

MET + FGFR effect (%)

MET − FGFR effect (%)

C

FGF1

FGF2

FGF3

FGF4

FGF5

FGF6

FGF7

FGF8

FGF9

FGF10

FGF16

FGF17

FGF18

FGF19

FGF20

FGF21

FGF22

FGF23

HGF

0.00.40.8 **

******

ns ns * ns ****

ns ns ****

** ns * ns * ns ns ****

ns

Enrichment with tumor stroma in TCGA-LUADD

Relative expression (PDF over cancer, L2CF)

FGF1

FGF2

FGF3

FGF4

FGF5

FGF6

FGF7

FGF8

FGF9

FGF10

FGF16

FGF17

FGF18

FGF19

FGF20

FGF21

FGF22

FGF23-15

01530

***

****

ns ns ****

****

****

* ns ****

* * * ****

ns ns ***

*

E

H

p-EGFRp-ERK1/2p-S6

β-Actin

shorter exposure

p-S6 longer exposure

No drug

MG

H134

MG

H154

MG

H707

MG

H708

H400

6

AZD9291

MG

H134

MG

H154

MG

H707

MG

H708

H400

6

AZD9291 + FGF7

MG

H134

MG

H154

MG

H707

MG

H708

H400

6

I

KRT18FG

F7

KRT18HG

F

J

40

70

10

1 2 3 4 5 6 7 8 9 10110

10

20FGFR2+/FGF7- cellFGFR2+/FGF7+ cellFGFR2+ cells

FGF7+ / FGFR2- cellsP

erce

ntag

e of

tota

l cel

ls (%

)

FGFR2 FGF7

FGF7 + + + + + + + + + + —

FGFR2 — + — + + — — — + — —

Sample

K

L

0

4

8 nsACTA2

-2

3

8PDGFRA

**

-15

-5

5DES

**

-5

10

25ITGA1

**RN

A (L

2FC

)

Cor

rela

tion

(r s)

H3122

MGH00

6-1

MGH02

1-5

MGH04

8-1

0

25

50

0

40

80ALKi + HGF ALKi + FGF7

Res

ista

nce

(%)

H3122

MGH00

6-1

MGH02

1-5

MGH04

8-1

-4 -2 0 2 40

100

200

300

-log 10(

Padj

)

Fold of change (log2)

FGFR2

FGFR

2 fo

ld c

hang

e

Canceralone

Co-culturedwith PDF

No treatmentEGFR TKI

PDFs

0 50 100 0 50 10

0

ACAT2 PDGFRA DES ITGA1

UM

AP c

lass

Positive fibroblast single cells (%)

8765321

0 50 100 0 50 10

0

NSCLC CAF single cells

M N

Subtype I Subtype II Subtype III

<60

≥60

0

50

100 nsAge

Prop

ortio

n

Lung

Oth

ers

0

50

100 nsBiopsy site

EGFR

Others

0

50

100 nsOncogene

Subtype I Subtype II Subtype III

O

F G

0 100 200 3000

20

40

rs = 0.5235p = 0.0006CTRLAv

e. re

sist

ance

(%)

FGF7 (pg/mL)

PDFs p-value Overall <0.0001

EGFR PDFs 0.0072

ALK PDFs 0.0039

0 40 800

40

80EG

FRir

esis

tanc

e (%

)

ALKi resistance (%)

B

Cancer co-cultured with paired PDF

Cancer alone

No treatment

EGFR TKINo treatment

EGFR TKI

Cancer

PairedPDF

Biopsy

0

2

4

6*** ** ***

HGF+HGF- / FGF7+HGF- / FGF-

Subtype I: rescuing via MET +/- FGFRSubtype II: rescuing mainly via FGFRSubtype III: minimally rescuing

Perc

enta

ge

Lambrechts Kim Laughney Maynard Travaglini

PDFs Lung CAF / fibroblast scRNAseq

NSCLCEGFR

non-

EGFR

norm

al lun

g

EGFR

non-

EGFR

norm

al lun

g

EGFR

non-

EGFR

norm

al lun

g0

50

100

PDF

0

50

100

Cor

rela

tion

(r s)

represents a PDF model, with 13 PDFs from ALK+ NSCLC and 6 PDFs from EGFR+ NSCLC included. B. Breakdown of the PDF functional subtypes based on patients’ age, tumors’ site-of-biopsy, and oncogene status. Two-tailed Fisher’s exact test was used to compare the proportions of subtype I/II vs. subtype III PDFs depending on patient clinical features. C. PDFs are characterized based on their rescue via MET effect (rescue against EGFRi plus FGFRi) and FGFR effect (rescue against EGFRi plus METi) (top graph, MET effect plus FGFR effect are shown as stacked bars to estimate their rescue potential with both pathways) and their preference in using either of these pathways (bottom graph, MET effect minus FGFR effect, positive value indicates MET effect preference, negative value indicates FGFR effect preference). 60 PDFs (38 EGFR+ and 22 non-EGFR+ PDFs) were categorized into three distinctive functional subtypes (red, subtype I PDFs with significant MET-mediated rescue; green, subtype II PDFs with modest rescue through FGFR; blue, subtype III PDFs with minimum rescue effect). D. Correlation between expression levels of FGFs and HGF in lung adenocarcinoma with the stroma score in each tumor (relative quantification of stroma component abundance, previously published by Yoshihara et al., 2013) in the TCGA-LUAD dataset. Two-tailed spearman’s r is used. E. Relative expression of FGF family members in PDFs (n=26) versus cancer cells (n=10) determined by qRT-PCR. The PDFs’ FGF7 level over cancer cells are shown as -𝛥𝛥CT. Two-tailed t-test is used. F. Correlation between FGF ligands secretion in PDFs (n=12) and PDFs’ FGFR-dependent rescue against EGFRi plus METi. G. The correlation between FGF7 concentration assessed by ELISA and PDFs’ FGFR-dependent rescue against EGFRi plus METi. H. Western blotting analysis of EGFR, ERK and S6 phosphorylation status in the indicated conditions. FGF7 was used at 10ng/mL. I. The rescue effect of HGF and FGF7 on 4 ALK cancer models upon ALK inhibition (lorlatinib at 300nM). Mean value is shown with error bar showing 95% CI. J. Representative images of FGF7 and HGF detected by RNAscope in the stroma of xenograft tumor from co-injecting MGH707 cancer cells with CCD19-Lu human fibroblasts. K. Representative images of RNAscope staining of FGF7 and FGFR2 in EGFR+ lung cancer biopsies (n=11) A sample with more than 10% of cells showing FGF7 expression (without detectable FGFR2) was considered FGF7 positive and a sample with more than 10% of cells with detectable FGFR2 expression (regardless of FGF7 staining) was considered FGFR2 positive. L. In a cancer-PDF pair (MGH805-1 and PDF805-1) derived from the same patient biopsy (schematics on the left), only treating cancer cells with EGFRi in the presence of PDF leads to increased FGFR2 expression in cancer cells (RNAseq volcano plot and RNA expression on the right). M. The proportion of functionally classified PDFs (left, according to Figure 5A and Figure S6C) and the proportion of tumor and normal lung fibroblast single cells with high HGF, low HGF/high FGF7, and low HGF/low FGF7 levels (Right) in five previous tumor and normal lung scRNA-seq datasets (Lambrechts et al., 2018; Kim et al., 2020; Laughney et al., 2020; Maynard et al., 2020; and Travaglini et al., 2020). N. The mRNA level of four fibroblast markers ACTA2 (αSMA), PDGFRA, DES and ITGA1 across three subtypes of PDFs. Gene RNA expression is assessed by qRT-PCR (log2, normalized by house-keeping gene) (whiskers are maximum and minimum values). Two-tailed t test is used to compare the indicated single group to all other PDFs. O. The proportion of fibroblast single cells in the NSCLC scRNA-seq dataset from (Lambrechts et al., 2018) positive for key CAF markers per UMAP class. * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001.

Figure S7: Mechanistic insight of the three subtypes of PDFs. Related to Figure 6, Figure 7 and Figure 8. A. The protein level of TGF-β1 and three cytokines among three subtypes of PDFs (n=12) assessed by using a 448-multiplex ELISA array. Means with 95%CI are shown. Two-tailed Mann-Whitney U test is used to compare the indicated single group compared to all other PDFs. B. Western blotting showing phospho-SMAD2, αSMA, and phospho-STAT3 in three subtypes of PDFs at the baseline and after exposure to exogeneous TGF-β1 (10ng/mL for 24 hours). Lysates from this panel were also probed in Figure 6E. C.

#1 #2 #3 #4-2

0

2 ✱✱✱✱ ✱✱✱✱ ✱✱✱ ns

#1 #2 #3 #4-2

0

2 ✱ ✱ ✱ ns

#1 #2 #3 #4-2

0

2 ✱ ✱ nsns

#1 #2 #3 #4-2

0

2 ✱✱ ns ✱✱✱✱ns

#1 #2 #3 #4-2

0

2 ✱ns ✱✱✱ns

#1 #2 #3 #4-2

0

2 ns nsns ns

0

400

800 ns

0.0

0.3

0.6***

0

4

8 *

0

125

250 *

A BSubtype I: rescuing via MET +/- FGFR

Subtype II: rescuing mainly via FGFRSubtype III: minimally rescuing

p-SMAD2

α-SMA

p-STAT3

β-Actin

Subtype I Subtype II Subtype III Subtype I Subtype II Subtype III

Baseline TGF-β1 treated

TGFβ-1 sFas PTX3 THBS2

Std.

exp

ress

ion

C

CTRL

TGF-β1

0

10

20✱

CTRL

TGF-β1

0

30

60✱

CTRL

TGF-β1

0

30

60✱

PC9 MGH134 MGH707

Res

ista

nce

(%)

HGF RNA

FGF7 RNA

p-SMAD2

HGF High

FGF7 High/Low

p-SMAD2 Low

HGF Low

FGF7 High

p-SMAD2 Low

HGF Low

FGF7 Low

p-SMAD2 High

Functionalsubtype

Marker-basedsubtype I PDFs

Marker-basedsubtype II PDFs

Marker-basedsubtype III PDFs

Expression

Row max Row min

I II III0

5

10

Cas

e

Marker-basedPDF subtypes

p = 0.0060D

100μm Jurket cell PDF

No PDF Non-subtype III Subtype III

100μm THP-1 cell PDF

No PDF Non-subtype III Subtype IIIK L

siNEG

siETV1

-10

-5

0

5✱✱

siNEG

siETV1

-6

0

6✱

siNEG

siETV1

0

3

6

9ns

vehic

le

pLen

ti ETV1

-10

-5

0

5✱

vehic

le

pLen

ti ETV1

-6

0

6ns

vehic

le

pLen

ti ETV1

0

3

6

9ns

RN

A (L

2FC

)

RN

A (L

2FC

)

E G

Fold

cha

nge

(log2

) ETV1 HGF FGF7 TBX2 HGF FGF7

Additional ETV1 siRNAs

F H

Fold

cha

nge

(log2

)

Additional TBX2 siRNAs

ETV1 HGF FGF7 ETV1 HGF FGF7 I

Res

ista

nce

(%)

PDF190 PDF062-4c

Initial biopsy(pre-) RNAseq

Resistance (post-)RNAseq

Osimertinib(EGFRi)

Fold

cha

nge

HGF

0

25

50

p = 0.0535 ns

0

5

20 ✱ nsPFS: ≧12m <12m ≧12m <12m

Pre

Post

Pre

Post

Pre

Post

Pre

Post

FGF7

J

ns

N E TE+T

20

40

60

✱✱✱✱✱✱

N E TE+T

20

40

60 ns ✱✱

siRNAs

Subtyp

e I / I

I

Subtyp

e III

0

1

2

3

Mig

ratio

n sc

ore

Jurkat✱

Subtyp

e I / I

I

Subtyp

e III

0

1

2

3

4

Mig

ratio

n sc

ore

THP-1p = 0.0646

Three subtype I PDFs with and without ectopically activating their intrinsic TGF-β signaling are assessed for their resistance effect on three EGFR+ NSCLC cancer models. PDFs are pre-incubated with TGF-β1 (20ng/mL) for 72 hours, PDF media was then changed and was banked for five days for cancer cell resistance assessment. D. 21 PDFs are classified based on proposed markers (HGF, FGF7, and phospho-SMAD2, see also Figure 6I) and are compared to their functional classification (see also Figure 5A). HGF and FGF7 RNA are assessed by qRT-PCR, and phospho-SMAD2 is assessed by western blotting and quantified by using ImageJ. Fisher’s exact test is used. E-F. Gene expression change upon knocking down ETV1 in subtypes I (red) and II (green) PDFs. Both an siRNA pool (E) and a set of individual siRNAs (F) against ETV1 were used for validation. G. Gene expression change upon ectopically expressing ETV1 in subtype III (blue) PDFs. ETV1, HGF, and FGF7 RNA expression is assessed by qRT-PCR. H. Gene expression change upon knocking down TBX2 in subtypes I (red) and II (green) PDFs. A set of individual siRNAs against TBX2 were used for validation, in addition to an siRNA pool used in Figure 6M. (C, E-H), Mean and 95%CI are shown. Paired one-tail t-test is used. I. Subtype I PDFs treated with scramble siRNA (N), and siRNAs against ETV1 (E) and TBX2 (T) are assessed for their resistance effect on MGH707 (EGFR+ NSCLC cancer models). PDFs are pre-treated with an siRNA pool for 24 hours, then PDF media was changed and banked for five days for cancer cell resistance assessment. One-tail t-test is used with bars showing mean and 95%CI. J. RNAseq data of matched before (pre-) and after (post-) osimertinib treatment biopsies from 11 EGFR+ NSCLC patients (from Roper et al., 2020). The dynamic change of HGF and FGF7 RNA are shown according to patients’ progression-free survival (PFS). One-tailed t-test was used. K-L. Example images and summary of non-subtype III PDFs (subtypes I and II, n=4, example of a subtype I PDF is shown) and subtype III PDFs (n=4) in chemoattracting Jurkat cells (K) or THP-1 cells (L). One of the representative interface areas is shown, migration score is assessed across all interfaces in the chip and averaged by three replicates. Each bar is the average level with error bar showing 95%CI, one-tailed t-test is used. * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001.