Science

Where drugs act.

Many targeted therapies act on proteins. We are building models of how protein phosphorylation, abundance and glycosylation answer each drug, dose by dose, starting with pancreatic cancer.

The signalling layer

Measure signalling directly.

Phosphorylation can change rapidly after drug exposure. We are testing whether protein and PTM measurements improve drug-response decisions beyond RNA alone.

Response order and speed depend on the drug, the system and the measurement.

Two modifications, one sample

Phosphorylation and N-glycosylation, read together.

The signalling switch. A phosphate added to or removed from a serine, threonine or tyrosine turns enzymes and pathways on and off. One protein can gain and lose sites at once.

The cell-surface layer. Sugar chains built on asparagine shape how receptors fold, cluster and signal. A perturbation can change which glycan occupies a site, not only whether one is there.

Example data

One drug, eight doses.

An example of the dose-resolved data the model learns from. Tunicamycin blocks N-glycosylation and triggers ER stress. Glycosites lose signal above about 0.1 µM, and phosphorylation shifts from 0.5–1 µM as the ER-stress transcription factors XBP1s and ATF4 rise.

Phosphorylation

SEMA4B S803SEMA4B S803

NCL T76NCL T76

NPM1 S254NPM1 S254

NACC1 S188NACC1 S188

CEBPD S191CEBPD S191

MKI67 S2299 Ki-67MKI67 S2299

SPP1 S62 · S63SPP1 S62+

CLCC1 S509CLCC1 S509

VAPB T150VAPB T150

ADD1 T724ADD1 T724

MAP1B T744MAP1B T744

UHRF1 S639UHRF1 S639

ARHGEF5 S474ARHGEF5 S474

TMPO S424TMPO S424

SP3 S646SP3 S646

MTFR2 S328MTFR2 S328

ARHGAP9 S113ARHGAP9 S113

XBP1 S288 XBP1sXBP1 S288

TCEA1 S57TCEA1 S57

AARS1 S555AARS1 S555

TLN1 S1641TLN1 S1641

ATF4 S248ATF4 S248

CDCA2 S977CDCA2 S977

HAUS6 S906HAUS6 S906

decreaseincreasecolour saturates at ±3 log2

DMSO10 µM

XBP1 S288 XBP1s

DMSO10 µM

ATF4 S248

DMSO10 µM

MKI67 S2299 Ki-67

Each curve runs from DMSO to 10 µM on one scale (±4 log2).

Log2 change vs DMSO in the EGF-unstimulated arm, mean of two batches, with channels scaled to the global-proteome channel medians, as for the glycosites below. Confidently localised phosphopeptides (probability ≥ 0.75) with a consistent dose trend in each batch (Spearman |ρ| ≥ 0.85), no more than 0.6 log2 of variation up to 0.1 µM, and on average at least a two-fold change at 5–10 µM, one per protein: the 16 that fall most and the 8 that rise most. Signal is not adjusted for protein abundance, and ER stress raises XBP1s and ATF4 protein. Site numbers are from UniProt. The XBP1 peptide exists only in the stress-induced spliced form, XBP1s. A498 kidney cancer cells, 0.005–10 µM, 24 h. Re-analysed from the CC0 deposit PXD071440.

Show the numbers

Log2 change vs DMSO by tunicamycin dose (µM)
Phosphosite	0.005 µM	0.01 µM	0.05 µM	0.1 µM	0.5 µM	1 µM	5 µM	10 µM
SEMA4B S803	-0.31	-0.01	-0.52	-0.04	-1.07	-3.04	-3.52	-2.85
NCL T76	-0.32	-0.24	-0.46	-0.16	-1.12	-1.40	-2.44	-3.77
NPM1 S254	+0.18	-0.08	-0.18	+0.02	-1.01	-1.50	-2.44	-3.10
NACC1 S188	-0.03	-0.43	-0.41	-0.18	-1.28	-1.80	-2.41	-2.99
CEBPD S191	-0.02	-0.24	-0.16	-0.11	-1.56	-2.80	-2.89	-2.51
MKI67 S2299 (Ki-67)	-0.24	-0.04	-0.54	-0.14	-1.30	-1.56	-2.31	-2.98
SPP1 S62 · S63	-0.41	-0.38	-0.49	-0.53	-0.98	-2.48	-2.74	-2.52
CLCC1 S509	-0.40	-0.12	-0.45	-0.27	-1.04	-1.69	-2.44	-2.75
VAPB T150	-0.24	-0.14	-0.59	-0.14	-1.36	-1.85	-2.33	-2.64
ADD1 T724	-0.14	-0.25	-0.19	-0.24	-1.26	-1.49	-2.11	-2.83
MAP1B T744	-0.16	-0.21	-0.24	-0.12	-0.87	-1.28	-2.20	-2.67
UHRF1 S639	-0.04	-0.22	-0.24	-0.06	-1.38	-1.96	-2.35	-2.28
ARHGEF5 S474	-0.25	-0.17	-0.28	-0.30	-1.29	-1.86	-2.01	-2.50
TMPO S424	-0.05	-0.22	-0.30	-0.27	-0.81	-1.29	-2.01	-2.48
SP3 S646	+0.01	-0.03	-0.10	-0.06	-0.95	-1.39	-2.10	-2.34
MTFR2 S328	-0.02	-0.12	-0.20	-0.12	-1.19	-1.78	-2.23	-2.10
ARHGAP9 S113	+0.25	+0.32	+0.47	+0.44	+2.60	+3.44	+3.28	+2.76
XBP1 S288 (XBP1s)	+0.03	+0.10	+0.07	+0.21	+1.65	+2.45	+2.87	+3.04
TCEA1 S57	+0.01	-0.01	+0.12	+0.14	+1.66	+2.62	+2.95	+2.78
AARS1 S555	+0.08	+0.09	+0.17	+0.39	+1.26	+1.75	+1.94	+2.82
TLN1 S1641	+0.38	+0.06	0.00	+0.45	+1.41	+1.93	+2.10	+2.63
ATF4 S248	+0.21	+0.13	+0.42	+0.31	+0.85	+1.93	+2.20	+2.41
CDCA2 S977	+0.05	+0.05	+0.07	+0.21	+1.04	+1.85	+2.18	+2.18
HAUS6 S906	-0.14	-0.15	+0.15	+0.37	+0.57	+1.90	+2.27	+2.08

N-glycosylation

CD109 N1355 N4H6F0S1

CFI N464 N2H10F0S0

LAMA1 N2384 N2H10F0S0

PLOD2 N209 N2H9F0S0

MEGF8 N1115 N2H5F0S0

EFNA4 N33 N2H9F0S0

TLL2 N628 N2H8F1S0

PLXNB1 N1183 N2H9F0S0

LAMA5 N95 N2H10F0S0

CD63 N150 N5H6F1S3

LMF2 N616 N2H10F0S0

LRP1 N3788 N4H4F0S2

PLOD2 N365 N2H8F0S0

COLGALT1 N96 N4H4F1S1

LAMB1 N677 N2H7F0S0

CPD N1070 N2H8F0S0

IGFBP3 N116 N2H10F0S0

RNASET2 N212 N2H6F0S0

CD59 N43 N2H8F0S0

GRN N265 N2H10F0S0

BSG N268 N2H7F0S0

LGALS3BP N551 N4H5F2S2

GRN N236 N3H6F3S0

IGF2R N400 N3H6F0S1

decreaseincreasecolour saturates at ±3 log2

DMSO10 µM

GRN N236 N3H6F3S0

DMSO10 µM

LAMA5 N95 N2H10F0S0

DMSO10 µM

IGF2R N400 N3H6F0S1

Each curve: DMSO, then 0.005 → 10 µM.

Log2 change vs DMSO in the EGF-unstimulated arm, mean of two batches, with channels scaled to the global-proteome channel medians. The 24 glycosites with the strongest monotonic dose trend, one glycoform each. The three plotted as curves are in bold. A498 kidney cancer cells, 0.005–10 µM, 24 h. One site rises and is shown as measured. Re-analysed from the CC0 deposit PXD071440.

Show the numbers

Log2 change vs DMSO by tunicamycin dose (µM)
Site · glycoform	0.005 µM	0.01 µM	0.05 µM	0.1 µM	0.5 µM	1 µM	5 µM	10 µM
CD109 N1355 · N4H6F0S1	+0.22	+0.13	+0.05	+0.01	-0.21	-0.49	-0.64	-0.64
CFI N464 · N2H10F0S0	-0.01	+0.06	-0.06	-0.17	-0.60	-1.35	-2.11	-2.50
LAMA1 N2384 · N2H10F0S0	-0.04	+0.07	-0.17	-0.36	-1.25	-1.48	-2.24	-2.83
PLOD2 N209 · N2H9F0S0	+0.94	+0.91	+0.81	+0.79	+0.35	-1.63	-4.72	-4.95
MEGF8 N1115 · N2H5F0S0	+0.33	+0.29	+0.09	+0.28	-1.27	-1.66	-2.32	-2.34
EFNA4 N33 · N2H9F0S0	+0.07	-0.01	-0.16	-0.25	-1.09	-1.69	-2.23	-2.06
TLL2 N628 · N2H8F1S0	+0.11	+0.11	+0.09	+0.01	-1.12	-1.70	-2.18	-2.11
PLXNB1 N1183 · N2H9F0S0	-0.04	+0.15	-0.16	-0.19	-1.19	-1.75	-1.99	-2.00
LAMA5 N95 · N2H10F0S0	+0.27	+0.15	+0.08	+0.01	-1.17	-2.04	-2.45	-2.72
CD63 N150 · N5H6F1S3	-0.13	-0.15	-0.66	-0.24	-0.75	-2.05	-2.59	-2.69
LMF2 N616 · N2H10F0S0	+0.21	+0.13	+0.08	+0.10	-0.45	-2.09	-2.26	-2.71
LRP1 N3788 · N4H4F0S2	+0.06	-0.03	-0.34	-0.11	-0.82	-2.17	-3.11	-3.15
PLOD2 N365 · N2H8F0S0	+0.07	-0.05	-0.06	-0.08	-1.59	-2.27	-2.83	-2.66
COLGALT1 N96 · N4H4F1S1	-0.06	-0.06	-0.14	-0.18	-1.69	-2.39	-2.72	-2.81
LAMB1 N677 · N2H7F0S0	+0.04	-0.06	-0.09	-0.07	-0.82	-2.43	-3.02	-3.75
CPD N1070 · N2H8F0S0	+0.02	-0.09	-0.29	-0.10	-1.51	-2.60	-2.96	-3.00
IGFBP3 N116 · N2H10F0S0	+0.05	-0.02	-0.17	-0.18	-1.15	-2.61	-4.76	-4.38
RNASET2 N212 · N2H6F0S0	+0.10	+0.06	+0.03	+0.04	-1.17	-2.62	-2.75	-2.81
CD59 N43 · N2H8F0S0	+0.07	+0.01	-0.22	-0.06	-1.39	-2.73	-2.85	-2.98
GRN N265 · N2H10F0S0	+0.12	-0.02	-0.04	-0.08	-1.48	-2.90	-4.83	-5.05
BSG N268 · N2H7F0S0	+0.06	+0.03	-0.45	-0.09	-1.41	-3.24	-3.41	-3.82
LGALS3BP N551 · N4H5F2S2	+0.13	+0.08	+0.01	-0.10	-1.55	-3.28	-7.24	-3.83
GRN N236 · N3H6F3S0	+0.04	-0.04	-0.06	-0.09	-2.14	-4.36	-5.50	-6.12
IGF2R N400 · N3H6F0S1	+0.53	+0.62	+0.65	+0.67	+2.06	+3.08	+4.02	+3.99

Dose, not snapshots

A curve says more than a fold change.

A dose series shows how potently, how far and in what direction each site moves. Those are the quantities a model can learn, and be tested on.

Sites differ in potency and direction across one dose series.

Working on the protein layer?

We would like to hear from teams building models of signalling, and from programmes whose drugs act there.

Request access [email protected]