Secondary resistance to anti-EGFR therapy by transcriptional reprogramming in patient-derived colorectal cancer models

Background The development of secondary resistance (SR) in metastatic colorectal cancer (mCRC) treated with anti-epidermal growth factor receptor (anti-EGFR) antibodies is not fully understood at the molecular level. Here we tested in vivo selection of anti-EGFR SR tumors in CRC patient-derived xenograft (PDX) models as a strategy for a molecular dissection of SR mechanisms. Methods We analyzed 21 KRAS, NRAS, BRAF, and PI3K wildtype CRC patient-derived xenograft (PDX) models for their anti-EGFR sensitivity. Furthermore, 31 anti-EGFR SR tumors were generated via chronic in vivo treatment with cetuximab. A multi-omics approach was employed to address molecular primary and secondary resistance mechanisms. Gene set enrichment analyses were used to uncover SR pathways. Targeted therapy of SR PDX models was applied to validate selected SR pathways. Results In vivo anti-EGFR SR could be established with high efficiency. Chronic anti-EGFR treatment of CRC PDX tumors induced parallel evolution of multiple resistant lesions with independent molecular SR mechanisms. Mutations in driver genes explained SR development in a subgroup of CRC PDX models, only. Transcriptional reprogramming inducing anti-EGFR SR was discovered as a common mechanism in CRC PDX models frequently leading to RAS signaling pathway activation. We identified cAMP and STAT3 signaling activation, as well as paracrine and autocrine signaling via growth factors as novel anti-EGFR secondary resistance mechanisms. Secondary resistant xenograft tumors could successfully be treated by addressing identified transcriptional changes by tailored targeted therapies. Conclusions Our study demonstrates that SR PDX tumors provide a unique platform to study molecular SR mechanisms and allow testing of multiple treatments for efficient targeting of SR mechanisms, not possible in the patient. Importantly, it suggests that the development of anti-EGFR tolerant cells via transcriptional reprogramming as a cause of anti-EGFR SR in CRC is likely more prevalent than previously anticipated. It emphasizes the need for analyses of SR tumor tissues at a multi-omics level for a comprehensive molecular understanding of anti-EGFR SR in CRC. Supplementary Information The online version contains supplementary material available at 10.1186/s13073-021-00926-7.


Fig. S2
Primary response pattern of additional PDX models with primary resistance (activating mutation in KRAS, BRAF,or PIK3CA). Relative growth curves are derived from mean values ± SEM (error bars). PD, progressive disease; PR, partial response.

Fig. S3
Primary response pattern of additional PDX models with stable disease (KRAS,NRAS,BRAF,and PIK3CA wt). Relative growth curves are derived from mean values ± SEM (error bars). PD, progressive disease; PR, partial response.

Fig. S4
Primary response pattern of additional PDX models with partial or complete response (KRAS,NRAS,BRAF,and PIK3CA wt). Relative growth curves are derived from mean values ± SEM (error bars). BoC220 and 254, treatment starting volume was 400mm 3 . PD, progressive disease; PR, partial response.

Fig. S5
Primary response pattern of additional PDX models with primary resistance (KRAS,NRAS,BRAF,and PIK3CA wt). Relative growth curves are derived from mean values ± SEM (error bars).*, each asterisk represents a tumor that was taken out of the treatment cohort at the indicated time point either because the tumor reached the maximum size criteria or due to health issues of the animal. PD, progressive disease; PR, partial response.

Fig. S6
Anti-EGFR secondary resistant PDX models can be generated with high efficiency. The growth pattern of PDX models not shown in Figure. 2B with long-term cetuximab treatment and initial treatment response to anti-EGFR therapy. For models BoC20, 35, and 71, time to progression was defined by the mean growth curve crossing the PD borderline. In Boc106 treatment was discontinued for 3 out of 4 animals (6 of 8 tumors) between weeks 10 and 16 (dotted lines); for one animal treatment was continued in sub-therapeutic dose (1:100 of starting dose). Between week 16-23 treatment was continued in all animals in sub-therapeutic dose (1:100 or 1:10; dotted lines) and from week 24 onwards all animals received full dosage. This led to growth stabilization for another 9 weeks until secondary resistance could finally be detected during week 31 (tumor volume increased by 20% relative to volume measured in week 24). In BoC209, 184, 214, and 237 treatment was paused at the first dotted line. Animals were re-treated as soon as at least one tumor per animal reached a volume of approx. 400-500mm 3 .
BoC18 showed complete response and did not develop secondary resistance even after more than 600 days of intermittent treatment (periods of standard cetuximab treatment are shown as light blue shades areas). In BoC220 and BoC254 treatment was paused at week 5 and 9, respectively, following initial response. These models are still awaiting to reach 500mm 3 to reinitiate treatment. Arrows indicate the time point progressive disease was observed. PD, progressive disease; PR, partial response; gray shades area, stable disease; Relative growth curves are derived from mean values ± SEM (error bars). *, each star represents a tumor that was taken out of the treatment cohort at the indicated time point. Please note that this can lead to marked changes in mean tumor volumes (see also Additional file 4: Table S3 for individual growth curves)

Fig. S7
Additional genomic alterations identified in SR PDX models. Cancer gene mutations with unclear functional relevance identified in the 10 PDX models. Acquired mutations in the SR models are indicated by an asterisk. K, untreated control tumor; C, cetuximab SR tumor.

Fig. S8
Transcriptomic subtypes and expression of growth factor signaling pathway genes. The upper part shows the CRIS subtypes of the 10 models linked to their CET treatment and response status as well as to driver alterations. BL, baseline tumor untreated; ST, CET sensitive tumors; SR, secondary resistant tumor. The lower part shows expression values for relevant genes linked to growth factor signaling. Shown are expression values for untreated control tumors (K), tumors treated for 5 days with CET (5dC), and tumors that developed SR under chronic CET treatment (C). SR tumors harboring driver alterations are given in a separate column. mut, mutation in the indicated gene; amp, amplification in the indicated gene. Data shown as quantile normalized log2 values. Log2 expression values below 5 are considered background.

Fig. S9
DUSP6 expression in SR PDX tumors without driver gene mutations. Fold changes are given relative to untreated control tumors in SR PDX models compared to the corresponding cetuximab sensitive tumors treated for 5 days (5dC). Fold changes are calculated as mean values from at least three measurements. *, models with KRAS_SIGNALLING_UP" set found to be enriched, ndm, no driver mutation.

Fig. S10
Synopsis of confirmed candidate genes with a potential role in secondary resistance. Shown are fold changes in gene expression for the indicated genes derived from Agilent array data (A) and the corresponding fold changes detected via 3`RNA Seq (S) for individual SR tumors (C) relative to their expression in 5 days CET treated controls (5dC). n.d., not detected.

Fig. S11
qRT-PCR data confirming the overexpression of FGF3 in the indicated SR tumors. Relative FGF3 expression of the SR tumor (C) to the corresponding expression in the control tumor (5dC) normalized to either COX6C (BoC35) or GAPDH (BoC69).

Fig. S12
Course of body weights for animals treated with Trametinib mono-or Trametinib-CET combination therapy. Shown are body weights relative to the weights of the animals at treatment start corresponding to the experiments shown in Fig. 9a and c. Critical losses are highlighted with circles.

Fig. S13
Targeted treatment addressing SR. a pSTAT3 protein expression at Y705 [pSTAT3 (Y705)] in an untreated BoC32 tumor (K), a tumor treated for 5 days with cetuximab (5dC) and tumors which developed secondary resistance under chronic cetuximab treatment (C). b Relative quantification of pSTAT3 signal intensities to the pSTAT3 intensity of the 5dC tumor normalized with the corresponding beta-actin signal intensities using Image Lab (BioRad).