nature communications Article https://doi.org/10.1038/s41467-022-32828-6 High p16 expression and heterozygous RB1 loss are biomarkers for CDK4/6 inhibitor resistance in ER+ breast cancer Received: 22 February 2021 Marta Palafox1, Laia Monserrat1, Meritxell Bellet2,3, Guillermo Villacampa4, Abel Gonzalez-Perez 5,6, Mafalda Oliveira 2,3, Fara Brasó-Maristany7, Accepted: 17 August 2022 Nusaibah Ibrahimi 8,9, Srinivasaraghavan Kannan 10, Leonardo Mina11, Published online: 07 September 2022 Maria Teresa Herrera-Abreu12, Andreu Òdena1, Mònica Sánchez-Guixé 1, Marta Capelán2,3, Analía Azaro2,3, Alejandra Bruna13, Olga Rodríguez1, Check for updates Marta Guzmán1, Judit Grueso1, Cristina Viaplana4, Javier Hernández 14, 1234567890():,; 1234567890():,; Faye Su15, Kui Lin16, Robert B. Clarke 17, Carlos Caldas 18, Joaquín Arribas 19,20,21,22,23, Stefan Michiels 8,9, Alicia García-Sanz11, Nicholas C. Turner 12, Aleix Prat 7,24,25,26,27, Paolo Nuciforo 28, Rodrigo Dienstmann4, Chandra S. Verma10,29,30, Nuria Lopez-Bigas 5,6,23, Maurizio Scaltriti 31, Monica Arnedos32,33, Cristina Saura 2,3 & Violeta Serra 1,19 CDK4/6 inhibitors combined with endocrine therapy have demonstrated higher antitumor activity than endocrine therapy alone for the treatment of advanced estrogen receptor-positive breast cancer. Some of these tumors are de novo resistant to CDK4/6 inhibitors and others develop acquired resis- tance. Here, we show that p16 overexpression is associated with reduced antitumor activity of CDK4/6 inhibitors in patient-derived xenografts (n = 37) and estrogen receptor-positive breast cancer cell lines, as well as reduced response of early and advanced breast cancer patients to CDK4/6 inhibitors (n = 89). We also identified heterozygous RB1 loss as biomarker of acquired resistance and poor clinical outcome. Combination of the CDK4/6 inhibitor ribociclib with the PI3K inhibitor alpelisib showed antitumor activity in estrogen receptor-positive non-basal-like breast cancer patient-derived xeno- grafts, independently of PIK3CA, ESR1 or RB1 mutation, also in drug de- escalation experiments or omitting endocrine therapy. Our results offer insights into predicting primary/acquired resistance to CDK4/6 inhibitors and post-progression therapeutic strategies. The combination of cyclin D-dependent kinases 4 and 6 (CDK4/6) (BC)1–4. In the adjuvant setting, abemaciclib in combination with ET inhibitors (CDK4/6i) palbociclib, ribociclib and abemaciclib with improves disease-free survival compared with ET alone5. Abemaciclib endocrine therapy (ET) has been approved for the treatment of or palbociclib in combination with trastuzumab and ET has also shown patients with advanced estrogen receptor-positive (ER+) and human clinical activity in HER2-positive (HER2+)/ER+ BC6,7. Despite the clinical epidermal growth factor receptor 2 (HER2)-negative breast cancer success of these treatments in these subsets of BC, the identification of A full list of affiliations appears at the end of the paper. e-mail: vserra@vhio.net Nature Communications | (2022)13:5258 1 Article https://doi.org/10.1038/s41467-022-32828-6 biomarkers of response to CDK4/6i plus ET as well as designing ther- of drug. Both of these ribociclib-resistant models maintained the apeutic strategies for treating patients that escape from this therapy histopathological features of their respective sensitive counterpart remains a major clinical need. Previous studies have highlighted that (Supplementary Table 1). In summary, we tested the sensitivity to tumor sensitivity to CDK4/6 inhibition require a physiologically func- ribociclib in 23 BC PDXs, including two models that acquired resistance tional G1-S restriction point, as well as the absence of mechanisms that to ribociclib from the sensitive counterparts. activate the cyclin E/CDK2 complex8–11. More specifically, high CDK6 results in a reduced response to CDK4/6 inhibitors12–14. Amplification PDXs expressing high p16 are resistant to ribociclib and overexpression of FGFR1 has been associated with high expression To identify biomarkers of de novo resistance to ribociclib, we under- of cyclin D1, resistance to antiestrogens alone and in combination with took genetic, transcriptomic and proteomic approaches. We firstly CDK4/6i9. Loss of pRb itself implies the complete loss of cell cycle determined the intrinsic PAM50 subtype of the PDXs23. Most of the 23 regulation at the G1-S restriction point for which CDK4/6i are no longer PDXs showed concordant molecular and intrinsic subtypes (83%) effective15,16. Alternatively, high cyclin E results in resistance to CDK4/ except four PDXs expressing ER, that were categorized as basal-like 6i, as it bypasses the requirement of CDK4/6 for cell cycle (PDX313, PDX098 and STG201) or HER2-enriched (PDX225) instead of progression17. Luminal B, suggesting that they are not dependent on ER signaling. PI3K inhibitors (PI3Ki), in combination with fulvestrant, have been Unsurprisingly, all basal-like models (ER+ or TNBC) were resistant to approved for the treatment of ER+ metastatic BC with PIK3CA ribociclib (Fig. 2A)24. We then performed genetic analyses of these mutation18. The activity of combining PI3Ki and ET (NCT03056755) as PDXs employing a capture-based sequencing platform that detects well as triple combinations of CDK4/6i, PI3Ki and ET is being investi- genomic aberrations in ~410 cancer related genes (MSK-IMPACT™)25 gated in patients whose tumors progress after CDK4/6i treatment and analyzed whether the incidence of genetic alterations in thirteen (NCT01872260; NCT02088684; NCT03056755). Results from the PIPA cell cycle and PI3K-related genes correlated with ribociclib trial (NCT02389842) have also shown a 37.5% response rate of palbo- response8–10,12,13,16,17. We observed a trend towards ERBB2 amplification ciclib with the PI3Ki taselisib and fulvestrant in PIK3CA-mutant and CDKN2A/B loss-of-function mutations being more frequent in ER + BC19. ribociclib-sensitive PDXs, whereas CCND1 amplification and loss of In this study, we aimed to identify biomarkers of primary and function mutations in TP53 were identified amongst CDK4/6i-resistant acquired resistance to CDK4/6i in a panel of 37 patient-derived tumor models (Fig. 2A, S1B and Supplementary Table 3). models, using genetic, transcriptomic and proteomic approaches. Comparing the mRNA expression levels of 54 cell cycle- and Additionally, we explored if the combination of a PI3Ki plus a CDK4/6i apoptosis-related genes in 12 ER+ or HER2+ ribociclib-resistant versus 5 has therapeutic potential in ER+ and in HER2+ BC with primary or ribociclib-sensitive PDXs, we found that ribociclib-resistant models acquired resistance to CDK4/6i, and its association with the PIK3CA, expressed higher CCNB2 (p = 0.002) as well as a trend towards higher ESR1, RB1-mutation status. CDK1 and CDK7 (p < 0.1; Fig. S1C). Moreover, the pro-apoptotic genes BID and HRK were expressed at lower levels in ribociclib-resistant PDXs Results compared to the sensitive ones (p = 0.03 and 0.02, respectively). A Ribociclib monotherapy has higher antitumor activity than similar pattern of expression was observed in ribociclib-treated PDXs, other targeted agents in ER+ and HER2+ BC PDXs with higher levels of CCNB2 (p = 0.03) and CCND1 (p = 0.14) in Patient-derived xenografts (PDXs) are clinically relevant preclinical ribociclib-resistant PDXs (Fig. S1D). These results suggest that CDK4/ models for drug screening and biomarker identification20,21. We 6i-resistant tumors harbor high CDK1/cyclin B2 activity and/or obtained 58 BC PDXs from implanting 473 ER+ BC tumor specimens in undergo early adaptation to non-canonical cell cycle bypass via CDK2/ immune-deficient mice (12% of success rate; Fig. 1A). Among the cyclin D18. established PDXs, 12 from primary tumors and 9 from metastatic At the protein level, both ribociclib-resistant and sensitive biopsies were initially available for the study (Supplementary Tables 1, PDXs expressed comparable levels of ER, PR, CDK4, CDK6, cyclin 2). Overall, the PDXs recapitulated the molecular subtype of the D2, CDK2, and FGFR1 (Fig. S2A). Of note, none of the PDXs included corresponding original tumor, with the exception of PDX284, which in this panel harbored high-level gene copy number (CN) of FGFR1 lacked the expression of progesterone receptor (PR) and became (Supplementary Table 3)9,26. Conversely, higher p16 levels (p = 0.01) TNBC (Supplementary Table 1). and a trend towards low nuclear pRb levels (p = 0.09) were detected We then examined the antitumor activity of ribociclib in these 21 in resistant PDXs compared to the sensitive ones (Fig. 2B). In addi- PDXs and responses to therapy were classified according to the relative tion, even though the levels of cyclin E1 and cyclin D1 were similar in change in tumor volume upon treatment (similar to the Response ribociclib-responders vs. non-responders (p = 0.2 and 0.4, respec- Evaluation Criteria In Solid Tumors (RECIST) criteria)20,22. We observed tively), PDXs expressing high levels of either protein were resistant a complete response (CR) in one model (5%), a partial response (PR) in to ribociclib. We further computed the accuracy of a complex bio- two (9.5%), a stable disease (SD) in two (9.5%) and a progressive disease marker composed of p16, pRb, cyclin D1 and cyclin E1 expression. (PD) in 18 (76%; Fig. 1B) for a total of 14% of preclinical response rate This composite biomarker showed higher sensitivity (87%) and (pRR; CR + PR) and a 24% of preclinical benefit rate (pCB; CR + PR + accuracy (85%) for the identification of ribociclib-resistant models SD). The degree of response to ribociclib was independent of the compared to single or binary biomarkers (Fig. S2B). Two out of tumor growth rate of the untreated tumors, ruling out a potential bias 3 ER+, basal-like PDX had high p16 levels with concomitant low pRb for slower growing tumors being more sensitive. The three TNBC expression (PDX313, PDX098, Fig. 2B), which was expected given models were, as expected, refractory to CDK4/6 inhibition, whereas the generally reciprocal relationship between p16 and pRb some ER+ and HER2+ PDXs responded, and others did not. We subse- (Fig. 2C)27–29. In addition, these two models also expressed high quently tested the sensitivity of 17 available ER+ or HER2+ PDXs to cyclin E1, consistent with their basal-like intrinsic subtype. In line endocrine (fulvestrant or letrozole) or anti-HER2 (trastuzumab) with these observations, analysis of the ER + TCGA dataset showed therapies, respectively. We observed that PD was the best response in the co-occurrence between low pRb and high p16 protein/mRNA all but one case (PDX191 with SD on fulvestrant), including the 5 expression (p = 0.003; p < 0.001; Fig. S2C), and between low pRb models sensitive to ribociclib (Fig. 1C and Fig. S1A). and high cyclin E1 protein/mRNA expression (p = 0.08; p < 0.001). Resistance to ribociclib was generated from PDX244 (ER+, The latter was also observed in METABRIC. Of note, 17% of PDX244LR) after prolonged drug exposure (Fig. 1D)8 and from PDX153 p16-high tumors in TCGA had normal levels of pRb, suggesting that (HER2+, PDX153LR), spontaneously after 7 serial passages in the absence high levels of p16 might indicate pRb functional loss, or be the result Nature Communications | (2022)13:5258 2 Article https://doi.org/10.1038/s41467-022-32828-6 A Vehicle Drug Step 1. Paent fresh biopsy Step 2. PDX establishment and Step 3. PDXs expansion for drug implantaon in NSG mice serial maintenance (n = 58) antumoral assay (n = 23) (n = 473) B C Ribociclib PDX response 1000 1000 P R /C R S D PD 313 F u lv e s tra n t ET L e tro z o le R e s p o n s e t o r ib o c ic lib 800 T ra s tu z u m a b 600 98 600 STG 201 400 400 % C h a n g e fro m B a s e lin e 153LR 200 225 39 222 251 191 131 2 8 7 .2 200 118 PD 0 2 8 7 .3 343 SD P R /C R 153 244 173 0 -2 0 0 -2 0 0 0 200 400 600 1000 -2 0 R e s p o n s e to E T /a n ti- H E R 2 D 10 P D X 244 -4 0 9 -6 0 L u m in a l B ( E R + /H E R 2 -) 8 R e la tiv e T u m o r V o lu m e H E R 2 + ( E R + /H E R 2 + ) 7 H E R 2 + ( E R - /H E R 2 + ) -8 0 TNBC 6 PD -1 0 0 5 S D /P R /C R (n = 4 ) (n = 9 ) (n = 4 ) (n = 3 ) (n = 6 ) (n = 6 ) (n = 5 ) (n = 4 ) (n = 6 ) (n = 8 ) (n = 1 ) (n = 5 ) (n = 8 ) (n = 1 ) (n = 6 ) (n = 1 ) (n = 2 ) (n = 2 ) (n = 7 ) (n = 2 ) (n = 2 ) (n = 1 ) (n = 2 ) P D X 153LR 4 PD X251J PD X313 PD X347 PD X098 PD X284 PD X270 PD X225 PD X222 PD X039 PD X251 PD X131 PD X191 PD X118 PD X004 PD X343 PD X153 PD X244 PD X173 S TG 2 0 1 P D X 2 8 7 .2 P D X 2 8 7 .3 PD X244LR PD X153LR 3 2 PD X251J 1 V e h ic le P D X 173 0 E R + ( H R + /H E R 2 -) H E R 2 + (H R + /H E R 2 + ) 0 20 40 60 80 100 120 140 H E R 2 + ( H R - /H E R 2 + ) T N B C ( H R -/H E R 2 -) D a y s o f tre a tm e n t Fig. 1 | Ribociclib monotherapy has higher antitumor activity than other tar- other targeted agents (x-axis; endocrine therapy (ET) or trastuzumab) in PDXs geted agents in ER+ and HER2+ BC PDXs. A Workflow depicting the generation of represented as the percentage of tumor volume change compared to the initial BC PDX models from BC patient samples and its subsequent expansion for targeted tumor volume. Symbols represent the different targeted therapies. D Spaghetti treatment screening. B Waterfall plot representing the growth of n = 23 PDX treated plot showing the relative tumor volume change along time in 23 BC PDX treated with ribociclib 75 mg/kg (bars and black dots) and vehicle (white circles). The with ribociclib 75 mg/kg. Ribociclib-sensitive models are represented with filled percentage change from the initial volume is shown at day 35 of treatment. Dashed symbols and ribociclib-resistant with open symbols. Symbols represent the PDX’s lines indicate the range of PD (>20%), SD (20% to −30%) and PR/CR (<−30%). The molecular subtype. Dashed lines indicate the range of PD (>1.2), SD (1.2 to –0.7) and number of tumors treated per model is indicated in brackets (n). Data represent PR/CR (<−0.7). Acquisition of ribociclib resistance in PDX244 (PDX244LR) is shown. mean values and error bars ± SEM. C Antitumor response of ribociclib (y-axis) vs. Source data are provided as a Source Data file. of a pRb/cyclin E1-independent oncogenic stress. In line with this, high p16 protein expression is associated with CDK4/6i resistance two out of 8 Luminal B PDXs resistant to ribociclib (PDX039 and regardless of the pRb status. PDX287.2, 25%) expressed high p16 without concomitant loss of We further analyzed pharmacodynamic biomarkers by immuno- pRb or overexpression of cyclin E1 (Fig. 2C). This data suggests that histochemistry as in the PreOperative-Palbociclib (POP) trial30. We Nature Communications | (2022)13:5258 3 Article https://doi.org/10.1038/s41467-022-32828-6 Fig. 2 | PDXs expressing high p16 are resistant to ribociclib. A Summary of parametric t-test two-tailed p-value are indicated. The pictures underneath are genetic alterations in the PDX panel from Fig. 1B, including the PDX subtype representative bright-field images of high/low staining for each protein. Scale classification, based on IHC (molecular subtype) or PAM50 analysis (intrinsic bar = 100 µm. R: resistant; S: sensitive. C Concordance analysis of the PDXs subtype), and the response to CDK4/6 inhibitors. Genes with similar function responses to ribociclib based on the analyzed biomarkers (y-axis) vs. the in vivo such as TSC1/TSC2 or CDKN2A/CDKN2B were considered as one single feature. response (x-axis). Singleplex or multiplex biomarkers are represented by circles B Quantification of IHC staining for p16, pRb, cyclin E1 and cyclin D1 in n = 23 of different colors and the number of PDX within each category is indicated. The untreated PDX ribociclib-response analyzed in two independent experiments. intrinsic subtype of each PDX is represented by different font colors. D Consort Semiquantitative analysis was performed for pRb, p16, according to cyclin E1 flow diagram for classifying the PDX responses to ribociclib based on the and cyclin D1. Different colors indicate the PDX intrinsic subtype and $ indicates molecular subtype, p16, pRb and cyclin E1/D1 scores. Source data are provided the models harboring gene amplification. Mean values ± SEM and the unpaired as a Source Data file. Nature Communications | (2022)13:5258 4 Article https://doi.org/10.1038/s41467-022-32828-6 examined the percentage of KI67- and phospho-pRb (Ser807/811)- experiment, whereby T47D-p16 cells were seeded 1:20 with control positive cells in ribociclib-sensitive and -resistant PDXs without and cells, p16 expression levels were upregulated after 14 days of treatment with ribociclib treatment and observed a decrease of KI67-positive (p = 0.0006; Fig. 4C and Fig. S3B). Similar results were obtained with cells as well as an unexpected increase of phospho-pRb-positive cells in MCF7 cells, suggesting that pre-existing, low-abundant p16-over- ribociclib-sensitive PDXs (Fig. S2D). In summary, the basal-like intrinsic expressing cells were positively selected upon treatment with CDK4/6i subtype classification (PAM50) and an integrated biomarker com- and represent a reservoir of drug-resistant cells. posed of p16, pRb, cyclin E1 and cyclin D1 expression levels identified To further support these findings, we posited that P18IN00331, an models with resistance to ribociclib in PDXs (Fig. 2D). inhibitor of the p18-CDK4 interaction, would also impair the binding of p16 to CDK4/6 and sensitize p16-high PDCs to CDK4/6i. Using in silico Biomarker validation in short-term patient-derived tumor modeling, we observed that the ankyrin repeats 1, 2, 3 in p16 constitute cells (PDCs) a binding pocket that is relatively large and shallow. One part of this To validate the potential predictive biomarkers for ribociclib anti- pocket consists largely of hydrophobic residues (Val51, Met52, Met53 tumor activity, we measured the activity of ribociclib using patient- and Met54) and the other part contains charged residues (Asp74, derived cells (PDCs) grown as short-term three-dimensional (3D) Asp84, Glu88, Arg46, and Arg87). P18IN003 fits into this pocket, with ex vivo cultures on a laminin-rich extracellular matrix (Fig. 3A). PDCs one methoxyphenyl moiety and the dihydroimidazole moiety of were able to proliferate for at least 14 days (Fig. S2E). PDCs growth was P18IN003 involved in H-bond interactions with Asp74 and Glu88, monitored by measuring the spheroid area and the percentage of cells respectively, of p16. The other methoxyphenyl moiety of P18IN003 is in the S-phase of the cell cycle using the 5-ethynyl-2′-deoxyuridine solvent exposed. Comparing the predicted p16-P18IN003 interaction (EdU) incorporation assay. The antiproliferative activity of ribociclib with the in silico model of the p16-CDK4 interaction revealed that was determined using sixteen representative available ex vivo cultures P18IN003 would disrupt the binding of CDK4 to p16 by binding to this from the panel of 23 PDXs described above and showed that the pocket (Fig. 4D). In ex vivo cultures, P18IN003 combined with riboci- responses to ribociclib of PDCs ex vivo agreed with that of their cor- clib markedly reduced proliferation in PDC191 (p16-high, pRb- responding PDXs in vivo (p = 0.0001 and 0.02; Fig. 3B, Fig. S2F expressing model; p = 0.0009) but not PDC313 (p16 high, pRb low and S2G). model; p = 0.2) compared to ribociclib monotherapy (Fig. 4E), pre- We then extended this analysis using ex vivo cultures from 14 sumably due to increased binding of ribociclib to CDK4/632. additional ER+ BC PDXs, three of which concomitantly expressing The response to ribociclib, fulvestrant and their combination was HER2 (Supplementary Table 4). Based on p16, pRb, cyclin E1, and cyclin also evaluated in T47D and MCF7 cells overexpressing cyclin D1 (T47D- D1 protein expression levels, 9 of these 14 PDXs were predicted to be cyclin D1 and MCF7-cyclin D1). Cyclin D1 overexpression moderately resistant and the remaining 5 sensitive to ribociclib (Supplementary increased the IC50 values 3- to 5-fold in T47D and 2- to 3-fold in MCF7 Table 4 and Fig. 2D). Of note, 2 out of 9 models classified as ribociclib (Fig. 4F). In line with this, T47D and MCF7 cells overexpressing cyclin resistant due to high expression of p16 (PDX301 and PDX346) did not D1 showed an attenuated response to downmodulation of phospho- harbor loss of pRb. As expected, PDCs predicted as ribociclib sensitive pRb, cyclin E2 and phospho-CDK2 T160 upon treatment with ribociclib exhibited a greater reduction in the relative spheroid area upon (Fig. 4G and Fig. S3C). In a competition experiment, cyclin D1 expres- treatment than PDCs predicted as ribociclib-resistant (p = 0.005; sion was upregulated after 14 days of treatment in both T47D Fig. 3C). Moreover, with the exception of PDX350 and PDX399, two (p = 0.0004) and MCF7 cell lines (p = 0.01; Fig. 4H and Fig. S3D). This models with unexpectedly high ex vivo sensitivity, the responses to result suggests that pre-existing, low-abundant cyclin D1- ribociclib of all PDCs were in agreement with the predicted responses overexpressing cells were positively selected upon treatment with (Supplementary Table 4 and Fig. 3D). ROC curve analysis indicated that CDK4/6i and represent a reservoir of drug-resistant cells. In summary, the change of the spheroid area upon ribociclib treatment could dis- we conclude that overexpression of either p16 or cyclin D1 attenuates criminate between sensitive and resistant PDCs with 100% sensitivity the response of BC cells to CDK4/6i through activation of G1 check- and 87.5% specificity (p < 0.0001; cut-off = −25%; n = 37 models; point kinase activity. Fig. 3E). In summary, we identified ribociclib resistant ER+ BC PDCs using a composite biomarker of p16, pRb, cyclin E1 and cyclin D1. High p16 levels associated with lack of response to CDK4/6i in ER+ BC patients p16 and cyclin D1 overexpression attenuate the response to Given our preclinical results, we interrogated a potential association of ribociclib in ER+ BC cell lines the aforementioned biomarkers with response to CDK4/6i in ER+ BC Previous studies have already demonstrated the impact of RB1 loss and patients. In early-stage breast cancer, treatment with palbociclib or CCNE1 amplification in response to CDK4/6i8,13,16,17. Therefore, we abemaciclib significantly reduced proliferation and the CDK4/6 aimed at evaluating whether p16 and cyclin D1 overexpression can be downstream response, measured by KI67 and phospho-pRb (S807/ added to the list of candidates associated with resistance to ribociclib 811), respectively30,33. Here, we reanalyzed data from the abemaciclib in ER+ BC. We generated T47D and MCF7 cells (ER+) overexpressing p16 preoperative (ABC-POP) clinical trial33. In total, 72 patient samples were with a doxycycline-inducible system (T47D-p16, MCF7-p16). We mea- analyzed, 33 Luminal A and 39 Luminal B. As expected, the majority of sured the response of these cells to ribociclib, fulvestrant and the the Luminal A tumors were sensitive to CDK4/6 inhibition in terms of combination and found that T47D-p16 cells had 20-fold higher IC50 drop of KI67 at day 15 of treatment, a “gold standard” biomarker of (half-maximal inhibitory concentration) for ribociclib and 6-fold endocrine sensitivity in this patient population34. On the contrary, we higher for the combination of ribociclib plus fulvestrant than MOCK found that similar proportions of Luminal B tumors could be classified control cells; and MCF7-p16 cells had 17- and 24-fold IC50 values, as responders (56%) and non-responders (44%; Fig. 5A). In line with respectively (Fig. 4A). Biochemical analysis revealed that T47D-p16 results obtained in PDXs (Fig. 2B), the H-score levels of p16, but not cells had higher levels of phospho-pRb (S780, p = 0.0004; S807/811, pRb, cyclin E1 and cyclin D1, were significantly higher in resistant p = 0.003), cyclin E2 (p = 0.003) and phospho-CDK2 T160 (p = 0.0009) tumors treated with abemaciclib compared to sensitive tumors compared to control cells (Fig. 4B and Fig. S3A). Also, ribociclib (p = 0.008; Fig. 5B and Fig. S4A). In the multivariable analyses, high p16 treatment resulted in CDK6 upregulation in T47D cells while p16- expression alone was significantly associated with a decreased prob- overexpression resulted in reduced CDK6 expression in MCF7 cells, so ability of KI67-response (p = 0.004 for All tumors and p = 0.03 for it cannot be excluded that CDK6 modulation contributes to the CDK4/ Luminal B tumors) and we did not identify independent prognostic 6i resistance phenotype observed in vitro. In a competition association of the other three biomarkers (Fig. 5C and Fig. S4B). Nature Communications | (2022)13:5258 5 Article https://doi.org/10.1038/s41467-022-32828-6 Fig. 3 | Biomarker validation in short-term patient-derived tumor cells (PDCs). 100 µm. C Relative spheroid area in n = 14 PDC models classified as resistant (mar- A Workflow depicting the generation of BC PDCs short-term ex vivo cultures from oon) or sensitive (orange) according to the composite biomarker after treatment PDXs and the subsequent analysis of ribociclib response using two different read- with 1 µM ribociclib for 7 days. Relative data to the vehicle control (100%) is repre- outs. B Correlation analysis of the ex vivo response of PDCs (y-axis) vs. the in vivo sented as mean values of three independent experiments ± SEM; unpaired para- response of the corresponding PDXs (x-axis), measured as change in spheroid area metric t-test two-tailed p-value are indicated. D Concordance analysis of PDXs (open dots) or the change in EdU incorporation (filled dots) after ribociclib treat- responses to ribociclib based on biomarker prediction (y-axis) vs. the ex vivo ment. The Spearman’s coefficient (r), two-tailed p-value and 95% of confidence response (x-axis). Biomarkers are represented by circles with different colors and the interval (95% CI) for each read-out are summarized below the graph. Representative number of PDX within each category is indicated. E ROC curve of the spheroid area images of one ribociclib-resistant (PDC039) and one ribociclib-sensitive (PDC244) increment for ribociclib response prediction in n = 37 BC PDCs. Unpaired t-test two- model treated with vehicle or ribociclib are shown on the right panel, namely EdU/ tailed p-value (<0.0001), the sensitivity, 95% confidence interval (95% CI) and spe- K18 staining by confocal microscopy and organoid size by bright field microscopy. cificity are summarized below the graph. Source data are provided as a Source EdU is shown in green, cytokeratin 18 (K18) in red and DAPI in blue. Scale bar = Data file. Nature Communications | (2022)13:5258 6 Article https://doi.org/10.1038/s41467-022-32828-6 Next, we tested if high p16 was also associated with lack of three CDK4/6i-resistant patients with high p16 levels did not response to abemaciclib as single agent in the metastatic setting exhibit low pRb. In this cohort, multivariable analyses was not (n = 17). Higher p16 levels (p = 0.04) and high cyclin E1 levels feasible due to the limited number of samples. In summary, our (p = 0.03) were detected in resistant tumors (n = 6) compared to data shows that high protein levels of p16 are associated with the sensitive ones (n = 11) whereas levels of pRb and cyclin D1 were resistance to CDK4/6i in both early-stage and metastatic ER+ BC not associated with CDK4/6i resistance (Fig. 5D). Interestingly, the tumors. Nature Communications | (2022)13:5258 7 Article https://doi.org/10.1038/s41467-022-32828-6 Fig. 4 | p16 and cyclin D1 overexpression attenuate the response to ribociclib in hydrogen bonds made between P18IN003 and p16 shown as black dashed lines. ER+ BC cell lines. A Half-maximal inhibitory concentration (IC50) and IC50 fold- E Relative spheroid area of PDC191 and PDC313 after treatment with 1 µM ribo- change (fc) values for ribociclib, fulvestrant and the combination of drugs in ciclib, 20 nM P18IN003 and the combination for 7 days. Data are presented as T47D and MCF7 cells overexpressing p16 compared to control cells (MOCK), mean values of three independent experiments ± SEM. Two-tailed p-values are treated for 6 days and evaluated as shown underneath. Data are presented as based on the one-way ANOVA test with Tukey’s method correction compared mean values ± SEM of at least three independent experiments. B Immunoblot of with the vehicle (black line). Dashed line indicates the optimal cut-off estab- the indicated proteins in MOCK and p16 overexpressing T47D and MCF7 cells lished in Fig. 3E. p16 and pRb scores of each PDC are indicated. F IC50 and IC50 fc untreated or treated with ribociclib for 5 days at the indicated concentrations. values for ribociclib, fulvestrant and the combination of drugs in T47D and C Immunoblot of the indicated proteins in a cell competition assay. Cells were MCF7 cells overexpressing cyclin D1, compared to control cells (MOCK), treated treated with vehicle or ribociclib 1 µM for the indicated days. D Comparison of for 6 days and evaluated as shown underneath. Data are presented as mean structural models for the complexes of p16 bound to P18IN003 and p16 bound values ± SEM from at least three independent experiments. G Immunoblot of to CDK4. Gray cartoon is p16, orange spheres represent P18IN003, cyan cartoon indicated proteins in control (MOCK) and cyclin D1-overexpressing T47D and is CDK4; zoomed view are the binding pocket with p16 shown as gray surface MCF7 cells untreated or treated with 0.5 µM ribociclib for 48 h. H Immunoblot as and highlighting the residues in the binding pocket of p16 as sticks and the indicated for panel (C). Source data are provided as a Source Data file. Acquisition of subclonal RB1 mutations as mechanism of mutation of the RB1 gene appears preferentially amongst patients who acquired resistance to ribociclib in tumors with RB1 hetero- had received CDK4/6 inhibitors, we applied a multivariable logistic zygous loss regression. We found a significant association between the double hit We next investigated mechanisms of acquired resistance in PDXs and (mutation and heterozygous deletion) and the prior exposure of the posited that tumors with an underlying RB1 heterozygous loss tend to patient to CDK4/6i as part of their treatment (p = 0.003, Fig. 6D). No acquire RB1 point mutations that result in CDK4/6i resistance. This significant association was found between any alteration (only muta- hypotesis was based on the evidence from one patient with matched tion, only deletion or both) affecting RB1 and the previous exposure to pre/post-CDK4/6i tumor sample and the relatively low frequency of CDK4/6 inhibitors (Fig. S4D), implying that only patients with a double acquired RB1 point mutations after treatment with palbociclib15,16,35. We hit in RB1 have an increased likelihood of having received CDK4/6i. generated eight derivatives from PDX244 that became refractory to Altogether, these data suggest that tumors harboring heterozygous ribociclib treatment overtime8. Sequencing data from the sensitive RB1 loss are susceptible of acquiring a second hit in the RB1 gene and tumor (2R) revealed a CDKN2A homozygous loss and concomitant RB1 becoming resistant to CDK4/6i. heterozygous loss (Fig. 6A). Protein analysis confirmed the lack of p16 and normal pRb levels in this model (Fig. 6A). Three out of 8 ribociclib The PI3K inhibitor alpelisib sensitizes non-basal-like BC PDX to resistant tumors (16L, 16R, and 18R) acquired deleterious mutations in ribociclib RB1 (p.M695Nfs*26, p.K810* and p.X180_splice, respectively) and Preclinical studies37 and a phase III clinical study (SOLAR-1, tumors 19L and 19R underwent a further reduction in the RB1 copy NCT0243731818) have shown that PI3K-α inhibitors, such as alpelisib, number (from −0.9 to −1.5 and −2.0, respectively), suggesting the are able to sensitize PIK3CA-mutant tumors to ET. Moreover, pre- acquisition of homozygous RB1 loss. Protein analysis by IHC confirmed liminary results from the BYLieve clinical trial have shown that the the total or subclonal loss of pRb expression in these tumors (16L and proportion of patients treated with alpelisib plus fulvestrant without 19L vs. 16R and 18R). Intriguingly, a sixth tumor (15R) also showed disease progression at 6 months (CBR) was 50.4%, showing clear effi- partial loss of pRb expression without a detectable underlying genetic cacy of PI3K-α inhibitors post-CDK4/6i. Moreover, this benefit is alteration in RB1, suggesting alternative mechanisms that regulate RB1 maintained regardless of the duration of response to the previous gene expression. Regarding 15L and 17L, we observed an increment in CDK4/6i-based treatment or presence of CDK4/6i resistance genes in CDKN2A CN log ratio (from −4.2 to −0.4) along with the restoration of circulating tumor DNA (NCT0305675538–40). Here, we interrogated p16 expression (Fig. 6A), suggesting that selection of tumor cells whether alpelisib can sensitize ribociclib-resistant PDXs to CDK4/6i. retaining normal CDKN2A CN was favored upon treatment. Cyclin E1 In our PDX panel, alpelisib monotherapy resulted in a pCB of 43% and cyclin D1 did not show any alteration at the gene or protein levels compared to 25% of ribociclib alone (Fig. 7A). All HER2+ PDXs tested in any of the acquired-resistant tumors (Fig. 6A). were sensitive to alpelisib, including the ribociclib-resistant PDX153LR, In addition to the aforementioned model, we developed paired PDX222, and PDX118. All the PIK3CA mutated tumors were also PDXs from a patient who received palbociclib plus letrozole before sensitive to alpelisib except PDX287.3, which was derived from a treatment initiation (PDX476.1) and after 12 cycles of treatment at the patient’s tumor that progressed while being treated with the PI3Ki time of disease progression (PDX476.2). Similar to the respective GDC-0032 plus letrozole (Fig. S4E). As expected, all 3 TNBC PDXs were patient tumors, PDC476.1 was sensitive to palbociclib whereas resistant to alpelisib since none of them harbored a PIK3CA mutation. PDC476.2 was resistant (Fig. S4C). Genetic analysis showed that Combined treatment of ribociclib and alpelisib resulted in a pCB PDX476.1 harbored a heterozygous RB1 loss (CN = 0.9) but still of 78% (Fig. 7A). We noticed that all basal-like PDXs by PAM50 exhib- expressed pRb. In contrast, PDX476.2 lost pRb protein expression, ited PD or SD as best response. Similarly, combination of palbociclib suggesting that it may be the mechanism responsible of tumor pro- plus GDC-0032 also showed improved antitumor activity compared to gression (Fig. 6B). either one alone in PDX287.2 and PDX287.3 that derived from a Next, we interrogated the prognostic implications of RB1 het- patient’s tumors collected on-treatment or after progression with erozygous loss in ER+ BC patients. In two out of the three cohorts GDC-0032 plus letrozole, respectively (Fig. S4F). Analysis of pharma- analyzed, patients with tumors harboring RB1 heterozygous loss codynamic biomarkers showed that both KI67 and phospho-pRb showed significantly poorer clinical outcome compared to patients S807/811 were downregulated in PDXs that responded to the drug with unaltered RB1 tumors in terms of disease-free survival (DFS), combination (p < 0.001 and p < 0.01, respectively), but not in the overall survival (OS) or days of treatment (DOT; Fig. 6C). In order to resistant ones (Fig. 7B). Because of concerns regarding the safety of analyze the role of RB1 heterozygous loss as biomarker of resistance to the combination of alpelisib plus ribociclib, we conducted de- CDK4/6i in patients, we obtained genomic and clinical data of meta- escalation experiments showing that dose reduction of either riboci- static BC patients included in the Hartwig Medical Foundation (HMF) clib or alpelisib resulted in similar antitumor activity as the full dose cohort36. Out of 582 patients diagnosed with metastatic BC, 71 received tested in both PDX039 and PDX191. Dose reduction of both drugs, CDK4/6i. To test whether concomitant heterozygous deletion and however, resulted in an attenuated efficacy (Fig. 7C). Of note, both Nature Communications | (2022)13:5258 8 Article https://doi.org/10.1038/s41467-022-32828-6 Fig. 5 | High p16 levels associated with lack of response to CDK4/6i in ER+ BC with abemaciclib for all tumors (n = 72; left panel) or Luminal B tumors (n = 39; patients. A Percentage distribution of sensitivity to abemaciclib after 15 days of right panel). Two-sided p-values from Wald tests in logistic models are provided. treatment in the neoadjuvant setting in the ABC-POP trial (n = 72). Tumors were D Quantification of p16, pRb, cyclin E1 and cyclin D1 in a cohort of n = 10 classified as Luminal A if %KI67 < 15 or as Luminal B if %KI67 ≥ 15. Tumors advanced BC tumors detected by IHC semiquantitatively (pRb p16 cyclin E1 and showing ln KI67 < 1 at day 15 were considered sensitive and those with ln KI67 ≥ 1 cyclin D1) displayed according to the patient’s response to abemaciclib. Sym- were resistant to the studied drug. B Box and whiskers plot showing a logistic bols indicate different CDK4/6i treatments. Mean values ± SEM and unpaired model to evaluate the effect of p16 on the response to abemaciclib (n = 39 parametric t-test two-tailed p-value are indicated R: resistant; S: sensitive. Luminal B tumors). Box represents the median and the 25th and 75th percen- E Forest plot displaying the odds ratios (OR; black squares) ± 95% confidence tiles, whiskers show the largest and smallest value. The mean value of each intervals (CI; horizontal lines) of the association between the indicated bio- subgroup is indicated. C Forest plots displaying the odds ratios (OR; black markers and the patient’s response to the study treatments (n = 10 tumors). squares) ±95% confidence intervals (CI; horizontal lines) of the association Fisher’s exact test two-tailed p-values are indicated. Source data are provided as between the indicated biomarkers and the percentage of KI67 after treatment a Source Data file. dose reductions were within the range of doses being tested/recom- palbociclib plus letrozole: treatment of PDC476.2 with palbociclib plus mended in the clinic for these combinations41. alpelisib resulted in higher reduction of the spheroid area than palbo- Importantly, we demonstrated that ribociclib plus alpelisib (or the ciclib plus fulvestrant (Fig. S4C). The triple combination of ribociclib triple combination with fulvestrant) was effective in two PIK3CA-wt, plus alpelisib and fulvestrant was the most efficacious controlling pro- ESR1-mut PDXs with primary resistance to ribociclib plus fulvestrant liferation and, biochemically, increasing the levels of PARP1 cleavage (PDX131 and PDX244LR#18 R, Fig. S4G and Fig. 7D) and in a PDX from a (Fig. 7F). Altogether, these results suggest that the combination of a patient who showed an early progression when treated with palbociclib PI3Ki with a CDK4/6i, with or without ET, is a valid therapeutic option for plus letrozole (PDX450, PIK3CA and ESR1 mutant; Fig. 7E). Similar results the treatment of ER+ BC tumors after progression on CDK4/6i plus ET, were obtained ex vivo with a PIK3CA-wt PDX from a patient who received independently of PIK3CA, ESR1 mutation status or pRb expression. Nature Communications | (2022)13:5258 9 Article https://doi.org/10.1038/s41467-022-32828-6 Fig. 6 | Acquisition of subclonal RB1 mutations as mechanism of acquired is provided. Scale bar = 100 µm. C Clinical outcome expressed in months for the resistance to ribociclib in tumors with RB1 heterozygous loss. A On the top, indicated clinical endpoint in patients with ER+/HER2-, RB1 unaltered tumors vs. copy number and mutation status of CDKN2A, RB1, CCNE1 and CCND1 in tumors those harboring tumors with RB1 heterozygous loss. Data and statistical analysis derived from PDX244, including an untreated sensitive tumor (2R) or tumors were extracted from the cBioportal. HT hormone therapy, WES whole-exome with acquired resistance to ribociclib (15L to 19R). Deep-Del (homozygous loss), sequencing, TS targeted-sequencing, n number of patients, DFS disease-free CN < −1; Shallow-Del (heterozygous loss), −1 ≤ CN < −0.4; Unaltered, CN ≥ −0.4. survival, OS overall survival, DOT days of treatment. D Association between RB1 Hashtags indicate tumors that acquired deleterious mutations in RB1. On the double hit alterations (concomitant mutation and deletion) and prior exposure bottom, representative pictures showing IHC staining of p16, pRb, cyclin E1 and to CDK4/6 inhibitors in metastatic BC patients (n = 582 tumors). Black square cyclin D1 from the indicated tumors (bottom). Dashed-red lines mark off areas represent the logit value. Multivariable logistic regression two-tailed unad- with different protein staining intensity and protein score is indicated. For 19R justed p-value and the 95% confidence intervals (CI; horizontal segment repre- there was no FFPE tumor available. Scale bar = 100 µm. B IHC staining of p16 and sents) for the test are shown. Source data are provided as a Source Data file. pRb in representative FFPE sections from PDX476.1 and PDX476.2. Protein score Discussion 6i plus ET to avoid unnecessary toxicity, to reduce unnecessary costs 15–30% of ER+ metastatic BC progress rapidly when treated with CDK4/ and to provide more effective alternative treatments. Targeting 6i plus ET. In general, the subsequent line of therapy has a short mTORC1 with everolimus or PI3K with alpelisib in combination with duration, lasting less than 6 months42,43. Therefore, it is important to aromatase inhibitors (AI) or fulvestrant, respectively, are currently identify the group of patients that are not likely to benefit from CDK4/ available therapies but have some limitations. Everolimus approval for Nature Communications | (2022)13:5258 10 Article https://doi.org/10.1038/s41467-022-32828-6 metastatic ER+ BC is in combination with an AI, but AI is not active in a PI3Ki is effective in CDK4/6i-resistant ER+, non-basal-like PDXs, patients with ESR1-mutant tumors44. On the other hand, alpelisib is independently of the PI3KCA or ESR1 mutation status. Moreover, triple only recommended for patients with BC harboring PIK3CA mutations18. combinations of a CDK4/6i with a PI3Ki and ET are active in PDXs In this work, we explored biomarkers of resistance to CDK4/6i and the harboring biallelic RB1 mutation (or pRb protein loss). efficacy of a therapeutic strategy based on CDK4/6i plus PI3Ki. We Although CDK4/6i are mostly administrated in combination with identified that overexpression of p16 or cyclin D1 is associated with an ET, abemaciclib has shown activity in ER+/HER2-negative metastatic BC impaired response to CDK4/6i and that a CDK4/6i in combination with that progressed on ET and chemotherapy4. Also, palbociclib has shown Nature Communications | (2022)13:5258 11 Article https://doi.org/10.1038/s41467-022-32828-6 Fig. 7 | PI3K inhibition sensitizes non-basal-like BC PDX to CDK4/6i. A Waterfall p-values are based on Mann-Whitney U test for all biological replicates. Different plot showing the growth of n = 23 PDX treated with ribociclib 75 mg/kg plus alpe- colors represent the tumor’s intrinsic subtype. R, resistant; S, sensitive; Alp, alpe- lisib 35 mg/kg (bars and black dots) and vehicle (white circles). The percentage lisib; Rib, ribociclib. C Relative tumor growth of PDX039 and PDX191 after treat- change from the initial volume is shown at day 35 of treatment. Dashed lines ment with the indicated drugs and time. Dashed lines indicate the range of PD indicate the range of PD (>20%), SD (20% to −30%) and PR/CR (<−30%). The number (>1.2), SD (1.2 to −0.7) and PR/CR (<−0.7). Two-tailed p-values are based on the two- of tumors treated per model is indicated in the brackets (n). Hashtags (#) indicate way ANOVA test with Bonferroni’s method correction. V, vehicle; R, ribociclib; A, models harboring mutations in PIK3CA. Data represent mean values and error bars alpelisib. D, E Relative tumor growth of the ribociclib-resistant PDX244LR#18R and ± SEM. Boxes underneath show the molecular and intrinsic tumor subtype as well as PDX450 after treatment with the indicated drug(s) for the indicated period of time, their responses to the indicated treatments. The preclinical benefit to each drug is respectively. Dashed lines indicate the range of PD (>1.2), SD (1.2 to −0.7) and PR/CR indicated as percentage in brackets. n.d, not determined. B IHC analysis of KI67 (left (<−0.7). F Immunoblot of the indicated proteins in PDC476.2 treated with vehicle or graph; p-value < 0.001) and phospho-pRb S807/811 (right graph; p-value < 0.01) in 500 nM palbociclib as single-agent or combined with 100 nM fulvestrant and/or vehicle and alpelisib plus ribociclib-treated PDXs according to the PDX response to 2.5 µM alpelisib in ex vivo cultures for 48 h. At least two independent experiments alpelisib plus ribociclib. Data are presented as mean values of each PDX. Two-tailed were conducted. Source data are provided as a Source Data file. a signal of clinical activity in post-menopausal women with advanced blockade due to acquisition of a second hit in RB1. In patients, we ER+/HER2-negative BC having received prior ET45. This data with abe- observed that RB1 heterozygous loss is associated with worse maciclib and palbociclib monotherapy support the examination of prognosis and that a double hit in RB1 is more likely to occur in CDK4/6i monotherapy-specific mechanisms of sensitivity. We patients who received a CDK4/6i than in those with ET alone. At observed that CDKN2A/B were more frequently disrupted in ribociclib- least in part, this might be explained by the haploinsufficiency of sensitive models and the best PDX responder lacked expression of p16. pRb in its contribution to DNA repair in the S-phase checkpoint Similarly, in the Phase I trial of abemaciclib the best BC responder and initiation of DNA replication; cancer cells with only one copy harbored a concomitant deletion of CDKN2A and CDKN2B in her of RB1 exhibit a genomic instability phenotype52. Therefore, tumor46. This observation is consistent with the increased sensitivity to tumors harboring RB1 heterozygous loss might be prone to palbociclib observed in 544 cancer cell lines in association with acquire a second hit in RB1 and resistance to CDK4/6 blockade. As CDKN2A inactivation by DNA methylation and with the limited capacity we have shown, tumors with RB1 loss are sensitive to the triple of current CDK4/6i to inhibit its target when it is bound to INK4 pro- combination of CDK4/6 plus PI3Ki and ET or might be sensitive to teins or to p27Kip1 14,32,47. other targeted treatments such as inhibitors of the Aurora kinases FGFR1 gain and TP53 mutation detected in circulating tumor DNA AURKA and AURKB53. have recently emerged as markers of early progression on CDK4/6i In closing, therapies for metastatic BC that progress on or after treatment, albeit they are also associated with poor prognosis irre- CDK4/6i treatment currently being tested in the clinic include: (1) ET or spective of palbociclib treatment9,26. Although CCND1 and CDKN2A more potent SERDs in endocrine sensitive patients, albeit this patient individually have not been associated with lack of response to CDK4/6i population is difficult to identify (NCT02338349, NCT03616587); (2) plus ET, results from the BioItaLee (NCT03439046) and PALOMA-1 continuation of CDK4/6i with a different ET backbone (NCT03616587; (NCT00721409) trials suggest that genes in the CDK4/6 pathway are NCT03809988); (3) continuation of ET with a different CDK4/6i; for also biomarkers of de novo resistance2,48. High p16 protein has been example with abemaciclib, because it has additional targets including previously described as a surrogate biomarker of pRb loss in several CDK1/2 complexes54; (4) different ET combined with a PI3Ki in PI3KCA tumor types28,49. Here, we found similar results analyzing 814 ER+ mutant ER+ BC18,38–40; or (5) different ET combined with an AKT inhi- invasive breast carcinomas from the TCGA dataset. However, we also bitor in PIK3CA/AKT1/PTEN altered tumors (NCT04305496). In this observed that 17% of the p16-high tumors from TCGA do not express context, our previously reported preclinical data suggests that PTEN low pRb, suggesting that p16 might independently predict CDK4/6i- alterations alone do not result in benefit to AKT inhibitors55. Here, we resistance. This phenotype might have been relevant in CDK4/6i- report that the combination of ribociclib with the PI3K inhibitor resistant tumors from the ABC-POP trial33 in which p16, but not pRb, alpelisib (or palbociclib with GDC-0032) has antitumor activity in non- was associated with resistance to abemaciclib. The identification of p16 basal-like PDXs independently of PIK3CA or ESR1 mutation, extending overexpression as biomarker of resistance to CDK4/6i is in contrast to the evidence from cell line-derived xenoimplant models11,56,57. Of note, the results observed in PALOMA-2 and MONALEESA-2 and could be PAM50-based intrinsic subtyping has become a potential indicator of due to the specific cut-offs applied or the specimens utilized50,51. benefit to CDK4/6i and our data highlights the continued translational However, our data is in line with recent obsevations showing that INK4 relevance of PAM50 which is not currently used in standard practice proteins such as p16 and p18 induce resistance in CDK6- for early-stage or metastatic breast cancer58–60. In addition, we overexpressing tumors14. Specifically, INK4 proteins preferentially demonstrated antitumor efficacy of the triple combination (CDK4/6i bind CDK6 over CDK4 and distort the ATP/drug binding pocket of plus PI3Ki and ET) in PDXs harboring RB1 mutation, which may be an CDK6 in a manner that binding of ATP is favored over palbociclib, appropriate first-line treatment strategy for patients harboring het- inducing drug resistance. These findings are relevant for the clinic erozygous RB1 loss. Finally, we provide rationale for de-escalation because multiple genetic alterations lead to CDK6-mediated resistance strategies, either by omission of ET or reduction of the PI3Ki/CDK4/ (e.g. FAT1, PTEN). A PROTAC drug to specifically degrade CDK6 was 6i dose. able to overcome CDK4/6i resistance. Our results further show that Altogether, this study identifies high p16 protein levels and het- targeting the p16-CDK4/6 interaction sensitizes p16-overexpressing erozygous RB1 loss as biomarkers of resistance to CDK4/6i treatment tumor cells to CDK4/6i. and suggests that CDK4/6i plus PI3Ki may be effective in non-basal-like Acquired or pre-existing RB1 mutations has also been asso- tumors that progress to CDK4/6i and ET, independently of the PIK3CA, ciated with resistance to ET plus CDK4/6i13,16,26. Wander et al. ESR1 or RB1 status. described one potential mechanism being the biallelic disruption in RB1 in a patient who progressed to ET plus palbociclib, and Methods harbored an RB1 heterozygous loss in her pre-treatment tumor Study design sample35. This evidence is in line with our results showing that This study was designed to identify predictive biomarkers of response most of the tumors from the ribociclib-sensitive PDX244, that to ribociclib that can be effectively used for patient stratification. We harbors an RB1 heterozygous loss, became resistant to CDK4/6 assessed ribociclib activity in a cohort of 37 patient-derived xenograft Nature Communications | (2022)13:5258 12 Article https://doi.org/10.1038/s41467-022-32828-6 models from primary/metastatic breast cancer patients. All animal whole antibody (NA931) from Sigma-Aldrich. All secondary anti- procedures were approved by the Ethics Committee of Animal bodies were diluted 1:2000. Ribociclib (LEE11) and alpelisib Research of the Vall d´Hebron Institute of Oncology and by the Catalan (BYL719) were provided by Novartis. Commercial trastuzumab Government (FUE-2020-01541918) and were conformed to the princi- (Herceptin) was obtained from a pharmacy. Fulvestrant and letro- ples of the WMA Declaration of Helsinki, the Department of Health and zole were purchased from Selleckchem. P18IN003 was purchased Human Services Belmont Report and following the European Union’s from Glixxlabs. Lenti ORF clone of human cyclin D1 (CCND1) mGFP animal care directive (2010/63/EU). Use of PDXs from other labora- tagged (RC204957L2) and Lenti-C-mGFP tagged empty vector tories was approved by the National Research Ethics Service, Cam- (PS100071) were purchased from Origene. pLX401-INK4A vector bridgeshire 2 REC (RED reference number: 08/H0308/178 and http:// was a gift from William Hahn (Addgene plasmid#121919; http://n2t. caldaslab.crik.cam.ac.uk/bcape/) or by the Central Office for Research net/addgene:121919;RRID: Addgene 121919). Ethics Committee study number 05/Q1402/25. For ethical issues, in vivo experiments were ended when the total tumor volume of a Cell lines mouse surpassed the maximal tumor size permitted by the ethics 293T (CRL-3216), MCF7 (HTB-22) and T47D (HTB-133) cell lines were committee, namely 1500 mm3, or a decline in mouse welfare was obtained from ATCC and maintained according to the manufacturer’s observed. Tumors were harvested and formaldehyde and flash-frozen instructions. Cell lines were authenticated utilizing short tandem for posterior proteomic and genomic analyses. repeat (STR) profiling (FTA Sample Collection Kit for Human Cell For PDX generation, we obtained fresh tumor samples from the Vall Authentication Service; ATCC services). The submitted profile was an d´Hebron University Hospital and following the institutional guidelines. exact match for the ATCC human cell lines in the ATCC STR database. Informed written patient consent, approved by the Ethics Committee Mycoplasma test was performed every 5 passages (MycoAlert™ for Clinical Research and Animal Research of Vall d’Hebron Hospital Mycoplasma Detection Kit; LONZA). Both MCF7 and T47D cell lines (PR(AG)130/2015), was obtained for the use of these patient samples. were grown in RPMI 1640 with GlutaMAX medium (Gibco) supple- To study the predictive value of cell cycle biomarkers, 72 patients mented with 10% of heat inactivated fetal bovine serum (Gibco) and of the 105 enrolled in the ABC-POP trial (NCT02831530) were subject to 1 nM of β-estradiol (Sigma-Aldrich), and HEK293T cell line was grown in analysis of pRb, p16, cyclin E1 and cyclin D1 in their tumors. This ana- DMEM (Gibco) supplemented with 10% of heat inactivated fetal bovine lysis was blinded to the study endpoint, namely antiproliferative serum (Gibco). Cell lines were banked in multiple aliquots on receipt to response defined as natural logarithm of Ki67 expression at day 15 < 1. reduce risk of phenotypic drift and were used for a maximum of 15 Briefly, untreated females aged 18 years or older who were diagnosed passages. All cells have been cultured at 37 °C with 5% CO2 atmosphere. with HR-positive, non-metastatic invasive breast carcinoma and signed an informed consent were selected. Patients received abemaciclib Generation of PDXs (150 mg twice daily for 14 days) or no treatment before surgery. FFPE Tumor pieces of 30–60 mm3 obtained from patient primary tumors or tumor samples were obtained at baseline and upon surgery for bio- metastatic lesions at time of biopsy were immediately implanted into marker analysis. Regarding the metastatic cohort, FFPE material the mammary fat pad (surgery samples) or the lower flank (metastatic representative of the disease was obtained from HR-positive meta- samples) of 6-week-old female NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ mice (M. static breast cancer patients aged 18 years or older who signed an IRB- musculus, Charles Rives). Mice were housed in air-filtered flow cabinets approved informed consent form and received treatment at the Vall with a 12-hours light cycle at 18–23 °C, 40–60% of humidity and food d’Hebron University Hospital with a CDK4/6 inhibitor as monotherapy and water ad libitum. Mice were continuously supplemented with 1 or in combination with HT. Patients were considered sensitive to the µmol/L 17β-estradiol (Sigma-Aldrich) in their drinking water, an treatment in the metastatic setting if a clinical benefit (defined by amount shown to be sufficient to reach serum levels and uterine RECIST criteria) was achieved and maintained for a period ≥ 6 months growth in ovariectomized female mice similar to the ones obtained and/or ≥ 10 cycles of treatment. pRb, p16, cyclin E1 and cyclin D1 were with other mechanism of 17β-estradiol supplementation. Upon growth analyzed in their tumors. of the engrafted tumors, a tumor piece was implanted into the lower flanks of new recipient mice for the model perpetuation. In each pas- Antibodies and reagents sage, flash-frozen and formalin-fixed paraffin embedded (FFPE) sam- Primary antibodies used for immunohistochemistry (IHC) were cyclin ples were taken for genotyping and histological studies. STG201 was D1 (RM9104, 1:100) from ThermoScientific; pRb (554136, 1:100) from generated in CRUK/UCAM as previously reported21 and PDXs BB3RC31, BD Pharmigen; phospho-pRb S807/811 (8516, 1:300) from Cell Signal- BB6RC39, BB6RC87, and BB6RC160 were generated in Manchester ing Technology; cyclin E1 (05-363, 1:300) from Millipore; p16 (725- Breast Center. Both laboratories are members of the EuroPDX 4713), ER (790-4324), PR (790-2223), Ki67 (790-4286) and HER2 (790- consortium. 2991) from Ventana Medical Systems, Roche. Primary antibodies used for immunofluorescence (IF) were Cytokeratin 18 (ab133263; 1:500) Molecular characterization of PDXs and Alexa Fluor® 568 anti-Vimentin (ab202504; 1:500) both from For molecular subtyping, immunohistochemical staining was per- Abcam and secondary antibody was Alexa Fluor® 488 goat anti-rabbit formed on formalin-fixed paraffin embedded (FFPE) PDXs tissue sec- IgG (A48282, 1:2000). Primary antibodies used for Western blot were tions (3 μm). Staining of estrogen receptor (ER), progesterone CDK4 (12790), cyclin D2 (3741), phospho-pRb S807/811 (9308), receptor (PR) and human epidermal growth factor receptor 2 (HER2) phospho-pRb S780 (9307), phospho-CDK2 T160 (2561), phospho-AKT were undertaken following the protocol provided by Ventana Medical T308 (2965), phospho-AKT S473 (9271), AKT (9272), PARP (9542) and Systems, Inc. In short, the slides were heated in the oven at 75 °C for FGFR1 (3472) from Cell Signaling Technology; CDK6 (ab124821), cyclin 28 min and deparaffinized with EZ prep solution (Ventana Medical D1 (ab40754) and cyclin E2 (ab40890) from Abcam; Tubulin (T-9026) Systems). Then, antigen retrieval was performed at slightly basic pH at from Sigma; cyclin E1 (sc-481), CDK2 (sc-163) and, human GAPDH 95 °C for 56 min. Primary antibodies were incubated for 40 min for ER (sc137179) from Santa Cruz Biotechnology; p16-INKA (10883-I-AP) and HER2 using the Cell Conditioning 1 buffer (CC1; Ventana Medical from ProteinTech; pRb (554136) from BD Pharmigen; ER-alpha (MS- Systems), and with CC2 buffer for the PR antibody. Finally, the slides 315-P0) from Neomarkers. All primary antibodies were diluted 1:1000 were counterstained with Hematoxylin II and Bluing Reagent (Ventana except for human GAPDH and Tubulin that were diluted 1:5000. Sec- Medical Systems) and mounted with xylol based mounting medium. ondary antibodies used for Western blot were goat anti-rabbit IgG HRP An investigator blinded to identify the samples quantified the per- linked whole antibody (NA934) and goat anti-mouse IhG HRP linked centage of positively stained cells. Nature Communications | (2022)13:5258 13 Article https://doi.org/10.1038/s41467-022-32828-6 Flash-frozen pieces of tumor xenograft were used for RNA The tumor volume was calculated using the ellipsoid formula: sequencing and PAM50-molecular subtype classification. All the tumor V = (length × width2) × (π/6). Mice were euthanized when tumors samples used in this study were pieces of patient-derived xenografts. reached 1500 mm3 or in case of severe weight loss, in accordance with After surgical resection, the tumors were dissected, and a piece was institutional guidelines. All the efficacy experiments contained an quickly frozen in liquid nitrogen and stored at −80 °C. A frozen tumor untreated control arm with a percentage of change in tumor growth specimen was then homogenized in RNAse-free containing lysis buffer superior to 20% from the initial volume. The antitumor activity was and mRNA was prepared by using a PerfectPure RNA Tissue Kit-50 determined by comparing tumor volume at last day of treatment to its from 5 Prime and protocol. 250 ng of total RNA were used to measure baseline (day1): % tumor volume change = (V35days –Vbaseline)/Vbaseline the expression of 50 genes of the PAM50 intrinsic subtype predictor x100. The antitumor response of subcutaneous implants was classified assay and 5 housekeeping genes (ACTB, MRPL19, PSMC4, RPLP0, and according to the relative change in tumor volume upon treatment SF3A1) using the nCounter platform (NanoString Technologies). Data (similar to the Response Evaluation Criteria in Solid Tumors (RECIST) was log base2-transformed and normalized to the housekeeping genes and labeled as mRECIST20,22. Complete response (CR) was set as best using the nSolver 4.0 software and the script developed by61 R (3.4.3) response ≤ −95%; partial response (PR) as −95% +20%. The models that displayed a like or Normal-like) using the PAM50 subtype predictor. preclinical benefit from ribociclib (SD, PR, and CR) were categorized as Flash-frozen pieces of tumor xenografts were used for targeted ribociclib-sensitive. All PD models were categorized as ribociclib- exome sequencing by the MSK-IMPACTTM (Integrated Mutation Pro- resistant. At the end of the experiment, animals were euthanized using filing of Actionable Cancer Targets), a hybridization capture-based CO2 inhalation. Tumor volumes are plotted as mean values ± SEM. next-generation sequencing assay for targeted deep sequencing Ribociclib-sensitive models were chronically treated with ribociclib designed to capture all protein-coding exons and selected introns of until progression to acquire ribociclib-resistance. Tumor growth was 410 commonly implicated oncogenes, tumor suppressor genes, and measured once per week and mice weights were recorded twice per members of pathways deemed actionable by targeted therapies25. week. If mouse welfare was compromised before tumor progression, Barcoded sequence libraries were prepared using 100–250 ng geno- tumors were harvested and implanted into another recipient mouse. mic DNA (Kapa Biosystems) and combined into equimolar pools of Dosing schedule was reinitiated 10 days post-surgery and lasted until 13–21 samples. The captured pools were subsequently sequenced on progression. an Illumina HiSeq 2000 as paired-end 100-base pair reads, producing a median of 588-fold coverage per tumor. Sequence data were demul- PDC ex vivo cultures tiplexed using CASAVA, and reads were aligned to the reference Patient-derived tumor cells (PDC) were isolated from PDX through human genome (hg19) using BWA and post-processed using the Gen- combination of mechanic disruption and enzymatic disaggregation21. ome Analysis Toolkit (GATK) according to GATK best practices. Briefly, PDX tumors not bigger that 500mm3 were freshly collected in MuTect and GATK were used to call single-nucleotide variants and DMEM/F12/HEPES (GIBCO) after surgery resection, minced using small indels, respectively. Candidate mutations were manually sterile scalpels and dissociated for a maximum of 90 min in DMEM/ reviewed using the Integrative Genomics Viewer (IGV) to eliminate F12/HEPES (GIBCO), 1 mg/ml collagenase (Roche), 100 u/ml, hyalur- likely false positive calls. Because matched normal DNA was not onidase (Sigma-Aldrich), 5% BSA (Sigma-Aldrich), 5 µg/ml Insulin and available, tumors were compared to a pool of unmatched normal 50 µg/ml gentamycin (GIBCO). This was followed by further dissocia- samples to eliminate common polymorphisms and systematic tion using trypsin (GIBCO), dispase 5 mg/ml (StemCell technologies) sequencing artifacts. and DNase 1 mg/ml (Sigma-Aldrich). Red blood cell lysis was done by washing the cell pellet with 1X Red Blood Cell (RBC) Lysis Buffer con- In vivo experiments taining ammonium chloride (Invitrogene). Then, cells were resus- To evaluate the sensitivity to the different targeted therapies each PDX pended in MEGM™ Mammary Epithelial Cell Growth Medium was implanted subcutaneously in six-week-old female athymic nude Bulletkit™ (LONZA) supplemented with 2% of fetal bovine serum and HsdCpb:NMRI-Foxn1nu mice (M. musculus, Janvier) or NOD.Cg- 10 µM of ROCK inhibitor (Sigma-Aldrich). To test drug antiproliferative PrkdcscidIl2rgtm1Wjl/SzJ mice (M. musculus, Charles Rives) and supple- responses and for immunoblotting analysis, cells were seeded on mented with 1 µmol/L 17β-estradiol (Sigma-Aldrich) in their drinking collagen-enriched matrix Corning® Matrigel® growth factor reduced water. Mice were housed at 18–23 °C with 40–60% of humidity and 12 (GFR) basement membrane matrix (Corning, INC) at 2 × 105 cells/ml in light/12 dark cycle. Food and water were accessible all time. 8 well-chamber slides (NUNC) or 1 × 106 cells/ml in 6 well plates (BD Upon xenograft growth, tumor-bearing mice were randomized into Biosciences), respectively. For antiproliferative analysis, PDCs were treatments group with tumors ranging 100–300 mm3 (for drug effi- treated with vehicle (DMSO), 1 µM of ribociclib, 500 nM palbociclib, cacy experiments) or ∼500 mm3 (for short-term pharmacodynamic 2.5 µM alpelisib, 100 nM fulvestrant, 20 µM P18IN003 or the combina- experiments). Ribociclib was administrated by oral gavage once daily, tions and cultured at 37 °C in 5% of CO2. Medium and treatments were six days/week, at 75 mg/kg (total daily dose) dissolved in distilled water refreshed every 2–3 days. For immunoblotting analysis, cells were 0.5% hydroxymethyl cellulose. Alpelisib was dosed with the same treated for 24 h. Matrigel® was melted in PBS-EDTA 1 mM on ice for schedule at 35 mg/kg dissolved in distilled water 0.5% methylcellulose. 20 min, the spheroids were collected into a conical tube and cen- The combination was administrated with one-hour delay between trifuged at 450 × g for 5 min at 4 °C. Pellets were stored at −20 °C until ribociclib (first) and alpelisib (second). Fulvestrant was administered protein lysates were prepared for immunoblotting analysis subcutaneously twice weekly 10 mg/mice dissolved in peanut-oil, (see below). letrozole by oral gavage three times per week (1 day on and 1 day off) 20 mg/kg dissolved 0.5% methylcellulose and trastuzumab by intra- Analysis of PDCs area peritoneal injection twice weekly 10 mg/kg in PBS. Cell suspensions generated from a 500 mm3 PDX were plated in Tumor growth was measured bi-weekly blinded to the treatment duplicated at 60.000 cells/well into 8 well-chambers slides. Drugs and effect with a caliper the first day of treatment and to day 35 (for the vehicle (DMSO) were added after 24 h. To quantify the drug response efficacy assays of ribociclib, alpelisib and their combination), day 15 in PDCs, representative bright field pictures of each well were obtained (for the efficacy assays of fulvestrant and trastuzumab) or day 12 7 days post-treatment and normalized against untreated (vehicle). A (pharmacodynamic assays). Mice weights was recorded twice weekly. minimum of three different biological replicates (different tumors) Nature Communications | (2022)13:5258 14 Article https://doi.org/10.1038/s41467-022-32828-6 from each model were assayed. For bright field images analysis ImageJ triplicates at 300 K using standard protocols63. Simulation trajectories (http://rsb.info.nih.gov/ij/) was used. Two representative areas of sin- were visualized using VMD (1.9.1) and figures were generated using gle spheroids were manually quantified individually from at least two Pymol (2.3.2). independent wells. The mean spheroid area for every treatment was calculated and normalized to untreated controls (vehicles). Relative Immunohistochemistry (IHC) and image analysis mean spheroid areas for every treatment condition and the ± SEM were PDX and patient tumors were fixed immediately after excision in 10% plotted. buffered formalin solution for a maximum of 24 h at room tempera- ture before being dehydrated and paraffin embedded (FFPE). IHC was Analysis of S-phase entry cells by EdU incorporation performed on FFPE tissue sections (3 µM). For pRb, cyclin D1 and cyclin Cell suspensions generated from a 500mm3 tumors were plated in E1, sections were dewaxed, rehydrated and antigen retrieved using a duplicated at 60,000 cells/well into 8 well-chambers slides. After 24 h, microwave at maximum power (Whirlpool JT479/WH) in EDTA 1 mM drugs and vehicle (DMSO) as well as 10 µM of 8-ethynyl–2´-deoxyur- pH 8 (pRb), citrate pH 6 (cyclin D1) or pH 9 (cyclin E1) for 20 min. After idine (EdU) were added and the cells were incubated for 2 days. EdU peroxide blocking, the slides were stained with the corresponded staining was performed using the Click-iT™ EdU Alexa Fluor™ 488 primary antibodies at dilution 1:100 (pRb and cyclin D1) or 1:300 Imaging kit (ThermoFisher Scientific) adapting the manufacturer’s (cyclin E1), then with ready-to-use Mouse (pRb and cyclin E1) or Rabbit instructions. Briefly, the cells were fixed with 3.7% paraformaldehyde (cyclin D1) EnVisionTM + System-HRP labeled polymer and finally with for 15 min and permeabilized with 1% Triton X-100 for 20 min, all at Liquid DAB + Substrate Chromogen System (DAKO). Harris´ hema- room temperature. After 1 h of 5% BSA in PBS blocking, cells were toxylin was used to counterstain the nuclei and mounted with xylol incubated with the Click-iT™ reaction cocktail and primary antibodies based mounting medium. Positive and negative controls were run (mouse Vimentin 1:500 or human CK18 1:100) overnight at room along the tested slides per each marker. temperature. The following day, cells were washed 3x with 3% BSA in Staining of p16, phospho-pRb (Ser807/811), and Ki67 was under- PBS and incubated with secondary antibodies for 1 h at room tem- taken following the protocol provided by Ventana Medical System, Inc. perature. Finally, cells were washed 3x with 3% BSA in PBS, mounted Briefly, the slides were heated in the instrument at 75 °C for 28 min and with Prolong™ Antifade Reagent Mountant with DAPI (Molecular deparaffinized with EZ prep solution (Ventana Medical System). Then, Probes) and stored at −20 °C until analysis. the antigen retrieval was performed at slightly basic pH at 95 °C using Confocal microscopy analysis was carried out using the Nikon the Cell Conditioning 1 buffer (CC1; Ventana Medical System) followed confocal microscope C2+equipped with LU-N4S laser unit and the NIS- by staining with anti-rabbit HQ (KI67; 1:500) or anti-mouse HR (p16 and Elements software (5.10) was used for capturing representative images phospho-pRb (Ser807/811) ready-to-use and 1:600, respectively) and of spheroids. Number of both DAPI positive and EdU positive cells in anti-HQ-HRP and DAB staining (Roche). The slides were counter- each spheroid was manually obtained using ImageJ 1.51 (http://rsb.info. stained with Hematoxylin II and bluing reagent (Ventana Medical nih.gov/ij/). The percentage of Edu positive cells per spheroid was System) and mounted with xylol based mounting medium. calculated and the mean of every treatment was relativized to the A pathologist scored the different proteins expression in each untreated (vehicle). Relative percentage of S-phase entry cells and ± sample. Total pRb and p16 were scored semi-quantitatively onto life SEM were plotted. images with very strong (4+), strong (3+), moderate (2+) weak (1+) or negative staining (0). Allred scores of cyclin D1 and cyclin E1 were Molecular modeling of the complex between p16 and P18IN003 calculated from 0 to 8 taking into account the percentage of positive We constructed a structural model of the complex between p16 and a cells (0 to 5) plus the staining intensity (0 to 3). The percentage of cells known inhibitor P18IN003 using computational methods of docking with nuclear Ki67 or phospho-pRb (Ser807/811) staining was quantified and molecular dynamics simulations. p16 has four Ankryin repeats and in samples at baseline and after ribociclib treatment from pharmaco- we used the only available apo structure of p16, an NMR structure dynamic experiments. (PDB: 2A5E), which is very similar to the crystal structure of p16 bound to CDK6 (PDB: 1BI7, root mean squared deviation of <1 Å, confined In vitro cell line assessment largely to the loop regions). An homology model62 was constructed to For half-maximal inhibition concentration (IC50) analysis, cells (2 × 103/ model the interactions between p16 and CDK4 based on the p16-CDK6 well) were seeded into 96-well plates (BD Bioscience) and, after 24 h, crystal structure, given that similarity between CDK4 and CDK6 is 81%. were treated for 6 days with different concentrations of ribociclib, This crystal structure was then used to identify the region of interac- fulvestrant or the combination of drugs. The treatments and media tion between p16 and CDK4 and this region was used to define a were refreshed every 3 days. Cell proliferation was measured at day 0 binding pocket on the surface of the NMR structure of apo p16 (a and day 6 by fixing with 4% glutaraldehyde (MERCK) in PBS, staining similar method was used to identify inhibitors of p1831) to which the with 0.1% of crystal violet (Sigma-Aldrich) in methanol, solubilizing inhibitor P18IN003 was docked. For docking, the 3D structure of the with 10% of acetic acid in PBS and measuring the absorbance of each inhibitor P18IN003 was built using the Maestro module and minimized well at 560 nm. Values at day 6 were normalized with values at day 0, using the Macromodel module, employing the OPLS-2005 force field, relativized to controls (vehicle-treated cells) and plotted as the per- in the program Schrodinger 12.0. The minimized P18IN003 inhibitor centage inhibition against the log concentration of ribociclib. IC50 was was docked into the binding pocket of p16 defined above with the determined using a sigmoidal regression model and was defined as the program Glide using standard docking protocols63. Out of the top 10 concentration of drug required for a 50% reduction in growth. Each lowest energy poses of the binding of P18IN003 to p16, 8 poses of the experiment was repeated three times with three technical replicates. inhibitor were very similar to each other and so we chose the top pose For biochemical analysis, cells (1.5 × 106/well) were seeded into and subject the complex to further refinement using molecular p100 dishes (BD Bioscience) and, the following day, were treated for dynamics (MD) simulations. The simulations were carried out using the 24 h or 5 days with 0.5 μM, 1 μM or 1.5 μM ribociclib, 100 nM fulves- Amber (18) program, using protocols that we have shown to be suc- trant or the combination of drugs. Next, cells were harvested and cessful in previous studies63. The partial charges and force field para- prepared for immunoblotting analysis (see below). T47D-p16 cells meters for P18IN003 were generated using the Antechamber module were incubated with 1 µg/ml doxycycline 48 h prior to add the treat- in Amber. All atom versions of the Amber ff14SB force field and the ments for inducing p16 expression. general Amber force field (GAFF) were used for the protein and the For competition experiments cells (5.0 × 105/well) were seeded inhibitors respectively. Simulations were carried out for 100 ns in into p100 dishes (BD Bioscience). 5% of MCF7- or T47D- cells Nature Communications | (2022)13:5258 15 Article https://doi.org/10.1038/s41467-022-32828-6 overexpressing p16 or cyclin D1 (2.5 × 104/well) were mixed with 95% of Statistical analysis MCF7- or T47D-MOCK cells (4.75 × 105/well), seeded together and, the GraphPad Prism (6.00) for Windows was used for statistical analysis. A following day, were treated with 1 μM ribociclib for 0, 5 or 14 days. Cells bootstrap resampling procedure (n = 2000) was used to calculate the were harvested and prepared for immunoblotting analysis (see below). standard error in the percentage of change in tumor volume relative to untreated. D’Agostino-Pearson omnibus test was performed to check Immunoblotting the normality assumption in all comparative studies. If the null Both flash-frozen tumor pieces from pharmacodynamic assays and hypothesis of normality was not rejected, we assumed Gaussian dis- harvested cells were lysed in ice-cold buffer containing Tris-HCl pH7.8 tribution of the samples, but if the sample size was too small or the 20 mmol/L, NaCl 137 mmol/L, EDTA pH 8.0 2 mmol/L, NP40 1%, gly- hypothesis was rejected, we did not assume it. cerol 10%, supplemented with NaF 10 mmol/L, Leupeptin 10 mg/mL, For the comparative experiments of biomarkers between riboci- Na2VO4 200 mmol/L, PMSF 5 mmol/L, and Aprotinin (Sigma-Aldrich). clib and vehicle groups, we used paired t-test or Wilcoxon signed-rank Tissue homogenization was performed on ice with a POLYTRON® test, as appropriate after checking normality assumption. For the system PT 1200 E (Kinematica). Lysates were centrifuged at 18,000 × g comparative between ribociclib-sensitive (S) and -resistant (R) sam- 4 °C for 10 min and the supernatants were collected. Protein con- ples, we used an unpaired t-test or Mann-Whitney U test, as appro- centration was calculated using DCTM Protein Assay (Bio-Rad). A total priate after checking normality assumption. Adjustment for multiple of 30 µg of protein were separated on 12% SDS-PAGE acrylamide gels at testing was performed in each biomarker by controlling the false dis- 100 V and transferred to nitrocellulose membrane for 1.5 h at 100 V. covery rate at 5% according to the Benjamini and Hochberg method. Membranes were blocked for 1 h in 5% milk in Tris-buffered saline Univariate logistic models were used to obtain odds ratios (OR) of (TBS)-Tween and then hybridized using the corresponding primary studied biomarkers in PDXs. To quantify the level of association antibodies in 5% BSA TBS-Tween. All primary antibodies were used at between a qualitative factor and response we calculated accuracy, dilution 1:1000. Mouse and rabbit horseradish peroxidase (HRP)-con- sensitivity, specificity, positive and negative predictive values (PPV and jugated secondary antibodies (1:2000; GE Healthcare) were diluted in NPV). The optimum cut-off points established (p16 high ≥ 2, pRb low ≤ 5% milk in TBS-Tween and proteins were detected with Immobilon 2 and Cyclin E1/D1 high > 4/6) in this study were selected by the Youden Western Chemiluminescent HRP substrate (Millipore). Images were index, which maximizes the sum of the sensitivity and specificity in captured with FUJIFILM LASS-4000 camera system and quantified with each biomarker analyzing the ROC curve. ImageJ (1.51). Uncropped and unprocessed scans of the most impor- For the ABC-POP trial we performed a logistic regression of bio- tant blots are supplied in the Source Data file. markers and KI67 percentage, univariate logistic regressions to esti- mate the odds ratio for a standard deviation change in continuous Lentiviral infection H-score biomarkers levels of p16, pRb and cyclin D1 and absolute KI67 Lentiviral infection was done following the manufacturer ś indications response and a multivariable logistic regression analysis. For the (Sigma-Aldrich). Briefly, 293FT cells were used for the production of comparative between ribociclib-sensitive (S) and -resistant (R) meta- the virus. 293FT cells (5 × 106) were transfected with lentivirus and static patient samples, we used an unpaired t-test or Mann-Whitney U packaging (gag-pol, vsvg, rev) plasmids (Addgene) using poly- test, as appropriate after checking normality assumption. We also ethyleneimine (Sigma-Aldrich) in a DNA-PEI ratio of 1:3. Viral produc- applied univariate logistic regressions to estimate the odds ratio. The tion was induced by adding 10 mM Na Butyrate the following day. Virus analysis was performed with R (4.0.3). were harvested 48 h post transfection. p16-INK4A and cyclin D1 lenti- For analyzing the correlation between double genetic hit in RB1 viral stocks were tittered using colony forming assay (Hela cells). MCF7 locus and treatment with CDK4/6i, genomic data of metastases from and T47D cells were infected with doxycycline-inducible pLX401- 800 BC was obtained from the Hartwig Medical Foundation (HMF36); INK4A (MOI 1:2) for overexpressing p16 or pLenti-CCND1 (MOI 1:2) and (DR-110). Patients with ‘None’, ‘Other’ or ‘Unknown’ treatment were the control plasmid (pLenti-tGFP) for overexpressing cyclin D1. 8 µg/ml filtered out, yielding 582 patients who received known treatments. All of polybrene (Sigma-Aldrich) were added, plates were centrifuged 1 h metastatic samples bearing single base substitutions and indels caus- at 450 × g at 37 °C to improve the infection and incubated overnight. ing frameshift variants, stop gained variants, splice acceptor variants, For p16-INK4 cell selection, 2 µg/ml of puromycin (Invitrogene) were splice donor variants, start lost variants, stop lost variants, missense added to the cultured media 48 h post-infection. After 5 days, all cells variants, inframe deletions or inframe insertions affecting the RB1 gene in the control plate (non-infected cells) were dead and the con- were identified. Metastatic samples with minor and major allele copy centration of puromycin was reduced to 1 µg/ml (maintenance dose). number of the genomic region containing the RB1 locus smaller than For p16 expression, 1 µg/ml of doxycycline (Sigma-Aldrich) was added 0.01 were deemed to carry homozygous deletion of the RB1 gene, to the culture media 48 h before treatments were added. while those with minor allele copy number of this genomic region smaller than 0.01, but greater major allele copy number were deemed Real Time-qPCR ready custom panels to carry a heterozygous deletion of RB1. We used multivariable logistic RNA was extracted from flash-frozen control and ribociclib-treated regression to assess the association between the alteration status of PDX samples (15–30 mg) by using the PerfectPure RNA Tissue kit (five the metastatic breast tumor and the likelihood probability that the Prime). The purity and integrity were assessed by the Agilent 2100 patient received CDK4/6i. Two logistic regression formulas modeling Bioanalyzer system, and cDNA was obtained using the PrimeScript RT different interactions between the variables (mutation and deletion) Reagent kit (Takara). Quantitative RT-PCR was performed in the were applied. LightCycler® 480 (Roche) using LightCycler® 480 Probes Master (Roche) and ready-to-use custom 384-plates panels containing pre- Reporting summary designed human specific primers and UPL Probes for each gene Further information on research design is available in the Nature (Supplementary Table 5). Tbd probes were designed and tested spe- Research Reporting Summary linked to this article. cifically for this assay. The comparative CT method was used for data analysis, in which geoNorm algorithms were applied to select the most Data availability stably expressed housekeeping genes (GAPDH and ACTB) and geo- The annotated DNA sequencing data of the PDX models generated in metric mean values were calculated to obtain normalized CT values64. this study is provided in the Source Data file. Access to the raw qPCR assay IDs are shown in Supplementary Table 5. sequencing data can be made available for academic research only Nature Communications | (2022)13:5258 16 Article https://doi.org/10.1038/s41467-022-32828-6 from Dr Violeta Serra (vserra@vhio.et) and after completing a Data 16. Condorelli, R. et al. Polyclonal RB1 mutations and acquired resis- Transfer Agreement with Memorial Sloan-Kettering Cancer Center tance to CDK 4/6 inhibitors in patients with metastatic breast can- (New York, USA). Genomic and clinical data of metastatic breast cer. Ann. Oncol. 29, 640–645 (2018). cancer patients from the Hartwig Medical Foundation (HMF) cohort 17. Turner, N. C. et al. Cyclin E1 expression and palbociclib efficacy in and ABC-POP trial analyzed used in this manuscript were previously previously treated hormone receptor-positive metastatic breast published and requested to the corresponding main authors [doi: cancer. J. Clin. Oncol. 37, 1169–1178 (2019). 10.1038/s41586-019-1689; 10.1016/j.annonc.2020.08.283]. Clinical 18. Andre, F. et al. Alpelisib for PIK3CA-mutated, hormone receptor- data generated in this study from metastatic breast cancer patients positive advanced breast cancer. N. Engl. J. Med. 380, treated at Vall d´Hebron Hospital are provided in the Source Data 1929–1940 (2019). file. Clinical data shown in Fig. 6C and supplementary 2C of this 19. Pascual, J. et al. Triplet therapy with palbociclib, taselisib, and ful- manuscript were obtained from cBioportal [https://www.cbioportal. vestrant in PIK3CA-mutant breast cancer and doublet palbociclib org/]. The 3D protein and compound structures employed in this and taselisib in pathway-mutant solid cancers. Cancer Discov. 11, manuscript are deposited in the PDB (Protein Data Bank) archive: 92–107 (2021). 2A5E, 1BIN and P18IN003 [https://pubchem.ncbi.nlm.nih.gov/ 20. Gao, H. et al. High-throughput screening using patient-derived compound/9994705]. Source data are provided with this paper. tumor xenografts to predict clinical trial drug response. Nat. Med. 21, 1318–25 (2015). References 21. Bruna, A. et al. A biobank of breast cancer explants with preserved 1. Hortobagyi, G. N. et al. Ribociclib as first-line therapy for HR-posi- intra-tumor heterogeneity to screen anticancer compounds. Cell tive, advanced breast cancer. N. Engl. J. Med. 375, 167, 260–274 e22 (2016). 1738–1748 (2016). 22. Therasse, P. et al. New guidelines to evaluate the response to 2. Finn, R. S. et al. Palbociclib and letrozole in advanced breast cancer. treatment in solid tumors. European Organization for Research and N. Engl. J. Med. 375, 1925–1936 (2016). Treatment of Cancer, National Cancer Institute of the United States, 3. Goetz, M. P. et al. MONARCH 3: abemaciclib as initial therapy for National Cancer Institute of Canada. J. Natl Cancer Inst. 92, advanced breast cancer. J. Clin. Oncol. 35, 3638–3646 205–16 (2000). (2017). 23. Cejalvo, J. M. et al. Intrinsic subtypes and gene expression profiles 4. Dickler, M. N. et al. MONARCH 1, a phase II study of abemaciclib, a in primary and metastatic breast cancer. Cancer Res. 77, CDK4 and CDK6 inhibitor, as a single agent, in patients with 2213–2221 (2017). refractory HR(+)/HER2(−) metastatic breast cancer. Clin. Cancer Res. 24. Finn, R. S. et al. PD 0332991, a selective cyclin D kinase 4/6 inhibitor, 23, 5218–5224 (2017). preferentially inhibits proliferation of luminal estrogen receptor- 5. Johnston, S. R. D. et al. Abemaciclib combined with endocrine positive human breast cancer cell lines in vitro. Breast Cancer Res. therapy for the adjuvant treatment of HR+, HER2−, node-positive, 11, R77 (2009). high-risk, early breast cancer (monarchE). J. Clin. Oncol. 38, 25. Cheng, D. T. et al. Memorial sloan kettering-integrated mutation 3987–3998 (2020). profiling of actionable cancer targets (MSK-IMPACT): a hybridiza- 6. Tolaney, S.M. et al. Abemaciclib plus trastuzumab with or without tion capture-based next-generation sequencing clinical assay for fulvestrant versus trastuzumab plus standard-of-care chemother- solid tumor molecular oncology. J. Mol. Diagn. 17, 251–64 (2015). apy in women with hormone receptor-positive, HER2-positive 26. O’Leary, B. et al. Circulating tumor DNA markers for early progres- advanced breast cancer (monarcHER): a randomised, open-label, sion on fulvestrant with or without palbociclib in ER+ advanced phase 2 trial. Lancet Oncol. 21, 763–775 (2020). breast cancer. J. Natl Cancer Inst. 113, 309–317 (2021). 7. Ciruelos, E. et al. Palbociclib and trastuzumab in HER2-positive 27. Serrano, M., Hannon, G. J. & Beach, D. A new regulatory motif in cell- advanced breast cancer: results from the phase II SOLTI-1303 cycle control causing specific inhibition of cyclin D/CDK4. Nature PATRICIA trial. Clin. Cancer Res. 26, 5820–5829 (2020). 366, 704–7 (1993). 8. Herrera-Abreu, M. T. et al. Early adaptation and acquired resistance 28. Shapiro, G. I. et al. Reciprocal Rb inactivation and p16INK4 to CDK4/6 inhibition in estrogen receptor-positive breast cancer. expression in primary lung cancers and cell lines. Cancer Res. 55, Cancer Res. 76, 2301–13 (2016). 505–9 (1995). 9. Formisano, L. et al. Aberrant FGFR signaling mediates resistance to 29. Dublin, E. A. et al. Retinoblastoma and p16 proteins in mammary CDK4/6 inhibitors in ER+ breast cancer. Nat. Commun. 10, carcinoma: their relationship to cyclin D1 and histopathological 1373 (2019). parameters. Int J. Cancer 79, 71–5 (1998). 10. Costa, C. et al. PTEN loss mediates clinical cross-resistance to 30. Arnedos, M. et al. Modulation of Rb phosphorylation and anti- CDK4/6 and PI3Kalpha inhibitors in breast. Cancer Cancer Discov. proliferative response to palbociclib: the preoperative-palbociclib 10, 72–85 (2020). (POP) randomized clinical trial. Ann. Oncol. 29, 1755–1762 (2018). 11. Jansen, V. M. et al. Kinome-wide RNA interference screen reveals a 31. Gao, Y. et al. Small-molecule inhibitors targeting INK4 protein role for PDK1 in acquired resistance to CDK4/6 inhibition in ER- p18(INK4C) enhance ex vivo expansion of haematopoietic stem positive breast cancer. Cancer Res. 77, 2488–2499 (2017). cells. Nat. Commun. 6, 6328 (2015). 12. Yang, C. et al. Acquired CDK6 amplification promotes breast cancer 32. Green, J. L. et al. Direct CDKN2 modulation of CDK4 alters target resistance to CDK4/6 inhibitors and loss of ER signaling and engagement of CDK4 inhibitor drugs. Mol. Cancer Ther. 18, dependence. Oncogene 36, 2255–2264 (2017). 771–779 (2019). 13. Li, Z. et al. Loss of the FAT1 tumor suppressor promotes resistance 33. Arnedos, M. et al. Randomized preoperative window of opportinity to CDK4/6 inhibitors via the hippo pathway. Cancer Cell 34, (WOO) study with the CDK4/6 inhibitor abemaciclib in early breast 893–905 e8 (2018). cancer (EBC) patients and differential gene expression pathway 14. Li, Q. et al. INK4 tumor suppressor proteins mediate resistance to analysis with palbociclib. Ann. Oncol. 31, S303–S339 (2020). CDK4/6 kinase inhibitors. Cancer Discov. 12, 356–371 (2021). 34. Dowsett, M. et al. Short-term changes in Ki-67 during neoadjuvant 15. O’Leary, B. et al. The genetic landscape and clonal evolution of treatment of primary breast cancer with anastrozole or tamoxifen breast cancer resistance to palbociclib plus fulvestrant in the alone or combined correlate with recurrence-free survival. Clin. PALOMA-3 Trial. Cancer Discov. 8, 1390–1403 (2018). Cancer Res. 11, 951s–8s (2005). Nature Communications | (2022)13:5258 17 Article https://doi.org/10.1038/s41467-022-32828-6 35. Wander, S. A. et al. The genomic landscape of intrinsic and acquired receptor 2-negative (HER2-) advanced breast cancer (ABC): efficacy resistance to cyclin-dependent kinase 4/6 inhibitors in patients with by baseline tumor markers. Cancer Res. 77, CT045 (2017). hormone receptor-positive metastatic breast. Cancer Cancer Dis- 52. Coschi, C. H. et al. Haploinsufficiency of an RB-E2F1-Condensin II cov. 10, 1174–1193 (2020). complex leads to aberrant replication and aneuploidy. Cancer 36. Priestley, P. et al. Pan-cancer whole-genome analyses of metastatic Discov. 4, 840–53 (2014). solid tumours. Nature 575, 210–216 (2019). 53. Gong, X. et al. Aurora A kinase inhibition is synthetic lethal with loss 37. Bosch, A. et al. PI3K inhibition results in enhanced estrogen of the RB1 tumor suppressor gene. Cancer Discov. 9, receptor function and dependence in hormone receptor-positive 248–263 (2019). breast cancer. Sci. Transl. Med 7, 283ra51 (2015). 54. Gelbert, L. M. et al. Preclinical characterization of the CDK4/6 38. Rugo, H. S. et al. BYLieve: A phase II study of alpelisib (ALP) with inhibitor LY2835219: in-vivo cell cycle-dependent/independent fulvestrant (FUL) or letrozole (LET) for treatment of PIK3CA mutant, anti-tumor activities alone/in combination with gemcitabine. Invest. hormone receptor-positive (HR+), human epidermal growth factor N. Drugs 32, 825–37 (2014). receptor 2-negative (HER2–) advanced breast cancer (aBC) pro- 55. Gris-Oliver, A. et al. Genetic alterations in the PI3K/AKT pathway and gressing on/after cyclin-dependent kinase 4/6 inhibitor (CDK4/6i) baseline AKT activity define AKT inhibitor sensitivity in breast cancer therapy. J. Clin. Oncol. 36, TPS1107–TPS1107 (2018). patient-derived xenografts. Clin. Cancer Res. 26, 3720–3731 39. Chia, S. et al. Effect of duration of prior cyclin-dependent kinase 4/6 (2020). inhibitor (CDK4/6i) therapy (≤6 mo or >6 mo) on alpelisib benefit in 56. Vora, S. R. et al. CDK 4/6 inhibitors sensitize PIK3CA mutant breast patients with hormone receptor-positive (HR+), human epidermal cancer to PI3K inhibitors. Cancer Cell 26, 136–49 (2014). growth factor receptor 2-negative (HER2-), PIK3CA-mutated 57. O’Brien, N. A. et al. Targeting activated PI3K/mTOR signaling advanced breast cancer (ABC) from BYLieve. In: SABCS Annual overcomes acquired resistance to CDK4/6-based therapies in pre- Meeting P1-18-08 (2021). clinical models of hormone receptor-positive breast cancer. Breast 40. Juric, D. et al. Alpelisib+endocrine therapy (ET) in patients with Cancer Res. 22, 89 (2020). hormone receptor-positive (HR+), human epidermal growth factor 58. Ma, C. X. et al. NeoPalAna: neoadjuvant palbociclib, a cyclin- receptor 2-negative (HER2-), PIK3CA-mutated advanced breast dependent kinase 4/6 inhibitor, and anastrozole for clinical stage 2 cancer (ABC) previously treated with cyclin-dependent kinase 4/6 or 3 estrogen receptor-positive breast cancer. Clin. Cancer Res. 23, inhibitor (CDK4/6i): Biomarker analyses from the Phase II BYLieve 4055–4065 (2017). study. In: SABCS Annual Meeting P5-13-03 (2021). 59. Finn, R.S. et al. Comprehensive gene expression biomarker analysis 41. Juric, D. et al. Phase Ib/II study of ribociclib and alpelisib and of CDK 4/6 and endocrine pathways from the PALOMA-2 study. letrozole in ER+, HER2– breast cancer: Safety, preliminary efficacy Cancer Res. 78, P2-09-10 (2018). and molecular analysis. Cancer Res. P3-14-01 (2016). 60. Prat, A. et al. Correlative biomarker analysis of intrinsic subtypes 42. Turner, N. C. et al. Overall survival with palbociclib and fulvestrant and efficacy across the MONALEESA phase III studies. J. Clin. Oncol. in advanced breast cancer. N. Engl. J. Med. 379, 1926–1936 39, 1458–1467 (2021). (2018). 61. Parker, J. S. et al. Supervised risk predictor of breast cancer based 43. Sledge, G.W. Jr. & Frenzel, M. Analysis of overall survival benefit of on intrinsic subtypes. J. Clin. Oncol. 27, 1160–7 (2009). abemaciclib plus fulvestrant in hormone receptor-positive, ERBB2- 62. Sali, A. & Blundell, T. L. Comparative protein modelling by satis- negative breast cancer-reply. JAMA Oncol. 6, 1122–1123 faction of spatial restraints. J. Mol. Biol. 234, 779–815 (1993). (2020). 63. Kannan, S. et al. Probing the binding mechanism of Mnk inhibitors 44. Chandarlapaty, S. et al. Prevalence of ESR1 mutations in cell-free by docking and molecular dynamics simulations. Biochemistry 54, DNA and outcomes in metastatic breast cancer: a secondary ana- 32–46 (2015). lysis of the BOLERO-2 clinical trial. JAMA Oncol. 2, 1310–1315 (2016). 64. Vandesompele, J. et al. Accurate normalization of real-time quan- 45. Malorni, L. et al. Palbociclib as single agent or in combination with titative RT-PCR data by geometric averaging of multiple internal the endocrine therapy received before disease progression for control genes. Genome Biol. 3, RESEARCH0034 (2002). estrogen receptor-positive, HER2-negative metastatic breast can- cer: TREnd trial. Ann. Oncol. 29, 1748–1754 (2018). Acknowledgements 46. Patnaik, A. et al. Efficacy and safety of abemaciclib, an inhibitor of The authors are grateful to the patients who consented for the use of CDK4 and CDK6, for patients with breast cancer, non-small cell their tumors and the personnel involved in sample collection from the lung cancer, and other solid tumors. Cancer Discov. 6, Breast Surgical Unit, Breast Cancer Center, the Department of Radiology 740–53 (2016). Vall d’Hebron University Hospital; the Molecular Oncology Group at Vall 47. Guiley, K. Z. et al. p27 allosterically activates cyclin-dependent d’Hebron Institute of Oncology (VHIO). Emmanuelle di Tomaso from kinase 4 and antagonizes palbociclib inhibition. Science 366, Novartis Pharmaceuticals Corporation kindly provided LEE011 and eaaw2106 (2019). BYL719 for in vivo and ex vivo work. Genentech, Inc. kindly provided 48. De Laurentiis, M., et al. BioItaLEE-Biomarker analysis on liquid GDC-0032 for in vivo and ex vivo work. We are also grateful to Hong biopsies of patientes treated with ribociclib and letrozole as first- Shaoh at the MSKCC and the Molecular Biology Group at Vall d´Hebron line therapy for advanced breast cancer (ABC). In: San Antonio University Hospital for providing technical support with PDXs sequen- Breast Cancer Sympsium (P5-01-07) (2019). cing and immunostainings, respectively. This manuscript was edited by 49. Lukas, J. et al. Retinoblastoma-protein-dependent cell-cycle inhi- Life Science Editors. Artwork by vector.me. The ABC-POP received bition by the tumour suppressor p16. Nature 375, 503–6 funding from the Breast Cancer Research Foundation and research (1995). grants from Pfizer and Eli-Lilly, respectively (to M.A.). This publication 50. Finn, R., et al. Biomarker analyses from the phase 3 PALOMA-2 trial and the underlying research are partly facilitated by Hartwig Medical of palbociclib (P) with letrozole (L) compared with placebo (PLB) Foundation and the Center for Personalized Cancer Treatment (CPCT) plus L in postmenopausal women with ER + /HER2– advanced which have generated, analyzed, and made available data for this breast cancer (ABC). Ann. Oncol. 27, LBA15 (2016). research. We acknowledge Novartis, Genentech, the GHD-Pink pro- 51. Andre, F. et al. Ribociclib + letrozole for first-line treatment of hor- gram, the FERO Foundation and the Orozco Family for supporting this mone receptor-positive (HR+), human epidermal growth factor study [to V.S.]. This study has been supported by the Susan G. Komen Nature Communications | (2022)13:5258 18 Article https://doi.org/10.1038/s41467-022-32828-6 Foundation (CCR15330331) and by the Catalan Agency AGAUR (2017 Partners and consultancy honoraria from Menarini. A.P. reports that his SGR 540) [to V.S.]. V.S. received funds from the Instituto de Salud institution received research funding from Nanostring Technologies, Carlos III: grants PI13/01714, CP14/00228, MV15/00041, CPII19/00033 Roche and Novartis and reports consulting and lecture fees from and PI20/00892. M.P. received a Juan de la Cierva Grant from the Nanostring Technologies, Roche, Novartis, Pfizer, Oncolytics Biotech, Ministerio de Economía y Competitividad (FJCI-2015-25412), L.Mo. a Amgen, Elli-Lilly, MSD and PUMA. P.N. has consulted for Bayer, Novartis, grant from FI-AGAUR (2019 FI_B 01199), F.B-M. a grant from the Funda- and MSD and received compensation. R.D. is on advisory role of Astra- ción Científica Asociación Española Contra el Cáncer (AECC_Postdoc- Zeneca, Roche and Boehringer-Ingelheim and has received speaker’s toral17-1062) and M.S-G, a Marie Slodowska-Curie Innovative Training fees from Roche, Symphogen, IPSEN, Amgen, Servier, Sanofi, MSD, and Networks PhD fellowship (H2020-MSCA-ITN-2015_675392). This work research support from Merck. M.S. is on the scientific advisory board of was supported by Breast Cancer Research Foundation (BCRF-19-08), Menarini Ricerche and the Bioscience Institute, has received research Instituto de Salud Carlos III Project Reference number AC15/00062 and funds from Puma Biotechnology, Daiichi-Sankio, AstraZeneca, Targim- the EC under the framework of the ERA-NET TRANSCAN-2 initiative co- mune, Immunomedics and Menarini Ricerche, and is a cofounder of financed by FEDER, Instituto de Salud Carlos III (CB16/12/00449 and Medendi.org. M.A. received a research grant from Eli-Lilly, honoraria PI19/01181) and Asociación Española Contra el Cáncer (to J.A.). R.B.C. from Novartis, Astrazeneca, Seattle Genetics, Abbvie and Pfizer and laboratory is supported by Breast Cancer Now (grant numbers: MAN-Q1 travel grants from Novartis, Roche, Pfizer. S.M. has provide punctual and MAN-Q2), NIHR Manchester Biomedical Research Centre (IS-BRC- statistical advice to IDDI and Janssen Cilag and participated to data and 1215-20007) and EdiREX Horizon 2020 grant No.731105. The xenograft safety monitoring committees of clinical trials (Hexal, Steba, IQVIA, program in the C.C. laboratory is supported by Cancer Research UK and Roche, Sensorion, Biophytis, Servier, Yuhan), outside the submitted also received funding from an EU H2020 Network of Excellence (Euro- work. C.S. has served as consultant, participated in advisory boards or CAN). This work has been supported by NIH grants P30 CA008748 and received travel grants from AstraZeneca, Celgene, Daiichi Sankyo, RO1CA190642-01, the CDMRP grant BC171535P1, and the Breast Cancer Roche, Genomic Health, Merck, Sharp and Dhome España S.A., Novartis Research Foundation [to M.S.]. A.P. received funds from Instituto de Odonate Therapeutics, Pfizer, Philips He. C.Ve. & S.K. are founders of Salud Carlos III—PI16/00904 and PI19/01846, Breast Cancer Now— Sinopsee Therapeutics and Aplomex; neither company has any conflict 2018NOVPCC1294, Breast Cancer Research Foundation-AACR Career with the current work. VHIO has had funding (paid directly to the Insti- Development Awards for Translational Breast Cancer Research 19-20- tution) from AstraZeneca, Daiichi Sankyo, Eli Lilly and Company, Gen- 26-PRAT, Fundació La Marató TV3 201935-30, the European Union’s entech, Immunomedics, Macrogenics, Merck, Sharp and Dhome España Horizon 2020 research and innovation program H2020-SC1-BHC-2018- S.A., Novartis, Pfizer, Piqur Therapeutics, Puma, Roche, Synthon and 2020. IRB Barcelona is a recipient of a Severo Ochoa Centre of Excel- Zenith Pharma. The remaining authors declare no competing interests. lence Award from the Spanish Ministry of Economy and Competitive- ness (MINECO; Government of Spain) and is supported by CERCA (Generalitat de Catalunya). C. S. Verma reports grants from MSD Inter- Additional information national and grants from Ipsen outside the submitted work. Supplementary information The online version contains supplementary material available at Author contributions https://doi.org/10.1038/s41467-022-32828-6. M.P.: collection and/or assembly of data, data analysis and interpretation, and manuscript writing. L.Mo., S.K., M.T. H.-A., A.O., M.S.-G., O.R., M.G., Correspondence and requests for materials should be addressed to and J.G.: collection and/or assembly of data, data analysis and inter- Violeta Serra. pretation. G.V., A.G.-P., F. B.-M., N.I., S.M., N.L.-V., and R.D.: statistical design and clinical trial data analysis. V.S.: conception and design of the Peer review information Nature Communications thanks Sarah Sam- study, financial support, collection and/or assembly of data, data analysis mons and the other, anonymous, reviewer(s) for their contribution to the and interpretation, and manuscript writing. J.H and P.N.: Pathology peer review of this work. Peer reviewer reports are available. assessment of the biological samples. L.Mi., M.C., A.A., C.Vi., and A.G.-S.: collection and/or assembly of clinical trial and clinical samples data. A.B., Reprints and permission information is available at F.S., K.L., R.B.C., C.C., J.A., A.P., C.Ve., M.B., M.O., N.C.T., M.S., M.A., and http://www.nature.com/reprints C.S.: data analysis and interpretation. All: interpretation of the data ana- lysis and final approval of the manuscript. Publisher’s note Springer Nature remains neutral with regard to jur- isdictional claims in published maps and institutional affiliations. Competing interests V.S. received non-commercial research support from Novartis and Open Access This article is licensed under a Creative Commons Genentech. M.B. reported receiving honoraria for speaker activities and Attribution 4.0 International License, which permits use, sharing, advisory role from Pfizer, Novartis and Elli-Lilly and support for travel adaptation, distribution and reproduction in any medium or format, as expenses from Roche and Pfizer. M.O. declares grant/research support long as you give appropriate credit to the original author(s) and the (to the Institution) from AstraZeneca, Philips Healthcare, Genentech, source, provide a link to the Creative Commons license, and indicate if Roche, Novartis, Immunomedics, Seattle Genetics, GSK, Boehringer- changes were made. The images or other third party material in this Ingelheim, PUMA Biotechnology, and Zenith Epigenetics; consultant article are included in the article’s Creative Commons license, unless role for Roche, GSK, PUMA Biotechnology, AstraZeneca, and Seattle indicated otherwise in a credit line to the material. If material is not Genetics; and has received honoraria from Roche, Seattle Genetics, and included in the article’s Creative Commons license and your intended Novartis. G.V. reported receiving honoraria for speaker activities from use is not permitted by statutory regulation or exceeds the permitted MDS and advisory role from Astrazeneca. F.S. is employee of Novartis. use, you will need to obtain permission directly from the copyright K.L. is employee of Genentech. C.C. is a member of AstraZeneca’s holder. To view a copy of this license, visit http://creativecommons.org/ External Science Panel, of Illumina’s Scientific Advisory Board, and is a licenses/by/4.0/. recipient of research grants (administered by the University of Cam- bridge) from AstraZeneca, Genentech, Roche and Servier. J.A. has © The Author(s) 2022, corrected publication 2022 received research funds from Roche, Synthon, Menarini, and Molecular Nature Communications | (2022)13:5258 19 Article https://doi.org/10.1038/s41467-022-32828-6 1 Experimental Therapeutics Group, Vall d’Hebron Institute of Oncology, Barcelona, Spain. 2Breast Cancer and Melanoma Group, Vall d’Hebron Institute of Oncology, Barcelona, Spain. 3Department of Medical Oncology, Hospital Vall d’Hebron, Barcelona, Spain. 4Oncology Data Science Group, Vall d’Hebron Institute of Oncology, Barcelona, Spain. 5Institute for Research in Biomedicine (IRB Barcelona), Barcelona, Spain. 6Research Program on Biomedical Infor- matics, Universitat Pompeu Fabra, Barcelona, Spain. 7Translational Genomics and Targeted Therapies in Solid Tumors, August Pi i Sunyer Biomedical Research Institute (IDIBAPS), Barcelona, Spain. 8Service de Biostatistique et d’Epidémiologie, Gustave Roussy, Villejuif, France. 9Oncostat U1018, Inserm, University Paris-Saclay, Villejuif, France. 10Bioinformatics Institute (A*STAR), Singapore, Singapore. 11Medica Scientia Innovation Research (MedSIR), Barcelona, Spain. 12The Breast Cancer Now Research Centre, London, UK. 13Preclinical Modelling of Pediatric Cancer Evolution Group, The Institute of Cancer Research, London, UK. 14Translational Molecular Pathology, Vall d’Hebron Institute of Research (VHIR), Barcelona, Spain. 15Novartis Pharmaceuticals, East Hanover, NJ, USA. 16Genentech, Inc., South San Francisco, California, USA. 17Breast Biology Group, Manchester Breast Centre, Manchester, UK. 18Cancer Research UK, Cambridge, UK. 19CIBERONC, Vall d’Hebron Institute of Oncology, Barcelona, Spain. 20Growth Factors Laboratory, Vall d’Hebron Institute of Oncology, Barcelona, Spain. 21Department of Biochemistry and Molecular Biology, Universitat Autònoma de Barcelona, Barcelona, Spain. 22IMIM (Hospital del Mar Medical Research Institute), Barcelona, Spain. 23Institució Catalana de Recerca i Estudis Avançats (ICREA), Barcelona, Spain. 24University of Barcelona, Barcelona, Spain. 25Department of Medical Oncology, Hospital Clinic, Barcelona, Spain. 26SOLTI Breast Cancer Research Group, Barcelona, Spain. 27 Department of Oncology, IOB Institute of Oncology, Barcelona, Spain. 28Molecular Oncology Group, Vall d’Hebron Institute of Oncology, Barcelona, Spain. 29 School of Biological Sciences, Nanyang Technological University, Singapore, Singapore. 30Department of Biological Sciences, National University of Singapore, Singapore, Singapore. 31Departments of Pathology and Human Oncology and Pathogenesis Program, Memorial Sloan-Kettering Cancer Center, New York, USA. 32Department of Medical Oncology, Gustave Roussy, Villejuif, France. 33Inserm Unit U981, Villejuif, France. e-mail: vserra@vhio.net Nature Communications | (2022)13:5258 20