A Precision Medicine Drug Discovery Pipeline Identifies Combined CDK2 and 9 Inhibition as a Novel Therapeutic Strategy in Colorectal Cancer

Colorectal cancer is the third most common cancer in the world, with approximately 150,000 new cases in the United States each year and the third leading cause of cancer-related deaths (1). At initial diagnosis, approximately 20% of patients will have distant metastasis, and another 25%–30% of patients with stage II/III disease will develop metastasis (2, 3). Currently, the use of chemotherapy in the metastatic setting can palliate symptoms and improve survival, but cannot cure patients. If left untreated, patients with colorectal cancer metastasis have an overall survival of just 6–9 months (4), but with combination therapy, survival can be improved to greater than 24 months (5, 6). However, despite these improvements, metastatic colorectal cancer remains an incurable and debilitating disease.

The concept of precision medicine is to apply the most effective therapeutic strategy to each patient at the right time to improve efficacy and minimize toxicity for patients. However, prior to the incorporation of new therapeutic agents in the clinical setting, these drugs must be assessed for their therapeutic potential in predictive preclinical models. Patient-derived models of cancer, such as cell lines, organoids, and patient-derived xenografts (PDX), are increasingly being accepted as “standard” preclinical models to facilitate the identification and development of new therapeutics. For example, large-scale drug screens of cancer patient–derived cell line panels have been used to identify sensitivity to a large number of potential therapeutics (7). Similarly, tumor organoid cultures from colorectal cancer specimens have also been used to perform drug screens (8), and colorectal cancer PDXs are being used to predict drug response (9) and identify novel drug combinations (10). Combinations of these patient-derived models of cancer are also being explored to develop precision medicine workflows for cancer care (11).

In this study, we developed a precision medicine strategy for patients with metastatic colorectal cancer. Specifically, we developed a series of patient-matched early-passage cell lines and PDXs to identify the most efficacious agents for metastatic colorectal cancer. In our pipeline, our early passage cell lines were first used to perform high-throughput drug screens across multiple cell lines to identify potential therapeutic targets, and the top targets were subsequently validated in matching PDXs. Using this approach, we identified the cyclin-dependent kinase (CDK) family as promising targets in the treatment of metastatic colorectal cancer. Further interrogation of this class of inhibitors in our screens revealed combined inhibition of CDK2 and CDK9 as the primary driver of efficacy.

Our findings suggest that the use of matched low-passage cell lines and PDXs is a useful platform to identify novel inhibitors. This platform led to the identification of dual CDK2/9 inhibition as a promising therapeutic strategy to treat metastatic colorectal cancer.

Patient tumor samples were collected in partnership with the Duke BioRepository & Precision Pathology Center at the time of the patient's surgical resection. Work was performed under a Duke Institutional Review Board (IRB)-approved protocol (Pro00002435). Development of PDXs was performed as described in our previous studies (10, 12). Briefly, surgically resected tumors were washed with phosphate-buffered saline (PBS) and minced to small (<2 mm³) fragments. Tumor fragments were further dissociated with a tissue dissociation kit (gentleMACS Dissociator). Subsequently, 150 μL of homogenized tumor tissue suspensions (150 mg/mL concentration) were injected into the flanks of 8- to 10-week-old severe combined immunodeficient (SCID) mice (obtained from the Duke University Rodent Genetic and Breeding Core). Mice were monitored a minimum of thrice weekly. Upon identification of initial tumor appearance, tumor size was measured by digital calipers. When tumors reached 1 cm², tumors were harvested and used to create matched cell lines. All animal procedures were performed in accordance with the Duke University Institutional Animal Care and Use Committee.

As described previously (10), colorectal cancer cell lines were generated from early-passage PDXs. After harvesting PDXs, homogenized single-cell solutions were prepared and grown in 10 cm² tissue culture dishes in the presence of complete cell culture media containing DMEM, 10% FBS, 100 IU/mL penicillin and 100 μg/mL streptomycin at 5% CO₂ and 37°C. To acquire pure cell lines for each PDX, a single colony of cancer cells was isolated within an O ring, trypsinized, and placed into 24-well plates. A total of six cell lines were established, including CRC057 (liver metastasis), CRC119 (liver metastasis), CRC16–159 (primary colon tumor), CRC240 (liver metastasis), CRC247 (liver metastasis), and CRC401 (primary rectal tumor). Cell lines were authenticated by the Duke University DNA Analysis Facility using polymorphic short tandem repeat (STR) genetic profiling (GenePrint 10 kit from Promega).

High-throughput drug screens were performed in collaboration with the Duke University Functional Genomics Shared Resource. Briefly, each of 2,100 compounds from the BioActive small-molecule library (SelleckChem) was stamped in triplicate onto 384-well plates for a final concentration of 1 μmol/L using an Echo Acoustic Dispenser (Labcyte). Next, 500–1,000 cells per well was distributed using a Well mate (Thermo Fisher Scientific). Cells and compounds were incubated for three days, after which cell viability was assessed using the CellTiter-Glo Luminescent Cell Viability Assay (Promega) on a Clariostar plate reader (BMG Labtech). Percent killing was quantified using the formula 100 × (1-(average CellTiterGlo^drug/average CellTiterGlo^DMSO)) where the value average CellTiterGlo^DMSO was the average DMSO CellTiterGlo value across each plate.

In vitro drug sensitivity assays were performed by treating each cell line with a CDK1/2/9 inhibitor (AZD5438), CDK9 inhibitor (LDC067), CDK2 inhibitor (CVT-313), or a CDK1 inhibitor (RO-3306). To do this, cells at 70%–80% confluency were plated into 96-well plates at 4,000 cells/well and incubated for 24 hours in drug-free media. After 24 hours of incubation, cells were treated with drug at a range of doses, starting from either 50 μmol/L or 100 μmol/L, depending on the cell line, with a serial dilution factor of three and a 10-point dose curve. A total of five replicates were included for each drug dose. Cells were incubated in the presence of drug for two days, after which the number of metabolically viable cells was estimated using CellTiter-Glo Luminescent Cell Viability Assay kit (Promega). All drugs were purchased from SelleckChem. All in vitro drug sensitivity assays were performed at least two independent times. The in vitro drug sensitivity assay results were reported as half maximal inhibitory concentration (IC₅₀).

Data to create Kaplan–Meier curves were collected the Cancer Genome Atlas for colorectal adenocarcinoma (COADREAD). The UCSC Xena browser (13) was used to extract expression data for CDKs, pathologic stage, days to last fallow up and days to death data. Data was divided by median or quartiles, and GraphPad Prism software was used to draw Kaplan–Meier graphs.

From a 10 μmol/L working stock, 10 μL of nonsilencing or CDK9 siRNAs (Qiagen) were added to 1 mL of OptiMem (Sigma). In a separate tube, 20 μL of Lipofectamine RNAiMax (Thermo Fisher Scientific) was added to 1 mL of OptiMem. The two solutions were mixed and incubated at room temperature for 20 minutes. During this time, 1.5 mL of cell suspension (333,000 cells/mL) was plated into each well of a 6-well plate. After 20 minutes, 1 mL of the siRNA mixtures was added to each well of the plate and incubated overnight. The following day, cells were plated in a 96-well plate, while the media on the remaining wells was switched to normal DMEM supplemented with 10% FBS and 1% penicillin/streptomycin. After 24 hours, cells were incubated in triplicate in the presence of vehicle (dimethyl sulfoxide; DMSO) or a range of drug concentrations serially diluted 1:3 starting from 50 μmol/L. The dose curves were incubated for 48 hours, after which CellTiter-Glo (Promega) was added to the wells and imaged on a microplate reader (Molecular Devices). Results were uploaded to Prism, normalized to DMSO treated wells, and plotted using a 4-point nonlinear regression analysis.

For knockdown validation, siRNA-treated cells were lysed in radioimmunoprecipitation assay lysis buffer supplemented with phosphatase and protease inhibitor (Thermo Fisher Scientific). The concentration of the protein lysates was determined using the BCA Protein Assay (Bio-Rad). A total of 50 μg of lysate was electrophoretically separated on 4%–20% SDS polyacrylamide gels using the Bio-Rad Miniprotean Tetra system and transferred to polyvinylidene difluoride membranes (Bio-Rad). Subsequent to blocking in StartingBlock T20 (Thermo Fisher Scientific), membranes were incubated with primary antibodies against CDK9 (Thermo Fisher Scientific; catalog number: MA5–14912) or β-actin (Santa Cruz Biotechnology), followed by the appropriate secondary antibodies at manufacturer-recommended dilutions. Blots were scanned and analyzed using a LI-COR Odyssey imaging system.

A total of 300,000 cells were plated in 6-well plates and incubated at 37 °C with 5% CO₂ until they reached approximately 60% confluence. Cells were treated with 0.1% DMSO or three times the IC₅₀ dose of AZD5438 in full growth media. After 24 hours, 2 × 10⁶ cells were harvested and washed twice with PBS, fixed in in 80% ethanol on ice for 30 minutes, washed twice more with PBS, and resuspended in cell staining buffer [0.1% Triton X-100, 0.1 mmol/L EDTA disodium, 50 μg/mL Rnase A and 50 ug/mL propidium iodide (PI) in PBS]. Flow cytometry was performed in conjunction with the Duke University Flow Cytometry Shared Resource.

A total of 100,000 cells were plated in 6-well plates and incubated for 48 hours. Cells were treated for 16 hours as described for the cell-cycle analysis, after which, cells were fixed in 4% paraformaldehyde for 15 minutes and permeabilized for 30 minutes with 0.1% Triton-100X in PBS. Subsequent to blocking with normal goat serum, cells were treated with a 1:5,000 dilution of mouse anti α-tubulin antibody (Cell Signaling Technology) for 16 hours at 4°C, and stained with Hoechst dye (1:2,000; Thermo Fisher Scientific) and goat anti-mouse secondary antibody (1:5,000; LI-COR Biosciences). Cells were analyzed using an Olympus epifluorescence microscope at 400× total magnification. Cells were quantified from a total of 10 microscopic fields per treatment group.

To evaluate the effect of CDK inhibition on CRC240 PDXs in vivo, homogenized PDX tissue–PBS suspension (150 μL at 150 mg/mL concentration) was subcutaneously injected into the right flanks of n = 20 male mice (JAX NOD.CB17-PrkdcSCID-J, 8–10 weeks) and randomized into four groups (n = 5 per group) to receive either vehicle (DMSO), CDK2 inhibitor (CVT-313), CDK9 inhibitor (LDC067), or the combination (CVT-313 + LDC067). Tumor volumes were measured regularly, and treatments were started when the tumor reached a volume of 250 mm³. PDXs were treated with intraperitoneal injections of either CVT-313 (0.625 mg/kg/every other day), LDC000067 (5 mg/kg/every other day), CVT-313 + LDC067 (0.625 mg + 5 mg/kg/every other day, respectively) or vehicle (2% DMSO every other day). Both drugs were diluted in DMSO according to company protocol and further diluted in sterile pure water to reduce the DMSO concentration to less than 2%. Mice were examined regularly, and tumor volume was charted until endpoint of either maximal tumor size of 4,000 mm³ or death for 14 days. All mice were euthanized after completion of experiment. Tumor sizes were measured using a digital Vernier caliper, and tumor volumes were estimated using the following formula: (length × (width)²)/2.

All statistical analyses were performed in GraphPad Prism software. t tests (two groups) and ANOVA (three or more groups) with Tukey post hoc correction was used to compare differences between groups. A P < 0.05 was considered statistically significant.

In our previous work, we established a series of matched low-passage cell lines and patient-derived xenografts (PDX) to study the biology and genomics of colorectal cancer (10). These studies revealed important parallels between patient-derived models and the tumors from which they were originally derived (10, 12, 14). Here, we sought to leverage these valuable patient-derived models to establish a precision medicine pipeline to identify and validate promising new therapies for colorectal cancer and other cancers. To do this, we capitalized on the unique strengths of each component of the pipeline, using the low-passage cell lines for high-throughput chemical screens and mechanistic studies then using the PDXs to validate efficacy for promising agents in vivo (Fig. 1A).

To implement our colorectal cancer precision medicine pipeline, we first developed patient-derived models of cancer including low passage cell lines and patient-derived xenografts (PDX). Samples were obtained from patients undergoing resection of their colorectal cancer (either metastatic or primary disease) at Duke University under an Institutional Review Board (IRB)- and Institutional Animal Care and Use Committee (IACUC)-approved protocol. For each patient, matching cell lines and PDXs (CRC057, CRC119, CRC240, CRC247, CRC401 and CRC16–159) were developed as described previously (12, 14) for this project. Patient demographics are described in Table 1, and Fig. 1B shows the histologic features of PDXs and the matching cell lines. There were two primary samples (CRC401 and CRC16–159) and four samples from colorectal cancer liver metastasis (CRC057, CRC119, CRC240, and CRC247; Table 1; Fig. 1B). Three samples were from patients self-reported to be of African ancestry (CRC119, CRC247, and CRC16–159) and three were from patients self-reported to be of European ancestry (CRC057, CRC240, and CRC401). There were four female patients (CRC057, CRC119, CRC401, and CRC16–159) and two male patients (CRC240 and CRC247). All were diagnosed with adenocarcinoma of the colon, with two found to be poorly differentiated (CRC240 and CRC401). One patient (CRC16–159) had a tumor with microsatellite instability that also had a BRAF mutation, and 4 had KRAS-mutated tumors (CRC057, CRC119, CRC248, and CRC401). Overall, this set of models represents a diverse population of colorectal cancer specimens.

To identify potential therapeutic agents, a high-throughput chemical screen was performed using the BioActives compound library (SelleckChem). This library contains 2,100 compounds and is comprised of both FDA-approved and non-FDA–approved bioactive small molecules. The BioActives library is fully annotated by target and pathway, enabling interrogation of single agents, rational combination therapies, and efficacious targets or pathways. High-throughput screens were performed on six early-passage colorectal cancer cell lines (CRC057, CRC119, CRC16–159, CRC240, CRC 247, and CRC401; Supplementary Fig. S1). Overall, across the entire panel of drugs, CRC16–159 and CRC119 were most broadly sensitive while CRC247 and CRC401 were the most resistant (Supplementary Fig. S1A). In addition, the vast majority of drugs had no effect on colorectal cancer cells, with over 90% of compounds showing lower than 25% cytotoxicity among all six cell lines and only 7.9% (n = 116) of drugs with average cytotoxicity of more than 50% (Supplementary Fig. S1B). We next utilized the annotations available for the BioActives library to interrogate the pathways for which multiple drugs demonstrated efficacy. To pinpoint drugs with broad efficacy across multiple patient-derived models, we focused on compounds in which at least five out of six cell lines were inhibited by more than 50% (Fig. 2A). This analysis produced 94 compounds, and to further understand the targets and pathways that may potentially represent the most efficacious therapeutic strategies for colorectal cancer, the drugs were categorized by their mechanisms of action and pathway-level annotations. At the pathway level, drugs targeting histone deacetylases (HDAC), heat shock proteins (HSPs), the MEK pathway, the proteasome, PI3K/mTOR pathway, and cyclin-dependent kinases (CDK) had the highest average percent killing across the entire panel (Fig. 2B). Several of these pathways have been previously identified as promising targets for colorectal cancer, such as MEK (15, 16) and the PI3K/mTOR pathway (17–19), while others remain investigational, such as the CDKs (20, 21).

Although CDK inhibitors have been studied extensively in breast and other cancers and are currently considered standard-of-care therapy in breast cancer, their use in colorectal cancer remains investigational (22–25). To validate the utility of targeting CDKs in colorectal cancer, we first compared the efficacy of CDK inhibitors against other pathway-specific inhibitors studied in colorectal cancer, such as mTOR/PI3K inhibitors, STAT inhibitors, and Wnt/β-catenin inhibitors (Fig. 2C), as well as standard-of-care therapy in metastatic colorectal cancer, including oxaliplatin, irinotecan, 5-FU/Xeloda, lonsurf and regorafenib (Fig. 2D). These analyses revealed that, in general, CDK inhibitors have higher antitumor activity in comparison with other pathway-specific inhibitors and even many standard-of-care therapies.

CDKs are a protein family of 21 genes, several of which exhibit functional redundancies with other members of the gene family (26). As such, inhibition of a single CDK could be compensated for by others, especially in the context of the ever-evolving landscape of metastatic colorectal cancer. Therefore, it was not surprising to find that almost all potent CDK inhibitors target multiple CDKs. We therefore sought to determine the CDK combination with the highest efficacy. To do this, the CDK inhibitors and their targets were sorted from highest to lowest percent killing (Fig. 2E). These analyses revealed that drugs targeting combinations of CDK 1, 2, and 9 were among the most potent CDK inhibitors across the six patient-derived cell lines. Both CDK 1 and 2 are involved in cell-cycle regulation: CDK1 is involved in G₂–M progression while CDK2 promotes G₁ and S-phase progression (27, 28). CDK9, on the other hand, serves an independent function in transcription and can inhibit cancer cell proliferation beyond disruption of the cell cycle (29). As studies have shown mechanistic synergy between these classes of CDKs (30, 31), we concluded that the combination of CDK1, 2, and 9 may be an effective combination strategy in the treatment of metastatic colorectal cancer.

Consistent with our drug screen results, which pinpointed CDK1, 2, and 9 as being novel targets for colorectal cancer, analysis of CDK1, 2, and 9 expression in colorectal cancer from The Cancer Genome Atlas suggests that, while the individual expression of most of these CDKs is not a prognostic marker for early-stage disease (stages I–III, Supplementary Fig. 2), increased expression of these three CDKs in combination is correlated with poor prognosis for overall survival in advanced (stage IV) colorectal cancer (Fig. 2F), reinforcing the potential utility of targeting these CDKs in colorectal cancer.

Next, to determine the efficacy of CDK1, 2, and 9 inhibition on colorectal cancer both as single agents and in combination, we performed IC₅₀ dose–response curves on our early-passage colorectal cancer cell lines (CRC057, CRC119, CRC16–159, CRC240, CRC 247, and CRC401) with a CDK1 inhibitor (RO-3306), CDK2 inhibitor (CVT-313), CDK9 inhibitor (LDC067), and a CDK1,2,9 inhibitor (AZD5438; Fig. 3A–F). Single-agent inhibition by RO-336, CVT-313 or LDC067 showed minimal efficacy, with estimated IC₅₀ values ranging from 1.7 μmol/L (CDK9 inhibitor, LDC067) to 7.6 μmol/L (CDK2 inhibitor, CVT-313). In contrast, we observed that the CDK1/2/9 inhibitor, AZD5438, had significantly lower IC₅₀ values ranging from 110 nmol/L to 3 μmol/L (P < 0.001). Specifically, CRC057 (Fig. 3A) and CRC119 (Fig. 3B), both with IC₅₀ of 110 nmol/L, were found to be most sensitive to AZD5438, whereas CRC16–159 (Fig. 3C), with IC₅₀ of 1.2 μmol/L, and CRC401 (Fig. 3F) with an IC₅₀ of 3.0 μmol/L, were found to be the most resistant. Together, these results showed that combined CDK1, 2, and 9 inhibition could be an efficacious treatment strategy for patients with metastatic colorectal cancer.

Inhibition of CDK 1 or 2 induces potent cell-cycle arrest (32). Therefore, to confirm target-specificity of AZD5438, we analyzed cell-cycle progression for CRC057, CRC119, CRC16–159, and CRC240 cells treated with vehicle or AZD5438 (Fig. 4). Specifically, CDK2 promotes S-phase progression while CDK1 promotes G₂ progression and entry into mitosis (33, 34). Consistent with the known mechanisms of CDK1 and 2 inhibition, treatment of CRC057 and CRC16–159 with AZD5438 led to a significant, dose-dependent increase in S and G₂–M phase arrest (Fig. 4A and B). There was also a significant increase in S and G₂–M phase arrest in CRC119 and CRC240 upon treatment with AZD5438 (Fig. 4C and D). These analyses confirm target specificity for AZD5428 in our cell lines.

To further examine the mechanism of action of AZD5438 in promoting cell death, we tested whether AZD5438 was inducing anaphase catastrophe. Anaphase catastrophe is a pro-apoptotic pathway that has been shown to be the mechanism of action upon CDK2 inhibition (35). Because the inhibitory effect of AZD5438 is, in part, due to CDK2 inhibition, we investigated whether the same antitumor activity could be seen in our colorectal cancer cell lines. To do this, CRC057 cells were treated with two different doses of AZD5438 (the IC₅₀ dose and three times the IC₅₀) for 12 hours. Treatment with AZD5438 induced anaphase catastrophe as observed by multipolar spindles during anaphase (Fig. 4E). AZD5438 treatment led to a significant increase in the number of cells undergoing anaphase catastrophe in the treatment group compared with vehicle-treated control cells (Fig. 4F).

Our data suggest combined inhibition of CDKs 1, 2, and 9 outperforms single-agent inhibition of any one of these CDKs in isolation. However, because CDK1 and 2 have partially-redundant functions in cell-cycle regulation, it was unclear which of the CDKs is responsible for the synergistic activity of the CDK1/2/9 inhibitor, AZD5438. To further dissect the relative importance of CDK1 or CDK2 inhibition in combination with CDK9 inhibition we performed siRNA-mediated knockdowns of CDK9 in combination with inhibition of either CDK1 or 2. We first assessed the efficacy of three independent siRNAs targeting CDK9 by western blots in CRC057 (Fig. 5A), CRC16–159 (Fig. 5B), and CRC119 (Fig. 5C). Among these three siRNAs, siRNA_8 provided the most complete knockdown of CDK9 all three cell lines (Fig. 5A–C). We next performed siRNA-mediated knockdowns of CDK9 with or without drug-mediated inhibition of CDK1 (RO-3306) or CDK2 (CVT-313). Interestingly, while there was no synergistic effect between CDK9 knockdown and treatment with the CDK1 inhibitor, R0–3306 (positive delta score) (Fig. 5D–F), we observed a synergistic effect of CDK9 knockdown upon CDK2 inhibition (CVT-313) for all cell lines (Fig. 5G–I). These results suggest that CDK1 inhibition is dispensable for the efficacy of AZD5438 and that the mechanism of action for AZD5438 is likely due predominantly to combined CDK2 and CDK9 inhibition.

To validate the efficacy of dual inhibition of CDK2 and 9, we tested whether combined inhibition of CDK2 and 9 was capable of reducing tumor growth in vivo. To do this, matched PDXs of CRC240 were treated with either vehicle (DMSO), the CDK2 inhibitor, CVT-313, (5 mg/kg), the CDK9 inhibitor, LDC000067 (0.625 mg/kg) or in combination. Figure 6A shows that treatment with CVT-313 or LDC000067 resulted in minimal tumor growth inhibition; however, consistent with our in vitro observations using both small molecules and siRNAs, the combined inhibition of both CDK2 and 9 led to synergistic and significant tumor growth inhibition. Importantly, no additional toxicities, such as weight loss, diarrhea, limb weakness, anxiety/restlessness, or hair loss, were observed in the mice treated with the combination therapy (Fig. 6B). Collectively, these in vitro and in vivo results indicate that dual CDK2 and 9 inhibition could be a promising and tolerable therapy for patients with metastatic colorectal cancer.

Patient-derived models of cancer, such as early passage cell lines, patient-derived organoids, and PDXs are being accepted as efficient tools for the development of the cancer therapeutics and development of precision medicine pipelines (36, 37). Specifically, morphologic and molecular similarity between these models and the original patient tumors facilitates the evaluation of anticancer drug responses and the identification of mechanisms of resistance. Indeed, previous studies have demonstrated that PDXs can recapitulate the patient response to therapy (38, 39) and major efforts by both commercial and academic institutions are being developed to determine how patient-derived models of cancer can be best used to guide patient care. In our study, we have developed a novel precision medicine strategy for patients with colorectal cancer using matched early-passage cell lines and PDXs. We coupled these matched patient-derived models with high-throughput drug screens as a reliable means to identify (i) response to standard-of-care agents, (ii) patient specific target and pathway vulnerabilities, (iii) effective single-agent therapies, and (iv) rational combination therapies. These analyses led to the discovery of a novel combination of cyclin-dependent kinases (CDK2/9) as a treatment of colorectal cancer. Although a small dataset, we feel that this combination is potentially applicable across the general colorectal cancer population as Table 1 showed that our samples consisted of males and females, European and African ancestry, mostly MSS tumors, but contained both KRAS WT and mutant samples, although it must be noted that one KRAS WT sample (CRC16–159) had a BRAF-mutant tumor. Indeed, the CDK2/9 combination was effective even in CRC-247 and CRC-401, both of which were broadly resistant to standard-of-care agents and were among the most resistant in the drug screen (Fig. 2). While the mechanisms of resistance to standard-of-care agents in these lines are unknown, both of these lines harbor a KRAS G12D mutation, which has been shown to promote therapy resistance in non–small cell lung cancer (40, 41). It is also worth noting other cell lines in the screen were also KRAS mutant, suggesting additional mechanisms of resistance in CRC-247 and CRC-401.

There has been considerable progress in our understanding about specific function of CDKs, which are involved in cell-cycle regulation and transcription. CDKs can be divided into two subgroups that directly or indirectly regulate the cell cycle, including CDKs1–6, 11 and 14–18, or transcription, including CDKs7–13, 19, and 20 (42). CDK1, CDK2, CDK4, and CDK6 and their associated Cyclins (A, B, D, and E) are considered fundamental cell-cycle regulators. CDK 2, 4, and 6 promote expression of genes essential for G₁ completion and progression through S-phase, while CDK 1 promotes G₂ progression and entry into mitosis (28). In contrast, CDK7/cyclin H and CDK9/cyclin T promote phosphorylation of the carboxy-terminal domain of RNA polymerase II, facilitating initiation and elongation of RNA transcription (28).

CDKs are best known for their remedial effect in breast cancer therapy, as FDA has approved three CDK4/6 inhibitors, including palbociclib, ribociclib, and abemaciclib, for the treatment of hormone receptor–positive and HER2-negative breast cancer (23–26). Although clinical benefit of other combinations of CDK inhibitors in various cancers has been studied in clinical trials (43–46), concerns have always been raised in terms of toxicity and tolerability (47). Unfortunately, few studies evaluating CDKs in colorectal cancer have been performed; however, these have shown minimal efficacy in colorectal cancer (48) and overall, the efficacy of CDK inhibitors in colorectal cancer remains unknown.

Similar to clinical trials in colorectal cancer that showed limited activity for single-agent CDKs, our data also suggest minimal activity for any of the CDKs as single agents. However, we observed significantly improved efficacy as combinatorial therapy. This is in agreement with studies that showed flavopiridol (CDK1, 2, 4 and 9 inhibitor) had promising antitumor activity in colorectal cancer xenografts (49). However, flavopiridol showed no activity in a phase II study in advanced colorectal cancer when used as a single agent (50). This lack of activity of flavopiriol was due to its toxicity in humans and the inability to achieve a therapeutic dose. Therefore, in our study we attempted to determine the minimal combination of CDK inhibition needed, while minimizing the toxicity to determine the most efficacious combination. The results of our study suggested that the combination of CDK2/9 inhibition, which targets both the cell cycle and transcription, is potentially the most optimal and minimal combination of CDK inhibition needed for the treatment of colorectal cancer.

Mechanistically, we showed that dual CDK2/9 inhibition led to cell cycle arrest and anaphase catastrophe. The degree of cell cycle arrest did not correlate with the degree of sensitivity to CDK2/9 inhibition, which may be due to the fact that CDK2 and CDK9 play different functional roles, with CDK2 inhibition inducing cell-cycle arrest (51, 52), and CDK9 inhibition preventing transcription elongation (53). In addition to cell-cycle arrest, CDK2/9 inhibition also induced anaphase catastrophe. This is consistent with a previous study showing that the combination of CKD2/9 inhibition by CCT068127 exhibited antiproliferative activity by decreasing retinoblastoma protein (RB1) phosphorylation, reducing phosphorylation of RNA polymerase II, and inducing cell-cycle arrest and apoptosis in colorectal cancer and melanoma cell lines (54). Given that similar results were also observed in lung cancer (55), this suggests that targeting these two independent pathways of CDK inhibition could have synergistic effects in several cancers. The presence of supernumerary centrosomes is a common characteristic of many cancers (56). Normally, an increased centrosome number would lead to multipolar cell division and trigger cell death. Cancer cells overcome this problem by active clustering of supernumerary centrosomes at two poles, which ensures bipolar mitosis (57, 58). CDK2 inhibition has been shown to antagonize supernumerary centrosome clustering in lung cancer, which can lead to multipolar cell division and subsequent apoptosis (59, 35). Consistence with these observations, our data showed that CDK2 inhibition led to anaphase catastrophe in colorectal cancer cells.

However, as described above, a major concern of combinatorial CDK inhibition and potentially the combination of CDK2/9 inhibition is drug toxicity. Despite their efficacy, common adverse effects of CDK inhibitors include hematologic abnormalities, fatigue, and gastrointestinal symptoms, such as diarrhea, which can limit their clinical use. However, as our vivo PDX study showed that combined CDK2/9 inhibition acted synergistically, we were able to reduce the dose of both CDK2 and 9 inhibitors while maintaining improved efficacy and simultaneously minimizing drug toxicity. These encouraging properties of the CDK2/9 combination in vivo will allow us to further explore this combination as a viable clinical combination.

Finally, our models will also now allow us to study additional therapies that can be combined with CDK inhibitors to improve efficacy and overcome resistance both in vitro and in vivo. Indeed, studies have shown that CDK inhibitors can enhance apoptosis when combined with chemotherapy to overcome resistance in various cancers, including colorectal cancer (60, 61). Other studies have shown that potentiation of the p53 tumor suppressor by 5-fluorouracil had a synergic killing effect with a CDK7 inhibitor in colorectal cancer (62). Similarly, the BRAF^V600E mutation in colorectal cancer is associated with therapeutic resistance secondary to reactivation of MEK/ERK signaling cascade, and Zhang and colleagues, showed that dual inhibition of multiple CDKs by dinaciclib (CDK1, 2, 5, 9 inhibitors) with MEK inhibition by cobimetinib (MEK inhibitor) can overcome resistance in BRAF-mutant colorectal cancer (63).

One limitation of the use of matched patient-derived cell lines and PDXs is the considerable time and resources needed to develop and apply these systems to drug discovery efforts. To address these limitations, researchers have turned to organoid systems that recapitulate the biology of patient tumors. These systems have been developed across a range of solid tumors, including colorectal cancer (8), pancreatic cancer (64), prostate cancer (65), liver cancer (66), breast cancer (67), and lung cancer (68). Advances in organoid culture technology have the potential to substantially increase the speed with which novel candidate therapies can be identified and translated into real-time clinical decision-making (reviewed in Kondo and colleagues; ref. 69).

In summary, the generation of our colorectal cancer precision medicine pipeline provides a robust platform to identify and validate potential new therapies with several advantages. First, a standardized workflow that combines matched patient-derived cell lines and PDXs provides maximal benefit by enabling a rapid screening and mechanistic validation tool. Second, capitalizing on the matched in vivo models facilitates validation of candidate therapies and mechanisms of response/resistance in matched PDXs in vivo. Studies from these models will pave the way to more rapidly identify, characterize, and validate new drugs for translation into the clinic.

S.J. McCall reports other from AstraZeneca (consultant) outside the submitted work. D.S. Hsu reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

J.A. Somarelli: Conceptualization, formal analysis, supervision, writing-original draft, writing-review and editing. R.S. Roghani: Resources, validation, investigation, writing-original draft, writing-review and editing. A.S. Moghaddam: Formal analysis, investigation, writing-review and editing. B.C. Thomas: Formal analysis, investigation, visualization, writing-review and editing. G. Rupprecht: Investigation, visualization, writing-review and editing. K.E. Ware: Supervision, investigation, writing-review and editing. E. Altunel: Investigation, writing-review and editing. J.B. Mantyh: Investigation. S.Y. Kim: Resources, formal analysis, investigation, writing-review and editing. S.J. McCall: Resources, writing-review and editing. X. Shen: Resources, writing-review and editing. C.R. Mantyh: Resources, writing-review and editing. D.S. Hsu: Conceptualization, resources, supervision, funding acquisition, investigation, writing-original draft, writing-review and editing.

We acknowledge the Duke BioRepository & Precision Pathology Center (BRPC, a shared resource of the Duke University School of Medicine and Duke Cancer Institute) for their regulatory and technical assistance in the procurement of the patient samples utilized in this study. The BRPC is supported by the P30 Cancer Center Support Grant (P30 CA014236). This work was supported, in part, by P30 CA014236 (NIH, Bethesda, MD).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.