Anlotinib

This was a first-in-human, phase I, open-label study of anlotinib in advanced refractory solid tumors. The primary objective was to establish the safety profile of anlotinib by identifying DLT, MTD, the recommended phase II dose, and schedule. Secondary objectives included description of single-dose and multi-dose pharmacokinetics of oral anlotinib and assessment of preliminary antitumor effect.
All patients provided written informed consent. The study protocol and amendments were reviewed and approved by the Institutional Review Board, in accordance with the Declaration of Helsinki.

Patients and Samples: In total, 440 advanced NSCLC patients were enrolled in the ALTER‐0303 study (https://clinicaltrials.gov/NCT02388919). Of the 440 patients, 126 patients (placebo: 15 patients; anlotinib: 111 patients) with qualified samples (including white blood cell (WBC), blood, and tissue) were analyzed in the present study (Figure S1, Supporting Information). All refractory advanced NSCLC patients were enrolled in Shanghai Chest Hospital, Chinese Academy of Medical Sciences Cancer Hospital, Peking Union Medical College Hospital, etc. All patients had received at least two lines of targeted therapy or chemotherapy, and had failed prior therapies. The patients were orally administered with anlotinib as a third‐line therapy or over third‐line therapy with a dosage of 12 mg day⁻¹ for two consecutive weeks that was then discontinued for one week. If PD or intolerable toxicity occurred, anlotinib therapy was terminated immediately. Multicenter plasma and tumor collection was performed as previously described.9, 11 Clinical information of each patient is shown in Tables S4 and S5 in the Supporting Information. Informed consent was obtained from all subjects following the ALTER‐0303 study.
Pathological Type and Staging: EGFR driver gene mutations were detected in tissue DNA by ADx‐ARMS method, and ALK fusion or ROS1 rearrangement were detected in tissue RNA via RT‐qPCR method. The patient harboring any one of these positive mutations in EGFR, ALK, and ROS1 was defined as driver gene positive. Tumor volume and metastases were evaluated on the basis of CT scans by at least one radiologist. Stages for each patient were determined by at least one investigator.
Tissue DNA Extraction and Sequencing: A customized targeted capture assay panel (168 cancer genes, Burning Rock Dx) was used to capture target DNA.24, 25, 26 Briefly, DNA was extracted from tumor tissue slides according to the standard procedures. Targeted capture was performed on at least 200 ng of input DNA for each sample. After amplifying captured DNA, high‐throughput sequencing was performed to collect raw data for genomic information. Trimmomatic (version 0.36) was used to trim low quality bases of raw reads.36 Cleaned data were aligned to the latest human genome assembly hg38 using Burrows–Wheeler Aligner (BWA) with default parameters.37 Mutations were called with Varscan2 with default parameters for each sample.38Circulating DNA Extraction: Blood samples for each patient were collected in a 10 mL K2‐EDTA tube. All plasma samples were collected within 2 h of collection by centrifugation of blood samples at 1600 × g for 10 min. Then, the upper plasma was transferred to 1 mL cleaned Eppendorf tubes using a pipette, and the tubes were sequentially marked. Plasma was stored at −80 °C. Up to 5 mL of plasma from each patient was available for this study (range, 3–5 mL). cfDNA and ctDNA were extracted from the entire volume of plasma using the QIAamp Circulating Nucleic Acid Kit (Qiagen). All cfDNA and ctDNA samples were eluted in 50 µL of DNA buffer (0.05 m, pH: 7.5). cfDNA and ctDNA quantification was performed by the Qubit fluorescence quantitative method (Invitrogen).
Library Preparation: Tumor tissue DNA (200–300 ng) or plasma cfDNA and ctDNA (10–100 ng) for each sample was used for targeted exome capture. Library preparation was performed as previously described.24, 25, 26 Captured DNA for each sample was end‐repaired and adaptor ligated, and then amplified for no more than 12 cycles in a thermal cycler (Applied Biosystems). Finally, the PCR products were quantified using Qubit (Thermo), and underwent paired‐end sequencing using a 2*150 model.
Plasma SNV Calling: Quality analysis of raw sequencing data were performed based on the authors' and other previous studies.24, 25, 26 The SNV calling algorithm was performed as previously reported.23 WBC samples were used to estimate the error parameters for calling SNVs. Germline and somatic mutations were obtained via calculating sequencing depth (≥100×) and VAF. All germline and somatic mutations were annotated, and then the genes that were not included in the scope of 168 genes were filtered. According to the methods reported in a previous study, the mutations were filtered with VAF > 20%,23 the mutations were deleted with low effect (MODIFIER and LOW), and finally the mutations were obtained with relatively high effect (MODERATE and HIGH). These mutations were defined as somatic mutations (synonymous mutations and nonsynonymous mutations). The synonymous mutations were filtered, and then nonsynonymous mutations were remained. The germline and somatic mutations, the somatic mutations, and the nonsynonymous mutations were sequentially obtained, for each patient.
Acquired Mutation Analysis: Totally 42 DCB patients were performed to compare the genetic alteration (nonsynonymous mutations with high affect) between BL and PD. Acquired mutation analysis was performed on the subgroups of driver gene (EGFR, ALK, and ROS1) negative lung adenocarcinoma patients (n = 14), lung squamous carcinoma patients (n = 6), and driver gene positive LUAD patients (n = 19), respectively. The types of acquired mutations, the numbers of acquired mutations, and the mutation frequency of acquired mutational genes were analyzed.
Analysis of Acquired Mutations in NB Patients: Totally 26 NB patients were performed with the same cfDNA and ctDNA profiling at BL. The mutations with top frequency were compared to the landscape of each NB patient. The correlation between acquired mutations and initial anlotinib resistance was discussed based on the data generated in 40 anlotinib DCB patients and 26 anlotinib NB patients.
Ward Method for Cutoff Determination: Survival analysis was performed to obtain significance P values by calculating the correlation between predictors (G+S MB, N+S MB, and UMS) and PFS/OS, and Kaplan–Meier plots were made with the R package “survival” or GraphPad Prism 5. According to mutation burden or TMI (from low to high), the P value of stratification was obtained sequentially. The P values were compared, and then the lowest P value set as the cutoff was selected out. This method is suitable for all PFS and OS analysis.
Clinical Efficacy Analysis: Objective response to anlotinib was evaluated by at least one investigator according to CT scan. Here, the patients with stable disease or partial response lasting 130 days were defined as DCB, while those patients with 45 days < PD ≤ 130 days were defined as NDB, and the patients with PD ≤ 45 days or anlotinib intolerance were defined as NB. For patients with ongoing response to anlotinib therapy, PFS was censored at the date of the most recent imaging evaluation. For the factor of alive or death, OS was censored at the date of last known contact.
G+S MB and N+S MB for Anlotinib Response: Kaplan–Meier curve analysis was performed to evaluate the correlation between mutation burden and anlotinib response. The cutoff P value was determined by the “Ward method.” Determination of ongoing response and living status was described as “clinical efficacy analysis.” The significance P value was obtained by comparing the median PFS or median OS between those with a high mutation burden and with a low mutation burden. The ROC curves for predicting PFS and OS were generated by the cutoff P value of mutation burden using GraphPad Prism. AUC (95% CI) and null hypothesis test P were determined by ROC.
UMS Used for Anlotinib Response Analysis: The mutation tables were generated with a custom Python script, in which each row indicated a specific mutation and each column indicated a sample. Each cell of the mutation table denoted the sequencing depth and VAF of the corresponding mutation. Survival analysis was performed for the samples of the discovery cohort at BL using R package “survival” for each single mutation. Patients were classified into 2 groups (positive or negative) based on whether the patient had this mutation. Each mutation was examined against the PFS to test whether this mutation could significantly reduce the PFS for the mutation‐positive group. Then the Wilcoxon P value was adjusted by the BH method. A total of 120 candidate mutations passed the cutoff with the adjusted P value. These mutations served as candidates that could significantly decrease PFS.
Finally, based on the 120 candidate mutations, a scoring system was developed to evaluate the risk of the patient. Each positive mutation shared the same weight and was scored as 1. For example, one patient would receive a score of 10 if the patient had 10 such mutations. Then, patients were grouped into 2 groups based on the scoring system, namely, the negative (no such mutation) and high‐risk (more than 1 mutation) groups. Then, Kaplan–Meier survival analysis was performed against with PFS or OS using the same method as above to test whether such a scoring system could differentiate low‐ and high‐risk patients.
TMI Generation: The process of generating the TMI is shown in Figure S4 in the Supporting Information. TMI is based on three different anlotinib predictors (G+S MB, N+S MB, and UMS). Distinguishing anlotinib responders and anlotinib nonresponders using above three predictors, each anlotinib responder will score 50 points as BL. According to the significance of Kaplan–Meier curve analysis and ROC curve analysis upon different predictors, the significantly different P values < 0.05 scored 1, P values < 0.01 scored 2, and P values < 0.001 scored 3, in Kaplan–Meier curve analysis for PFS and OS. AUC values > 0.7 scored 1, and a null hypothesis test P value < 0.05 scored 1, < 0.01 scored 2, and < 0.001 scored 3 in ROC curve analysis for sensitivity and specificity. A score was allocated to each subgroup according to the above standards. Each patient obtained a score based on the characters of demographic data (such as gender, smoking status, LUAD, negative driver gene, and ≤3 metastases). Under the scoring approach, each patient obtained three independent BL scores and subgroup scores based on three predictors (G+S MB, N+S MB, and UMS). The six values above were added together to obtain a total score for each patient. Homogenization was performed according to the formula TMI = 100 × (300−score)/300, and then the TMI score was obtained for each patient. The TMI was used as a predictor, the “Ward method” was performed to determine the cutoff, Kaplan–Meier curve analysis was used to test anlotinib response stratification, and ROC curve analysis was performed to evaluate the predictive value.
Composition Analysis: According to the demographic characteristics, all the patients were divided into 10 subgroups (male, female, smoking, non smoking, LUAD, LUSC, driver gene positive, driver gene negative, >3 metastases, and ≤3 metastases). The composition of each subgroup was distinguished by the obtained predictors, such as the proportion of men and women in G+S MB <4000, and among others.
Anlotinib Response Analysis in Subgroups Using the Predictors of G+S MB, N+S MB, UMS, and TMI: In the discovery cohort, anlotinib response analysis was performed on different subgroups using the above predictors. After these analyses, the P values of Kaplan–Meier curve analysis of the PFS/OS, the AUC values of area under ROC curve, and null hypothesis test P values were used for subgroup response analysis.
Data Availability: Clinical information and predictor scores for this cohort can be found in NCBI database. The BioSample accession address is https://www.ncbi.nlm.nih.gov/biosample, Submission ID: SUB1189225. Mutation list called by Varscan2 with default parameters appeared in GTR database. The Laboratory accession number is GTR000568272; the Submission ID is SUB5954608. These data are also shown in Tables S4–S6. Providing access to the raw sequencing reads was not possible due to the restrictions of the project supporter (Chia‐tai Tianqing Pharmaceutical Co. Ltd.). Raw sequencing data sharing was upon request to Dr. Baohui Han (xkyyhan@gmail.com, 18930858216@163.com).
Statistical Analysis: The Wilcoxon test was used to compare Kaplan–Meier curves during TMI generation. A log‐rank test was used to compare Kaplan–Meier curves in the validation cohort and subsequently stratify the analysis. Unpaired t test was used to compare the mutation burden between DCB and NDB. The ROC curve was determined by plotting the rate of DCB at various cutoff settings of predictors. That is, the proportion of all DCB patients with a mutation burden above any given cut point (sensitivity) was plotted against the proportion of the NDB patients who would also exceed the same cutoff point (1− specificity). The AUC and exact 95% confidence intervals were reported. To examine the credibility of stratification, null hypothesis test was performed to analyze the ROC curve. Statistical analyses were performed using GraphPad Prism 5. Differences were considered significant at *P < 0.05, **P < 0.01, and ***P < 0.001.

For patients in the surgery plus anlotinib cohort, the starting dose of anlotinib was 8 mg once a day for 2 weeks of treatment followed by the cessation of treatment for 1 week. Preoperative and postoperative medication should be used for at least 4 courses, respectively. Postoperative medication should be prolonged to 1 year as far as possible, and the medication regimen and specific time of postoperative withdrawal should be comprehensively determined according to whether the patient had recurrence and adverse reactions. The dose was reduced to 6 mg if the patient experienced intolerable or uncontrolled pharmaceutical-induced toxicity. If a patient had relapsed following surgery combined with anlotinib and was observed with progressive disease or clinical symptoms, the dose will be increased to 10 mg (Eisenhauer et al., 2009 (link)); if a patient developed refractory adverse reactions during the subsequent two cycles of 10 mg, the drug would be permanently discontinued and other treatment approaches would be employed.

Mice bearing 4T1 tumors with a size of ~100 mm³ were treated by anlotinib or saline by daily intraperitoneal injection for 10 consecutive days. Once the tumors reached 2000 mm³ in volume, the experiment ended, and the mice were sacrificed. Tumor-draining LNs as well as lungs were isolated for H&E staining and imaging.