To screen patients with cardiovascular diseases for their risk of cancer, our project aims to develop a LIRA-based system to target miRNA that are both highly expressed in CVD patients and capable of promoting cancer. LIRA is a designed RNA sequence with stem-loop structures, which could be activated by its target RNA to initiate the expression of reporter gene. To establish an ideal LIRA system, our model group mainly need to solve the following four questions.

To answer , we applied differential expression analysis in GEO database to identify miRNA that are highly expressed in patients post myocardial infarction (MI) or with heart failure. After differential analysis of miRNA in cancer database from TCGA, we identified miR-210-3p and miR-142-3p as ideal targets, as they are both highly expressed in multiple types of cancer and in post MI patients. Additionally, we also verified that these two miRNAs could promote migration and invasion of cancer cells with wet lab experiments.

To address and , we designed single-arm LIRAs to target miR-210-3p or miR-142-3p initially. Next, we designed double-arm LIRAs to target miR-210-3p and miR-142-3p, which could only be activated in the presence of both miRNAs. To optimize the design of LIRA, we analyzed multiple indicators of LIRA, and used Multinomial Logistic Regression to evaluate the potential contribution of these indicators to the expected performance to LIRA. To improve model performance, we introduced Random Forest to identify key indicators that could affect the stability and function of LIRA.

To answer , we conducted kinetic modeling to simulate the reaction of LIRA-mediated expression of reporter gene to obtain predicted time and concentration curves of the reaction. Furthermore, using 3D curves, we simulated the reaction with miR-210-3p and miR-142-3p at different concentrations, and estimated suitable concentration of LIRA for detecting miRNAs at a variety of concentration.

In summary, our model work laid the ground for our wet lab experiments that generates the LIRA to target miR-210-3p and miR-142-3p, and paved the way for the future application of the LIRA in the screening of cancer risk in post-MI patients.

Epidemiology studies have revealed that patients with CVDs, such as myocardial infarction (MI), heart failure (HF) and atherosclerosis, have increased risk of getting cancer^[1-4]. For instance, MI patients exhibit increased prevalence of a variety of cancers and also elevated mortality that are caused by cancer^[1]. HF patients also show higher cancer risk compared to healthy individuals^[5]. Experiments using animal models reveal that MI enhances tumor progression through circulating factors and immune reprogramming^[6,7]. Especially, MI leads to aberrant miRNA secretion from the inflicted heart into blood, which could circulate to distant organs and contribute to cancer progression at distant organs^[8]. To find good targets for our cancer screening for CVD patients, we analyzed database of CVD, including MI and HF, and databases of multiple cancers, to identify miRNA that show increased expression in both CVD and cancer, and then verified the function of the identified miRNA on cancer progression.

The whole process of our analysis of MI and cancer database is shown in Figure 1. First, we searched Gene Expression Omnibus (GEO) database using myocardial infarction and miRNA as key words and collected 318 datasets. Since our aim is to identify highly expressed miRNA in the blood of MI patients, we manually excluded 305 irrelevant datasets based on the sample sources, species and disease descriptions. To the remaining 13 datasets, we calculated the log₂FC value of the miRNAs based on the ratio of the mean values of the patient samples and control and applied logarithmic transformation to the ratio^[9]. We used Student's t-test to obtain P-values for each miRNA. Furthermore, to reduce the occurrence of false positives, we adjusted the P-values by comparing these P-values with the threshold selected for an acceptable false discovery rate (FDR)^[10], and used adj.P value to refer FDR. We used the log₂FC > 1 and adj.P Value < 0.05 as our screening criteria^[11], and got 9 datasets that contained significantly upregulated miRNA.

Second, we summarized multiple epidemiology studies and identified 33 types of cancers showing faster progression or higher risk in CVD^[12-16]. We conducted differential expression analysis of miRNA expression of these 33 types of cancer using the Cancer Genome Atlas Program (TCGA) database and obtained 22 cancer datasets that show significantly upregulated expression of miRNA using our screening criteria (log₂FC > 1 and adj.P Value < 0.05). The names of these 22 types of cancers are listed in Table 1.

We compared the significantly upregulated miRNAs in MI and cancers to identify miRNAs that are significantly upregulated in both diseases and obtained 24 miRNAs. We think that two criteria are especially important for the potential effects on cancer promotion of these 24 miRNAs during CVD:

1. log₂FC values in MI. A high log₂FC value of the miRNA in MI demonstrates a large increased amount of this miRNA in the blood of MI patients, which indicates that a large amount of this miRNA is likely to reach distant organs through circulation system and exerts its potential cancer-promoting effects at distant organs.

2. Types of cancers that show significantly upregulated expression of these miRNAs. MiRNAs that are highly expressed in multiple types of cancers could potentially influence the progression of multiple types of cancers. We would like to design a screening kit to identify MI patients with increased risk of getting of a variety of cancers, instead of a certain type of cancer.

We summarized this two key information of the 24 miRNAs in Table 2.

To further evaluate the log₂FC in MI and types of cancers of these 24 miRNAs, we applied a scatter dots graph to visualize these data (Figure 2). MiR-210-3p and miR-142-3p distinguish from other miRNAs in this graph, as they show both much higher log₂FC values in MI and much more types of cancers that showed significantly upregulated expression of them than other miRNAs.

As shown in Table 2, miR-210-3p ranks the highest in log₂FC value in MI (6.18), which demonstrate 73 folds of increased expression in MI patients, and show significantly upregulated expression in 13 types of cancers, with log₂FC values in cancer ranged from 1.02 to 5.04. The log₂FC values of miR-210-3p in 13 types of cancers are shown in Table 3.

miR-142-3p ranks the second highest in log₂FC value in MI (3.51), which demonstrated 11 folds of increased expression in MI patients, and show increased expression in 10 types of cancers, with log₂FC values in cancers range from 1.04 to 6.61. The log₂FC values of miR-142-3p in 10 types of cancers are shown in Table 4.

To summarize, our analysis shows that miR-210-3p and miR-142-3p are the best target miRNAs to identify MI patients with high risk of getting cancers.

HF is a condition resulting from the impairment of cardiac contractility and/or diastolic function^[17].Current study also reveals that HF can promote cancer^[6]. Hence we also analyzed miRNAs that are highly expressed in HF patients in GEO database. The analysis process is shown in the flowchart (Figure 3).

We collected 158 datasets from GEO database using heart failure and miRNA as key words. After we manually excluded 151 irrelevant datasets based on the sample sources, species and disease descriptions, we got 7 datasets to analyze miRNA expression. Using log₂FC > 1 and adj.P Value < 0.05 as screening criteria, we got 6 datasets that contain significantly upregulated miRNAs in HF. We compared the significantly upregulated miRNAs in HF with those in cancers, which was described in the previous section, we got 35 miRNAs that show significantly upregulated expression in both diseases. We summarized the log₂FC in HF and types of cancers that show significantly upregulated expression of these miRNAs in Table 5.

To further evaluate the log₂FC in HF and types of cancers of these 35 miRNAs, we applied a scatter dots graph to visualize these data (Figure 4). Unlike the results of analysis in MI, these miRNAs all clustered in the lower-left part of the graph, demonstrating that no miRNA showed superior performances on both log₂FC values in HF and types of cancers that showed significantly upregulated expression.

In summary, we did not identify ideal miRNAs as targets for cancer screening from our analysis of HF database.

From our analysis of MI database, we found that miR-210-3p and miR-142-3p are two outstanding candidates biomarkers to identify MI patients with high risk of cancer. We used an ROC curve to evaluate the diagnostic performance of miR-210-3p and miR-142-3p on our collected raw dataset samples from MI (Figure 5). We found that combined ROC curve of miR-210-3p and miR-142-3p (AUC=0.78) showed better results than ROC of miR-210-3p alone (AUC=0.66) and ROC of miR-142-3p alone (AUC=0.60), which demonstrates that using miR-210-3p and miR-142-3p as target miRNA simultaneously could decrease the probability of false discovery rate of detecting MI than using these 2 miRNAs as target miRNA separately.

We also revisited our HF screening dataset to specifically analyze miR-142-3p and miR-210-3p. MiR-142-3p was identified from our analysis of HF dataset with log₂FC value (2.13), which demonstrated a 5 folds increased expression (Figure 4). MiR-210-3p failed to be selected from our analysis of HF dataset due to the log₂FC value (0.67) did not pass our screening criteria.

From our analysis, we suggest using both miR-210-3p and miR-142-3p together as targets to screen MI patients to identify those with high risks of getting cancer.

Through bioinformatic analysis, we identified miR-210-3p and miR-142-3p as potential targets for screening cancer risk in MI patients. We did literature research of these two miRNA, and found that they are involved in the regulation of migration and invasion of cancer cells^[18-21] .To verify whether miR-210-3p and miR-142-3p can promote migration and invasion of cancer cells, mimics and inhibitors of miR-210-3p or miR-142-3p were transfected into human renal clear cell carcinoma cell line 786-o cells, and the transfected cells were tested by Wound Healing and Transwell assays.

To verify the function of miR-210-3p in cancer cells, we did Transwell and Wound Healing assays with 786-o cells transfected with mimics of miR-210-3p (mimics) and inhibitor of miR-210-3p (inhibitor). In Wound Healing assay, we evaluated the relative migration levels through measuring the wound area of 786-o cells transfected with control, mimics and inhibitor of miR-210-3p. Statistical analysis of the Wound Healing assays shows that 786-o cells transfected with mimics of miR-210-3p migrated faster than control cells and 786-o cells transfected with inhibitor of miR-210-3p migrated slower than control cells (Figure 6). Next, we evaluate the invasion levels through counting the amounts of 786-o cells that colonized in the lower well of transwell assays. Statistical analysis demonstrated that 786-o cells transfected with mimics of miR-210-3p showed higher invasion capability than control cells and 786-o cells transfected with inhibitor of miR-210-3p showed lower invasion ability than control cells (Figure 7). To summarize, 786-o cells with overexpression of miR-210-3p have increased migration and invasion capabilities, demonstrating that miR-210-3p could promote the migration and invasion of cancer cells.

Similarly, to verify the function of miR-142-3p in cancer cells, we did Transwell and Wound Healing assays with 786-o cells transfected with mimics of miR-142-3p (mimics) and inhibitor of miR-142-3p (inhibitor). In Wound Healing assay, we evaluated the relative migration levels through measuring the wound area of 786-o cells transfected with control, mimics and inhibitor of miR-142-3p. Statistical analysis of the Wound Healing assays shows that 786-o cells transfected with mimics of miR-142-3p migrated faster than control cells and 786-o cells transfected with inhibitor of miR-142-3p migrated slower than control cells (Figure 8). Next, we evaluate the invasion levels through counting the amounts of 786-o cells that colonized in the lower well of Transwell assays. Statistical analysis demonstrated that 786-o cells transfected with mimics of miR-142-3p showed higher invasion capability than control cells and 786-o cells transfected with inhibitor of miR-142-3p showed lower invasion ability than control cells (Figure 9). In summary, 786-o cells with overexpression of miR-142-3p have increased migration and invasion capabilities, demonstrating that miR-142-3p could promote the migration and invasion of cancer cells.

Together, the results of transwell and Wound Healing assays demonstrated that increased expression of miR-210-3p and miR-142-3p in MI patients might enhance the invasion and migration of cancer cells, which supports our idea of developing a screening kit that targets miR-210-3p and miR-142-3p to MI patients with high risk of cancer.

To design a screening approach that can target miR-210-3p and miR-142-3p, we adapted from an innovative approach named Loop-Initiated RNA Activator (LIRA) that was used to detect viral RNA^[22]. As shown in Figure 10, LIRA forms a stem-loop structure that mainly contains a recognition region, a ribosome binding sequence (RBS) and a start codon AUG. Without interaction with target RNA, the RBS and start codon in LIRA are kept in a close status in the stem structure, making them impossible to initiate translation of downstream reporter gene. The recognition region is a 31 nucleotides (nt) sequence that is reverse complimentary to the target RNA of LIRA. While 21 nt of the reverse complimentary sequence is designed to locate in the loop structure of LIRA, which is named as a*, 10 nt of the reverse complimentary sequence is designed to locate in the stem structure of LIRA, which is named as b*. As a result, the target RNA of LIRA can bind to both a* and b*, and the binding of the target RNA to b* destabilize the stem structure that locks RBS and start codon, which makes them to be exposed in an open status for the translation initiation of downstream reporter gene. Due to the complexity of secondary structure and complicated hybridization process of RNA, we used NUPACK software to design and analyze LIRA sequences in the subsequent study.

To design LIRA sequences that target miR-210-3p and miR-142-3p separately, we first determined the composition and structure of each part of LIRA, and replaced the recognition region with reverse complementary sequence of miR-210-3p and miR-142-3p. We used the design function of NUPACK to design the rest of LIRA sequences. Next, we used the analysis function of NUPACK to simulate the hybridization process of LIRA, and the nucleotide bases in the mismatch region of the LIRA with abnormal structure were modified to achieve our design expectation, which is to keep RBS and AUG at a closed status in a stem region without interactions with target miRNA and release RBS and AUG to an open status in a single strand RNA with interactions with target miRNA. The procedure of our LIRA designing is listed below.

During the process of designing, we found that our situation differed from the case reported in the literature in that the target RNA used in the literature was 31nt, while the target miRNAs used by our team were 22nt and 23nt respectively^[18]. Due to the difference in the length of the target RNA, we were unable to use the way of LIRA design reported in the literature directly. To determine the distribution of the recognition region on our LIRA, we defined a concept of stem-loop ratio: the ratio of the number of nucleotides of recognition region in the stem structure to the number of nucleotides of recognition region in the loop structure. We tried to design LIRA structure using different stem-loop ratios. We found that when the number of nucleotides of the recognition region in the stem is too large, it will partially overlap with the complementary sequence of RBS, which will lead to the exposure of the RBS sequence. Therefore, we designed LIRA to miR-210-3p with the stem-loop ratios from 0/22 to 13/9 and LIRA to miR-142-3p from 0/23 to 13/10. The hybridization of LIRA to their target miRNAs was also analyzed with NUPACK. The pairing results and related data are shown in Table 6 and Table 7.

We think that the greater the free energy of a LIRA, the easier it is for LIRA structures to be paired with miRNAs for translation; the smaller the free energy of a LIRA, the more stable the structure of LIRA is. Therefore, we chose free energy as an important index to analyze the hybridization of LIRAs to their target miRNAs. We calculated the changes of free energies of the LIRA before and after hybridization and fitted the changes of free energy to stem-loop ratio. From the fitting curve, it seems that there is no regularity between the changes of free energy and stem-loop ratio (Figure 12 and Figure 13).

Since the LIRA sequence designed by NUPACK algorithm in the unrestricted region is random, which causes large variations among these designed LIRA. Therefore, we designed 10 sequences for each stem-loop ratio and averaged the free energy values of these LIRA. A new fitting was performed with the averaged free energy and stem-loop ratio (Figure 14 and Figure 15). The new fitting curve demonstrates that there is no linear relationship between LIRA free energy and stem-loop ratio, indicating that LIRA would not become more or less stable as stem-loop ratio increase or decrease.

Next, we calculated the changes of free energy as the differences of free energy of LIRA before and after hybridization with its target miRNA, and analyzed the relation between LIRA free energy and the changes of free energy. We observed that the data had a clear linear relationship, so we linearly fitted it. The fitting results are shown in Figure 16 and Figure 17. The slope of the fitted curve is approximately equal to -1, which indicates that the free energy of LIRA after hybridization with its target miRNA is stable at a constant value. The free energy of LIRA hybridized with miR-210-3p is around -54 kcal/mol, and the free energy of LIRA hybridized with miR-142-3p is around -44 kcal/mol. These results show that free energy of LIRA before hybridization might be a more important indicator to determine the quality of a LIRA.

Furthermore, we averaged 10 LIRA free energies at the same stem-loop ratio, which could represent the LIRA free energy and hybridization free energy at certain stem-loop ratio. Subsequently, we performed a clustering with these 28 points, and observed a linear relationship between LIRA free energy and the averaged changes of free energy (Figure 18 and Figure 19).

From our analysis of LIRA to miR-210-3p and miR-142-3p, we concluded that stem-loop ratio has little effect on the free energy of LIRA. For our subsequent wet lab experiments, we chose 3 LIRA sequences for miR-210-3p with stem-loop ratios of 2/20, 7/15, and 10/12, and 3 LIRA sequences for miR-142-3p with stem-loop ratios of 3/20, 8/15, and 11/12. We selected these 6 LIRA sequences to verify whether stem-loop ratio indeed had no effect on the performance of LIRA. Meanwhile, since the free energies of these 6 LIRAs are very different, the influence of the LIRA free energy on the functions of LIRA can also be investigated (Table 8). In this regard, we expect that a stable LIRA structure is often accompanied by a lower free energy.

To target miR-210-3p and miR-142-3p simultaneously, we designed LIRAs with double-arm structure that can target two target RNA^[22]. As shown in Figure 20, double-arm LIRA has two recognition regions, A* and B*. While A* is located on the short arm of the double-arm structure, B* is located on the long arm of the double-arm structure, which is connected to RBS and AUG. We think that putting one miRNA reverse complimentary sequence in A* and the other miRNA reverse complimentary sequence in B* would make our LIRA targeting two miRNAs. Our expectation is that only when the LIRA binds to the target miRNA A and then bind to the target miRNA B, it would expose the RBS and AUG in an open status to initiate the translation of the downstream reporter gene, and otherwise the RBS and AUG would be locked to at a closed status in stem region.

We identified the composition of each part of double-arm LIRA and replaced the recognition region with sequences that were reverse complimentary to miR-210-3p and miR-142-3p. Utilizing the design function of NUPACK, we designed a primary double-arm LIRA structure and obtained the sequence. The following is the design procedures of LIRA sequences to detect miR-210-3p and miR-142-3p in NUPACK.