Close this search box.

In vivo phage display identifies novel peptides for cardiac targeting – Scientific Reports

In vivo phage display to identify peptides binding to the ischemic cardiac region

Phage display technology can be employed to identify targeting peptides. We performed selection of the phages in a mouse model of ischemia–reperfusion (Fig. 1A). The ischemia–reperfusion model used involved temporary restriction of blood supply to the heart followed by its restoration and mimicked the clinical situation in patients with MI9. We conducted our experiments four days after the surgery to increase the chances of identifying peptides that specifically recognized the diseased heart tissue. To evaluate heart function after the surgery, echocardiography was performed and the ejection fraction (EF) was reduced to less than 50% for all examined mice except one, which had EF at the level of a healthy heart (70–75%) (Fig. 1B,C).

Figure 1
figure 1

Identification of peptides targeting ischemic cardiac regions using phage display. (a) Illustration of the experimental workflow. Ninety-six hours after ischemia–reperfusion injury was induced, mice were injected with phage display libraries via the tail vein for in vivo phage display. Animals were terminated 15 min later and their damaged left ventricular (LV) region, remote heart, kidney, and liver were collected. Phages from the damaged LV tissue were amplified and injected into new animals with ischemia–reperfusion surgery for the next round of phage display. After the third round of phage display, phages from liver, kidney, healthy heart, and scar tissues were amplified and sequenced. Data analysis was followed by peptide validation. (b) To evaluate heart condition after ischemia–reperfusion surgery, echocardiography was performed across LV long-axis; at the level of the ventricle base; at approximately the level of the papillary muscles; and at the apex level. (c) Ejection fraction (EF) of the animals at day 3 after ischemia–reperfusion injury. Ejection fraction of healthy heart is indicated by a pink bar. N = 2 for groups treated with the phages and N = 4 for the healthy control. (d) Phage libraries overview. TriCo-20 includes peptides of 20 amino acids attached through a serine linker to the phage coat protein. PhD-12 consists of 12 amino acids peptides with a glycine-serine linker. (e) Schematics for data filtration steps. First, reads with stop codons in the peptide sequences were eliminated. Next, reads that were not translating to amino acid sequences were removed from the initial dataset. Peptide length and library patterns allowed us to sort out unrelated reads. Last, duplicated amino acid sequences were merged. (f) Total number of peptides at each filtration step as described in (a). (g) Filtered data set was used to calculate the number of reads in each tissue per panning round of the phage display experiment.

With the aim for identifying peptide sequences specifically associated with myocardial infarction, we subjected phage display libraries to three rounds of selection against ischemic left ventricular (LV) cardiac tissue. We used two commercial phage display libraries, TriCo-20 and PhD-12, that displayed random peptide sequences in the pIII coat protein of M13 bacteriophage (Fig. 1D). These libraries were administered to mice after ischemia–reperfusion by intravenous tail vein injection (Fig. 1A). These phages were allowed to circulate within the bloodstream to encourage interaction with various cell types, including cardiomyocytes in the damaged heart region, for 15 min then were washed away by perfusing animals before termination. Damaged LV cardiac tissue was isolated and retained phages were recovered and amplified for the next round of screen. We additionally collected remote heart tissue (the rest of the heart after removing damaged LV and without aorta), liver and kidney as controls for binding specificity. After the last round of screen, phages from tissues of interest were collected and the DNA sequences encoding the displayed peptides were determined by NGS.

Sequential rounds of phage display increase the proportion of peptides in the heart over other tissues

The NGS data obtained from the phage display experiments underwent a series of filtration steps to remove artifacts, noise in order to improve the quality of the data (Fig. 1E). Sequences containing stop codons or non-amino acid reads were eliminated first, the remaining sequences were then checked for conformity to the library length and pattern to ensure the presence of the correct amino acids preceding the targeting peptides. Finally, any duplicated peptide sequences originated from distinct DNA sequences were merged, ensuring a unique representation of the identified peptides in the final dataset (Fig. 1F).

The total peptide count changed over the rounds of phage display selection. The total peptide count was high in the first round (322761 for PhD-12, 279910 for TriCo-20), reflecting the diversity of peptides presented in the starting phage library (Fig. 1G, Supplementary Fig. S1A). As the selection progressed, peptides that did not bind specifically to the target of interest were gradually depleted from the population, leading to reduced total peptide count of approximately threefold (104660 for PhD-12, 96038 for TriCo-20). Conversely, the total peptide count of the phage population from the previously ischemic LV region increased by 41% due to the amplification of phages displaying peptides with higher affinity for the target (Fig. 1G). Both libraries showed a comparable total number of peptides in different tissues (Supplementary Fig. S1A,B), while in the last round over three times more phages were detected in the library PhD-12 in the remote heart region (Supplementary Fig. S1C).

Bioinformatic approaches to identify cardiac targeting peptides in the phage display data set

For further data analysis, we applied three complementary approaches Tissue-Oriented Peptide Identification and Clustering of the top 20 hits (TOPIC20), Tissue-Oriented Peptide Identification and Clustering (TOPIC) and Specificity Evaluation by Clustering (SPEC) to identify cardiac targeting peptides (CTPs) within the obtained dataset (Fig. 2A). TOPIC and TOPIC20 are more traditional approaches, that considered peptide count and uniqueness of sequences for the ischemic LV region, while SPEC is more advanced approach that additionally employed clustering techniques to identify potential similarities between peptides and investigated their tissue specificity.

Figure 2
figure 2

Bioinformatic analyses of the phage display data to identify hits. (a) Illustration of bioinformatic approaches for analyzing the in vivo phage display data. The key parameters included in the analysis are high count, unique sequences, clustering, and tissue specificity. (b) Description of the TOPIC20 data analysis approach, which combines filtering and manual clustering of the top 20 peptides. (c) Consensus sequence analysis of the top 20 hits identified in TOPIC20. (d) Description of the TOPIC data analysis approach, which includes filtering and clustering of all filtered peptides. (e) Overview of the clustering of the peptide sequences after the filtering step in TOPIC. (f) Consensus sequences corresponding to top clusters identified as hits in TOPIC. Amino acids differences between consensus sequences identified by TOPIC20 and TOPIC (peptides AZ#2 vs. AZ#2.1, AZ#3 vs. AZ#3.1) are highlighted in bold. Count assigned to the consensus sequence is a sum of peptide counts in the cluster.

Tissue-oriented peptide identification and clustering of the top 20 hits (TOPIC20): tissue-specific filtering and manual clustering of the top hits

To identify potential CTPs, we initially considered the unique peptide count, which provided an indication of the abundance of peptides in a tissue compared to others. In this approach, we evaluated the peptides that were found in the ischemic LV region starting from round 3 (Fig. 2B). Then, we examined if these peptides were also present in other tissues with a counts greater than 0.01% of total counts in the tissue in the round 3, this threshold allowed us to account for rare events of nonspecific binding. If a peptide was detected in other tissues above this threshold, it was considered non-specific to the previously ischemic LV tissue and was removed from the dataset.

To further analyse the remaining peptides, we selected the top 20 hits based on their counts. These selected peptides were then subjected to manual clustering based on sequence similarity to identify consensus sequences (Fig. 2C) and three distinct clusters were identified. To generate a consensus sequence from each identified peptide cluster, we assigned each amino acid in the cluster a “weight” corresponding to the count of the peptide it belonged to. This weighting scheme allowed us to prioritize amino acids that appeared more frequently, indicating their higher representation and potential importance, within the cluster. Subsequently, for each position within a peptide sequence, the amino acid with the highest count in that cluster was selected as the representative for that position in the consensus sequence. Based on this approach, we identified three consensus peptides from the TriCo-20 library as hits (Fig. 2C), but no hits with comparable counts were found in the PhD-12 library.

Tissue-oriented peptide identification and clustering (TOPIC): tissue specific filtering and Clustering of the filtered peptides

TOPIC20 approach has an issue of potentially losing relevant peptides by focusing solely on the top twenty hits. To mitigate this, we employed a different strategy that involved clustering all the peptides after applying the filtering step (Fig. 2D). After applying the same filtration strategy of removing peptides that were also present in other tissues at a frequency greater than 0.01% as in the previous approach, we clustered all the remaining peptides together as described in “Materials and methods”. This approach allowed us to capture a broader range of potentially relevant peptides, including those that had lower abundance but still exhibited important sequence similarities. Following the clustering step, we examined the clusters with most counts, which represented groups of peptides that shared common features or conserved regions (Fig. 2E). From these clusters, we could identify consensus sequences, which were representative sequences that summarized the shared characteristics of the peptides within each cluster. Count assigned to consensus sequence was a sum of peptide counts in the cluster.

We took the consensus sequence from the clusters with the highest count. Three hits belonged to the TriCo-20 library and were identical to hits identified in the previous approach, apart from a single amino acid replacement at the terminus (C-terminal E → D for AZ#2.1 and N-terminal N → H for AZ#3.1) (Fig. 2F). In addition, two hits were identified from the PhD-12 library (Fig. 2F). We did not identify these hits in the PhD-12 library using the previous approach because all peptide candidates had low counts, which became high once we clustered similar peptides together and assigned sum of the peptide counts to the consensus sequence.

Specificity evaluation by clustering (SPEC): clustering of all peptides and tissue specificity of the clusters

In TOPIC and TOPIC20 approaches we used pre-defined threshold to address peptide tissue specificity. In order to ensure that potentially important peptides were not overlooked applying this selection criteria we developed an alternative comprehensive approach that does not relying on this selection criteria (Fig. 3A). We first clustered all 7 988 uniquely identified peptides (Fig. 3B), we then focused on allocating these clusters within specific tissues, including the damaged LV region, remote heart, kidney and liver. We investigated each cluster representation across the tissues (Fig. 3C, Supplementary Fig. S2A). We prioritized clusters that were exclusive to the previously ischemic LV and those that showed a high number of reads in the whole heart (damaged LV + remote heart) for further investigation and downstream experiments (Fig. 3D). With this approach, we identified 8 hits, all of which were from the PhD-12 library. None of the clusters identified in the TriCo-20 library were specific for the damaged LV or for the remote heart (Supplementary Fig. S2A). There were a lot of clusters identified in the remote heart in library PhD-12 (Fig. 3C), we therefore additionally included an extra cluster from this area to the in vitro validation.

Figure 3
figure 3

Implementation of global clustering to bioinformatic analyses of the phage display data. (a) Description of the SPEC data analysis approach. All peptides from panning round 3 were clustered and the clusters’ tissue representation was examined afterward. (b) Clustering of all peptides identified in panning round 3 from libraries PhD-12 and TriCo-20. (c) Visual representation of identified clusters across different tissues in library PhD-12. The sum of the count of each peptide from each cluster is represented on the y-axis, cluster number on the x. Magenta circles indicate damaged LV area-specific clusters with the top 6 counts. The grey circles indicate clusters with specificity to the damaged LV region and the remote heart. The blue circle indicates the best hit among remote heart-specific clusters. (d) Consensus sequence of the clusters marked in (c). Count assigned to the consensus sequence is a sum of peptide counts in the cluster.

We also assessed whether it is necessary to consider clusters behaviour across all three rounds of the phage display experiment to confidently identify hits (Supplementary Fig. S2B). Thus, clusters specific for the damaged heart regions should get more counts in this region and less counts in the other tissues over the rounds. To explore this, we first clustered all the peptides identified in all the rounds of the phage display. Next, we visualized the presentation of clusters in the damaged myocardium and remote heart, the kidney and the liver over the course of three rounds (Supplementary Fig. S2C,D), which allowed us to identify clusters that were enriched specific to the target tissue. From the previously identified hits, three from the PhD-12 library showed enrichment in the damaged myocardium and remote heart, while number of counts reduced in the kidney and liver over the rounds (Supplementary Fig. S2E). This way of analysing the data appeared to be quite stringent, as only a few candidates from previously identified clusters satisfied the enrichment criteria.

Collectively, these approaches allowed us to comprehensively analyze the data obtained from different tissues, focusing on the identification of CTPs specific to the ischemic heart region. By considering factors such as peptide count, uniqueness, clustering, tissue specificity, and enrichment patterns, we uncovered novel potential CTPs.

In vitro validation of identified CTPs

After identifying hits through various bioinformatic analyses of the data, our next objective was to validate the results in in vitro experiments. To allow the visualisation of the peptides, we synthetised selected peptides AZ#1–AZ#12 with an additional lysine residue at the C-terminus to serve as the linker to the FITC fluorophore. We used human liver cells HepG2 as a negative control to assess non-specific peptide binding and used human induced pluripotent stem cell (iPSC)-derived cardiomyocytes and immortalized mouse cardiomyocytes HL1 to evaluate the targeting potential of the peptides toward cardiac cells. Human induced pluripotent stem cell (iPSC)-derived cardiomyocytes provided us with a relevant model of human cardiomyocytes. We confirmed the cardiomyocyte state of the iPSC-derived cardiomyocytes by immunofluorescence analysis of cardiac troponin T (Supplementary Fig. S3A). In order not to miss hits that are specific for mouse cardiomyocytes we also used immortalized mouse cardiomyocytes HL1.

We incubated cells with 10 μM of the selected peptides and Myc peptide as a non-targeting control for one hour, removing the peptides by replacing the media with imaging media, which also facilitated clear visualization of the cells, and acquired images using a high-throughput confocal microscope.

We first assessed the binding of the peptides to HL1 cells, which served as a model system closely resembling in vivo conditions. Peptides AZ#6, AZ#10 and AZ#12 gave the strongest fluorescent intensity signals (Supplementary Fig. S3B). The untargeted control peptide Myc bound to the plates, leading to high background signal, which was also observed at similar levels for peptides AZ#2, AZ#3, AZ#4, AZ#9, and AZ#11. Importantly, none of the tested peptides exhibited binding to liver cells above the background level, demonstrating the successful selection against liver binding in our assay (Supplementary Fig. S3C). The staining pattern we observed with the best hits suggested that 1 h incubation is sufficient to detect peptide internalization and intracellular accumulation. Subsequently, to provide insights into the potential translatability of the results obtained from the mouse model to humans, we employed human iPSC-derived cardiomyocytes to further evaluate the targeting properties of the peptides. In line with previous results, peptides AZ#6, AZ#10 and AZ#12 gave the strongest fluorescent intensity signals (Fig. 4A,B, Supplementary Fig. S4A). Peptides AZ#7 and AZ#11 showed weaker staining, suggesting a relatively low affinity for the cells (Supplementary Fig. S4A).

Figure 4
figure 4

In vitro validation of identified peptides with human iPSC-derived cardiomyocytes. (a) Representative confocal fluorescence microscopy images of live human iPSC-derived cardiomyocytes incubated for 1 h with indicated peptides labeled with the FITC fluorophore. Hoechst33342 dye was used for nuclei visualization. Blue lines indicate approaches used to identify the indicated peptide as a hit. Scale bar = 50 μm. (b) Human iPSC-derived cardiomyocytes were fixed after 1 h incubation with FITC-labelled peptides. Cell mask and Hoechst33342 dyes were used for membrane and nuclei visualization, respectively. Scale bar = 50 μm. (c) Graphical summary of the peptide evaluation. Blue lines at the right indicate approaches where hits were found. Magenta—positive hits, pink—hits with weaker interaction with the cells, green – hits with no binding/uptake detected in the in vitro assay. Grey and light grey lines mark the conditions where cardiomyocytes were hypoxia or ischemia-stimulated. Peptide AZ#4 was not evaluated in hypoxia and ischemia assays as indicated as NA.

We observed different background signals for different peptides. In order to address if these differences are coming from the unspecific binding of the peptides to the fibronectin coating of the plates, we tested peptide binding to coated and uncoated plates (Supplementary Fig. S4B,C). Indeed, peptides AZ#6, AZ#7, AZ#11, AZ#12 and Myc bind to fibronectin to a higher extent than the others, while once applied to the uncoated plates the pattern changes and Myc and AZ#9 give the most background signal.

After evaluating peptide specificity towards cardiomyocytes over the liver cell model we aimed to check peptide’s ability to interact with the “damaged” cardiomyocytes. We incubated cells in a low oxygen environment that would more closely mimic the changes in gene expression networks following a myocardial infarction. Thus, we kept iPSC-derived cardiomyocytes for 24 h in the hypoxia condition (0.8% oxygen) and followed with 6 h of reoxygenation. Afterward, peptides were added, and cells were imaged as previously. Peptide AZ#4 was not included in the validation, since we discovered its tendency to aggregate during the storage which makes it difficult for any future applications from the practical point of view. In line with previous results, peptides AZ#6, and AZ#10 gave the strongest fluorescent intensity signals (Supplementary Fig. S5A–C). In contrast, peptide AZ#12 did not demonstrate affinity to cardiomyocytes in the post-hypoxia condition.

Next, we aimed to evaluate peptides in the in vitro model that even more closely resembles in vivo MI conditions. We additionally changed cell media to nutrition-depleted as recommended in Hakli et al.22 during 24 h incubation with low oxygen to mimic the ischemic condition. Afterward cells were returned to maintenance media during 6 h of reoxygenation and treated with peptides as described before. Same as in the hypoxia stimulation experiment peptides AZ#6, AZ#10 gave the strongest fluorescent intensity signals, AZ#12 did not demonstrate affinity to cardiomyocytes (Supplementary Fig. S6A,B).

Altogether it suggests that AZ#6, AZ#10 keep their binding properties to iPSC-derived cardiomyocytes in three in vitro models: with “healthy” cardiomyocytes, with hypoxia-stimulated cardiomyocytes, and with ischemia-stimulated cardiomyocytes. We summarised these results in Fig. 4C.

Notably, all peptides that demonstrated binding to cardiomyocytes were identified using SPEC, emphasizing its effectiveness in identifying heart-specific peptides.