Introduction
We have been analyzing some single cell Next-generation sequencing (NGS) and multi-target cross-reactivity assay data from the Beacon B-cell workflow in PipeBio. In this post, we will summarize some of the analyses performed in PipeBio.
The data is from a study to detect antigen-specific and cross-reactive antibodies against vascular endothelial growth factor receptor (VEGFR) antigens in an immunized mouse. These are important cell surface protein receptors that mediate cellular responses to VEGF signaling proteins, most often in endothelium tissue. VEGF-VEGFR signals play an important role in physiological and pathological angiogenesis in diseases such as preeclampsia, aged macular degeneration, nephrotic syndrome, and cancer1.
This is a perfect example of how DNA sequence data analysis can be more powerful when integrated with functional assay data, and visa-versa.
The Beacon Optofluidic System
The Beacon Optofluidic System from Bruker Cellular Analysis (BCA; formerly Berkeley Lights) is an exciting technology that enables workflows for single cell analysis and functional characterization. Applications range from cell line development to T-cell profiling and antibody discovery. See Bruker’s webpage on the Beacon Optofluidic System for a more detailed overview of its capabilities.
This technology offers an alternative to lengthy B-cell and hybridoma workflows for antibody discovery, dramatically reducing timelines and complexity (Fig. 1). Early adopters have proven its efficacy for rapid VHH discovery directly from natural camelid immune repertoires2 and neutralizing antibody discovery against Covid-193, 4, 5.
Functional and NGS sequencing data from the Beacon B-cell workflow at single cell resolution
Researchers at BCA designed an antibody discovery campaign to investigate the ability to immunize a mouse with multiple related antigens (VEGFR1, VEGFR2, and VEGFR3) to yield both antigen-specific and cross-reactive antibodies. This study investigated the specificity and sensitivity of the system, as well as demonstrated a method to decrease immunization and screening campaign costs. This could help to validate the accuracy of the Beacon workflow when posed with a challenging cross-reactive multiplex assay, while also demonstrating the need for both functional assay and sequencing data to provide better insight about recovered sequences to carry forward for lead characterization and optimization.
In brief, optoelectrical positioning was used to move single mouse B-cells into individual Nanopen® chambers on OptoSelect® chips, on the Beacon platform (Fig. 2). Secreted antibodies were diffused from the Nanopen chambers into a common channel and their antigen binding specificity and cross-reactivity against VEGFR1, VEGFR2, and VEGFR3 were measured using time-lapse fluorescent imaging assays. As a positive control, fluorescence against mouse IgG was also tested. All screening assays were performed on cells across three chips, totaling ~35,000 cells screened.
A total of 3,106 Nanopen chambers showed antigen specific binding activity (Table 1), most pens showing specificity for one VEGFR antigen.
After the cellular assays, 1,142 B-cells were randomly selected for NGS. The mRNA encoding the variable heavy (VH) and light (VL) chains of IgG were sequenced using Illumina NGS (Fig. 3). Paired-end 150bp reads were assembled into full-length heavy and light variable regions (contigs) for each single cell using Bruker’s proprietary PrimeSeq™ BCR software.
Analysis of Beacon Data in PipeBio
An overview of the analysis pipeline performed is shown in Figure 4.
A total of 2,466 assembled contig sequences (1,041 VH and 1,425 VL) were imported into PipeBio along with the binding assay data, which was associated with each cell. Sequences were annotated using an IgG scaffold and the mouse antibody germline database with IMGT definitions as a reference. Almost all sequences (2,422 in total, 98.2%) were correctly annotated, and only two cells had no correct sequences.
VH and VL chains were paired together using a unique single cell identifier, resulting in 1,166 VH-VL pairs. Cells with only VH or VL sequences only were excluded from downstream analysis (199 cells). An additional filtering step further excluded cells with more than one VH-VL pair, resulting in 781 cells with a single VH-VL pair. Note, the cells with multiple VH-VL pairings can also be retained for further re-expression analysis. Figure 5 shows a summary of the number of cells and sequences at each step of the analysis.
To explore the chain pairing, a chain network diagram was generated for the 1,562 sequences (781 VH + 781 VL) found in the 781 cells (Fig. 6).
Most heavy and light chains form unique pairs, as shown by the sparse grouping pattern in the plot. Some larger nodes indicate the same VH and/or VL chains are used by multiple cells, but the grouping indicates variation in how specific sequences are being paired. For example, the grouping in Figure 7 shows a VH chain that appears in seven cells. In those seven cells, it is paired with different VL chains: the same two VL chains in two cells, another two VL chains in two other cells, and three additional VL chains in three other cells.
Clustering
The 1,562 sequences found in the 781 cells with a single VH-VL pair were clustered based on VH and VL independently with a 100% identity threshold, resulting in 627 VH clusters and 591 VL clusters (Fig. 8). The independent cluster assignments were combined to give 682 unique VH-VL clusters. This means that all sequences belonging to the same combined (union) cluster all have the same VH and VL sequence pairs. Most VH-VL clusters had one VH-VL sequence pair and 76 had more than one (Fig. 9).
Combining antibody sequences with binding assay data to identify reliable candidates
We aimed to find reliable antibody sequences that could be good candidates for further testing. If the same VH-VL sequence pair in different cells show the same binding profile, we can be more confident about the sequence and binding result. If, however, the same sequence shows different binding results in different cells, the sequence (or binding data) may not be so reliable.
To look for variability in binding profiles between cells with identical VH-VL sequence pairs, the clusters with one VH-VL sequence pair were filtered out. The remaining 76 clusters (175 VH-VL sequences) were used to generate honeycomb plots to group cells with the same sequences together and inspect their binding for the three antigens (Fig. 10).
The vast majority (88% on average) of VH-VL clusters show a consistent binding profile for cells with identical sequences (Table 2). The largest cluster of 5 identical VH-VL sequence pairs (toward the center of the honeycomb plots) shows a consistent positive binding for VEGFR2 and negative binding for VEGFR1 and VEGFR3 for all five member cells.
| VEGFR1 | VEGFR2 | VEGFR3 | Average | |
|---|---|---|---|---|
| Number of clusters where all cells agree |
64 | 67 | 70 | 67 |
| Number of clusters where not all cells agree |
12 | 9 | 6 | 9 |
The clusters that show a discrepancy in binding profiles (12, 9, and 6 clusters for VEGFR1, VEGFR2, and VEGFR3, respectively) can be considered as either unreliable sequences and/or unreliable antigen-specific calling on the Beacon system. It may be prudent to examine those pens again in more detail and/or possibly exclude these sequences from downstream analyses. Focusing downstream lead discovery efforts on the clusters with more consistent and reproducible binding profiles may be better.
Combining antibody sequences with binding assay data to examine cross reactivity
Eighty-eight of the original set of B-cells showed evidence of cross-reactivity (Table 1). If the cross-reactivity can be verified in independent cells with the same VH-VL sequence pair, we can be more confident that the cross reactivity is a genuine biological signal. If, however, cells with the same sequence show different cross reactivity profiles, it could be a technical artifact.
To examine cross-reactivity in some more detail, a multiple alignment of the VH sequences from the 76 clusters with more than one VH-VL sequence pairs (i.e. 175 VH sequences) was performed using the MAFFT algorithm6. and a phylogenetic tree was generated for visualization (Fig. 11). The binding assay data was plotted against the sequences in the tree.
Only two cells (D82744_13464 and D82744_614) showed evidence of cross reactivity between VEGFR1 and VEGFR2, in this subset of sequences. These two cells belong to clusters 132 and 656, respectively. All of the sequences in these clusters were extracted and re-aligned for a closer look (Fig. 12).
The cross-reactive cell D82744_13464 is part of a cluster of three cells (cluster id = 134), all with identical VH and VL sequences. The cross-reactivity is not observed in the other two cells. Indeed, all three cells differ in their binding profiles. D82744_9959 is negative for all three antigens and D82749_8272 is positive for VEGFR2 only.
The cross-reactive cell D82744_614 is part of a cluster of two cells (cluster id = 656). The identical VH and VL sequences are showing different binding profiles, with D82747_11281 showing positive binding for VEGFR2 only.
The lack of concordance between cells of the same sequence and their cross-reactive binding profiles suggests the cross-reactive signal might be a technical artifact in these specific cases. For example, we might need to look again at the choice of assay reagents, fluorophores, and/or how the Beacon binding assay activity was assessed.
Conclusion
In this post, we demonstrate that the Beacon system can efficiently recover high quality variable heavy and light chain sequences from cells that have been pre-screened for functional binding. Furthermore, the sequence and assay data can be brought together and analyzed in tandem in the easy-to-use PipeBio platform.
Relying on single cell assay data alone for calling antigen specificity and/or cross-reactivity can be challenging due to cryptic technical artefacts. Likewise, relying on sequence data alone can be equally challenging, because of a lack of phenotypic context. When analyzed together, there are new analyses that can be performed to accelerate decision making toward more promising leads and allowing us to dive more deeply into the accuracy of the initial assay scoring.
This post demonstrates some of those new analyses. In some of the clusters listed above where identical sequences were recovered from multiple cells/pens, but assay results did not align, this could prompt deeper investigation into those assay results, which may not have been the case without the sequence data.
It is worth pointing out that the same data set can be analyzed in different ways within the flexible PipeBio platform. To learn more about other software tools to uncover different patterns in your data, please look at our Tools and Features section.
Ready to try analyzing your Beacon data in PipeBio? Click here to sign up for a trial for free!
Other recent posts
