Machine-driven parameter screen of biochemical reactions

Abstract

The development of complex methods in molecular biology is a laborious, costly, iterative and often intuition-bound process where optima are sought in a multidimensional parameter space through step-by-step optimizations. The difficulty of miniaturizing reactions under the microliter volumes usually handled in multiwell plates by robots, plus the cost of the experiments, limit the number of parameters and the dynamic ranges that can be explored. Nevertheless, because of non-linearities of the response of biochemical systems to their reagent concentrations, broad dynamic ranges are necessary. Here we use a high-performance nanoliter handling platform and computer generation of liquid transfer programs to explore in quadruplicates 648 combinations of 4 parameters of a biochemical reaction, the reverse-transcription, which lead us to uncover non-linear responses, parameter interactions and novel mechanistic insights. With the increased availability of computer-driven laboratory platforms for biotechnology, our results demonstrate the feasibility and advantage of methods development based on reproducible, computer-aided exhaustive characterization of biochemical systems.

Figure 1.

Method overview. This conceptual drawing shows a 384-well plate where each individual well is recognized by a combination of 64 different ‘barcodes’ and six different ‘indexes’. The studied reaction is a reverse-transcription where a mRNA is converted to cDNA by a reverse transcriptase (RT) using a primer. In addition, a template-switching (T-S) oligonucleotide is present in the reaction to extend the sequence of the mRNA. After incubation in the plates, reactions prepared with six different starting amounts of RNA are amplified together in six different polymerase chain reactions using indexed primers, pooled in a multiplexed sequencing reaction and then demultiplexed in silico using the unique combination of barcode and index sequences. Multi-factorial quantitative analysis of the sequencing reads then follows, according to the initial reaction parameters decided for each well in the experiment design.

Figure 2.

Randomization of nanoCAGE reverse transcription reactions. (A–N) Distribution of reagent concentrations (A, C, E, H, J, L) and related liquid transfer volumes (B, D, F, G, L, K, M, N) in a 384-well plate before (A–G) and after randomization (H–N). Primer concentrations are displayed on a mM scale (9 different TSO concentrations ranging between 0.625 and 160 mM, and six different RT primer concentrations ranging between 1 mM and 24 mM), RNA masses are shown on a picogram scale (6 different RNA quantities ranging from 1 pg to 100 ng) and reagent transfer volumes are indicated on a nanoliter scale. Wells displayed in gray correspond to 60 negative control reactions performed without RT primers (see C and J) or without RNA (see E and L). There were thus 324 reagent combinations studied on each plate using two different enzymes (SSIII and SSIV), which represents a total of 648 data points used to explore the parameter space of the reaction. (O–P) Example of reverse transcription barcode ID distribution before (O) and after randomization (P) for a set of 64 reactions corresponding to one RNA mass (wells indicated in yellow on E and L). The 64 barcoded cDNA samples obtained after the reverse transcription were pooled, amplified and tagged by a specific index sequence (Q) representative of the corresponding group of samples in the sequencing library. Six different indexes characteristic of the six RNA masses were used for each plate. Randomized plates were prepared in quadruplicate for each reverse transcriptase using 24 different indexes, making a total of 1536 samples (64 barcoded samples x 24 indexes) that were pooled and sequenced together for each enzyme.

Figure 3.

Block diagram summarizing the MOIRAI workflow used to process the sequencing data. The full data for each sequencing run has been deposited to ZENODO (https://doi.org/10.5281/zenodo.1683162).

Figure 4.

Parameter screen at high dynamic range. (A) Log-scaled relative yields of demultiplexed read counts (in arbitrary units) as a function of the molarity of the template-switching oligonucleotide, for different molarities of the reverse-transcription primers (RTP), using the SSIII enzyme and 100 pg RNA. (B) Same data displayed for all RTP molarities, colour-encoded as a categorical variable, and the SSIII and SSIV enzymes. In this panel, each replicate is displayed as a dot. (C) Median relative yields of demultiplexed read counts (in arbitrary units) over each set of four replicates, displayed as a contour plot on a surface where each axis represents the molarity of one oligonucleotide. The colour scale indicates better (blue) or worse (red) performance.

Figure 5.

Quality and specificity across the parameter space. (A–E) percentage of (A) oligonucleotide artefacts, (B) reads mapping to reference ribosomal RNA sequences, (C) reads mapping to the genome, (D) reads aligning to promoter regions, (E) premature ‘strand invasion’ artefacts of the template switching reaction. (F) Richness index. The colour scale indicates better (blue) or worse (red) performance. Stars indicate oligonucleotide molarities of the standard nanoCAGE protocol. Half-circles indicate molarities of the modified protocol tested on single cells.

Figure 6.

Test of alternative protocol on low RNA inputs (single cells). Stacked barplots summarizing the quality control statistics for single cell libraries made with the standard nanoCAGE protocol (SSIII standard, n = 136), or with the SSIV enzyme and the standard oligonucleotide concentrations (SSIV standard, n = 184), or with the SSIV enzyme with 45 μM template-switching oligonucleotide and 0.4 μM reverse-transcription primer (SSIV alternative, n = 83). (A) proportions of the reads discarded during data processing until obtaining unique molecule tag counts. (B) annotation statistics of the aligned molecule tags.