Robotic search for optimal cell culture in regenerative medicine
- Laboratory for Retinal Regeneration, RIKEN Center for Biosystems Dynamics Research, Japan
- Laboratory for Biologically Inspired Computing, RIKEN Center for Biosystems Dynamics Research, Japan
- Robotic Biology Institute Inc., Japan
- Epistra Inc., Japan
- VCCT Inc., Japan
- Laboratory for Molecular Biology of Aging, RIKEN Center for Biosystems Dynamics Research, Japan
- Vision Care Inc., Japan
- Graduate School of Media and Governance, Keio University, Japan
- Graduate School of Frontier Biosciences, Osaka University, Japan
- Department of Life Science and Biotechnology, Cellular and Molecular Biotechnology Research Institute, National Institute of Advanced Industrial Science and Technology, Japan
Figures
Figure 1 with 2 supplements
Robotic search for optimal experimental conditions.
(A) Overall workflow for the optimization of experimental procedures using combined experimental robotics and Bayesian optimization. The user defines the target experimental protocol, subject …
Figure 1—figure supplement 1
Schematic diagram of iPSC-RPE transplantation.
(A) Schematic diagram of the process from iPSC stock to iPSC-derived RPE cell transplantation. The steps are roughly divided into iPSC culture, iPSC-RPE differentiation, purification, storage, …
Figure 1—figure supplement 2
System components.
The system platform. The robotic search using this system starts with the provision of queries for round 1 from the optimizer to the protocol compiler and the provision of the base protocol from the …
Figure 2 with 9 supplements
Robotization of iPSC-RPE differentiation protocols.
(A) Schematic diagram of the standard iPSC-RPE differentiation procedures. DDay indicates the differentiation day. Filled circles represent days when the robot operated, solid circles represent days …
-
Figure 2—source data 1
Acquired pigmented images of the baseline experiment.
Images acquired on Day 34 of the baseline experiment; images of the bottom of the well with cultured cells, cropped to the size of the well. These 8-bit images were adjusted to a minimum and maximum contrast value of 100 and 150, respectively. IDs on the bottom indicate 'B (baseline) - Plate No. - Well No.'.
- https://cdn.elifesciences.org/articles/77007/elife-77007-fig2-data1-v1.zip
-
Figure 2—source data 2
Executed parameters and scores of the baseline experiment.
Related to Figure 2E. Raw values of the parameter candidates and pigmentation scores in the baseline experiment. *KSR concentration was lowered in a systematic fashion, unlike the KP parameter. Specific values: DDays 1–3, 20% KSR; DDays 4–7, 15% KSR; from DDay 8, 10% KSR.
- https://cdn.elifesciences.org/articles/77007/elife-77007-fig2-data2-v1.xlsx
-
Figure 2—source data 3
Pipetting volume and pipette combination.
Related to Figure 2. Given the limitations of the LabDroid, setting the micropipette to an arbitrary volume was difficult. Therefore, we pseudo-implemented a fine volume setting for the transfer of 0–1000 µL by combining nine micropipettes with pre-set volumes (3000, 1000, 450, 300, 200, 80, 30, 10, and 5 µL). The number of combinations was limited to three or fewer. The numbers in the table indicate the micropipettes and the number of times they had to be used to achieve the desired volumes. For example, 260 µL indicates that the 200 and 30 µL micropipettes had to be used once and twice, respectively.
- https://cdn.elifesciences.org/articles/77007/elife-77007-fig2-data3-v1.xlsx
Figure 2—figure supplement 1
LabDroid Maholo booth.
(A) Exterior photograph of the LabDroid booth. The LabDroid consists of an acrylic box of W2500 x D2000 x H2200 (mm). (B) Plan view of the LabDroid booth (3D-CAD) and layout of the equipment. (C–D) …
Figure 2—figure supplement 2
Workflows of the experimental operation.
(A) Seeding (DDay −7). (B) Medium exchange type I for preconditioning (DDays −6 to −1) and the first part of RPE induction (DDays 1–19). (C) Medium exchange type II for the second part of RPE …
Figure 2—figure supplement 3
Representative LabDroid execution of a seeding experiment (round 3, DDay −7).
(A) Entire image. (B) Enlarged image of the dotted rectangle from panel (A).
Figure 2—figure supplement 4
Representative LabDroid execution of a preconditioning experiment (round 3, DDay −6, 1st run).
(A) Entire image. (B) Enlarged image of the dotted rectangle from panel (A).
Figure 2—figure supplement 5
Representative LabDroid execution of a passage experiment (round 3, DDay 0, 1st run).
(A) Entire image. (B) Enlarged image of the dotted rectangle from panel (A).
Figure 2—figure supplement 6
Representative LabDroid execution of an RPE differentiation experiment (round 3, DDay 10, 1st run).
(A) Entire image. (B) Enlarged image of the dotted rectangle from panel (A).
Figure 2—figure supplement 7
Representative LabDroid execution of an RPE maintenance experiment (round 3, DDay 32, 1st run).
(A) Entire image. (B) Enlarged image of the dotted rectangle from panel (A).
Figure 2—figure supplement 8
Image processing to calculate the pigmentation scores.
(A–C) The 6-well plates were placed on top of a white LED lighting plate and photographed with a camera fixed to the mount (A). The raw images acquired (B) were cropped in a circle to the size of …
Figure 2—video 1
Representative LabDroid movements.
Example of LabDroid movements extracted from actual cell culture operation: handling of a 50 mL tube, aspirator, micropipette, microscope, and CO2 incubator. The speed factor was 100%.
Figure 3 with 4 supplements
Optimization module.
(A) Definition of the target parameters and corresponding steps in the protocol: PC, preconditioning concentration; PP, preconditioning period; DP, detachment trypsin period; DS, detachment …
Figure 3—figure supplement 1
Demonstration of how Gaussian process regression updates a Bayesian posterior.
Test of a Bayesian posterior updated in a Gaussian process (GP). (A) Sample paths from the zero-mean GP prior. (B) Sample paths from the GP posterior after executing some experiments (indicated by …
Figure 3—figure supplement 2
Demonstration of Bayesian optimization generation of experimental parameters.
Sequential Bayesian optimization was tested by determining the maximum of a one-dimensional toy objective function. This figure illustrates the Bayesian optimization procedure over several …
Figure 3—figure supplement 3
Toy testing function based on domain knowledge.
An approximate prediction of how much the value of each variable affects the score based on domain knowledge before performing a series of optimizations. The toy testing function was simply …
Figure 3—figure supplement 4
Preliminary testing of the Bayesian optimization under different conditions.
The vertical axis shows the value, and the horizontal axis shows the number of circles. The blue horizontal line represents the ground truth (optimal value). In each series, three experiments were …
Figure 4 with 6 supplements
Robotic search for optimal parameters in iPSC-RPE differentiation.
(A–C) Parameter candidates sorted in order of the pigmentation score in optimization rounds 1 (A), 2 (B), and 3 (C). The ID label on the left represents 'Round No. - Plate No. - Well No.'. For …
-
Figure 4—source data 1
Acquired pigmented images of the round 1 experiment.
Images acquired on Day 34 of the round 1 experiment; images of the bottom of the well with cultured cells, cropped to the size of the well. These 8-bit images were adjusted to a minimum and maximum contrast value of 100 and 150, respectively. ID labeling on the bottom indicates '1 (round 1) - Plate No. - Well No.'.
- https://cdn.elifesciences.org/articles/77007/elife-77007-fig4-data1-v1.zip
-
Figure 4—source data 2
Acquired pigmented images of the round 2 experiment.
Images acquired on Day 34 of the round 2 experiment; images of the bottom of the well with cultured cells, cropped to the size of the well. These 8-bit images were adjusted to a minimum and maximum contrast value of 100 and 150, respectively. ID labeling on the bottom indicates '2 (round 2) - Plate No. - Well No.'.
- https://cdn.elifesciences.org/articles/77007/elife-77007-fig4-data2-v1.zip
-
Figure 4—source data 3
Acquired pigmented images of the round 3 experiment.
Images acquired on Day 34 of the round 3 experiment: images of the bottom of the well with cultured cells, cropped to the size of the well. These 8-bit images were adjusted to a minimum and maximum contrast value of 100 and 150, respectively. ID labeling on the bottom indicates '3 (round 3) - Plate No. - Well No.'.
- https://cdn.elifesciences.org/articles/77007/elife-77007-fig4-data3-v1.zip
-
Figure 4—source data 4
Executed parameters and scores of the optimization experiments.
Related to Figure 4. Raw values of the parameter candidates and pigmentation scores in the experiments from rounds 1 to 3.
- https://cdn.elifesciences.org/articles/77007/elife-77007-fig4-data4-v1.xlsx
Figure 4—figure supplement 1
Parallel coordinate plot (PCP) of the robotic search experimental results.
The experimental results were visualized in rounds 1–3 using a parallel coordinate plot (PCP) and represented in 8-dimensional space (seven-dimensional parameters +pigmentation scores) as a colored …
Figure 4—video 1
Seeding operation.
Video recording an example of a seeding operation (round 3, DDay −7). The speed factor was 6000%.
Figure 4—video 2
Preconditioning operation.
Video recording an example of a preconditioning operation (round 3, DDay −6, 1st run). The speed factor was 6000%.
Figure 4—video 3
Passaging operation.
Video recording an example of a passage operation (round 3, DDay 0, 1st run). The speed factor was 6000%.
Figure 4—video 4
RPE differentiation operation.
Video recording an example of an RPE differentiation operation (round 3, DDay 10, 1st run). The speed factor was 6000%.
Figure 4—video 5
RPE maintenance operation.
Video recording an example of an RPE maintenance operation (round 3, DDay 32, 1st run). The speed factor was 6000%.
Figure 5 with 2 supplements
Quality evaluation of robot-induced RPE cells.
(A) The pigmentation score evaluation of the pre-optimized conditions (n=3) and the top five conditions from round 3. Error bars represent the standard error of the mean (SEM). The numbers 1–5 in …
-
Figure 5—source data 1
Acquired pigmented images of the validation experiment.
Images acquired on Day 34 of the validation experiment: images of the bottom of the well with cultured cells, cropped to the size of the well. These 8-bit images were adjusted to a minimum and maximum contrast value of 80 and 125, respectively. Sample names on the top correspond to Figure 5A. ID labeling on the bottom indicates 'V (validation) - Plate No. - Well No.'. Wells 2, 5, and 6 were not subjected to the validation experiments. Plates 1 and 5 were used for cell biological analysis and were not evaluated using images.
- https://cdn.elifesciences.org/articles/77007/elife-77007-fig5-data1-v1.zip
-
Figure 5—source data 2
Executed parameters and scores of the validation experiment.
Related to Figure 4. Raw values of the parameter candidates and pigmentation scores in the validation experiments. The sample names on the top correspond to Figure 5A. Well numbers 2, 5, and 6 were not subjected to the validation experiments. *Plate numbers 1 and 5 were used for cell biological analysis and were not validated using images.
- https://cdn.elifesciences.org/articles/77007/elife-77007-fig5-data2-v1.xlsx
-
Figure 5—source data 3
ELISA scores.
Related to Figure 5D and E. Raw values of ELISA scores from the validation experiments.
- https://cdn.elifesciences.org/articles/77007/elife-77007-fig5-data3-v1.xlsx
-
Figure 5—source data 4
Immunohistochemistry images.
Related to Figure 5F. Images without contrast processing. TIFF files with three channels merged for each sample image: channel 1, MITF (magenta); channel 2, ZO-1 (green); channel 3, DAPI (gray).
- https://cdn.elifesciences.org/articles/77007/elife-77007-fig5-data4-v1.zip
-
Figure 5—source data 5
Robot log.
List of job names, start times, end times, time required, number of commands, and errors (if any) for all experiments performed by the LabDroid in this study.
- https://cdn.elifesciences.org/articles/77007/elife-77007-fig5-data5-v1.xlsx
Figure 5—figure supplement 1
Video monitoring.
Example pictures of the monitoring cameras. (A–B) Wide-angle cameras for 24×7 monitoring. Front camera (A) and back camera (B). (C–D) Magnifying cameras that record only when the robot is running. …
Figure 5—figure supplement 2
Errors in the robotic operations.
Robotic experiments were conducted for 185 days, with a total robot operating time of 995 hr, 39 min, and 21 s (Figure 5—source data 5). Each round consisted of 73 jobs, and five rounds were …
Tables
Table 1
Definition of optimized parameters.
Parameter names, parameter name codes, description, parameter ranges, parameter units, correspondence between experimental procedure and parameters used (related to Figures 2A, 3A and B).
| Parameter name | Code | Description | Range | Unit | Protocol step |
|---|---|---|---|---|---|
| Preconditioning concentration | PC | FGFRi concentration in medium | 0–505 | nM | Preconditioning |
| Preconditioning period | PP | FGFRi duration in medium | 1–6 | day | Preconditioning |
| Detachment trypsin period | DP | Trypsin incubation duration at room temperature after incubation at 37 °C, 20 min. | 5, 8, 11, 14, 17, 20, 23 | min | Passage |
| Detachment pipetting strength | DS | Pipetting strength during cell detachment | 10–100 | mm/s | Passage |
| Detachment pipetting length | DL | Bottom surface area to be pipetted | short / long | N/A | Passage |
| KSR period | KP | KSR concentration and duration in medium: KSR concentration is decreased linearly every day so that KSR becomes 10% on DDday of KP value | 1–19 | day | RPE differentiation |
| Three supplements period | 3P | Three chemical supplements duration | 3–19 | day | RPE differentiation |
Key resources table
| Reagent type (species) or resource | Designation | Source or reference | Identifiers | Additional information |
|---|---|---|---|---|
| Cell line (Homo-sapiens) | hiPSC 253G1 | RIKEN BRC | HPS0002 | |
| Antibody | Anti-ZO-1 (Rabbit polyclonal) | Thermo Fisher Scientific Inc. | 61–7300 | IHC (1:500) |
| Antibody | Anti-MITF (Mouse monoclonal) | Abcam plc. | ab80651 | IHC (1:1000) |
| Antibody | Alexa Fluor 488 Goat Anti-rabbit IgG (Goat polyclonal) | Thermo Fisher Scientific Inc. | A-11034 | IHC (1:1000) |
| Antibody | Alexa Fluor 546 Goat Anti-mouse IgG (Goat polyclonal) | Thermo Fisher Scientific Inc. | A-11030 | IHC (1:1000) |
| Sequence-based reagent | BEST1 (+) | This paper | RT-PCR primers | TAGAACCATCAGCGCCGTC |
| Sequence-based reagent | BEST1 (−) | This paper | RT-PCR primers | TGAGTGTAGTGTGTATGTTGG |
| Sequence-based reagent | RPE65 (+) | This paper | RT-PCR primers | TCCCCAATACAACTGCCACT |
| Sequence-based reagent | RPE65 (−) | This paper | RT-PCR primers | CCTTGGCATTCAGAATCAGG |
| Sequence-based reagent | CRALBP (+) | This paper | RT-PCR primers | GAGGGTGCAAGAGAAGGACA |
| Sequence-based reagent | CRALBP (−) | This paper | RT-PCR primers | TGCAGAAGCCATTGATTTGA |
| Sequence-based reagent | GAPDH (+) | This paper | RT-PCR primers | ACCACAGTCCATGCCATCAC |
| Sequence-based reagent | GAPDH (−) | This paper | RT-PCR primers | TCCACCACCCTGTTGCTGTA |
| Sequence-based reagent | RNeasy Micro Kit | QIAGEN | 74004 | |
| Sequence-based reagent | SuperScript III | Thermo Fisher Scientific Inc. | 18080–044 | |
| Commercial assay or kit | VEGF Human ELISA Kit | Thermo Fisher Scientific Inc. | BMS277-2 | |
| Commercial assay or kit | PEDF Human ELISA Kit | BioVendor | RD191114200R | |
| Chemical compound, drug | PD 173074 | Merck & Co., Inc. | P2499-5MG | |
| Chemical compound, drug | CultureSure Y-27632 | FUJIFILM Wako Pure Chemical Corporation | 036–24023 | |
| Chemical compound, drug | SB 431542 hydrate | Merck & Co., Inc. | S4317-5MG | |
| Chemical compound, drug | CKI-7 dihydrochloride | Merck & Co., Inc. | C0742-5MG | |
| Software, algorithm | LabDroid_optimizer | This paper | Available at our Github (see Data and code availability) | |
| Other | StemFit AK02N | Ajinomoto Co., Inc. | AK02N | see Materials and Methods >Reagents |
| Other | knockOut serum replacement (KSR) | Thermo Fisher Scientific Inc. | 10828028 | see Materials and Methods >Reagents |
| Other | FBS | Nichirei Corporation | 12007C | see Materials and Methods >Reagents |
