key: cord-0291218-ynjwdv80
authors: Ohlsson, Jonas A; Leong, Jia Xuan; Elander, Pernilla H; Dauphinee, Adrian N; Ballhaus, Florentine; Johansson, Johan; Lommel, Mark; Hofmann, Gero; Betnér, Staffan; Sandgren, Mats; Schumacher, Karin; Bozhkov, Peter V; Minina, Elena A
title: SPIRO – the automated Petri plate imaging platform designed by biologists, for biologists
date: 2021-03-24
journal: bioRxiv
DOI: 10.1101/2021.03.15.435343
sha: 3dd5740f4d48d8e63de458c08daa555b1b7b7a5a
doc_id: 291218
cord_uid: ynjwdv80

The imaging of plant seedlings, fungal mycelia and bacterial colonies grown on Petri plates is common in phenotyping assays, and is typically done manually despite the procedures being time-consuming and laborious. The main reason for this is the still limited availability of existing automated phenotyping tools and facilities. Additionally, constructing a custom-made automated solution is a daunting task for most research groups specializing in biology. Here, we describe SPIRO, the Smart Plate Imaging Robot, an automated platform that acquires time-lapse photos of up to four vertically oriented Petri plates in a single experiment. SPIRO was designed for biologists by biologists; thus, its assembly does not require experience in engineering or programming and its operation is sufficiently intuitive to be carried out without training. SPIRO has a small footprint optimal for fitting into standard incubators for plants and microbes, and is equipped with an LED light source for imaging in the dark, thus allowing acquisition of photos under optimal growth conditions. SPIRO’s web-based user interface allows setting up experiments and downloading data remotely, without interfering with samples growth. SPIRO’s excellent image quality is optimal for automated image processing, which we demonstrate with two semi-automated assays for analysis of commonly used phenotypic traits: seed germination and root growth. Moreover, the robot can be easily customized for a specific use, as all information about SPIRO, including the models for 3D-printed structural components, control software, and scripts for image analysis, are released under permissive open-source licenses.

Manual imaging of Petri plates using cameras or scanners is a common 54 practice in biology experiments that require phenotyping of plant seedlings or 55 microbial colonies. However, manual imaging necessitates removing the 56 plates from the growth conditions and increases the risk of introducing 57 unwanted variables into the experiment, e.g., changes in temperature, 58 humidity, illumination, vector of gravity, and mechanical stress. Such The result of our efforts is SPIRO, the compact Smart Plate Imaging Robot 79 for time-lapse imaging of vertical Petri plates, which fits into standard 80 plant/microbe incubators. SPIRO was designed for biologists by biologists, 81

introducing end-user insight into its development. We ensured that no prior 82 knowledge of mechanical engineering, electronics, or computer science is 83 necessary for its assembly and operation. SPIRO comprises the absolute 84 for a specific application. We successfully printed and tested four SPIRO 136 prototypes in two independent laboratories using black matte PLA filament 137

(for detailed information about printing, see Table S2 and the SPIRO  138 Hardware Repository 15 ). The hardware proved to be easy to reproduce, robust 139 and durable. 140

To facilitate use of SPIRO for a broad range of experiments we designed plate 141 holders compatible with the most commonly used plate formats: a 12 cm 142 square (Greiner Bio-One International, Item: 688102) and a 9 cm round Petri 143 plate (Sarstedt, 82.1473.001), and enabled adjusting distance between the 144 camera and the plate holders by moving the camera along the vertical and 145 horizontal aluminum profiles (Fig. 1A) . 146 Camera 147 SPIRO is equipped with a single 8 MP (3280×2464 pixels) color camera, and 148 saves images as RGB PNG files. Image files are stored in a user-defined 149 experiment folder and are automatically sorted into four sub-folders 150 corresponding to each plate holder. Metadata useful for further analysis is 151 included into the file names, i.e., the names contain the plate holder number, 152 date and time of acquisition, and information about day or night mode (for 153 detailed information, please refer to File S4). SPIRO acquires excellent quality 154 images regardless of ambient illumination conditions, which is crucial for 155 downstream automated data analysis ( Fig. 1E and F Stepper motor and positional sensor 187

The cube-shaped stage of SPIRO is rotated by a stepper motor (i) during 188 imaging, to position each of the four plate holders in front of the camera and 189 (ii) between the imaging cycles, to ensure the plates are evenly exposed to the 190 ambient conditions and that there is no shading effect from SPIRO's light 191 screen (Fig. 1) . 192

The 12 V unipolar stepper motor we recommend provides sufficient force to 193 reproducibly move the weight corresponding to the stage with 4 square Petri 194 plates containing medium and plants and a holding torque that stably holds the 195 stage position during imaging. Importantly, the motor movement is smooth and has no impact on Arabidopsis root growth under normal conditions 197

The current layout of SPIRO requires two power supply units: a 5 V supply 199 for the Raspberry Pi computer and a 12 V supply for the stepper motor and 200 LED illuminator (Fig. 1A, Table S1 , File S2). We decided for such 201 configuration, as it drastically simplifies the assembly and maintenance of the 202 robot, in comparison to implementing a single power supply unit. automatically assessed by SPIRO's camera immediately prior to acquiring 227 each image. If sufficient illumination is detected, SPIRO takes an image in the 228 "day" mode (Fig. 1C, E) , otherwise the robot turns on the LED light source 229 and acquires a "night" image ( Fig. 1D, F) . ISO and shutter speed for image 230 acquisition can be adjusted individually for day and night modes in the web-231 based interface of the SPIRO software (Fig. 1B) . 232

During prototyping, we tested night imaging using IR LEDs and the IR-233 sensitive Raspberry Pi NoIR camera. However, this increased the cost of the 234 robot, while significantly complicating its electronics layout and focusing 235 procedure, and did not provide satisfactory quality of images suitable for 236 automated image analysis. 237

Typically, color camera detectors are most sensitive to green light, as they 238 contain double the number of green pixels compared to red or blue. Hence, we 239 speculated that using a green light source for illumination would be most 240 efficient while using the color Raspberry Pi camera for imaging in the dark. 241

Additionally, we took into consideration use of SPIRO for plant imaging. whether this was the case, we compared germination rates and root growth of 250

Arabidopsis thaliana seeds and seedlings, respectively, imaged using two 251 SPIRO systems, each with and without green LEDs. Germination was 252 assessed using an ANOVA test where germination rate at t=50 h was set as 253 the dependent variable. Germination rate was compared at a single time point 254 (50 h) due to the constraints imposed on germination detection by not having 255 night image data available. For comparing the effect of LED illumination on 256 root growth, plates with identical genotypes and media were imaged with and 257 without LED illumination. Two models were fitted: the first was the standard 258 root growth model as described in the SPIRO Assay Manual (File S4), and the 259 second was the same model but with the fixed effect of LED and all of its 260 interaction effects added. Models were then compared using the anova 261 function in R. Our analysis confirmed that green light indeed had no effect on 262 germination and root growth (Tables 1 and 2) . 263 illumination has no effect on root growth 274

Log likelihood Deviance χ 2 χ 2 D.F.

Without LED

With LED 65281 -130563 0 30 1.00 To aid the use of SPIRO we designed a set of essential accessories (Table S2  278 and the SPIRO Hardware GitHub Repository 15 ). 279 3D-printable seed plating guides help the user to place Arabidopsis seeds on 280 plates at regular distances from each other and from the edges of the plate. The 281

latter is important to avoid overlaying seed and seedling images with 282 reflections and shadows caused by the Petri plates' rims. The marked regular 283 distance positions for seeds were optimized for the SPIRO Germination and 284

Root Growth Assays described below. 285

The anti-reflection lids are designed to reduce reflections from seeds and 286 seedlings that are usually visible in the Petri plates' lids. Although such 287 reflections might not be an issue during imaging, their presence is detrimental 288

for automated image processing, as some of them are difficult to automatically 289 differentiate from actual biological samples (for more information see list of components, step by step instructions, and tutorial videos for assembly 295 are provided in Tables S1 and S2, and Files S1 and S2. However, we highly 296 recommend to check the SPIRO Hardware Repository 15 for potential updates 297 before assembly. 298 SPIRO hardware includes a set of standard parts, like aluminium profiles, 299 screws and electronic components that need to be purchased. The complete 300 list of these components and links with suggestions on where to purchase them 301 is provided in Table S1 and the SPIRO Hardware Repository 15 . Our 302 experience of ordering hardware to build SPIRO prototypes in Germany and 303

Sweden reproducibly showed that the most challenging part to acquire are the 304 correct screws. At the moment, we cannot provide a plausible explanation for this peculiar phenomenon and will try to upgrade the SPIRO specifications to 306 reduce the requirements for screws. Notably, approximately one quarter of 307 purchase costs were covering shipping expenses. Furthermore, some parts, 308 such as the LED strips, had to be ordered in excess. Thus, building several 309

SPIROs lowers the price per robot. 310

The STL, 3MF and F3D files for 3D-printable parts of SPIRO and the printing 311 settings recommendations are provided in Table S2 SPIRO's software has an intuitive web-based graphical user interface that can 338 be easily accessed from any web browser (Fig. 1B) . The layout of the software 339 was optimized with the help of several beta-testers to ensure that the interface 340 is sufficiently self-explanatory, does not require training prior to use and was detected in order to place the stage in the "home" position.

• Image acquisition is preceded by assessment of illumination intensity. 367

If the average pixel intensity on a sampled image is less than 10 (out 368 of a maximum of 255), the software triggers acquisition under user-369 defined "night" settings and the LEDs are switched on for the duration 370 of acquisition (less than one second). If the average pixel intensity on 371 the sample image is higher than 10, the image is acquired with user-372 defined "day" settings. 

To thoroughly assess the data quality acquired using SPIRO and further 387 enhance applicability of the imaging platform for the plant biology 388 community, we developed complete pipelines for two commonly used 389 phenotyping assays that would greatly benefit from automated imaging: seed 390 germination and root growth assays. The assays comprise image processing 391 steps carried out by designated macro scripts in FIJI 24 (distribution of ImageJ) 392 and quantitative data processing steps carried out by custom R scripts.

Step by 393 step instructions and scripts are provided in Files S4 and S5. Please note that 394 updates are published in the SPIRO Assays Repository 23 . 395

Each assay starts with pre-processing of SPIRO raw data to combine 396 individual images into time-lapse stack files with a set scale bar. The 397 preprocessed data is then subjected to semi-automated image segmentation 398 with identification of the objects of interest and measurement of their physical 399 parameters, e.g., perimeter, length, and area, at successive time points. The 400 quantitative data is then processed by R scripts to first ensure data quality and 401 then apply custom-designed algorithms that determine seed size, germination 402 time, root length and perform suitable statistical analyses. 403

The assays were designed to enable applicability for a broad range of 404 experiment layouts and customization for specific uses, thus we introduced 405 several user-guided steps that allow combining seeds or seedlings into groups perimeter of a seed will steadily increase after germination, i.e., the radicle emergence 26 , took place. Hence, the image segmentation part of the assay 425 detects individual seeds on the photos ( Fig. 2A, Movie S2) , and tracks changes 426 in the perimeter of each seed within a user-defined time range. The data is then 427 gathered into user-defined groups, e.g., genotypes or treatments, and subjected 428 to a clean-up using a designated R script. After this, for each seed the earliest 429 time point of steady increase in perimeter is detected and identified as the time 430 of germination. The assay is optimized for an imaging frequency of every 30 431 minutes and thus allows tracking minute differences in germination times. To 432 take into account the effect of imbibition on the seed perimeter and also to 433 compensate for the natural variation in Arabidopsis seed sizes, the germination 434 algorithm normalizes perimeter changes for each seed by comparing it to the 435 same seed perimeter averaged over the first five images in the time-lapse data. 436

The significance of difference between mean germination times for the user-437 selected groups of seeds is then assessed by the Kaplan-Meier test (Fig. 2B) . 438

Furthermore, the assay provides information about the size of individual seeds 439 and the results of t-test comparing seed sizes for user-defined groups (Fig.  440   2C) . Additionally, we implemented calculation of other germination 441 parameters that might be valuable for the user, such as rate-of-germination 442 curve, time at maximum germination rate; time required to achieve 50% of 443 total germination efficacy, time required for 50% of total seeds to germinate 444 (for detailed information see File S4 and the SPIRO Assays repository 23 ). While developing the assay we optimized the SPIRO hardware and protocol 453 for seed plating, and introduced seed plating guides that demark positions for 454 placing seeds at optimal distance from each other and plate rims. As a result, 455 when using four 12 cm square Petri plates, it is possible to detect germination for up to 2300 seeds in a single experiment. Additionally, we strongly 457 recommend using SPIRO anti-reflection lids to reduce image segmentation 458 artifacts caused by reflections (Fig. 2 in the File S4) . 459 Additional parameters for germination are calculated using the germinationmetrics package for R 27 (for more information see SPIRO Assay manual, File S3).

C. The assay also provides a t-test comparison of the mean seed size for the analyzed groups. Boxes represent interquartile range, the horizontal line denotes the median value, circles indicate outliers. Means of the groups that are not significantly different are annotated with the same small case letters, n = 684 seeds.

Automatic detection of seed germination using SPIRO Seed Germination Assay provides results very similar to manual assessment of the germination (n=172).

Quantifying primary root length of Arabidopsis seedlings is frequently used 462 as a readout for physiological response to mutations or environmental 463 stimuli 13,28,29 . SPIRO is an excellent platform for seedling root phenotyping. 464

We first tested processing of SPIRO images by existing automated image 465 analysis tools for detection of single roots and root systems on time-lapse 466 data 13,30,31 . As these algorithms were optimized for a certain type of imaging 467 data, their applicability for SPIRO-acquired images was limited. 468 Therefore, we developed the designated SPIRO Root Growth Assay, which 469 uses SPIRO time-lapse data to track primary root length for individual 470 seedlings starting from the germination time-point of the corresponding seed 471 (Fig. 3A, Movie S2) , builds a root growth rate model for user-defined groups 472 of seedlings (Fig. 3B) , and then performs statistical analysis comparing root 473 lengths and growth rates for the groups (Fig. 3C, File S4 , and the SPIRO 474

Assays repository 23 ). Similar to the SPIRO Seed Germination Assay, the Root 475

Growth Assay provides the user with a graphical output that show the results 476 of image segmentation for each user-selected group (Fig. 3A) and a 477 quantitative output. The latter comprises (i) measurements of the segmented 478 objects performed by the ImageJ macro; (ii) the measurements data cleaned 479 up using a designated R script; (iii) germination time detected for each seed; 480 (iv) curve charts for seedling's root lengths plotted vs absolute time or 481 normalized to individual germination times; (v) curve charts generated by 482 models for root growth rates of user-selected groups of seedlings (Fig. 3B) ; 483 (vi) bar charts showing predicted root lengths for the groups of seedlings at 484 24-h intervals (Fig. 3C) and (vii) results of the statistical analysis comparing 485 growth rates and root lengths between user-defined groups. For more details 486 please refer to the assay manual in File S4 and the SPIRO Assays repository 23 . 487

Comparison of manual and automated measurements of 141 Arabidopsis 488 roots, revealed that the SPIRO Root Growth Assay provides accuracy 489 comparable with human performance (Fig. 3D) . also improves the quality of the acquired data and provides remote access to it 507 from anywhere in the world. The latter feature turned out to be surprisingly 508 useful during COVID-19 lockdown regulations. 509

We strongly believe that reducing the barrier of entry is crucial to the adoption 510 of open science hardware, especially in non-engineering disciplines. In our 511 experience, one of the significant bottlenecks in implementing engineering 512 solutions custom-designed for laboratory use is the requirement of expertise 513 in subjects that are not necessarily popular among biologists, such as 514 mechanical engineering and electronics. SPIRO was developed specifically 515 for biologists, and its design has at its core the concept of being simple and 516 intuitive enough to be assembled and operated with no training in engineering, 517 3D printing, programming or using the SPIRO per se. Furthermore, we strived 518 to develop an imaging platform that can be built anywhere in the world without 519 requiring access to rare specialized infrastructure and would be affordable also 520 for research groups with limited funding. During preparation of this 521 manuscript, multiple SPIROs have been already constructed in six laboratories 522 located in four different countries using only the provided instructions, thus confirming the need for such platform and validating our approach to making 524 its construction accessible. 525

The current configuration of SPIRO is optimized for image acquisition under 526 conditions commonly used in plant biology and microbiology. Moreover, the 527 underlying Raspberry Pi platform is ideal for further expanding the system. A 528 large variety of Raspberry Pi-compatible sensors and other input/output 529 modules could be incorporated into a custom-built SPIRO system to 530 accommodate different research needs. For example, cheap sensors for 531 temperature and humidity 18 can be connected to unused general-purpose 532 input/output (GPIO) pins on the computer board. Such upgrades can be 533 valuable when using SPIRO in a growth chamber that does not provide control 534 or logging of these parameters. 535

We based our design on the use of 3D-printed structural components to further 536 facilitate customization of SPIRO. We provide F3D model files that can be 537 easily modified using Autodesk Fusion 360 software (File S1). As of 2021, 538

Autodesk offers a free license tier for academic users, which includes training 539 material. Furthermore, use of 3D-printed parts warrants reproducibility and 540 robustness of the structural components, enables their easy replacement for 541 maintenance and ensures that assembly can be done without access to rare 542 specialized infrastructures. For instance, not only 3D printers are becoming 543 increasingly affordable, with the cheapest models costing less than €200, but 544 prints can also be ordered using commercial services 17 or at public Makerspace 545 facilities 16 . 546

We demonstrate analysis of high-throughput SPIRO-acquired data in the two 547 semi-automated assays for seed germination and seedling root growth. The 548 detailed protocols for the assays comprise the complete procedure from 549

preparing Petri plates to statistical analysis of the data. Both assays provide 550 data analysis accuracy closely matching human performance ( Fig. 2D and  551   3D) . Moreover, the SPIRO Seed Germination Assay enables very high 552 temporal resolution enabling the detection of minute differences in 553 germination times. Importantly, despite its small footprint, the platform still 554 provides practical throughput: a single SPIRO can image up to 2300 seeds in a single experiment for the Germination Assay, and 190 seedlings for the Root 556

Growth Assay. In their current versions, the assays are implemented using 557 ImageJ and R, software commonly used in biology labs, which should make 558 their use and customization relatively uncomplicated. We have developed 559 them as "plug-and-play" image analyses that are optimized for SPIRO data 560 and provide outstanding results for the intended purpose. Implementation of 561 the recent advances in machine learning approaches 32-34 will require thorough 562 training of appropriate models for image segmentation, but will eventually 563 enable a more advanced analyses of SPIRO-acquired data such as detangling 564 of crossing roots, measurement of lateral roots, hypocotyls and cotyledons 565 under day and night conditions. 566

The demand for affordable automated platform for Petri plate imaging is 567 clearly illustrated by the number of publications describing various prototypes 568 of such system 7,9,12,32,33,35 . However, we could not find a solution that would 569 be simultaneously affordable, compatible with standard growth cabinets, 570

provide high-quality data and come with instructions comprehensible for a 571 person not trained in engineering. Hence, we pursued developing SPIRO to 572 establish a platform that would enable universal access to automated high-573 quality imaging for all research groups, independent on their training or 574 funding situation, and enable easy integration of the automated approach into 575 ongoing research. 576 assay. We hope that SPIRO will help alleviate some pains of routine lab work 586 and will also become a stepping stone for advancement of users' interests in 587 developing further solutions. We encourage users to further customize the platform, develop image-analysis pipelines suited for their own research and 589

share optimization with the scientific community. 590 Table S1  Table S1 . List of components to be purchased.pdf

List of components to be purchased and suggestions of vendors 

University) who provided valuable advice, resistors and terminal ports for 595 SPIRO hardware development; all members of Prof. Karin Schumacher's 596 research group (Heidelberg University, Germany) for their patience and moral 597 support and especially to Rainer Waadt for the name SPIRO

Sweden) for their contribution to the development of the early prototype

University) for their contributions to the SPIRO software

This study was supported by the funding from EU Horizon 2020 MSCA IF 603 (799433 to EA Minina), Carl Tryggers Foundation

DFG (CRC1101, TPA02 to K Schumacher), Formas (2019-605 01565 to AN Dauphinee), Formas (2016-20031 to M Sandgren)

0026 to PV Bozhkov), the Swedish 607

Foundation for Strategic Research (Oil Crops for the Future), and Crops for 609 the Future Research Programme at the Swedish University of Agricultural 610

Molecular and genetic control of plant 615 thermomorphogenesis

Temperature variability is integrated by a 617 spatially embedded decision-making center to break dormancy in 618 Arabidopsis seeds

Analysis of spatio-temporal fungal growth 621 dynamics under different environmental conditions

Return on investment for open source scientific 624 hardware development

Haves and have nots must find a better way: The 626 case for open scientific hardware

Economic savings for scientific free and open source 628 technology: A review

High throughput phenotyping of root 630 growth dynamics, lateral root formation, root architecture and root 631 hair development enabled by PlaRoM

A high throughput 634 robot system for machine vision based plant phenotype studies

The platform GrowScreen-Agar enables 637 identification of phenotypic diversity in root and shoot growth traits 638 of agar grown plants

Raspberry Pi-powered imaging for plant 642 phenotyping

A scalable open-source pipeline for large-scale root 644 phenotyping of Arabidopsis

Automated high-throughput 646 root phenotyping of Arabidopsis thaliana under nutrient deficiency 647 conditions

The circadian clock. A plant's best 655 friend in a spinning world

Imaging 657 the Photosystem I/Photosystem II chlorophyll ratio inside the leaf

Green light: A signal to slow down 660 or stop

Fiji: An open-source platform for biological-664 image analysis

Seed Germination and Dormancy

Seed Germination Indices and Curve Fitting

Nitrate-regulated glutaredoxins control 672 arabidopsis primary root growth

SHORT-ROOT regulates primary, lateral, and 675 adventitious root development in Arabidopsis

MyROOT: a method and software for the 679 semiautomatic measurement of primary root length in Arabidopsis 680 seedlings

An online database for plant 682 image analysis software tools

ChronoRoot: High-throughput phenotyping by 684 deep segmentation networks reveals novel temporal parameters of 685 plant root system architecture

A deep 691 learning-based approach for high-throughput hypocotyl phenotyping. 692 Plant Physiol

PetriJet 694 Platform Technology: An Automated Platform for Culture Dish 695 Handling and Monitoring of the Contents

We would like to express our deepest gratitude to: Sebastian Mai (Heidelberg 594

The authors declare no competing interests. 592