Antiviral Functions of ARGONAUTE Proteins During Turnip Crinkle Virus Infection Revealed by Image-based Trait Analysis in Arabidopsis

RNA-based silencing functions as an important antiviral immunity mechanism in plants. Plant viruses evolved to encode viral suppressors of RNA silencing (VSRs) that interfere with the function of key components in the silencing pathway. As effectors in the RNA silencing pathway, ARGONAUTE (AGO) proteins are targeted of by some VSRs, such as that encoded by Turnip crinkle virus (TCV). A VSR-deficient TCV mutant was used to identify AGO proteins with antiviral activities during infection. A quantitative phenotyping protocol using an image-based color trait analysis pipeline on the PlantCV platform, with temporal red, green and blue (RGB) imaging and a computational segmentation algorithm, was used to measure plant disease after TCV inoculation. This process captured and analyzed growth and leaf color of Arabidopsis plants in response to virus infection over time. By combining this quantitative phenotypic data with molecular assays to detect local and systemic virus accumulation, AGO2, AGO3, and AGO7 were shown to play antiviral roles during TCV infection. In leaves, AGO2 and AGO7 functioned as prominent non-additive, anti-TCV effectors, while AGO3 played a minor role. Other AGOs were required to protect inflorescence tissues against TCV. Overall, these results indicate that distinct AGO proteins have specialized, modular roles in antiviral defense across different tissues, and demonstrate the effectiveness of image-based phenotyping to quantify disease progression. Author Summary Plant viruses caused substantial losses in crop production and quality worldwide. Precisely measuring plant health is critical for better understanding the mechanisms underlying plant virus and host interactions. Advances in high-resolution imaging technologies and deep-learning tools have made acquiring and analyzing “big data” of disease traits possible. In this study, we have developed a high-throughput, image-based trait phenotyping pipeline to quantify disease severity in Arabidopsis thaliana infected by Turnip Crinkle Virus (TCV). Our aim is to understand how the antiviral RNA silencing machinery is tuned to protect the host from invading virus infection. We focused on ARGONAUTE proteins, which are the effectors in the RNA silencing pathway. A mutant line of TCV with a dysfunctional silencing suppressor (P38) was used to investigate which ago mutation could compensate for the dysfunctional silencing suppressor and facilitate the development of disease symptoms. We demonstrated that specific AGO proteins contribute to protecting leaves from TCV infection in a non-additive manner. Our results also implied that distinct AGOs are required to function collectively to silence TCV in inflorescence tissues. More evidence is still needed to further understand how these antiviral AGOs interact with suppressor proteins molecularly during TCV infection.


49
Plants can protect themselves against invasive virus infection through RNA silencing by 50 targeting viral RNA for degradation [1]. This host silencing machinery is triggered by 51 viral double-stranded RNAs (dsRNAs), which are cleaved by Dicer-like (DCL) nucleases 52 associated with dsRNA binding proteins (DRBs) into 21-24 nucleotide RNA duplexes 53 called viral small interfering RNAs (vsiRNAs). The vsiRNAs are then methylated and 54 stabilized by HUA enhancer 1 (HEN1). One strand of these stabilized vsiRNAs is 55 recruited into the RNA-induced silencing complex (RISC) containing ARGONAUTE 56 (AGO) proteins, and then serves as the sequence-specific guide for specific AGOs to 57 slice cognate viral RNAs [2,3]. 58 59 Most plant viruses have evolved to encode viral suppressors of RNA silencing (VSRs) 60 that use varied mechanisms to target components in the silencing pathway [4]. One 61 such mechanism is interference of AGOs that mediate antiviral silencing. Accumulating 62 evidence indicates that various VSRs use diverse modes of action on AGO proteins, 63 such as promoting AGO degradation [5][6][7][8], inhibiting the slicing activities of AGOs [9], or 64 interfering with factors upstream of AGO activity such as RNA-dependent RNA 65 Polymerase (RDR)-dependent silencing [10], obstructing siRNA-loaded RISC activity 66 [11,12], or indirectly repressing AGO protein level [13]. Since functional VSRs can 67 mask host antiviral silencing effects, VSR-defective mutant viruses that can only 68 successfully infect immunocompromised plants have been constructed to identify 69 antiviral roles of key components in the silencing pathway during virus infection [14,15].

80
Turnip crinkle virus (TCV) is a positive single-strand RNA virus belonging to the 81 Carmovirus genus of the Tombusviridae family. The TCV genome encodes five 82 proteins, including two replicase proteins (P28 and P88), two movement proteins (P8 83 and P9) and the coat protein (P38). The coat protein (CP) is multifunctional, as it has 84 roles in virus movement [25,26], serves as a virulence factor [27], and functions as a 85 VSR to suppress antiviral silencing [28,29]. Other virus groups in the Tombusviridae 86 encode separate VSR proteins, such as P19 from Tomato bushy stunt virus (TBSV) 87 [19]. TCV systemically infects the susceptible Arabidopsis ecotype Columbia-0 (Col-0), 88 and causes disease symptoms that include severe chlorosis in leaves, stunted bolts and 89 reduced biomass. A previous study reported that replacement of a single amino acid 90 residue in the TCV CP P38 (R130T) disrupts its VSR function without affecting other 91 functions [26]. The VSR-deficient TCV is unable to suppress the host antiviral silencing 92 machinery, leading to a lack of disease symptoms in wild-type plants post-inoculation.

94
In this study, the roles of Arabidopsis AGO proteins in anti-TCV silencing were analyzed 95 using genetic and image-based quantitative phenotyping approaches. Turnip Crinkle Virus (TCV) and other carmoviruses use their coat proteins (CPs) as 106 VSRs to suppress host antiviral silencing [34,35]. By replacing a single amino acid in 107 TCV CP (P38) with its counterpart residue in TBSV CP (R130T) (Fig 1A), the VSR 108 function of TCV CP (P38) is abolished [26]. Introducing this mutation (R130T) in the 109 coat protein did not affect its ability to assemble functional virion particles (S1 Fig). We 110 confirmed that this mutant virus (TCV CPB) lost its capacity to suppress host antiviral 111 silencing by quantifying and comparing the effects of TCV CPB in wild-type (Col-0) and 112 dcl2-1 dcl3-2 dcl4-2 (dcl2 dcl3 dcl4) triple mutant plants. In the dcl triple mutant, three 113 DCL genes with roles in antiviral defense are dysfunctional, so it serves as a hyper-114 susceptible control genotype. Parental TCV infection caused severe chlorosis in 115 Arabidopsis leaves in both wild-type control (Col-0) and the dcl triple mutant plants, 116 while TCV CPB caused similar chlorosis symptoms only in the dcl triple mutant (Fig 1B).

118
Turnip crinkle virus was from samples of inoculated rosette leaves and non-inoculated 119 cauline leaves using immunoblot assays with CP antisera (anti-P38). High levels of CP 120 were detected in rosette leaves at 7 days post inoculation (dpi) and in cauline leaves at 121 14 dpi of both Col-0 and dcl2 dcl3 dcl4 plants infected with parental TCV, even at 122 1:1000 dilution ( Fig 1C). In contrast, low levels of CP were detected in rosette leaves of 123 wild-type control plants inoculated with TCV CPB at lower dilutions (1:1 to 1:50) but not 124 at higher dilutions (1:100 to 1:1000; Fig 1C). In non-inoculated cauline leaves of Col-0 125 plants, no observable CPB-P38 signal was detected at any dilution ( Fig 1C). However, 126 TCV CPB inoculation of dcl2 dcl3 dcl4 mutant led to high levels of CP protein 127 accumulation in the rosette and cauline leaves at each dilution ranging from 1:1 to 1:500 128 ( Fig 1C). Notably, in dcl2 dcl3 dcl4 mutants, no observable local or systemic CPB CP 129 protein was detected when the TCV CPB inoculum was highly diluted (1:1000) ( Fig 1C).

131
Parental TCV infection at any dilution had negative effects on the growth of both Col-0 132 and dcl2 dcl3 dcl4 plants ( Fig 1D). Morphologically, plants infected with TCV were 133 shorter compared to non-infected plants ( Fig 1D). In contrast, the TCV CPB mutant 134 affected growth of only dcl2 dcl3 dcl4 plants ( Fig 1D). These results confirmed that the 135 CPB (R130T) mutation in TCV CP attenuates VSR functions, and suggested that TCV 136 CPB could be used in a genetic analysis to identify components of the silencing 137 machinery necessary for antiviral defense. 138

Image-based analysis of disease symptoms 139 140
Before initiating a systematic screen of Arabidopsis ago mutants using TCV and TCV 141 CPB, a non-destructive, computer vision-based system using the PlantCV platform [33] 142 was developed for high-resolution, quantitative assessment of disease symptom 143 phenotypes over time (Fig 2A) inoculation to seventeen days post-inoculation. The images were analyzed using a 148 machine learning approach to segment plant from background and to classify plant 149 pixels as "healthy" pixels (green color) or "unhealthy" chlorotic/necrotic pixels (any non-150 green color; Fig 2B). The ratio of chlorotic/necrotic to total plant pixels was used to 151 calculate the extent and severity of symptomatic rosette leaf tissue.

153
To validate the approach, wild-type (Col-0) and hyper-susceptible dcl2 dcl3 dcl4 mutant 154 plants were inoculated with parental TCV or TCV CPB mutant virus, or were mock-155 inoculated (4 replicates per genotype and treatment combination). In the mock-156 inoculated groups, the majority of rosette area remained healthy (green) over time in 157 both Col-0 and dcl2 dcl3 dcl4 plants (Figs 3A-B). Infection of both plant genotypes with 158 parental TCV elicited local chlorosis at 7 dpi and nearly complete chlorosis/necrosis by 159 17 dpi (Fig 3A-B). Discoloration caused by parental TCV infection appeared to be more 160 severe in the dcl2 dcl3 dcl4 mutant plants than in wild-type controls ( Fig 3B). TCV CPB 161 inoculation led to parental virus-like discoloration symptoms in the dcl triple mutant 162 plants ( Fig 3B) but did not cause strong chlorosis symptoms in Col-0 plants ( Fig 3A).

163
Although an increased number of chlorotic pixels was observed in Col-0 rosettes at 15-164 and 17-days post TCV CPB inoculation, the increase was visually less than that caused 165 by parental TCV (Fig 3A).

167
Parental TCV infection led to reduced rosette area, based on total number of pixels 168 (healthy plus chlorotic), indicating a decrease in biomass. Parental TCV significantly 169 inhibited rosette growth over time in both wild-type and hyper-susceptible controls (Figs 170 3C-D) (Kolmogorov-Smirnov test, p<0.05). Similarly, TCV CPB led to a comparable 171 decrease in rosette area in dcl2 dcl3 dcl4 mutants after 7 dpi, but not in Col-0 (Figs 3B-172 D). At 17 dpi, a significant difference in rosette area was detected between the two 173 virus-treatments in Col-0, but not in dcl2 dcl3 dcl4 plants ( Fig 3E).

175
The percentage of unhealthy chlorotic/necrotic pixels in the whole rosette was also 176 calculated. Both parental TCV and TCV CPB infection caused a significant increase in 177 the percentage of unhealthy tissues over time in dcl2 dcl3 dcl4 plants (Fig 3G), from 178 approximately 0% at 5 dpi to nearly 100% at 17 dpi ( Fig 3G). In Col-0 plants, parental 179 TCV infection also led to a gradual increase in chlorotic tissue, from approximately 0% 180 at 5 dpi to nearly 60% at 17 dpi ( Fig 3F). In contrast, TCV CPB did not significantly 181 change the percentage of chlorotic tissue in wild-type control plants over time ( Fig 3F)

186
Measurement of hue value of leaf color has been used to estimate chlorophyll loss 187 caused by biotic and abiotic stress [36][37][38]. Hue value information extracted from the 188 RGB images was used as a parameter to quantify chlorosis in rosette leaves.  Fig 4B). In contrast, this green-to-yellow shift 202 was not observed in Col-0 plants inoculated with TCV CPB (right panel, Fig 4A).

204
The average hue value for whole plants in each treatment group was calculated.

205
Parental TCV infection caused a temporal decrease in hue value in both Col-0 (from 206 100° to 75°) and dcl2 dcl3 dcl4 plants (from 100° to 40°) (Figs 4C-D). TCV CPB induced 207 a similar decrease in average hue value over time in dcl2 dcl3 dcl4 mutants, but did not 208 significantly affect hue value in Col-0 plants (Figs 4C-E). These results were consistent 209 with the healthy/chlorotic classification-based results (Fig 3). These data indicate that 210 the image-based growth and color trait measurement protocols were effective in 211 quantifying virus-induced symptoms in Arabidopsis, and in distinguishing plant 212 susceptibility or virus virulence phenotypes. 213

Image-based analysis of TCV and TCV CPB infection in ago mutants 214 215
As with the dcl2 dcl3 dcl4 plants, we hypothesized that loss of ago genes with a function 216 in anti-TCV silencing would be revealed by gain of susceptibility to the VSR-deficient 217 TCV CPB mutant. A set of ten mutant plants with defects in each of the ten Arabidopsis 218 AGO genes was inoculated with mock solution, parental TCV or TCV CPB, and image-219 based traits and virus protein levels over an infection time-course were measured. 220 Pairwise Kolmogorov-Smirnov (K-S) tests were done to compare time-series data.

236
Focusing on the data on 17 dpi, the effects caused by TCV CPB on rosette chlorosis in 237 ago2-1 and zip-1 mutant could be directly displayed ( Fig 6A) and statistically tested ( Fig  238  6C) (Tukey post hoc test with α=0.05). Similarly, by using this statistic test, we further 239 confirmed that TCV CPB caused a significant decrease in ago2-1, ago3-2 and zip-1 240 rosette size at 17 dpi ( Fig 6B).

242
Another output of the phenotyping pipeline to evaluate leaf color is hue value (Fig 2A).

243
The temporal changes in hue value in the ten inoculated ago mutants and control plants 244 were measured over time (Fig 7). One primary peak at approximately 100° (green) was 245 observed in early stages for all genotype and inoculation combinations (-1 to 5 dpi in 246 each panel, Fig 7). In the mock-inoculated group, the primary peak remained at around 247 100° in all genotypes throughout the time course. The green peak gradually shifted to 248 yellow in each ago mutant and the two controls infected with parental TCV (7 to 17 dpi 249 in each panel, Fig 7). As expected, dcl2 dcl3 dcl4 mutant plants infected with TCV CPB 250 turned from green (peak at 100°) to yellow over time (blue area, dcl2/3/4 panel, Fig 7). 251 Similarly, a complete green-to-yellow shift was measured in ago2-1 mutant plants 252 infected with TCV CPB (blue area in ago2-1 panel, Fig 7). Gradual, partial shift from 253 green to yellow was measured in zip-1 plants infected with TCV CPB (blue area in zip-1 254 panel, Fig 7). All other ago mutants (ago1-27, ago3-2, ago4-2, ago5-2, ago6-3, ago8-1, 255 ago9-5 and ago10-5) inoculated with TCV CPB were similar to mock-inoculated plants 256 (blue area, Fig 7).

258
To quantify the color shift observed above, average hue value was calculated. Pairwise 259 Kolmogorov-Smirnov (K-S) tests were applied to compare time-series data. Parental 260 TCV infection led to a gradual decrease in average hue value in all genotypes (Fig 8A), 261 while TCV CPB caused a similar decrease only in ago2-1 and zip-1 mutants ( Fig 8A) In 262 addition, data at 17 dpi further confirmed the negative effects of TCV CPB on average 263 hue value in ago2-1 and zip-1 mutants ( Fig 8B) (Tukey post hoc test with α=0.05). 264

Analysis of viral coat protein levels in ago mutants 265
To further investigate the roles of anti-TCV AGO proteins identified above in different 267 tissues, immunoblotting assay was used to analyze viral coat protein (P38) levels in 268 inoculated, non-inoculated (systemic) tissues, and inflorescence tissues of ago mutants 269 and control plants. High levels of P38 were detected in leaves and inflorescence 270 tissues of all ten single ago mutants and two control plants inoculated with parental TCV 271 (left panel, Fig 9A; blue bars in Figs 9B-D). Local accumulation of CPB-P38 was only 272 detected in ago1-27, ago2-1, and zip-1 mutant rosette leaves (right panel, Fig 9A, red  273 bars in Fig 9B), at a comparable level with that in dcl2 dcl3 dcl4 leaves (red bars in Fig  274  9B). In non-inoculated cauline leaves, systemic accumulation of CPB-P38 was only 275 detected in ago2-1 and zip-1 mutants (Fig 9C), and at levels that were approximately 276 half of that in dcl2 dcl3 dcl4 mutants (Fig 9C). CPB-P38 in local and systemic leaves 277 was also detected in two of four biological replicates of ago3-2 mutants (right panel, Fig  278  9A), though at a significantly lower level than that in ago2-1 mutants (Figs 9B-C).

279
Notably, CPB-P38 was not detected in inflorescence tissues of any of the ten ago single 280 mutant and and/or zip-1 mutant allele (Fig 10). First, parental TCV CPs accumulated at comparable 287 levels in inoculated rosette, non-inoculated cauline leaves and inflorescence tissue in 288 the single, double and triple ago mutants (Figs 10A-C). In local and systemic leaves, no 289 significant differences in P38 level were observed between ago2-1, zip-1 and ago2-1 290 zip-1 double mutant (Figs 10B-C). These results suggested that AGO2 and AGO7 play 291 non-additive antiviral roles in Arabidopsis leaves during TCV infection. To examine if 292 AGO1, AGO2 or AGO7 play redundant antiviral roles, we checked if down-regulating 293 ago1 had any enhancing effects on P38 accumulation in ago2 or ago7 mutants. We 294 found no significant differences in local P38 level in rosette leaves between ago2-1 295 single, zip-1 single, ago1-27 ago2-1 double, ago1-27 zip-1 double and ago1-27, ago2-1 296 zip-1 triple mutant (Fig 10B). Similarly, combining the ago1-27 allele with an ago2-1 or 297 zip-1 allele, or ago2-1 and zip-1, did not significantly affect the CPB-P38 accumulation 298 level in systemic cauline leaves (Fig 10C). These results suggested that introducing the 299 ago1-27 allele does not enhance or suppress the antiviral activities of AGO2 and AGO7 300 in Arabidopsis leaves. Similar to the observation in ago single mutants, no TCV CPB 301 CPs were detected in the inflorescence clusters of the double and triple mutants tested 302 (Fig 10D). 303

Analysis of plant height during TCV infection in ago mutants 304
The effects of TCV infection on plant height in ten ago mutants, the dcl triple mutant and 306 wild-type plants inoculated with parental and VSR-deficient TCV were measured. 307 Growth curves were plotted from one day pre-inoculation to 21 dpi as described 308 previously [40]. Pairwise Kolmogorov-Smirnov (K-S) tests were done to compare time-309 series data.

320
The TCV CPB mutant also strongly inhibited plant height in ago2-1 and zip-1 mutants 321 ( Fig 11A). Bayes classifiers were defined: 1) green plant pixels ("healthy"); 2) chlorotic/necrotic 348 plant pixels ("unhealthy"); 3) blue mesh background pixels; 4) other background pixels. 349 Since the light intensity condition was relatively constant, the color information of pixels 350 from small number of sample images was sufficient for generating the training dataset 351 to cover the range of variation of all images. In the training session, four classes of 352 pixels could be segmented simultaneously. In addition, the naïve Bayes segmentation 353 process is robust across large number of images due to its probabilistic nature. This 354 pipeline is also simple and computationally expedient. The output of the pipeline could 355 be statistically analyzed to quantify plant size, the proportion of unhealthy tissue, and 356 leaf color. This non-destructive phenotyping pipeline enables visualization and 357 quantification of disease symptom development over time from large numbers of plants.

359
In this study, the image-based disease analysis method was put to the test in wildtype 360 and mutant Arabidopsis plants, with defects in AGO genes, infected by parental and 361 VSR-defective TCV. Detailed phenotyping of inoculated Arabidopsis ago mutant plants 362 was done. Ten AGO genes are encoded in the Arabidopsis genome and they are 363 grouped into three major clades: AGO1/5/10, AGO2/3/7, and AGO4/6/8/9 [3].

364
Combining disease phenotyping and biochemical results, AGO2 and AGO7 were 365 identified as prominent factors during anti-TCV defense, along with minor antiviral roles 366 of AGO3. The results with AGO2 and AGO7 are consistent with previous studies [14, 367 20]. Using the VSR-defective virus, the antiviral role of AGO7 was relatively minor 368 compared to that of AGO2. AGO2 and AGO7 were non-additive in leaves (Fig. 10). 369 However, movement of the TCV mutant to inflorescence tissues was still inhibited in 370 ago2 ago7 double mutant, but not in dcl2 dcl3 dcl4 mutants (Fig. 10). These results 371 implied that other AGOs not tested in the current genetic combinations are involved to 372 restrict TCV movement to, or accumulation in, inflorescence tissues. Regardless of precise antiviral mechanisms in play against TCV, the image-based 427 phenotyping system developed here was shown to be useful in delineating both major 428 and minor contributions of AGOs during virus infection over time. This should 429 encourage development and refinement of additional tools and readouts to measure 430 more effects and responses, even before visible symptoms of disease are detectable. 431 Closing the gap between knowledge at the genetic and molecular levels with phenotypic 432 effects during pathogen infection and non-pathogen colonization should yield 433 considerable new insights into host-microbe interactions. 434

436
Precise phenotyping methods to measure biotic and abiotic effects on plant health are 437 important to develop. A high-throughput, image-based disease trait phenotyping 438 pipeline to quantify virus-induced symptoms in Arabidopsis was developed. Combined 439 with infectivity experiments using both parental and VSR-defective TCV variants, AGO2 440 and AGO7 were identified as the most prominent antiviral AGO proteins against TCV, 441 while AGO3 was found to have a minor effect on antiviral silencing. These and previous 442 data support the idea that multiple AGOs are recruited and programmed during antiviral 443 defense in unique ways against different virus species. 444 445

Plant materials 447 448
All Arabidopsis plants used in this study (including all mutant lines) were in the 449 Columbia-0 (Col-0) background and were grown in growth chambers under long day 450 (16-hour light/8-hour dark) conditions, at 22°C and 75% relative humidity. The ten ago 451 mutants used in this study were described previously: ago1-27, ago2-1, ago3-2, ago4-2, 452 ago5-2, ago6-3, zip-1, ago8-1, ago9-5 and ago10-5 [15]. Double and triple ago mutants 453 were generated by crossing between single ago mutants. The soil in the growth pots was covered by a blue mesh (commercially available from 512 Amazon.com), leaving a hole in the center for the plant to grow out ( Fig 2B). Meshed 513 plants were bottom-watered to avoid water splash on plant tissues. Images of plant 514 rosettes were manually acquired using a Canon Digital Rebel XT DSLR camera with 515 EF-S 18-55mm f2.5-5.6 lens (0.60 s exposure, F/14, ISO100) on a Kaiser RS1 Copy 516 Stand. The flash function was kept off and fixed ambient light was used in a closed 517 room to minimize illumination discrepancies between images. An X-rite ColorChecker 518 Digital SG was placed next to the plant as a color reference and correction. Images 519 were stored in the native RAW format and also high-resolution JPG format. Plants were 520 removed from the growth chamber (10 am -11 am) and any bolts above the rosette 521 plane were removed before imaging. Stationary plant rosettes were imaged every other 522 day from one day before inoculation to 17 dpi. 523

531
The PlantCV naive Bayes machine learning method [31,32] was used to segment 532 image pixels into four classes: 1) green plant pixels ("healthy"); 2) chlorotic/necrotic 533 plant pixels ("unhealthy"); 3) background pixels from the blue mesh material used to 534 cover the soil, and 4) all other background pixels. The PlantCV naive Bayes classifier 535 was trained using pixel red, green, blue (RGB) color values from 1980 and 3779 pixels 536 manually sampled from the background and foreground classes, respectively, from 537 multiple images using the ImageJ pixel picking tool [64] as described in [32]. The 538 training data was used to calculate probability density functions (PDFs) using kernel 539 density estimation (KDE) for each class in the hue, saturation, and value (HSV) color 540 space [32]. During image analysis, the PDFs were used to parameterize the naive 541 Bayes classifier to segment images into the four output classes (one binary image per 542 class).

544
After segmentation, the blue mesh was used to automatically identify the position of the 545 pot within each image. Morphological erosion was used to reduce noise in the classified 546 blue mesh pixels. A universal region of interest was used to keep blue mesh connected 547 components in the center area of each image since the pot was consistently centered in 548 each image. The remaining blue mesh connected components were flattened into a 549 single object and were used to create a pot binary mask using the bounding rectangle 550 area of the blue mesh. Padding was added to the minimum bounding rectangle area 551 using morphological dilation. The resulting pot binary mask was used to mask the 552 "healthy" and "unhealthy" binary masks to remove misclassified background pixels. The 553 "healthy" and "unhealthy" binary images were further filtered to remove background 554 pixels misclassified as foreground pixels using a size-based filter that removed small 555 connected components (300 and 1000 pixels for "healthy" and "unhealthy" binary 556 images, respectively). A universal region of interest was used to keep connected 557 components in the center area of each image since the Arabidopsis plants were 558 consistently centered in each image. The resulting cleaned binary image for the 559 "healthy" and "unhealthy" classes were used to measure the area of green and 560 chlorotic/necrotic pixels in each image. 561 562 Additionally, the union of the "healthy" and "unhealthy" plant pixels was calculated to 563 create a combined plant mask. The input RGB images were converted to HSV color 564 space and the frequency distribution of hue color values for each plant at each timepoint 565 was extracted from the pixels that overlapped the combined plant mask. Author Contributions 586 X.Z. and J.C.C conceived and designed research. X.Z. and A.A. performed 587 experiments. X.Z., N.F. and J.B. analyzed and plotted data. J.C.C. and X.Z. wrote the 588 manuscript with contribution of all authors. 589