Soybean Tolerance to Off-Target Dicamba

CAIO CANELLA VIEIRA Author

04/05/2023 Added

15 Plays

Description

Assistant Professor of Soybean Breeding, Department of Crop, Soil and Environmental Sciences, University of Arkansas

Searchable Transcript

Search:

[00:00:00.810]The following presentation
[00:00:02.250]is part of the Agronomy and Horticulture seminar series
[00:00:05.850]at the University of Nebraska Lincoln.
[00:00:08.700]Good afternoon, everyone.
[00:00:11.700]It's my pleasure to introduce our distinguished guest today.
[00:00:16.406]Please join me in welcome Dr. Caio Canella Vieira today
[00:00:20.257]So, Dr. Caio joined University of Arkansas
[00:00:22.500]in January this year, 2023.
[00:00:25.560]Brandon leaving the university (indistinct)
[00:00:29.578]Before joining the University of Arkansas
[00:00:30.896]Dr. Caio Vieria obtained his master's and PhD,
[00:00:35.220]where he built genetics and genomics
[00:00:37.890]at the University of Missouri,
[00:00:39.561]where he was involved in development
[00:00:42.360]and released over 30 soybean (indistinct)
[00:00:45.330]and the lab (indistinct) of the soybean breeding problem
[00:00:49.050]by prioritizing the use of data analytics and predict model.
[00:00:53.160]Today, Dr. Caio will be speaking to us about soybean
[00:00:58.089](indistinct) dicamba.
[00:01:00.507]So please join me and give you a warm welcome to Dr. Caio.
[00:01:07.475](mumbling)
[00:01:09.840]Well, thank you.
[00:01:11.162]Thank you very much for the introduction.
[00:01:13.320]A little distinguished, little, little too much,
[00:01:15.540]but thanks for that.
[00:01:18.930]It's, it's a great pleasure to be here.
[00:01:21.262]I've been, you know,
[00:01:22.322]enjoying meeting everybody that I had the chance so far.
[00:01:25.290]Thanks for, for the reception in being welcoming.
[00:01:29.640]I made some my,
[00:01:31.670]my family types in Nebraska. My brother and his wife,
[00:01:34.560]they did their, their PhDs here, so it's nice to,
[00:01:37.650]nice to be back giving a talk here this time.
[00:01:41.160]So, like Marco said, my name is Caio.
[00:01:43.320]I actually just started this tradition at the
[00:01:45.420]University of Arkansas.
[00:01:46.860]I graduated from Missouri in December and been been working
[00:01:50.370]in Arkansas since January in our soybean breed program.
[00:01:53.820]What I'm sharing today is mostly what I've done during my
[00:01:57.270]part of my master's, but mostly during my PhD in Missouri,
[00:02:00.450]which is we were working on soybean natural tolerance 12
[00:02:04.830]target at campus. So just to give a little overall,
[00:02:06.344]the way I, the way we try to do this project,
[00:02:07.744]we have three major components. So the first one,
[00:02:13.650]we wanted to see if the damage caused any yield loss.
[00:02:16.740]There's so many different reports.
[00:02:18.390]We want confirm if there was loss and one was the magnitude
[00:02:22.142]of the yield loss. The second one,
[00:02:24.420]we wanted to use a type of pheno typing
[00:02:26.550]for differentiating the damage.
[00:02:29.220]Takes a lot of work to walk all the plots
[00:02:31.290]and if we could fly the drone, it would be be much easier.
[00:02:35.220]And the last one was to confirm whether this was really
[00:02:37.920]genetic. So we did our genetic analysis, we did our mapping,
[00:02:42.000]and the result of part seemed to be pretty interesting in
[00:02:46.838]to, to continue to pursue that.
[00:02:50.863]I just wanna make sure that I'm not,
[00:02:53.850]I'm not against any technology by any means. I think they,
[00:02:57.390]each of them have their role in agriculture.
[00:03:00.273]What we, what we're doing is, you know,
[00:03:01.800]reporting science in alternatives to any possible damage
[00:03:05.700]that can come from a specific technology.
[00:03:09.780]Okay.
[00:03:10.613]Alright. So just to, again, don't wanna bore everybody
[00:03:13.050]with the introduction, but I just thought it was worth
[00:03:15.690]mentioning this,
[00:03:16.523]what was the rationale of developing dicamba-tolerant crops
[00:03:20.550]way back in the, in the, the two thousands?
[00:03:22.590]This is an old technology,
[00:03:24.090]there's nothing really new about it.
[00:03:26.190]It was actually developed here in Nebraska
[00:03:27.780]so it's nice to, to talk about it here, but mainly the, the,
[00:03:31.200]the rationale was to preserve biotechnology.
[00:03:34.770]We manage it. If you recall in the early two thousands,
[00:03:38.100]you only had Roundup Ready and resistance was, you know,
[00:03:41.610]exploding within the weeds.
[00:03:43.650]And so this was a way of developing new technologies that
[00:03:47.153]farmers could deploy to do weed management.
[00:03:50.760]So they found, I don't wanna get too much into it,
[00:03:53.277]but it was this gene from,
[00:03:55.440]from bacteria in the soil where it converts the active
[00:03:58.950]dicamba into no active dicamba.
[00:04:01.230]They put that into the plants and as you can see there,
[00:04:03.930]one of the, the first varieties to be transformed
[00:04:06.023]was a Nebraska variety and another one from Ohio.
[00:04:10.470]And they were doing all the few trials and, you know,
[00:04:14.070]plants were highly resistant.
[00:04:15.750]They tried even higher doses than, than the label rate.
[00:04:19.440]And they were showing resistance, everything looking good,
[00:04:22.200]right? It's a, it's a beautiful paper published,
[00:04:24.180]I believe it's either science or nature.
[00:04:26.640]It's a great technology.
[00:04:28.470]Monsanto bought that technology and you had the first
[00:04:31.950]commercial variety came into market in 2016.
[00:04:35.550]And as you can imagine, there was a big desire for,
[00:04:38.460]for a new herbicide trait and quickly took over the acreage.
[00:04:42.390]So in the first year you have almost, you know,
[00:04:44.335]55% of the acreage coming from dicamba-tolerant sprays.
[00:04:49.320]I wanna add this here,
[00:04:50.280]if anybody's not very familiar with the,
[00:04:52.110]the timeline of the traits, but you know,
[00:04:54.900]you literally had the Roundup Ready varieties,
[00:04:57.690]which they were used for, for quite some time.
[00:05:00.270]But then you have the development of resistance that really
[00:05:03.690]effects things. For many years, all that we could rely on
[00:05:08.460]was Roundup Ready 1 and 2 and the liberty technology.
[00:05:12.480]So for many years there was that gap of new tech, excuse me,
[00:05:17.340]new technologies, hence the need of this dicamba tolerance.
[00:05:21.749]In 2016, like I mentioned,
[00:05:23.250]it was the first year where you had dicamba
[00:05:25.470]and glyphosate within the Xtend technology.
[00:05:28.530]And of course it's a,
[00:05:29.460]it's a competition between companies and then you get Enlist
[00:05:33.780]coming up in the market.
[00:05:35.160]Where we are now is just a com- combination of different
[00:05:39.360]technologies, so what we call the,
[00:05:40.710]the stacking multiple technologies. And that's,
[00:05:43.290]that's gonna be interesting to see where that goes.
[00:05:45.840]But the reason I, I put this here, so if you quite imagine,
[00:05:49.080]again, all the rational
[00:05:50.910]behind the development of this technology
[00:05:53.550]and how available this was at the time,
[00:05:57.390]but not everything is perfect.
[00:06:00.810]And you know,
[00:06:01.950]after this really came into multiple acres across,
[00:06:06.870]especially the Midwest, there was numerous reports of,
[00:06:10.380]of target damage to known dicamba tolerant crops and also
[00:06:15.270]different plant species, different dicot species,
[00:06:18.272]not necessarily only crops.
[00:06:19.500]That's the picture that I, I, I took this at the,
[00:06:23.642]the ASTA meeting and I thought it was,
[00:06:24.810]it was interesting to go it here, the,
[00:06:27.920]the way the advertisement for list was done, that
[00:06:31.770]you're hitting the target, not your, your neighbors view.
[00:06:35.220]But so between 17 and 21,
[00:06:37.560]the APA reported nearly about 10,000 cases of off target
[00:06:42.300]damage. But of course it's a very under reported number.
[00:06:46.380]And then, you know, there's a branch of USDA
[00:06:48.650]that does a lot of the statistics.
[00:06:50.070]They think the number's under reported by about 25 fold.
[00:06:55.050]If you're walking any farm around,
[00:06:57.180]you know, Missouri, Nebraska, Arkansas, all those regions,
[00:07:00.240]you probably would expect that the number is much higher
[00:07:02.200]than, than only 10,000 reports.
[00:07:06.180]One thing that to me is quite interesting and and again
[00:07:09.720]justify a lot of the research that we've done here is that
[00:07:12.983]all, all the acreage we're a fixed stand.
[00:07:16.890]You have almost 40% of that in every spray dicamba.
[00:07:20.460]So you have nearly half of the growers adopting this
[00:07:24.480]technology, but they never use the herbicide,
[00:07:27.360]which makes you think, are they doing this
[00:07:28.793]so they can protect themselves from all possible damage or,
[00:07:32.490]you know, it's really interesting to think about it.
[00:07:35.400]So there's, there's a little bit of affecting
[00:07:38.550]the freedom of choice of the crop system that you wanna
[00:07:41.910]pursue in your farm and,
[00:07:43.812]you know, it's not up to me to say if it's right or wrong,
[00:07:46.252]but I think it's a good expressive number when you consider
[00:07:49.800]all the (indistinct) wrong with this.
[00:07:52.890]So shouldn't be surprising for, for everybody.
[00:07:56.040]But soybean is highly sensitive to dicamba,
[00:07:59.271]they can be sensitive even to really small doses.
[00:08:01.830]So again, if you imagine this scenario
[00:08:03.930]where you're getting exposed to dicamba
[00:08:06.272]even multiple times during the season, it's,
[00:08:08.520]you know, but by frankly you're gonna expect some, some
[00:08:11.790]degree of yield loss, especially when you're those regions
[00:08:15.510]where you have a lot of damage growing.
[00:08:18.690]So the nice work done by weed scientists,
[00:08:22.410]they show multiple factors affecting the,
[00:08:25.287]the degree of the severity of the damage.
[00:08:28.200]But one major component that was never really considered in
[00:08:31.050]that, again that's where the symptom comes in,
[00:08:33.660]is how much the genetic background of the soybean varieties
[00:08:37.551]do affect severity of the damage.
[00:08:40.710]You know, I like to say that you know, you consider,
[00:08:43.380]imagine in this room you're for giving everybody a specific
[00:08:46.350]drug, the response are gonna be different.
[00:08:49.080]Not everybody responds the same. And when you think about
[00:08:51.270]plants and the amount of genetic diversity that you have,
[00:08:54.540]you would expect that your response is also different.
[00:08:58.500]And I thought it was interesting to add this because this
[00:09:00.780]ended up happening in, in Arkansas.
[00:09:03.084]It goes way beyond just yield loss. It affects social,
[00:09:08.370]social conflicts between neighbors and everything else.
[00:09:10.590]And if you're familiar with the reportings that were
[00:09:13.890]happening at the time, there's even fatal shootings that
[00:09:16.650]happened in just outside of Arkansas, excuse me.
[00:09:27.450]So holidays started for us.
[00:09:29.704]We were located at the time in (indistinct) Missouri.
[00:09:32.850]So that's really a area where you divided in acres between
[00:09:36.330]soybean and cotton. And as you can imagine,
[00:09:38.520]most of the cotton was also dicamba tolerant
[00:09:41.873]and my advisor at the time we were walking plots, you know,
[00:09:44.730]making gradient selections and we would walk by something
[00:09:47.940]like this where you would have the extent variety
[00:09:51.690]obviously no damage,
[00:09:53.460]and then we would've a conventional variety, so no traits,
[00:09:56.940]highly susceptible, you can see all the damage coming
[00:09:59.040]from that camera.
[00:10:00.330]And we would've something like this in the middle where they
[00:10:03.150]seem to have a lower degree of the symptoms from dicamba.
[00:10:07.860]And the first time we observed that it was,
[00:10:09.780]we were quite interested to see is this gen genetics or this
[00:10:13.590]is just differential response to dicamba or different doses
[00:10:17.880]being exposed to each plot.
[00:10:20.220]Started seeing that across multiple replications,
[00:10:22.380]multiple environments,
[00:10:24.270]thought it was worth investigating a little bit more.
[00:10:27.540]And then we went to progeny rows.
[00:10:29.280]So if you're familiar with breeding, progeny rows,
[00:10:32.310]you basically cross parent one, parent two,
[00:10:35.220]and then you have your progens.
[00:10:37.200]If you're seeing segregation for a specific trait,
[00:10:39.990]there's a high chance that this a genetic treat,
[00:10:43.140]and this is what we were observing in our oppressions.
[00:10:45.960]So these two had derived from the same parents,
[00:10:49.164]but their response to dicamba is completely different.
[00:10:53.610]And then we were seeing this consistently across different
[00:10:56.250]populations. So that one of the parents were tolerant,
[00:10:58.620]one was susceptive,
[00:11:04.050]and then we decided to investigate if that would be possible
[00:11:09.900]to observe in diverse accessions.
[00:11:11.880]So we brought in soybean accessions from all over the world.
[00:11:15.210]This is supposed to represent most of the genetic diversity
[00:11:18.631]available in soybean.
[00:11:20.940]And as you can see,
[00:11:22.560]they were showing differences the same way.
[00:11:25.020]And I I I like to mention that these are side by side rows,
[00:11:28.650]so it's a really small distance between them.
[00:11:31.230]You would expect that the doses are being exposed pretty
[00:11:34.050]much the same.
[00:11:36.840]When they start breeding for this,
[00:11:38.220]we start advancing varieties with higher tolerance,
[00:11:41.100]some with higher susceptibility.
[00:11:42.840]And this is one of the pictures that we,
[00:11:44.869]that we get one of our most advanced varieties
[00:11:47.760]with a higher tolerance.
[00:11:49.227]And this is just a regular run up ready check,
[00:11:52.410]same maturity, same everything,
[00:11:54.270]but it's almost like you draw a line in the middle,
[00:11:57.242]have have was sprayed, half was not,
[00:11:59.704]but those are being exposed to the same amount of dicamba.
[00:12:01.350]So you can really see at least on the visual aspect
[00:12:05.280]they were showing the difference.
[00:12:07.410]What we wanted to confirm is,
[00:12:08.940]is this visual translating into yield?
[00:12:13.080]The other thing that we've done is we got one of those
[00:12:15.660]diverse successions that we're showing higher tolerance
[00:12:18.210]cross to some of our susceptible elite lines such as lines
[00:12:21.780]from our program.
[00:12:23.370]And again, we started seeing segregation from the same way.
[00:12:26.940]Some of the projects from the cross,
[00:12:29.430]higher tolerance, side by side rows,
[00:12:31.950]grow completely hampered by dicamba.
[00:12:34.080]So every evidence that we were getting over the years
[00:12:37.710]were really confirming that
[00:12:39.030]this would be a true genetic treatment.
[00:12:42.630]Alright, so the first, like I mentioned,
[00:12:44.400]the first step of this work was to define
[00:12:48.420]does the damage cause yield loss?
[00:12:49.860]And if it does, what is the magnitude?
[00:12:52.625]Like I mentioned,
[00:12:53.580]one of the main differences of this work was
[00:12:56.490]relying on genetic diversity. So again,
[00:12:59.520]as much as as available as the previous work done,
[00:13:03.060]those are usually based on a handful of varieties.
[00:13:06.030]Here we used over 550 soybean varieties.
[00:13:09.030]So there's a lot of genetic diversity and we can get a good
[00:13:12.480]distribution of damage across them.
[00:13:15.420]It's also worth mentioning that they're coming from multiple
[00:13:18.531]different populations. So they are, you know, diverse.
[00:13:22.530]It's not all lines from the same parent.
[00:13:25.140]We did this for over three years and multiple locations,
[00:13:28.181]multiple replications. So they end up being a pretty
[00:13:30.450]large data set at the end of the year.
[00:13:35.010]So one of the main questions that we got when we were
[00:13:37.500]presenting the preliminary results was,
[00:13:39.810]how do you know your dicamba exposure?
[00:13:43.290]And it's a very fair question and, and to be honest,
[00:13:45.390]I think it's really hard to define your exposure because
[00:13:50.190]you don't know where it's coming from.
[00:13:52.410]You know, it may head with the immersion,
[00:13:54.510]you're getting dicamba from multiple directions
[00:13:57.180]and different timing of exposure, different doses.
[00:14:00.000]It's something really hard. You know,
[00:14:01.710]when you think about the statistics,
[00:14:03.630]the dosage doesn't necessarily matter that much as long as
[00:14:07.020]the exposure is homogeneous in a specific field. So we will,
[00:14:11.310]what we wanted to do is kind of like define a yield map
[00:14:14.370]of the performance of a known (indistinct) check
[00:14:18.780]with the (indistinct) check.
[00:14:20.490]So think about it like this.
[00:14:25.598]If you have two checks with the same yield potential,
[00:14:29.310]if there's no dicamba, they should be yielding the same.
[00:14:31.890]So the proportion between them is hundred percent.
[00:14:34.440]So that's what we've done.
[00:14:36.000]If there is dicamba and that proportion goes down,
[00:14:38.610]we cannot attribute that difference to the exposure
[00:14:39.562]to the dicamba.
[00:14:42.780]So that's kind of how the few layouts would look like.
[00:14:45.720]We would've, you know, a plot of a Roundup Ready customer.
[00:14:50.699]We got the proportion for the nearest extend plot.
[00:14:55.440]And then in between those we did the interpolation to really
[00:14:58.890]design that yield map distribution.
[00:15:02.310]And as you confirm if it was homogeneous or not,
[00:15:05.610]we used the Poisson model,
[00:15:07.170]which can account for a spatial distribution.
[00:15:10.200]Surprisingly or not,
[00:15:11.940]we found that distribution was highly homogeneous,
[00:15:14.160]which is what we've been observing over time. So it's,
[00:15:17.130]it's not necessarily a only a section of the field that is
[00:15:21.330]being exposed by dicamba, it's really almost like a cloud
[00:15:25.200]falling over your field. As you see,
[00:15:27.297]in the different fields that we tested,
[00:15:29.790]the distribution's pretty much homogeneous.
[00:15:31.980]So it shows that for, for what matters here,
[00:15:35.520]we know that the distribution was homogeneous.
[00:15:39.000]The other question that we got a lot was
[00:15:42.720]how do you know that your scoring is is accurate, right?
[00:15:45.980]We we basing this on visual scores.
[00:15:49.020]So we use two different metrics to decide if our phenotype
[00:15:52.590]was high quality or not.
[00:15:54.870]One of them is a simple correlation on among replications in
[00:15:58.080]among environments.
[00:15:59.670]As you can see the correlation was pretty high,
[00:16:01.901]which means that if one genotype is getting a low score,
[00:16:05.880]it's getting a low score everywhere. It's getting a high
[00:16:07.800]score, it's getting a high score everywhere.
[00:16:09.930]The other thing that we've done is the Cronbach's alpha.
[00:16:12.990]This is highly used on health science and medical field
[00:16:17.016]where they try to see the consistency
[00:16:18.930]of a specific response.
[00:16:20.700]For us here, the response was the dicamba damage.
[00:16:24.090]And this goes from zero to one.
[00:16:27.360]Obviously one is the highest consistency,
[00:16:30.000]zero no consistency.
[00:16:31.590]There's some quantitative geneticists that they're even
[00:16:33.570]considering this as a alternative credibility.
[00:16:37.320]Familiar with Rex Bernardo, he,
[00:16:38.870]he talks about this a little bit and for us the scores have
[00:16:42.960]very high consistency.
[00:16:45.480]So they're not only correlated among environments that we
[00:16:48.150]showing high consistency theme their own observations.
[00:16:51.390]So again, the two main limitations that we were faced,
[00:16:56.160]one was, is the damage homogeneous,
[00:16:58.707]and the second one is the scoring reliable.
[00:17:02.100]We're able to prove that using statistics.
[00:17:07.260]So when it comes to the damage, right?
[00:17:09.480]I've been been talking too much, but not getting to the
[00:17:11.460]point here, in a one to four scale we found that
[00:17:14.970]for every increment in that scale, you would see
[00:17:17.640]about 9% of yield loss. And again, I,
[00:17:21.000]I wanna reinforce that, but the size of this data setting,
[00:17:25.710]so again, we're not talking about a handful of varieties
[00:17:28.710]or a handful of observations.
[00:17:30.660]We're talking about 500 plus varieties in an almost 10,000
[00:17:35.220]data points data setting.
[00:17:38.290]As you would expect,
[00:17:39.840]the maturity was significant and that makes sense.
[00:17:43.440]The longer maturity
[00:17:44.550]tends to observe lower damage from dicamba,
[00:17:48.213]You have a longer window to detoxify naturally from there.
[00:17:52.980]But I think what is really nice about this and really gave
[00:17:56.160]us a lot of confidence to, to pursue the remaining
[00:17:59.460]of this research was that although maturity was significant,
[00:18:04.260]it was not, it didn't matter on our tolerant group.
[00:18:07.950]So let me explain this cause this is pretty important.
[00:18:10.883]On this graphic here, on the, on the X axis,
[00:18:16.200]we have our damage score. Like I said, it goes from 1 to 4
[00:18:19.732]and then each one of the colors of this bar represents
[00:18:22.860]one maturity group going from the earlier to the latest.
[00:18:27.030]And here is the yield related to the checks.
[00:18:30.420]When you look at the group of the tolerated genotypes,
[00:18:33.870]they all have the same yield potential around the 90%.
[00:18:37.530]So even though they may have a different maturity group
[00:18:40.050]that doesn't affect your yield potential.
[00:18:42.600]When you start going down on the scale of damage,
[00:18:45.960]look at the most susceptible ones than the maturity really
[00:18:49.380]plays in effect. So for us, I'm saying this is really nice
[00:18:52.740]because it shows that on the tolerant ones,
[00:18:55.860]whatever is given the tolerance has something to do with the
[00:18:59.250]physiology of the plant other than just the maturity.
[00:19:03.330]Whereas when you look at the susceptible ones, the
[00:19:05.880]difference in yield was really because of their maturity.
[00:19:10.560]Okay, so that finishes the first part of this,
[00:19:13.500]When we were looking at yield impact, we're seeing that it
[00:19:17.520]does cause a decent amount of yield loss,
[00:19:21.270]and the main goal when we, when we decided to do this
[00:19:25.230]was to provide an alternative so you could coexist
[00:19:31.020]in an environment with a lot of dicamba.
[00:19:34.530]This is not to be sprayed, that came on top of it.
[00:19:37.680]But if you're growing something else, then an extend
[00:19:40.470]technology you can survive in those environments.
[00:19:43.830]So the second part, like I mentioned earlier,
[00:19:46.170]we were trying to use a drone, it's an off shelf RGB drone,
[00:19:49.860]nothing to fancy about it, to differentiate the,
[00:19:52.920]the label of damage.
[00:19:55.170]Like I said, when you're walking the plots
[00:19:56.550]it's pretty clear you can see the difference between them.
[00:19:58.860]So if your eyes can tell that,
[00:20:00.630]most likely the drone can capture that as well.
[00:20:03.360]So we use a little smaller set of genotypes,
[00:20:06.480]about 230 soybeans.
[00:20:09.660]We have them in one environment, scoring pretty much the
[00:20:12.900]same as as the previous study.
[00:20:14.460]And then we use two different machine learning models here.
[00:20:17.130]One is Artificial Neural Network
[00:20:18.690]and the other one is a Random Forest.
[00:20:20.490]And our goal was to really see can we differentiate
[00:20:23.880]the damage just using the drone.
[00:20:27.060]To do this,
[00:20:28.170]we extracted image features from each one of the plus.
[00:20:31.260]So again, we fly the drone,
[00:20:33.630]you get your few plots, and then we finish plotting,
[00:20:36.330]we take features from that.
[00:20:39.450]The way we decided the features is basically
[00:20:41.610]reproducing what our eyes are seeing.
[00:20:44.220]So when you look at a tolerant plot,
[00:20:46.620]you know, the first thing that probably came to your mind is
[00:20:49.710]the canopy coverage is very different,
[00:20:51.417]and it's expected when you have the damage from dicamba,
[00:20:54.780]your canopy is much reduced.
[00:20:56.400]So it's the need that we use canopy coverage,
[00:20:59.490]that's one of the features. When you look at the color,
[00:21:01.920]and I hope it's making the justice there,
[00:21:04.410]you can see a pretty difference in the
[00:21:07.050]intensity of greenness. So again,
[00:21:09.938]you think about the damage from dicamba,
[00:21:12.600]it does end up changing the intensity of your greens color.
[00:21:17.520]And then the other thing, if you ever seen this in the field
[00:21:20.280]the texture from your plots are just different.
[00:21:23.880]You have the roughness from the dicamba damage (indistinct)
[00:21:27.030]if you have a tolerant, it's a smooth soybean leaf.
[00:21:30.960]For that we use entropy,
[00:21:32.610]which basically it captures like how complex,
[00:21:36.030]how much information you're getting from an image and we use
[00:21:39.240]that as a a texture feature and then we throw it away,
[00:21:45.030]throw in there two vegetation indexes to represent plant
[00:21:48.660]health conditions. And then like I said,
[00:21:50.760]this was across multiple environments.
[00:21:52.620]So we standardize all those features using (indistinct)
[00:21:56.426]simple Z scores. So you know,
[00:21:57.750]you're all under the same scale.
[00:22:01.103]When you look individually of those features,
[00:22:04.740]you can really,
[00:22:05.910]it's pretty nice to see how much that correlates to what
[00:22:09.720]we're seeing in the field. So for example,
[00:22:12.030]I'm gonna talk about a few, when you look at canopy coverage
[00:22:15.990]it's a very clear difference between
[00:22:18.270]the average of the tolerant, average of the moderator,
[00:22:21.540]and the average of the susceptible.
[00:22:22.890]So it does differentiate those classes pretty well.
[00:22:26.670]And it's nice because again it really replicates what we're
[00:22:30.570]observing in the field and a simple drone,
[00:22:33.300]off shelf drone, can do that as well.
[00:22:36.960]So when it comes to the performance of the model,
[00:22:39.503]the first one I used was the (indistinct) dashboard.
[00:22:42.870]It got a decent accuracy,
[00:22:44.280]I think especially giving how those scores were done,
[00:22:47.760]visuals scores, it does a pretty good job when you want to
[00:22:50.820]differentiate what is tolerant from what is susceptible.
[00:22:54.176]What we call on this paper is just extreme
[00:22:56.850]misclassification. So if you call something tolerant,
[00:22:59.610]but it's supposed to be susceptible or you call susceptible,
[00:23:01.680]it's supposed to be tolerant. It happened very few times,
[00:23:05.317]you know, you look at here and you looking here
[00:23:07.560]across all the samples, it's susceptible.
[00:23:09.630]It's not horrible when you consider how simple
[00:23:12.480]this entire pipeline is.
[00:23:14.722]When we use the Random Forest though, it seems to improve
[00:23:18.450]that misclassification quite a little bit.
[00:23:20.250]So you don't see any extremes on the prediction of tolerant
[00:23:24.720]as susceptible or vice versa.
[00:23:26.400]It just produces dramatically there. So the overall accuracy
[00:23:29.430]not too different, but it tends to identify
[00:23:32.190]those extremes a little bit better.
[00:23:34.500]One thing that is worth measuring, again,
[00:23:36.660]this is based on machine learning,
[00:23:39.060]so you gotta go through extracting the features.
[00:23:42.009]There's quite a little bit of work on that.
[00:23:44.550]One of the things that we're doing now,
[00:23:46.410]and hopefully the paper's coming out soon,
[00:23:48.660]we're using deep learning. So we're feeding to the model
[00:23:51.890]of the entire image of the plot.
[00:23:55.113]You don't have to go through extracting the features
[00:23:57.300]and that boosts the accuracy quite, quite substantially.
[00:24:00.660]So again, it should be coming out in a few weeks,
[00:24:03.838]but your overall accur, oops,
[00:24:06.480]your overall accuracy goes from, you know,
[00:24:08.797].75 to a .83, .84. So it's a much easier process,
[00:24:13.530]a little more computation intensive, but it,
[00:24:15.600]it's an easier process.
[00:24:17.220]So the main conclusion is that you can,
[00:24:19.860]again with an off shelf drone, it's an easy,
[00:24:22.980]easy process and it can really help you if you're interested
[00:24:25.830]in, in identifying higher tolerant varieties.
[00:24:28.890]Also can work for different crops
[00:24:30.270]so for the purpose of dicamba, that that seems to,
[00:24:32.910]to do a good job.
[00:24:34.770]And then the last part, which for me is, should be the,
[00:24:37.587]the most exciting, is when all of this come together.
[00:24:43.050]I think after we're done, you know,
[00:24:44.610]three years of researching, if we try to map the,
[00:24:48.270]the significant regions and,
[00:24:49.380]and we didn't find anything would be very disappointing
[00:24:52.230]because everything to this point has been indicating that
[00:24:55.770]this could be a, a genetic trait.
[00:24:58.440]So we just published that at the end of 2022.
[00:25:01.186]Here we use those diverse successions that I've shown
[00:25:04.472]on the previous slide,
[00:25:06.660]same way of doing the scoring,
[00:25:09.240]visual scores, multiple replications.
[00:25:11.192]And now we use two different models here.
[00:25:13.170]One is just, you know, the,
[00:25:14.580]the basic BLINK model that if you're working with with
[00:25:18.609]GWAS you probably see that.
[00:25:20.340]The other one we tried to account for GxE.
[00:25:24.180]So we were using the population structure in interaction
[00:25:27.720]with the environment. So the, the idea was
[00:25:30.240]if there is an instructor in that population of DIs,
[00:25:33.480]diverse lines, and the environment's different
[00:25:36.300]because of dicamba exposure.
[00:25:38.070]We're accounting for that difference using this model.
[00:25:42.060]So what we found is there's, there's two very clear peaks,
[00:25:46.710]one in chromosome 10 and one in chromosome 19
[00:25:49.620]that was detected with both, both models and they seem to
[00:25:54.270]have a pretty, pretty big impact on the dicamba response.
[00:25:58.980]But to me what was the most interesting is that
[00:26:02.490]the genes underlying those genomic regions,
[00:26:05.760]they appear to be related to the plant,
[00:26:08.310]the process of plant detoxification from the herbicide.
[00:26:11.640]So the, the weed scientists that help us on this work,
[00:26:14.670]when we shared the results,
[00:26:16.470]they were quite excited and I didn't know why at the time,
[00:26:20.460]but they said the two genes that we identified,
[00:26:22.680]they're directly related in the process of detoxified from
[00:26:25.740]herbicide. So for instance, the one in chromosome 19,
[00:26:29.882]the glycosyltransferase gene,
[00:26:31.830]it's directly associated with phase two
[00:26:34.334]of herbicide detoxification phase.
[00:26:37.170]And it's more interesting is that
[00:26:38.610]that specific genomic region
[00:26:40.590]is breaching of this type of genes.
[00:26:42.870]So it seems to really have a big impact there.
[00:26:45.660]And the one in chromosome 10 is right within a multidrug
[00:26:50.040]resistant protein,
[00:26:51.570]which you know, if you're familiar with the process,
[00:26:53.820]the phase three of detoxification,
[00:26:56.010]it moves the conjugated herbicide into the vacuole.
[00:26:58.800]So as the last step for the plants are getting rid of the
[00:27:01.697]herbicide side, so again, I don't think it was random
[00:27:06.000]that we identified those regions,
[00:27:07.530]it really seems to be having a real effect on the phenotype.
[00:27:13.440]So I wanted to add that here. You know,
[00:27:15.107]if you think about the, again, the three phases
[00:27:17.730]of detoxification, the conjugation of the herbicide
[00:27:21.690]is really catalyzed by UDP, you know, glycosyltransferases,
[00:27:25.920]which is the one in chromosome 19.
[00:27:28.470]And then when you're making a transportation of this
[00:27:30.900]conjugated herbicide into the vacuole,
[00:27:33.570]it's done by multi-drug resistant proteins,
[00:27:35.880]which is the one we found in chromosome 10.
[00:27:39.720]When you're looking a little deeper into the, you know, the,
[00:27:43.107]the phenotype distribution, so when you have those genes,
[00:27:47.370]how does that change your phenotype?
[00:27:49.440]So what we've done is we created classes based on the
[00:27:53.571]allelic version for those markers.
[00:27:55.920]So if you have the favorable, we call it one,
[00:27:59.192]you have the unfavorable, we call it zero.
[00:28:02.370]So this one here is based on just a classification
[00:28:06.840]tolerant, moderate or susceptible. When you look at the
[00:28:10.079]triple zeros, which would be all the no values,
[00:28:13.860]your higher distribution of genotypes is within the
[00:28:18.150]susceptible and the moderate.
[00:28:20.400]You have very little misclassified as tolerant.
[00:28:23.640]And I'm gonna be clear here,
[00:28:24.690]I think this is a highly quantitative trait.
[00:28:27.870]I don't think it's regulated by two or three genes,
[00:28:31.140]but as you're trying to do any sort of marker as a
[00:28:33.690]selection, you can probably identify the most tolerant
[00:28:37.140]or most susceptible with a handful of genes.
[00:28:40.200]And it's quite interesting where as you go down in the scale
[00:28:43.170]here, which means you start adding the favorable alleles,
[00:28:47.580]your, your concentration of susceptible really goes down
[00:28:51.690]when you start having at least, you know,
[00:28:53.880]one or two of the favorable alleles.
[00:28:55.770]And when you have all of them,
[00:28:57.480]you don't have any susceptible net classification.
[00:29:00.180]So for us it's really nice to see this,
[00:29:02.367]but we want to see on the (indistinct) damaged score,
[00:29:05.460]so not thinking about classification,
[00:29:07.560]but looking at the actual distribution of damage.
[00:29:10.920]And it's kinda the same trend.
[00:29:13.050]When you have all the unfavorable, your,
[00:29:16.410]your average is about, you know, three in our scale,
[00:29:20.130]when you start incorporating the favorable alleles
[00:29:22.860]that goes down, and when you have all the favorable,
[00:29:26.730]your score is above one and a half. So again, it's,
[00:29:30.360]it's on a trait, I think it's a pretty complex process
[00:29:33.386]of detoxify herbicide, but if you have those alleles
[00:29:38.040]in your population, you may be able to have
[00:29:40.440]higher levels of tolerance.
[00:29:43.410]So, so what do you get from from this, right?
[00:29:49.470]I think again, looking at the data that we have,
[00:29:53.366]it's, so far as we know,
[00:29:56.160]it's the largest study when it comes to genetic diversity
[00:29:59.100]and dicamba. The damage is not only cosmetic.
[00:30:03.007]I know there's been a lot of, you know,
[00:30:04.802]confusing reports at some point even there, you know,
[00:30:08.873]dicamba could boost your yield a little bit, right?
[00:30:12.180]It's hard to believe when you think about all the,
[00:30:15.750]all that it's causing to the plant.
[00:30:18.000]Back in the vegetative stage, you're affecting your
[00:30:20.640]vegetative growth, you're affecting your canopy coverage.
[00:30:23.430]Those things are directly related with the ability of
[00:30:25.961]capturing light and energy in the plant.
[00:30:29.040]So it's not surprising that you're seeing yield loss.
[00:30:32.520]I think we've provided enough evidence across a wide range
[00:30:37.560]of genotypes that you're gonna experience some degree
[00:30:41.460]of yield loss if you're being exposed continuously
[00:30:44.820]to dicamba over the years.
[00:30:47.621]One thing that's to me is, is very nice,
[00:30:49.260]especially if you're working with genetics, is that the
[00:30:51.120]genetic background, on top of everything else that weed
[00:30:53.910]science has found, it does affect your response.
[00:30:57.090]It's very nice for us because you know,
[00:30:58.740]the basics of breeding is you have genetic diversity or
[00:31:01.680]variability, you can improve the trait.
[00:31:04.110]That's what we are showing you.
[00:31:06.030]The third one is that the maturity is not-
[00:31:09.540]third one, this one, is that maturity is not
[00:31:11.367]the only thing affecting the response.
[00:31:14.250]So like we showed before,
[00:31:15.780]we think the same maturity group on the tolerant ones
[00:31:18.974]they yield the same. So there is more to the resistant.
[00:31:22.020]There is really physiological mechanism that are affecting
[00:31:24.930]the level of tolerance.
[00:31:27.570]You can use UAVs to to, to do this type of work.
[00:31:31.530]And if you're working with a different trade,
[00:31:33.570]I imagine that if it's something that you can see and easily
[00:31:36.360]differentiate, an RGB camera should, should do the work.
[00:31:40.620]I mean, if your your phenotype looks good,
[00:31:42.690]it shouldn't be that hard,
[00:31:44.550]You can see it, then you have other,
[00:31:46.290]other options, where this already seems to do a good job.
[00:31:50.067]And then lastly and probably the main takeaway message
[00:31:53.730]and that should apply to all traits and all crops
[00:31:56.700]is that if you have genetic resources that,
[00:31:59.823]well first you a better responsible for a specific trait,
[00:32:03.420]most likely you can improve that trait in a large scale.
[00:32:07.500]For us here, developing like a tolerant, nonresistant,
[00:32:12.540]tolerant soybeans, it really opens a lot of opportunities
[00:32:16.800]if you want to grow something else, you know,
[00:32:19.050]there's a growing market for conventional soybean, right?
[00:32:22.710]If you want develop a high oleic soybean or if you wanna
[00:32:25.504]develop for a specific, specific niche market,
[00:32:29.280]you may be able to get by and coexist.
[00:32:31.740]If you have this trade, you don't have to be, you know,
[00:32:35.280]hostage for a specific tolerance because you don't have
[00:32:38.370]options. That's our main goal with this,
[00:32:40.200]really deploy alternatives to farmers so they can
[00:32:44.520]really decide what they wanna do.
[00:32:46.296]They don't have to be forced to make any, any decision.
[00:32:49.996]I think that's all I have. I, I forgot to at my,
[00:32:53.910]my last slide, which was really the,
[00:32:55.950]the acknowledgement section.
[00:32:58.050]This is the work that was in between Missouri and the
[00:33:01.320]University of Illinois primarily. So we,
[00:33:03.480]we collaborated with Brian Dews in this work.
[00:33:06.404]So Jim Reachers and Brian Dews were the,
[00:33:08.997]the group from Illinois, the work with us,
[00:33:11.640]Missouri was of course our group at (indistinct),
[00:33:15.780]and then Doug (indistinct) at the University of Florida.
[00:33:18.720]He also contributed with some of the, some of the analysis.
[00:33:22.230]But again, that's my last one.
[00:33:25.758]Very nice to be here. Thanks.
[00:33:27.900]Thanks a lot for, for, for listening and I'm happy to,
[00:33:31.969]to take any questions. Thank you.
[00:33:34.980]You have time for questions and we can pass the
[00:33:37.050]mic around so those online can hear.
[00:33:59.577](indistinct)
[00:34:03.268]Thanks. Nice. Thank you.
[00:34:05.430]So what are you doing? Are you making use of the,
[00:34:12.859]the (indiscernible)?
[00:34:22.110]Yeah, so we, so my first cycle for us will be this summer.
[00:34:27.450]We're, we're licensing the material from Missouri
[00:34:31.230]to start putting that in our genetics because
[00:34:33.960]it may sounds really complicated to bring anything
[00:34:38.310]that is, you know, that came a resistance.
[00:34:40.350]We just got too much damage from, from multiple directions.
[00:34:44.757]The one thing that we're doing now,
[00:34:46.350]and I again should be coming out soon is
[00:34:48.750]we're confirming the, the markets that we found,
[00:34:51.810]but in the different population. So we found this in the,
[00:34:54.600]the PI span. We're now using the first study, you know,
[00:34:58.680]500 plots breeding lines to map the same thing and,
[00:35:02.877]and the warning chromosome 19 and 10.
[00:35:06.240]They keep appearing in a different population.
[00:35:08.777]So I think once we start making the process this summer,
[00:35:12.867]we're gonna be using that information
[00:35:14.520]for markets (indistinct).
[00:35:23.460]Thank you for this wonderful presentation.
[00:35:26.250]I think I have two questions. The first one is,
[00:35:29.340]I might have missed it in the beginning,
[00:35:30.960]but I didn't hear which those did you use,
[00:35:33.520]which those are the camera you used or exposed.

The screen size you are trying to search captions on is too small!

You can always jump over to MediaHub and check it out there.

Comments

0 Comments

Soybean Tolerance to Off-Target Dicamba

Description

Searchable Transcript

Comments icon comment

Related Channels

Comments