The Use of Gene Expression to Predict Gene Function in Maize and Sorghum

Vladimir Torres-Rodriguez, Research Assistant Professor, Department of Agronomy and Horticulture Author

01/03/2025 Added

5 Plays

Description

Vladimir joined the Schnable lab in October 2021 as a postdoctoral scholar, since then he has been working with quantitative genetic techniques to predict genes associated with a phenotype. He became a Research Assistant Professor in July of this year.

Searchable Transcript

Search:

[00:00:00.720]The following presentation is part of the Agronomy and Horticulture Seminar Series
[00:00:05.760]at the University of Nebraska-Lincoln.
[00:00:07.840]Good morning everyone. Thank you for attending to this seminar. As a reminder, we will have a
[00:00:16.960]presentation of about 45-50 minutes and after the presentation we have 10-15 minutes for questions.
[00:00:22.960]Those who are online, please use the chat to write the questions. Those who are in person will have
[00:00:28.800]opportunity to raise your hand and ask the questions directly to our speaker.
[00:00:33.280]So our speaker today is Vladimir Torres-Rodriguez. He's a research assistant professor
[00:00:39.360]in the Department of Agronomy and Horticulture here at UNL.
[00:00:43.600]He joined the Schnebel Lab in October 2021 where he has been using quantitative genetic techniques
[00:00:51.120]to predict genes associated with phenotypes. Today Vladimir will present on infusion gene function
[00:00:58.640]from expression levels in maize. Thank you guys for coming and thank you for all the folks who
[00:01:09.440]joined online. Thank you for the introduction, Walter. We are going to talk in this talk about
[00:01:19.360]how can we infer gene function and which is important to do it. To start this talk I would like to
[00:01:28.480]ask you a question. Let's imagine that we have an opportunity to get a million dollars. But then we
[00:01:35.760]just need to answer one question and it should be correctly answered. We need to choose which
[00:01:42.160]is the winning number between one and a million. And let's take a few seconds thinking a number.
[00:01:48.160]And this is the winning number. So how many of you get this number right? It's really
[00:01:58.320]difficult, right? So what if we change that million dollar for a trait of interest? And
[00:02:04.480]we can lower the number of chances. Now we need to choose between one and around 40,000, which is
[00:02:12.400]the number of genes that the MACE genome has. It's a bit easier. Now the range is smaller. But
[00:02:19.280]if we choose the wrong number, it can take two years of our lives and can take a bunch of money.
[00:02:28.160]Then it's not that easy, right? Okay, so if just we had a tool to narrow more this window,
[00:02:37.840]it would be great. Well, we actually have it. We have association studies,
[00:02:42.720]which are a powerful tool to link traits with genes. And probably the most popular association
[00:02:50.480]study is the genome-wide association study, or GWAS, in which we are linking genetic variants
[00:02:58.000]with a phenotype. And let me take a few minutes to briefly explain what GWAS is about.
[00:03:04.560]It will start with a diversity panel. There are multiple diversity panels published already
[00:03:12.320]there, so we can either choose one of those or we need to put a diversity panel by ourselves.
[00:03:18.480]One characteristic of these diversity panels is that they are sometimes coming
[00:03:24.080]from all over the world. And this is important because the more lines
[00:03:27.840]that we have, the more diversity that we have, and then we can achieve better results.
[00:03:32.560]Another important characteristic is that we need to have genetic variance information from those.
[00:03:40.320]Okay, once that we select our favorite diversity panel, we need to grow them to collect the trait
[00:03:50.560]in which we have interest. And this can be any trait. Once that we grow the plant,
[00:03:57.680]we collected the measurements of that trait, and we use this both type of information.
[00:04:03.920]The measurements of the trait and the genetic variance information, we combine them,
[00:04:08.320]and we ask to the program, to an algorithm that already exists,
[00:04:14.080]okay, which of these genetic variants are associated with the phenotype that we have.
[00:04:19.360]And because we can have millions and millions of these variants in the genome, it's difficult
[00:04:27.520]to see all of them at once. But one way that we have is with Manhattan plots. In a Manhattan plot,
[00:04:35.200]we have multiple dots. Each dot represents a genetic variant, for which we have the p-value,
[00:04:42.240]which is the probability that this genetic variant is associated with a phenotype. And also we have
[00:04:48.640]in the y-axis the position in the chromosomes. And in this example that I am showing you, we have two
[00:04:57.360]peaks that are significantly associated. They are significantly associated after the correction that
[00:05:02.880]we are doing, and that in this case is Bonferroni correction. Okay, great. Now we narrow our window
[00:05:09.040]to two peaks, but we need to know which is the gene that is causing or is associated with these
[00:05:15.360]changes in the phenotype. And that's not an easy question. Sometimes these windows are representing
[00:05:27.200]regions in the genome, and maybe there is not a single gene there, or maybe there is multiple
[00:05:32.960]genes there. So then we need to do extra analysis to retrieve more information to choose the right one.
[00:05:41.280]Okay, and there are multiple factors that can affect the phenotype, and I want to use one
[00:05:50.560]example to exemplify this. I want to use the color in the kernel in maize.
[00:05:57.040]So we know that genetic variance or DNA sequence is associated with changes in kernel color,
[00:06:08.560]and this we can see in this zoom in of a Manhattan plot that is multiple genetic
[00:06:15.840]variance associated in this region that corresponds to Y1. Y1 is yellow endosperm,
[00:06:21.920]which is calling for phytoin synthase, but then
[00:06:26.880]at this point we can see that we can use DNA sequence that are affecting the phenotype levels,
[00:06:34.640]but also the gene expression can be linked to changes in the phenotype. In this case,
[00:06:41.280]higher expression of Y1 is associated with yellow endosperm color in maize kernels.
[00:06:50.000]So we can use gene expression, which are the advantages
[00:06:56.720]of using gene expression to find association with the phenotype of interest. One is that
[00:07:04.560]a single gene expression, let me use this one,
[00:07:08.160]to write, okay with this one, a single gene expression can capture the signals coming from
[00:07:18.720]other genes. So even though that those other genes, the signals are quite small, they are all converging
[00:07:26.560]into a single one. So maybe maximizing this and then we can get, maybe we can get association
[00:07:33.040]with this single gene expression. Other advantage is that we have a constant communication between
[00:07:39.200]the environment with that gene expression. And also we keep a constant communication
[00:07:44.640]between the phenotype with the gene expression. Other advantage is that we can collect
[00:07:50.800]information coming from different haplotypes. In this sense, multiple haplotypes can be the
[00:07:56.400]fine, but multiple variants, not just biallelic. That is the one that we are capturing in GWAS.
[00:08:03.360]Okay. The use of gene expression in this association studies, it's called
[00:08:10.400]transcriptome-wide association studies or TWAAS. But
[00:08:15.760]not everything is good. There are also some limitations for use gene expression to find a
[00:08:26.240]gene expression. One, it's also very sensitive to changes in the environment. And here, it depends
[00:08:33.120]on how we are using this information, because we mentioned there is also an advantage. But when it
[00:08:38.560]becomes a limitation, it's what we don't control properly for this environment factor that are
[00:08:46.720]adding noise or could add noise to the gene expression.
[00:08:49.760]And then as an example of this, we can look at the genes that are having the urinal pattern.
[00:08:56.080]Around 30% of the main transcripts are showing that are affected by changes in the length of the
[00:09:04.560]day. And here in this example, we have gigantea, and we can see that as the day starts, the gene
[00:09:11.680]expression is higher and higher, and at some point, it decreases.
[00:09:16.960]So the more time it takes us to collect this data, we are going to capture more of this information.
[00:09:25.920]That is not necessarily reflecting the data that we want.
[00:09:31.360]The other limitation is that to create these is expensive.
[00:09:36.800]They are really, really expensive.
[00:09:38.080]Okay, having these limitations and disadvantages in mind, we in the Schnebel Lab take the challenge
[00:09:48.000]to collect one of the biggest gene expression data sets in maize.
[00:09:52.880]We collect 750
[00:09:55.760]unique samples and we'll do with it in less than two hours.
[00:10:00.560]This was happening in 2020.
[00:10:02.560]And here in the photo, we have an aerial view of the field in that year.
[00:10:07.680]And we have on your right side, all the people that helped to collect this data set.
[00:10:16.160]Everything was organized by John Tarkus, which I hope is online.
[00:10:20.640]And to explain this briefly, we collect
[00:10:25.600]a single maize plant, we identify the top four leaves, we collect five bunches,
[00:10:32.720]and we quickly froze those in liquid nitrogen. Then we extracted RNA and we did the sequencing.
[00:10:40.080]And I explained this with words, but I think it's going to be clearer if you see it in a video.
[00:10:45.440]This video was put together by Harshita Mangat.
[00:10:55.440]Thank you.
[00:10:57.440]Thank you.
[00:10:59.440]Thank you.
[00:11:29.280]Thank you.
[00:11:32.960]Okay, so that's how our team were collecting these data sets.
[00:11:39.120]And this was really smart because as you can see, we were collecting a triplicate of the samples at
[00:11:46.400]that time.
[00:11:47.440]So then we were having some backups in something that something happened during the sequencing
[00:11:53.760]process or other downstream steps.
[00:11:57.120]Okay.
[00:11:59.120]We collect the data.
[00:11:59.920]We did the RNA sequencing.
[00:12:02.960]How the data looks like.
[00:12:04.240]Okay.
[00:12:05.280]Here in your left side, we have a distribution of the gene expression of the samples in the
[00:12:12.640]first two principal components.
[00:12:14.160]And it is important to see that we are not having an obvious bias between these samples.
[00:12:20.720]But if we look at the distribution of these samples for each of the three principal components,
[00:12:28.960]according to the order of collection, we see that PC1 is capturing some bias.
[00:12:34.880]So this order of collection just represents which was the plant that we collected first,
[00:12:42.320]that was assigned number one, which was the plant that we collected second, which was
[00:12:47.040]the plant that we collected the hundredth plant that we collected, and the last plant
[00:12:51.200]that we collected.
[00:12:52.240]And we can see that for PC1, there is something that is explaining this by the order of
[00:12:58.800]collection.
[00:12:59.360]If we dig more into it, we see that there are some classical genes which are described
[00:13:06.320]with a diurnal pattern to have a good correlation with this order of collection, like late
[00:13:14.080]hypothetical elongation protein ortholog, the same gigantea that we were talking before.
[00:13:20.000]Here in the bottom of the plot, we had two housekeeping genes in which we see that they
[00:13:26.320]are not having an effect due to order of collection.
[00:13:28.640]Which is good.
[00:13:29.280]So despite that we have some genes affected by the order of collection, most of them,
[00:13:35.520]the values are really, really low, which is good for us.
[00:13:39.520]Okay, so now that we look at our data, how is this data able to find genes controlling
[00:13:48.800]a trait of interest?
[00:13:49.760]So first, we need to choose a trait of interest.
[00:13:53.360]In this case, we do flowering time.
[00:13:55.680]Why flowering time?
[00:13:57.280]Well, because
[00:13:58.480]it is associated with yield, because it's also
[00:14:01.760]helped with stress avoidance, and it's a really
[00:14:04.400]well-studied trait.
[00:14:06.640]This is important because we want to validate some
[00:14:09.280]of the results, and it's always better to do it
[00:14:11.520]with something that we know.
[00:14:12.720]In maize, flowering time is divided into parts.
[00:14:19.680]One is silking, and the other one is anthesis.
[00:14:22.480]Silking is just when we saw silks in the ears,
[00:14:26.800]and anthesis is when we see
[00:14:28.320]the pollen in the tassels.
[00:14:29.520]Okay, we choose our trait of interest.
[00:14:34.560]We have our gene expression data set.
[00:14:36.720]Let's run the analysis.
[00:14:37.840]After running the analysis,
[00:14:41.040]we have 18 genes related to flowering time.
[00:14:43.920]Three of these 18 genes were previously reported
[00:14:48.880]as a true positive gene.
[00:14:50.320]What is a true positive gene?
[00:14:53.600]It's a gene that has been,
[00:14:55.360]in which the expression has been
[00:14:58.160]modified, either loss of expression or overexpression,
[00:15:02.240]and we see that it is affecting the phenotype,
[00:15:06.080]a given phenotype.
[00:15:07.520]Okay, that's a true positive.
[00:15:08.880]We found three of these.
[00:15:10.240]We also can divide the other ones,
[00:15:13.360]the information of other genes,
[00:15:14.960]in three different categories.
[00:15:16.720]One can be genes that belongs to a family
[00:15:20.240]that has been reported to affect flowering time.
[00:15:24.320]So we have some example of some CCN
[00:15:28.000]genes or some match box transcription factors.
[00:15:32.000]The other category are genes that in which
[00:15:36.560]orthologs in another species are related to flowering time.
[00:15:40.880]Here we can talk about late gene, oh sorry, one here.
[00:15:48.000]And the third category are genes from which we don't know anything.
[00:15:52.960]These are completely novel genes which are also good to explore.
[00:15:57.840]Okay, so in the beginning, we start talking about the use of genetic variants to find
[00:16:03.920]genes associated with a trait of interest.
[00:16:12.800]To measure how good gene expression was respect to use genetic variants,
[00:16:20.720]we also use genetic variants to measure this trait, and we found two genes.
[00:16:27.680]One was MAT69 and the other one CCNA. So we are finding more genes using gene expression,
[00:16:34.640]which is good because if not, I will call this talk using genetic variants to infer the
[00:16:40.400]function of a gene. Okay, but gene expression data is expensive.
[00:16:49.360]Can we reuse them? Or we need to create a gene expression data set every single time?
[00:16:58.400]We were collaborating with people in Michigan and they shared some flowering time data with us. We
[00:17:06.160]ran the analysis and we found that many of the genes that we were finding here in Nebraska were
[00:17:13.440]similar. But we also find some genes that are different. And now we expand our 18 genes to 21.
[00:17:27.360]We can also identify some genes that are environment specific.
[00:17:30.640]Okay, conclusions up to here. We are able to find 10 times more genes associated with flowering time.
[00:17:39.920]It's easy to name genes associated. This is because we are working with gene expression
[00:17:46.960]since the beginning. So this gene expression has an ID of a gene attached to it.
[00:17:52.640]It's fast and that's really cool because
[00:17:57.200]let's imagine that we are training people or even ourselves when we are learning new technologies,
[00:18:02.400]new tools. It's much faster to get the error, which means that it will be fast, it's going to
[00:18:08.960]be faster for us to fix it. I was able to run another difference for this fast is that
[00:18:18.960]the files are easier to handle. So to handle a file with gene expression for 24,000 gene
[00:18:27.040]columns is easier than for 15 million columns. And gene expression datasets can be reused.
[00:18:35.760]Okay, so because of these characteristics, the use of gene expression has been becoming
[00:18:50.640]more and more popular. Up to date, at least 37 transcriptome-ware association studies have been
[00:18:56.880]published in plants. The first one was in 2014, and it was in maize. Now, in the most recent
[00:19:03.680]years, we see that we are having more species, including sorghum, rice, soybean, codon, wheat,
[00:19:11.200]canola, and Arabidopsis. And the dataset that we were talking is here. We have one of the
[00:19:18.080]datasets with a larger number of individuals. In the bottom of the panel, we have the methodology
[00:19:26.720]it employs in all these studies. We identify at least nine different methodologies.
[00:19:34.080]Okay. When I was talking about GWAS, we were talking about the numbers of individuals
[00:19:46.400]and how numbers were important to increase diversity.
[00:19:49.600]That will increase diversity within species, but also maybe we can increase diversity between
[00:19:56.560]species, including more species. And gene expression can function as a bridge to do it.
[00:20:03.040]So in this example, we have the gene from species A1, the gene A from species 1, we have the gene A
[00:20:11.760]from species 2, and we have numbers. These are -- gene expression is just numbers. So can we
[00:20:17.200]combine them to increase the diversity and thus increase the statistical power? Okay. MACE
[00:20:26.400]is closely related to sorghum. And if you look at the photo,
[00:20:30.960]it's not that easy to say which is MACE and which is sorghum.
[00:20:34.800]MACE, around 23,000 of MACE genes have an ortholog in sorghum, which means that we can
[00:20:46.800]use up to 23,000 genes, combine them with sorghum, and see if we can achieve better results. Okay.
[00:20:56.240]Having this question in mind, we decided to grow and to collect
[00:21:00.880]data to produce two new gene expression datasets on 2021. The first one was for MACE, with around
[00:21:11.440]750 individuals, and the second one was for sorghum, about the same number of individuals.
[00:21:18.080]If you want to know more about Tiwas in sorghum, Harshita Mangal is the expert. She's over there.
[00:21:26.080]So be sure to ask some questions. But more importantly, these provide to the University
[00:21:32.000]of Nebraska with three of the biggest gene expression data sets from MACE and sorghum.
[00:21:36.480]How these data sets look like? Well, once that we collect the gene expression, it's important
[00:21:43.840]to see that there is no bias there. Even if there are some bias, probably we still can work with
[00:21:50.080]them, but we need to process the data in different ways. Okay, but here we don't have bias.
[00:21:55.920]We were talking about methodologies, that there are a lot of methodologies there.
[00:22:03.600]Many of those methodologies are just simply adapted from GWAS, but the data that we are
[00:22:11.200]providing in GWAS is different from the data that we are using here. In GWAS, it's more of a discrete
[00:22:17.680]in which we have 0, 1, or 2 depending on the number of alleles, and here we have more a continuous
[00:22:26.720]variables. So, to answer new questions, we're required to develop new methodologies,
[00:22:31.760]new algorithms. So, to have this in mind, we set some collaboration with some friends from
[00:22:39.280]Peking University. We designed these new algorithms and we tried in our MACE gene
[00:22:45.600]expression dataset from 2021. Here, we were finding 26 genes with this new approach. From those 26
[00:22:55.600]genes, at least one was reported as a true positive, and five were part of the previous
[00:23:03.920]TWAS that we did with 2020 data. That gave us 21 new genes. And again, many of those genes belong
[00:23:16.320]to CCN family, which is a matchbox transcription factor, which are genes reported with an effect in
[00:23:24.640]flowering time.
[00:23:25.440]And I want to highlight two of them, which is CCN12, which possess a florigen activity.
[00:23:32.320]It's working together with CCN8. And the other one is CM24, which arise, orthologs, mutants,
[00:23:42.080]alter influences. So these two genes are part of the 21 new genes. Okay. So what happened if we
[00:23:55.280]combine this data from 2021 and maize with the sorghum one? Now we are increasing the number of
[00:24:04.080]genes that we observe. From 26, now we are having 43. And we can divide these 43 genes in four
[00:24:13.680]different categories that I will explain briefly. So if you look at panel B, these are the four
[00:24:25.120]genes. We have maize, sorghum, maize, sorghum, maize, sorghum, maize plus sorghum, and combined.
[00:24:31.520]Maize refers to those genes that they are showing significantly association, but this association
[00:24:40.240]is coming mostly from maize. That means if we run this analysis independently with maize,
[00:24:47.200]we are also finding these genes. Sorghum is the same information but with sorghum. And maize plus
[00:24:54.960]sorghum means that these genes are coming from, if we run the analysis independently, this same
[00:25:01.680]gene is coming from both of them. Okay, the category in which we are more interested is the
[00:25:09.600]combined one because those are genes that if we run the analysis independently, they are not showing
[00:25:16.640]association. But when we combine the information, we are having them with association. There are 15
[00:25:24.800]genes in this category, and in this Manhattan plot, they are represented with a star.
[00:25:30.480]Between these 15 genes, we are having two SPL genes, SPL13 and SPL29, which mutants
[00:25:44.000]are affecting flowering time. Okay, we are having more genes, new ones, but how reliable they are?
[00:25:55.280]It's not just about finding more genes, but also we want to be sure that we are finding
[00:26:00.320]the right ones. And we have, we could have different, different methodologies to verify this.
[00:26:09.920]One of the first one is just to look for genes that went under selection during domestication.
[00:26:16.160]Why domestication? Well, maize was domesticated from Teosinte parviglumis in Mexico. After that, it moved to
[00:26:24.480]higher latitudes in the U.S. and higher latitudes of Argentina. During domestication, maize suffer,
[00:26:32.560]or Teosinte suffer, a lot of change in morphology at plant level, but also in ear morphology,
[00:26:38.720]and here we have some photos. What is cool about this photo for the panel A is that this
[00:26:44.960]ear is one of the crosses that George Beadle did. George Beadle was, he won the Nobel Prize
[00:26:54.320]and he did his bachelor here in Nebraska. Okay, if we compare our 15 genes here that we are finding
[00:27:00.960]in this category, four of them are in this list of genes that were selected during domestication
[00:27:08.080]process. To see if those genes were just by chance, if we get those genes just by chance,
[00:27:15.360]or actually we have an enrichment, we set a control. For our control, we select 15 genes
[00:27:24.160]from our original data set. And from these, we were matching with all these lists from the gene
[00:27:31.760]of selected genes that were selected during domestication. And the results are in panel C
[00:27:38.320]here. We see that many of them are matching at least one gene. One gene reported as selected
[00:27:47.360]during domestication is matching sometimes two. But the probability to have four
[00:27:54.000]or more genes is 0.06, which is relatively low.
[00:27:59.280]But, and all you agree with me, the best way to know if a gene is actually affecting this
[00:28:07.600]is by affecting the function of the gene. For that, we are collaborating with amazing
[00:28:14.400]people from UNL from here, Jinglang Zhang Lab, and our team are working together. So Kyle
[00:28:23.840], Jingliang, Jen, Shen, Chidu, and Zach, thank you all for all the work that you are putting
[00:28:29.680]together. We know that this is not easy. But, and here I'm showing you a photo of the first plants
[00:28:37.600]that move to a growth chamber. So we are excited to validate one of these genes, some of those
[00:28:43.520]genes that we are finding with association. And this process is not easy. Apart from really skilled
[00:28:53.680]people, it takes a lot of time. So first we need to get the embryos, then we need to infect them
[00:29:01.040]with agrobacterium. We need to do some several other steps like the selection media,
[00:29:07.840]do the callus regeneration, shooting and rooting, hardening, and this process can take up to four
[00:29:13.920]months. Then we need to process the plants to get the T0.
[00:29:19.480]and that can take 10 months.
[00:29:21.780]And then we need to do crosses to actually have the seed
[00:29:26.780]to produce, to validate,
[00:29:29.880]to run the experiments to validate the data.
[00:29:32.180]So all these process make us think about different ways
[00:29:38.300]in which we can validate the data.
[00:29:43.000]And I want to explain some concepts before moving forward.
[00:29:48.880]In this talk, we have been talking about GWAS,
[00:29:53.880]which is the use of genetic variants
[00:29:59.160]to find association with a phenotype.
[00:30:01.700]We also have been talking about the use of gene expression
[00:30:06.660]to find association with a phenotype,
[00:30:09.040]which is called TWAS.
[00:30:10.800]But we can also use genetic variants
[00:30:13.440]to find association with a gene expression.
[00:30:16.580]And that process is called eQT.
[00:30:18.880]So eQTL information can be used to validate gene expression.
[00:30:23.880]And I will explain this project a little bit.
[00:30:31.820]So this project is in charge of Sofia Arora,
[00:30:34.620]which is also in the audience.
[00:30:36.300]And as I mentioned, the first step is to find
[00:30:39.780]which are the genetic variants
[00:30:41.560]that are regulating the expression of a gene.
[00:30:44.460]Once that we find those variants,
[00:30:47.980]we are splitting our groups in two,
[00:30:52.420]depending on the genotype that they have.
[00:30:54.960]In this example, they can have the CC genotype
[00:30:58.280]or the GG genotype.
[00:31:00.240]And we can see that the GG genotype has less expression
[00:31:03.300]than the CC genotype.
[00:31:05.300]So if you think about CRISPR-AIDS,
[00:31:08.760]with CRISPR-AIDS, we are doing kind of a similar thing,
[00:31:12.080]but just for the same background.
[00:31:14.760]We have our wild type expression,
[00:31:17.560]we want to have our edited expression,
[00:31:19.880]which is typically less low.
[00:31:23.180]And here we are seeing this,
[00:31:25.060]we are trying to see in this similar pattern.
[00:31:27.420]Then we are going to select some lines.
[00:31:31.620]We are going to grow them in a greenhouse.
[00:31:36.300]And we are going to see if we are able to replicate
[00:31:39.300]those results from mutants.
[00:31:42.240]Then for these, we are going to use some mutants
[00:31:45.180]as a control.
[00:31:46.840]And if you want to know more about this,
[00:31:47.140]Sophia has a poster in the PSI retreat.
[00:31:51.840]It's poster number four, so be sure to stop by
[00:31:55.220]and ask some questions.
[00:31:56.640]Okay, many genes are being regulated by eKTLs.
[00:32:03.680]And I want to use this information in two ways.
[00:32:08.260]The first one is that many of the relevant genes
[00:32:11.380]seems to have an eKTL, which is good for the project
[00:32:14.160]that I just talked, but also can help us
[00:32:16.720]to find more regions controlling trait of interest.
[00:32:21.060]And I want to use as an example, MATS1 and SAX6.
[00:32:25.440]So if we run the eKTL analysis,
[00:32:30.240]we are finding that is a cis eKTL for SAX6.
[00:32:35.040]So a cis eKTL is a eKTL that is close to the gene,
[00:32:40.040]but we are also having two trans eKTL.
[00:32:45.340]Trans eKTL
[00:32:46.300]are eKTLs that are far from the gene.
[00:32:49.640]So these are offered to new genes
[00:32:53.800]that we can look at that are controlling
[00:32:56.780]our threat of interest.
[00:32:58.040]The other example is for MATZ1.
[00:33:01.140]Again, we have a cis eKTL
[00:33:04.180]controlling the expression of MATZ1,
[00:33:06.620]and we have another region,
[00:33:10.180]another trans eKTL,
[00:33:11.900]which is close to MATZ69.
[00:33:15.240]MATZ69
[00:33:15.880]is also a true positive gene,
[00:33:18.360]which is affecting Florentine.
[00:33:21.640]And also shows some frequency shifts
[00:33:25.020]according to the lines of a region.
[00:33:28.060]We can see that the agenotype,
[00:33:30.100]which is having higher expression
[00:33:32.380]to MATZ1, seems to be more present
[00:33:35.500]and more frequently present in
[00:33:37.940]Teosinte and in tropical lines
[00:33:40.820]rather than in temperate ones.
[00:33:42.580]OK.
[00:33:45.460]So we have been using information
[00:33:51.000]from sorghum and from maize,
[00:33:52.420]and we see that that increased
[00:33:54.760]the diversity, and that diversity
[00:33:56.860]helped us to find more genes.
[00:33:59.620]So we are curious about if those genes
[00:34:04.100]that we are able to find
[00:34:06.080]in this combined analysis,
[00:34:07.700]how those genes are regulated.
[00:34:10.100]Are they regulated by the same factors
[00:34:14.140]or different ones?
[00:34:15.420]And that's a project that Carla Cuellar
[00:34:17.660]is working on.
[00:34:19.600]So she has two data sets.
[00:34:22.080]The first one is for maize.
[00:34:23.340]The first one is maize, and the second one is sorghum.
[00:34:26.820]For the maize data set, she's been
[00:34:29.040]running eQTLs for 24,000 genes.
[00:34:32.660]And she is using genetic markers of 10 million.
[00:34:37.680]For the sorghum pipeline, she is using 22 genes.
[00:34:41.620]And I will talk a little bit more
[00:34:43.660]about the numbers.
[00:34:45.380]The number of lines in the next slides.
[00:34:50.380]One of the first attempts that she
[00:34:53.880]had to run this eQTL in sorghum was
[00:34:57.700]using 700 lines and 170,000 genetic variants.
[00:35:04.720]And these are the results.
[00:35:06.720]So we were not having any significant peak here.
[00:35:13.120]But what happened?
[00:35:15.340]Is that Carla put herself together
[00:35:17.500]with Subhash from Young Lab.
[00:35:20.980]And then they were using genetic variants coming
[00:35:24.020]from whole genome for sequencing.
[00:35:26.620]Now they are using 400 lines and 4.7 million
[00:35:33.020]of genetic variants.
[00:35:36.320]And the results are much better.
[00:35:38.760]We are now having huge peaks in Cs affecting those regions.
[00:35:45.300]And luckily for us, they both are having a talk in the PSI
[00:35:51.420]retreat as well tomorrow.
[00:35:54.140]Carla in the morning and Subhash in the afternoon.
[00:35:56.660]So be sure to be there and ask some questions.
[00:35:58.820]I'll discuss more about science with them.
[00:36:06.340]And with this, I am concluding my talk.
[00:36:09.400]I would like to say that hopefully you
[00:36:12.920]have more overviews.
[00:36:15.260]How can we use gene expression to find genes that
[00:36:19.620]are associated with a phenotype?
[00:36:22.020]That gene expression can be reused.
[00:36:26.120]The gene expression controlling soaring time in maize
[00:36:29.980]show complex regulation.
[00:36:33.040]And also that cis and trans factor
[00:36:35.700]can offer a resource to validate these genes.
[00:36:40.260]Well, I want to say thank you to James, which
[00:36:42.780]has been a great mentor, also to our
[00:36:45.220]collaborators, Yumo King Lab at Peking University,
[00:36:50.020]King Liangjian Lab at UNL, and Nadi Thomsom Lab at MSU,
[00:36:54.700]to all the funding agencies, and to the incredible folks
[00:36:58.360]in James' lab here at UNL.
[00:37:01.540]Thank you, and--
[00:37:02.240]Thank you so much, very interesting presentation.
[00:37:12.840]Thank you.
[00:37:13.820]So we have--
[00:37:15.180]20 minutes for questions.
[00:37:17.220]So if someone here has questions, just raise your hand.
[00:37:21.220]We can start.
[00:37:21.720]Thank you, Wai, for the great presentation.
[00:37:27.700]So I'm curious about the nature of the, I guess,
[00:37:31.860]genes they are identifying with different expression level.
[00:37:34.800]So I guess in class, I learned about a lot of those expression
[00:37:38.280]differences is not necessarily associated with the gene
[00:37:41.320]itself, but with the promoter region.
[00:37:44.500]So I guess--
[00:37:45.140]In your experience and in your data set,
[00:37:47.600]do you find them more associated with the promoter region,
[00:37:50.300]with the gene itself?
[00:37:51.600]Or is it because of some other regulatory mechanism?
[00:37:56.820]You mean the C-CQTL or in--
[00:37:59.240]Yeah, for the expression data that you see
[00:38:02.420]between different lines, right?
[00:38:04.200]OK, so for the expression data, we
[00:38:06.440]are sequencing the transcript.
[00:38:09.600]So with just the expression data, we don't really know--
[00:38:15.100]which is originating this expression.
[00:38:18.200]With the C-CQTL analysis, we can have an idea of--
[00:38:22.140]depending on if this is in the promoter region, which
[00:38:25.040]is mostly in the promoter region,
[00:38:26.920]or can be other regions in trans.
[00:38:31.160]OK, sounds good.
[00:38:32.420]Can I ask a second question?
[00:38:33.500]Yeah, sure.
[00:38:34.220]We have 10 minutes for it.
[00:38:35.500]All right, so my second question.
[00:38:37.520]So since I'm more of a molecular person, I guess,
[00:38:40.620]the first thing I do is just go to aerobatopsis
[00:38:43.860]and see if the pathway--
[00:38:45.060]it's kind of all mapped out, right?
[00:38:46.620]And if it's mapped out, I go back to BLAST and BLAST
[00:38:49.200]whatever gene I'm working with and find genes that way, right?
[00:38:53.700]So I guess comparing that approach versus yours,
[00:38:58.500]can you talk a little bit about the pros and cons of what
[00:39:02.840]is better and what is worse, what we're missing,
[00:39:04.800]stuff like that?
[00:39:05.900]Yes.
[00:39:07.260]Yeah, that's a good question.
[00:39:08.920]So I agree that that's also another really useful
[00:39:13.320]methodology that we can migrate.
[00:39:15.020]We can migrate information, in this case,
[00:39:16.980]from Arabidopsis to other organisms.
[00:39:19.540]But in some other cases, let's talk about CCNA or CCN family
[00:39:25.260]of matchbox transcription factor.
[00:39:27.300]In maize, I think we have more than 70 matchbox
[00:39:30.980]transcription factor.
[00:39:32.640]So to know which of those 70 are the ones that
[00:39:38.380]are affecting the trait, we can select
[00:39:44.980]probably do our best guesses, depending
[00:39:48.560]on the similarity between all those matchbox transcription
[00:39:51.200]factors, because they are like a difference of families
[00:39:53.600]and something like that.
[00:39:54.800]And that obviously is going to give us
[00:39:56.900]some good sense of which are the genes that are doing that.
[00:40:01.500]But we won't be completely sure that it's
[00:40:06.960]affecting the phenotype.
[00:40:09.220]So here, we know that some of them
[00:40:11.820]are affecting the phenotype, at least with the model that we
[00:40:14.940]have.
[00:40:16.640]And as we can see, many of the genes that are finding
[00:40:20.620]are matchbox, but we are not finding the 17 matchbox.
[00:40:25.120]So I think two valuables are really used--
[00:40:28.800]two, sorry, two methodologies are really useful.
[00:40:34.060]All right.
[00:40:35.460]Thank you.
[00:40:35.960]Yeah.
[00:40:36.460]Very nice talk, Brad.
[00:40:47.200]I just wanted to thank you for the data set, too, the one
[00:40:51.220]that you made public for Maze.
[00:40:53.440]We use it, and it's kind of a relief
[00:40:57.400]to see that it was very carefully collected,
[00:40:59.400]because sometimes you download data sets
[00:41:01.220]and you don't really know what you're getting.
[00:41:04.600]I mean, if you get a result, you walk away.
[00:41:06.440]You're smiling, but most of the time,
[00:41:07.800]you always have that feeling that,
[00:41:10.360]I hope they're collected in the right way.
[00:41:12.820]So kind of going back to, do you have any advice
[00:41:15.960]on how you're seeing some bias from the PCA
[00:41:19.920]on the clock genes, LHY, Gigantia, I think?
[00:41:24.320]How are you, I mean, did you try to correct it,
[00:41:27.300]or how do typically people correct for that?
[00:41:30.240]Because as you said in the beginning,
[00:41:31.760]that there's a whole bunch of thousands of genes
[00:41:34.060]that are diurnal variations.
[00:41:36.440]So is there a method to transform and maybe that data set?
[00:41:41.440]- Yes, so there are two different ways
[00:41:45.520]or two different approaches that we have used
[00:41:47.840]to correct for those.
[00:41:49.620]For the original TWA that we run,
[00:41:53.060]we use the order of collection as a covariate.
[00:41:56.880]So we were providing that information there.
[00:42:00.340]For the second ones that we are using,
[00:42:02.900]we are correcting, especially correcting those
[00:42:06.380]and we are also adding the order of collection
[00:42:09.760]as a covariate in that process.
[00:42:12.200]So yeah, that's helping us to fix for that.
[00:42:17.200]- Okay.
[00:42:18.840]I have another question.
[00:42:19.820]So maize has a pretty remarkable latitudinal adaption.
[00:42:24.820]How does sorghum's domestication and spread compare
[00:42:30.300]in terms of the latitudinal spread that you see in maize?
[00:42:34.060]Are there any signatures that would
[00:42:36.320]be more distributed than sorghum?
[00:42:38.320]- That's a really good question.
[00:42:40.440]I'm afraid I don't have the complete information
[00:42:53.340]to answer it, but based on my observations,
[00:42:56.560]they are following kind of the similar distribution.
[00:43:02.020]Maybe maize is more distributed than sorghum,
[00:43:05.900]but mostly is, they are overlapping.
[00:43:10.900]That's my guess, but I don't fully know.
[00:43:15.180]- I think sorghum probably is more than maize.
[00:43:21.580]- Okay.
[00:43:22.420]Okay, yeah.
[00:43:28.000]- Could you see that the land grade
[00:43:29.720]is 15 minutes, 15 minutes, yeah, well,
[00:43:32.460]sorghum, maize,
[00:43:35.480]I think some part of that is brown.
[00:43:40.480]I mean, it's probably in the top four areas.
[00:43:46.940]- And I think temperate sorghum
[00:43:50.640]is maybe two to 300 years old.
[00:43:52.260]- Yeah, it's not very,
[00:43:53.480]it's been around for a very long time.
[00:43:55.740]- And that's a good point to help solve that.
[00:44:01.060]Maybe that's a good word to investigate.
[00:44:05.060]- Can I answer the question?
[00:44:07.140]Sorry, I'm not matching.
[00:44:09.040]- Yes.
[00:44:09.880]- Very great talk.
[00:44:20.380]And actually it's very inspiring.
[00:44:22.400]So my question is,
[00:44:25.120]so there's a lot of studies about
[00:44:27.980]from time in is determined,
[00:44:30.120]is polygenic determined by a lot of genes.
[00:44:33.440]After this study,
[00:44:34.640]do you have an impression like overall,
[00:44:38.180]how many genes that involved in determining
[00:44:41.640]the from time ballpark number is not necessarily
[00:44:44.040]be on 132 or something like that.
[00:44:46.720]Just how many of those are large effect
[00:44:50.620]and how many of those are small effects?
[00:44:53.360]And a single follow up question is for this large effect,
[00:44:57.660]so you can also do EQT analysis.
[00:45:00.740]So how many of those actually have a CCQT
[00:45:04.220]and how many of those have trans EQT?
[00:45:07.400]I think it might be next step will
[00:45:08.920]be try to connect those together, connect those dots
[00:45:11.320]together.
[00:45:11.820]Is this trans the gene that you identify the control for that?
[00:45:16.600]So maybe it's time to think about the network.
[00:45:21.160]Yes.
[00:45:23.080]Yep, I totally agree with that question and point of view.
[00:45:27.420]I think it would be great to connect all these network
[00:45:33.800]of genes in MACE, but also to get information from Sargon.
[00:45:39.980]So part of the question was, how many of these
[00:45:44.760]are having regulators in common, right, for MACE,
[00:45:49.120]if I understand correctly?
[00:45:50.340]And one of the reasons why I was putting MATS1 and SAX6
[00:45:55.040]is because they are paralogs in MACE.
[00:45:59.000]So we know here that MATS1 is regulated,
[00:46:03.380]but MATS69 in trans.
[00:46:07.380]So the paralogue of MATS1 is SAX6.
[00:46:12.020]And if you see this pic here, I believe it's MATS69.
[00:46:18.140]Although it's not significant, but it's--
[00:46:20.780]I wish it could be, because then the story could be better.
[00:46:24.760]Yeah.
[00:46:26.920]But then, I will be curious, if we look at this gene,
[00:46:31.320]they have one orthologue of this in Sargon.
[00:46:32.960]If we look at the orthologue in Sargon,
[00:46:35.820]it's also regulated, but MAT69 orthologue in Sargon.
[00:46:39.620]So that's the kind of questions that then we want to address.
[00:46:44.980]How many genes with high or low impact on plurin time?
[00:46:50.080]Probably hundreds.
[00:46:51.460]I'm not sure.
[00:46:52.840]At this point, we have the true positive list of genes
[00:47:01.540]that we have.
[00:47:02.540]It has 26, and all they are having an impact.
[00:47:09.300]There are four different pathways
[00:47:11.760]that are working for plurin time.
[00:47:14.220]So I guess they are all interacting,
[00:47:16.860]but also they have their own specific genes, so maybe
[00:47:20.580]hundreds.
[00:47:21.020]Yeah, I think Jinliang's comment about networks.
[00:47:30.620]So I wanted to ask that.
[00:47:32.120]But I forgot, because your talk's interesting,
[00:47:35.020]so I keep forgetting stuff.
[00:47:36.480]The MADs type 1, for sure, also form heterodimers.
[00:47:43.380]So essentially, proteins would work together
[00:47:46.020]to control transcription downstream.
[00:47:48.240]So is the 1 and 69 kind of--
[00:47:54.040]is there any evidence that they're type 1s?
[00:47:56.200]Or are they interacting?
[00:47:57.380]Or are they co-expressing?
[00:47:59.220]Because you could kind of parse a lot of those MADs
[00:48:01.700]that way.
[00:48:03.060]And if they're paralogs, we've kind of worked on one
[00:48:06.020]which is 78 and 79.
[00:48:08.180]They're paralogs and next to each other or duplicates.
[00:48:11.120]And they also interacted.
[00:48:12.940]So just kind of a prompt from what
[00:48:15.940]was saying about networks.
[00:48:17.300]So they form small networks.
[00:48:19.280]Oh, OK.
[00:48:20.480]Thank you.
[00:48:20.940]We have some questions in the chat.
[00:48:25.920]OK.
[00:48:26.420]OK.
[00:48:26.920]The first question says, are there bottlenecks observing
[00:48:36.580]when collecting 70, 150 unique samples in two hours?
[00:48:42.020]What software did you use to run the analysis?
[00:48:46.000]OK, so bottlenecks to--
[00:48:50.020]ah.
[00:48:52.900]I think it depends on--
[00:48:56.180]all the preparation that it takes
[00:49:00.300]previous to collect the data.
[00:49:03.000]So in our group, we have John.
[00:49:06.220]And he's really amazing to organize things.
[00:49:09.940]So he will have liquid nitrogen there, dry ice.
[00:49:15.220]He designed-- if I can go back--
[00:49:18.260]he designed all this equipment that we
[00:49:26.000]are using.
[00:49:26.160]So I think the only bottleneck that I can identify
[00:49:35.140]is after we collect the data.
[00:49:41.440]Let's say that we have seven samplers.
[00:49:43.440]All these seven samplers need to put--
[00:49:47.520]no, actually, no.
[00:49:48.540]Now that I-- no, no, forget it.
[00:49:50.760]That's not a bottleneck.
[00:49:52.400]Just if you prepare really well before collecting, I
[00:49:55.980]think everything should be pretty good.
[00:49:58.260]The software to analyze--
[00:50:03.120]To run the analysis.
[00:50:05.200]OK, to get gene expression, I was using Callisto.
[00:50:09.720]To run T-was, I was using Gapit.
[00:50:12.480]OK.
[00:50:14.420]The next question is, can T-was help
[00:50:16.720]to identify SMP variants for the genes responsible for traits
[00:50:20.740]which can be employed for marker-assisted selection
[00:50:23.580]in plant breeding?
[00:50:25.800]Maybe.
[00:50:27.040]Yes.
[00:50:27.540]Yeah.
[00:50:28.040]We have here says, was there on-farm evaluation
[00:50:36.820]of these genotypes across diverse environment
[00:50:39.280]to confirm its instability?
[00:50:40.940]That means stability of the lines
[00:50:47.600]in different environments?
[00:50:48.680]Or stability-- let's see if they write something.
[00:50:55.620]OK.
[00:50:57.780]We can wait.
[00:50:58.740]We can move to another one.
[00:50:59.840]Here says, UNL has the widest database for--
[00:51:06.680]links to access the databases for further study?
[00:51:12.940]Are they links for--
[00:51:14.640]Accessing the databases.
[00:51:16.840]Yes.
[00:51:17.760]For the first one, for the 2020, the link should be in the paper.
[00:51:25.440]And then-- or you can email me, and I can share the UNExpression
[00:51:34.180]data set already filtered that was published, so that's OK.
[00:51:39.380]For the new ones, I will suggest to email my advisor, James Schnabel.
[00:51:45.560]That's the way to get them.
[00:51:47.480]Thank you.
[00:51:49.720]We have another one says, is having only transcript--
[00:51:54.240]sorry.
[00:51:55.260]Is not eQTL after an eQTL analysis a reliable result?
[00:52:00.400]The target gene is a transcription factor gene.
[00:52:03.400]That's awesome.
[00:52:15.720]So--
[00:52:21.160]I think they're asking about their own--
[00:52:25.080]their own result, right?
[00:52:26.240]They're saying, I ran this analysis,
[00:52:28.160]I only got trans-eQTL.
[00:52:29.420]So gene--
[00:52:30.960]Oh, yeah, yeah.
[00:52:32.680]OK.
[00:52:33.180]OK.
[00:52:33.720]Yes.
[00:52:34.220]To have only trans-eQTL for a gene, yes,
[00:52:38.440]that's also a reliable result. So in many of the genes
[00:52:42.680]that we were having here, we are also only having a trans-eQTL.
[00:52:48.140]So for example, ccN12 or--
[00:52:50.960]yeah.
[00:52:54.900]It's sub one.
[00:52:58.800]Yeah.
[00:53:00.800]Yeah, OK.
[00:53:03.260]Another one, is the GWAS model flexible enough
[00:53:06.700]to incorporate the intermediate step of gene expression
[00:53:09.520]between genotype and phenotype?
[00:53:12.820]Yeah, it could be some method.
[00:53:14.260]It depends on how they develop their method.
[00:53:20.560]But I think now people is trying to unify
[00:53:24.720]a lot of this information with some deep learning
[00:53:28.100]models that are more able to handle
[00:53:31.720]that kind of information.
[00:53:34.480]OK, thanks.
[00:53:36.740]Another one.
[00:53:38.420]You show in one slide that many genes, 30%,
[00:53:42.260]are expressed in diurnal fashion.
[00:53:45.400]Does one have to gather genetic samples
[00:53:48.100]at two or more different diurnal time frames
[00:53:51.640]to ensure that the transcriptomic analysis captured
[00:53:54.540]the expression of all genes?
[00:53:57.840]How will you do that?
[00:53:58.860]Well, the transcriptomic or the RNA-seq analysis
[00:54:08.400]won't capture the expression for all genes.
[00:54:11.900]So that's one of the limitations as well.
[00:54:16.720]And the other one is that some of them
[00:54:19.300]maybe are really, really low expressed,
[00:54:21.600]but they are also more--
[00:54:24.360]are likely to be affected by errors during the sequencing.
[00:54:28.460]So that's why MACE genome has around 40,000 gene models.
[00:54:34.160]And that's why our analysis was done with 24,000
[00:54:38.200]after removing the ones that were low expressed
[00:54:41.660]and the ones that we don't have any information
[00:54:43.760]of gene expression.
[00:54:46.260]All right.
[00:54:47.420]Thank you.
[00:54:48.620]Do we have more questions?
[00:54:49.660]Yeah.
[00:54:54.180]Very nice talk.
[00:54:59.140]In your slides, you talk about you did a TWAS combined,
[00:55:04.160]sorghum and maize.
[00:55:07.140]I'm thinking if you leverage it as much as some public data,
[00:55:13.220]can you get a similar result?
[00:55:16.720]If you don't do the for sorghum,
[00:55:19.820]save some money, can we just use public data
[00:55:22.720]and we get a similar result?
[00:55:24.000]Or maybe I didn't get a point.
[00:55:30.100]Or if you think about it this way from some different angle
[00:55:37.420]using sorghum, you want to answer some evolution
[00:55:41.880]related questions?
[00:55:44.200]Yes, absolutely.
[00:55:45.380]So the use of public data sets can also
[00:55:50.060]answer the same question.
[00:55:53.820]But the pros about using that is that we won't spend any money.
[00:56:02.720]The cons is that probably we are going to have more noise
[00:56:06.100]according to how the data were collected,
[00:56:09.820]that if they were in different batches.
[00:56:14.760]And that can also add some noise to the analysis.
[00:56:18.740]And yeah, that was the question or no?
[00:56:22.640]Have you tried some public data combined together?
[00:56:27.900]OK.
[00:56:28.400]No, I haven't.
[00:56:30.200]Since you mentioned that you identified some new genes.
[00:56:34.060]Yeah.
[00:56:35.400]Yes.
[00:56:37.240]No, I haven't.
[00:56:37.960]But that's a good suggestion.
[00:56:39.640]OK, we have time for one last question.
[00:56:47.320]I'd like to shift things a little bit
[00:56:48.880]and have an application type of question here.
[00:56:52.320]With maize, for breeding hybrid corn,
[00:56:58.600]it's really important to coordinate the flowering time
[00:57:02.600]of anthesis and silking of the inbred lines.
[00:57:07.120]How can a plant breeder use this kind of information
[00:57:11.560]to better coordinate that application in the field
[00:57:17.080]compared to just growing out his inbred lines
[00:57:19.120]and figuring out from that what their flowering
[00:57:22.320]time is going to be like?
[00:57:23.880]- Yes, that's a really good question.
[00:57:26.380]So plants are really sensitive of flowering time
[00:57:31.380]are really sensitive to change in the environment
[00:57:35.400]or in the life for the period.
[00:57:37.040]So we have some examples that I'm from Mexico, for example.
[00:57:42.200]If you move some plants that are adapted from Mexico
[00:57:44.660]and you try to grow them here, they will never flower.
[00:57:51.280]So that happened also with plants.
[00:57:54.000]So to know which are the genes that are affecting
[00:57:58.600]or modulating this flowering time,
[00:58:00.440]they can try to narrow first to adapt these plants
[00:58:04.860]to move to other environments,
[00:58:06.940]and then to narrow that window,
[00:58:08.840]the difference between silking and anthesis
[00:58:11.060]to show that we are having more pollinization of the plants.
[00:58:16.060]So yeah, thank you.
[00:58:20.240]- We're running out of time.
[00:58:21.080]We will finish our seminar today.
[00:58:22.640]Thank you so much Vladimir again for your presentation,
[00:58:25.540]and thank you all of you for attending.
[00:58:27.280]If you still have questions, feel free to contact Vladimir.
[00:58:30.740]Thanks.
[00:58:31.580]- Thank you.
[00:58:32.420]-
[00:58:37.420]Thank you.

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

The Use of Gene Expression to Predict Gene Function in Maize and Sorghum

Description

Searchable Transcript

Comments icon comment

Related Channels

Comments