Added datasets and R analysis scripts

Analysis/02_Analysis_Dataset2.R

@@ -0,0 +1,388 @@
+##' Data Analysis of Electroantennogram Data from two different Stonefly Ecotypes
+##' (full-winged and wing-reduced)
+##'
+##' R routine to reproduce the Fig S2 from the manuscript entitled
+##' 'Rapid evolution of olfactory degradation in recently flightless
+##' insects'.
+##' 
+##
+
+# load R package
+library(dplyr)
+library(tidyr)
+library(ggplot2)
+library(tibble)
+library(scales)
+
+
+# clean up environment
+rm(list=ls()) 
+
+
+# set working directory
+currentpath <- dirname(rstudioapi::getSourceEditorContext()$path)
+setwd(currentpath)
+getwd()
+
+
+# load function to calculate various response dynamic measures
+source('./functions/responseDynamics.R')
+source('./functions/utils.R')
+
+
+fig_path <- "../Figures"
+
+
+# aspect ratio (as y/x) for trace plots
+ar <- 2/3
+
+### colors for stoneflies ###
+fw_sm_l <- "#000000" ## black
+fw_sm_d <- "red"
+
+
+
+data2 <- read.csv("../Datasets/Dataset2.csv", stringsAsFactors = FALSE)
+
+
+# automatically read in amplification from info file
+multiply_voltage <- 1/data2$Amplification
+
+## correct voltage values for the different amplification factor
+data2[, paste0("t", 1:5000)] <- data2[, paste0("t", 1:5000)] * multiply_voltage #in Volt
+data2[, paste0("t", 1:5000)] <- data2[, paste0("t", 1:5000)] * 1000  # in millivolt
+
+
+# convert ID to a factor
+data2$ID <- as.factor(data2$ID)
+data2$AntennaState <- as.factor(data2$AntennaState)
+data2 <- droplevels(data2)
+
+
+# reorder the factor levels of column Odor
+data2$Odor <- factor(data2$Odor, levels = c("Blank", "2Heptanone", "1Octanol", "PropionicAcid",
+                                            "2Butanone", "Water", "2HeptanoneTest"),
+                     labels = c("Blank", "2-heptanone", "1-octanol", "Propionic acid",
+                                "2-butanone", "Water", "2-heptanone (2nd)"))
+
+
+# data2$antpart[data2$Winged=="base"] <- "middle"
+data2$antpart[data2$Winged=="tip"] <- "distal"
+data2$antpart <- as.factor(data2$antpart)
+data2$AntennaState <- factor(data2$AntennaState,
+                             levels = levels(data2$AntennaState),
+                             labels = c("live", "dead"))
+data2 <- droplevels(data2)
+
+data2$NewGroup <- paste(data2$antpart, data2$AntennaState)
+data2$NewGroup <- as.factor(data2$NewGroup)
+
+
+
+# create median filter mean traces (only for non-10Hz stimulations)
+med <- as.data.frame(t(apply(data2[data2$PulseDuration != 50, paste0("t", 1:5000)], 1, runmed, k=11)))
+data2[data2$PulseDuration != 50, paste0("t", 1:5000)] <- med
+
+
+
+# add response characteristcs
+y1 <- as.data.frame(t(apply(data2[, paste0("t", 201:5000)], 1, responseDynamics)))
+
+colnames(y1) <- c("Timeto10Max",
+                  "Timeto90Max",
+                  "TimetoMax", 
+                  "TimeFromMaxto90",
+                  "TimeFromMaxto10",
+                  "Time10to90Max",
+                  "Time90to10Max",
+                  "Timeto10Min",
+                  "Timeto90Min",
+                  "TimetoMin", 
+                  "TimeFromMinto90",
+                  "TimeFromMinto10",
+                  "Time10to90Min",
+                  "Time90to10Min",
+                  "MeanMax",
+                  "MeanMin")
+
+
+data2 <- cbind(as.data.frame(data2), y1)
+
+data2$max <- apply(data2[, paste0("t", 201:5000)], 1, max)
+data2$min <- apply(data2[, paste0("t", 201:5000)], 1, min)
+data2 <- data2 %>%
+  mutate(mima = ifelse(Odor == "Propionic acid" |
+                         (Odor == "Water" & (abs(min) > max)), min, max))
+
+data2 <- data2 %>%
+  mutate(Mean = ifelse(Odor == "Propionic acid" |
+                         (Odor == "Water" & (abs(min) > max)), MeanMin, MeanMax))
+
+
+#######################################
+
+## generate plot function for difference in factor NewGroup
+
+plot_Antenna <- function(data, group) {
+  
+  d_mean <- data %>%
+    filter(Group==group) %>%
+    group_by(AntennaState, Odor, PulseDuration) %>%
+    summarise_at(.funs = mean, .vars=paste0("t", 1:5000))
+  
+  
+  # calculate the standard error over all odor-pulseduration configurations
+  d_se <- data %>%
+    filter(Group==group) %>%
+    group_by(AntennaState, Odor, PulseDuration) %>%
+    summarise_at(.funs = se, .vars=paste0("t", 1:5000))
+  
+  # calculate sample size (n) over all odor-pulseduration configurations
+  d_n <- data %>%
+    filter(Group==group) %>%
+    group_by(AntennaState, Odor, PulseDuration) %>%
+    summarise(N=n()) %>%
+    group_by(AntennaState) %>%
+    summarise(N=unique(N))
+  d_n <- unite(d_n, col = "AntennaState", sep = " (n = ")
+  d_n$t <- ")"
+  d_n <- unite(d_n, col = "AntennaState", sep = "")
+  
+  
+  d_all <- d_mean
+  data_se <- gather(d_se, time, se, t1:t5000, factor_key=TRUE)
+  
+  
+  d_all$PulseDuration <- factor(d_all$PulseDuration, levels = c(15, 30, 150, 300))
+  d_all$PulseDur <- factor(d_all$PulseDuration, levels = c(15, 30, 150, 300),
+                           labels = c("15 ms", "30 ms", "150 ms", "300 ms"))
+  
+  
+  data_long <- gather(d_all, time, Voltage, t1:t5000, factor_key=TRUE)
+  data_long$time <- (as.numeric(gsub("t", "", data_long$time)) - 200)/1000
+  data_long$SE <- data_se$se
+  data_long$AntennaState <- droplevels(data_long$AntennaState)
+  data_long$AntennaState <- factor(data_long$AntennaState,
+                             labels = t(d_n))
+  
+  data_long$high <- data_long$Voltage + data_long$SE
+  data_long$low <- data_long$Voltage - data_long$SE
+  
+  
+  ## generate dummy values for a the same scale of y, but different ranges (max and min)
+  ranges <- data.frame("NegResp" = range(data_long[data_long$Odor=="Water" |
+                                                     data_long$Odor=="Propionic acid", c("high", "low")]),
+                       "PosResp" = range(data_long[data_long$Odor!="Water" &
+                                                     data_long$Odor!="Propionic acid", c("high", "low")]),
+                       "Range"= c("min", "max"))
+  
+  roundUp <- function(x, to)  {
+    to*(x%/%to + as.logical(x%%to))
+  }
+  max_diff <- roundUp(max(diff(ranges$NegResp), diff(ranges$PosResp)), to = 0.01)
+  
+  
+  max_diff/2
+  ranges$NegResp_new <- c((ranges$NegResp[1] + (diff(ranges$NegResp)/2)) - max_diff/2,
+                          (ranges$NegResp[1] + (diff(ranges$NegResp)/2)) + max_diff/2)
+  ranges$PosResp_new <- c((ranges$PosResp[1] + (diff(ranges$PosResp)/2)) - max_diff/2,
+                          (ranges$PosResp[1] + (diff(ranges$PosResp)/2)) + max_diff/2)
+  
+  df <- as.data.frame(table(data_long$Odor, data_long$PulseDuration))
+  df <- df[df$Freq!=0,]
+  colnames(df) <- c("Odor", "PulseDuration", "Freq")
+  df$range <- "min"
+  df$time <- 1
+  
+  df2 <- df
+  df2$range <- "max"
+  
+  df <- rbind(df, df2)
+  
+  df$Voltage[(df$Odor=="Water" | df$Odor=="Propionic acid") & df$range=="min"] <- ranges$NegResp_new[1]
+  df$Voltage[(df$Odor=="Water" | df$Odor=="Propionic acid") & df$range=="max"] <- ranges$NegResp_new[2]
+  df$Voltage[(df$Odor!="Water" & df$Odor!="Propionic acid") & df$range=="min"] <- ranges$PosResp_new[1]
+  df$Voltage[(df$Odor!="Water" & df$Odor!="Propionic acid") & df$range=="max"] <- ranges$PosResp_new[2]
+  
+  diff(range(df[df$Odor=="Water", "Voltage"]))
+
+  mids <- df %>%
+    filter(PulseDuration=="300") %>%
+    group_by(Odor) %>%
+    summarise(mid=Voltage[range=="max"] - ((Voltage[range=="max"] - Voltage[range=="min"])/2))
+  
+  
+  data_segm1 <- data.frame(y=-0.02, yend=c(0), x=3, xend=3,
+                           PulseDuration=15,
+                           Odor=c("Water"))
+  label1 <- data.frame(x=3, y=c(-0.01),
+                       text=c("0.02 mV"),
+                       PulseDuration= 15,
+                       Odor=c("Water"))
+  data_segm2 <- data.frame(x=3, xend=4, y=-0.02, yend=-0.02,
+                           PulseDuration= 15,
+                           Odor=c("Water"))
+  label2 <- data.frame(x=3.5, y=-0.02,
+                       text=c("1 s"),
+                       PulseDuration= 15,
+                       Odor="Water")
+
+  
+  label3 <- data.frame(x=5, y=mids$mid,
+                       text=levels(data_long$Odor),
+                       PulseDuration=300,
+                       Odor=levels(data_long$Odor))
+  
+  col <- c(fw_sm_l, fw_sm_d)
+  
+  df_sub <- df[df$range=="min",]
+  df_sub <- droplevels(df_sub)
+  df_sub$PulseDuration <- factor(df_sub$PulseDuration, levels = c(15, 30, 150, 300))
+  
+  ggplot(data_long, aes(x=time, y=Voltage)) + 
+    scale_color_manual(aesthetics = c("colour", "fill"), values = col) +
+    facet_grid(Odor ~ factor(PulseDuration, 
+                             levels = c(15, 30, 150, 300),
+                             labels = c("15 ms", "30 ms", "150 ms", "300 ms")),
+               scales = "free_y") +
+    geom_rect(mapping=aes(xmin=0, xmax=as.numeric(as.character(PulseDuration))/1000,
+                                       ymin=-Inf, ymax=Inf),
+              fill="grey90", alpha=1) +
+    geom_hline(mapping = aes(yintercept = 0), color="black", size=0.3, linetype="dotted") +
+    geom_ribbon(aes(ymin=Voltage + SE, ymax=Voltage - SE, fill=AntennaState), alpha=0.3) +
+    geom_line(aes(colour = AntennaState)) +
+    theme_bw() +
+    geom_blank(df, mapping = aes(x=time, y=Voltage)) +
+    labs(x = "Time (s)", y="Voltage (mV)") +
+    theme(panel.grid.major = element_blank(),
+          panel.grid.minor = element_blank(),
+          axis.line = element_blank(),
+          axis.title = element_blank(),
+          axis.ticks = element_blank(),
+          aspect.ratio = ar,
+          plot.margin = unit(c(5.5, 5.5, 5.5, 5.5), "points"),
+          strip.background = element_rect(colour="NA", fill="NA"),
+          panel.border = element_blank(),
+          text = element_text(size=18),
+          axis.text=element_blank(),
+          strip.text.y = element_text(angle = 0, hjust = 0),
+          legend.position = c(0.2, 0.08),
+          legend.title = element_blank()) +
+    geom_segment(data=data_segm1,
+                 aes(x=x, y=y, yend=yend, xend=xend)) +
+    geom_text(data = label1, aes(x=x, y=y, label=text), size=5, hjust=1.1) +
+    geom_segment(data=data_segm2,
+                 aes(x=x, y=y, yend=yend, xend=xend)) +
+    geom_text(data = label2, aes(x=x, y=y, label=text), size=5, vjust=1.3) +
+    coord_cartesian(expand = TRUE,
+                    clip = 'off')
+}
+
+
+## plot real data
+p1 <- plot_Antenna(data = data2,
+                   group = "FenestrateZombie"); p1
+
+
+
+subdat <- data2 %>%
+  dplyr::select(Mean, Odor, AntennaState, ID_new, PulseDuration) %>%
+  group_by(AntennaState, ID_new, PulseDuration) %>%
+  pivot_wider(names_from = Odor, values_from = Mean)
+
+colnames(subdat) <- c("AntennaState","ID_new","PulseDuration","Blank","HPT",
+                      "OCT","PropionicAcid","BTN","Water","HPT2")
+
+subdat2 <- subdat %>%
+  group_by(ID_new, PulseDuration) %>%
+  summarise("HeptAlive"=HPT[AntennaState=="live"], "PropDead"=PropionicAcid[AntennaState=="dead"])
+
+subdat2$PulseDuration <- factor(subdat2$PulseDuration,
+                                levels = c(15, 30, 150, 300))
+
+subdat2$pvals <- NA
+
+## Create a list holding the p-values for regressions on each PulseDuration
+subdat2 <- subdat2 %>%
+  group_by(PulseDuration) %>%
+  mutate(HeptAlive_sc=scale(HeptAlive), PropDead_sc=scale(PropDead))
+
+subdat2$ID_new <- as.factor(subdat2$ID_new)
+
+NEWDAT <- as.data.frame(matrix(NA, ncol = 7, nrow = 4))
+colnames(NEWDAT) <- c("PulseDuration", "pval", "r2", "xmin", "xmax", "ymin", "ymax")
+
+NEWDAT$PulseDuration <- as.character(unique(subdat2$PulseDuration))
+
+par(mfrow=c(1,4))
+
+
+
+for (i in as.character(unique(subdat2$PulseDuration))) {
+  dat <- subdat2[subdat2$PulseDuration==i, ]
+  
+  mod <- lm(HeptAlive ~ PropDead, data = dat)
+
+  ry <- range(dat$HeptAlive)
+  rx <- range(dat$PropDead)
+  rxy <- max(abs(diff(ry)), abs(diff(rx)))
+  xlim <- c((rx[1] + abs(diff(rx))/2) - rxy/2, (rx[1] + abs(diff(rx))/2) + rxy/2)
+  ylim <- c((ry[1] + abs(diff(ry))/2) - rxy/2, (ry[1] + abs(diff(ry))/2) + rxy/2)
+  
+  plot(HeptAlive ~ PropDead, data = dat, pch=21, las=1,
+       cex.lab=1.2, xlim=xlim,
+       ylim=ylim, main=i)
+  abline(mod)
+  summary(mod)
+  m <- summary(mod)
+  print(m)
+  
+  NEWDAT[NEWDAT$PulseDuration==i, "pval"] <- round(m$coefficients[2, 4], digits = 2)
+  NEWDAT[NEWDAT$PulseDuration==i, "r2"] <- round(m$r.squared, digits = 2)
+  NEWDAT[NEWDAT$PulseDuration==i, c("xmin", "xmax")] <- xlim
+  NEWDAT[NEWDAT$PulseDuration==i, c("ymin", "ymax")] <- ylim
+}
+
+
+subdat2$PulseDuration <- factor(subdat2$PulseDuration,
+                                c(15, 30, 150, 300), 
+                                c("15 ms", "30 ms", "150 ms", "300 ms"))
+
+NEWDAT$PulseDuration <- factor(NEWDAT$PulseDuration,
+                                c(15, 30, 150, 300), 
+                                c("15 ms", "30 ms", "150 ms", "300 ms"))
+
+dependencePlot <- ggplot(subdat2, aes(PropDead, HeptAlive)) +
+  geom_point()  +
+  scale_x_continuous(breaks = pretty_breaks(n = 3)) +
+  scale_y_continuous(breaks = pretty_breaks(n = 3)) +
+  geom_blank(data=NEWDAT, mapping=aes(x=xmin, y=ymin)) +
+  geom_blank(data=NEWDAT, mapping=aes(x=xmax, y=ymax)) +
+  facet_wrap( . ~ PulseDuration, scales = "free", ncol=4) +
+  xlab("Response strength of dead antennae to propionic acid (mV)") +
+  ylab("Response strength \n of live antennae \n to 2-heptanone (mV)") +
+  theme_bw() +
+  theme(panel.grid.major = element_blank(),
+        panel.grid.minor = element_blank(),
+        panel.spacing = unit(2, "lines"),
+        plot.margin = unit(c(5.5, 5.5, 5.5, 5.5), "points"),
+        strip.background = element_rect(colour="NA", fill="NA"),
+        text = element_text(size=18),
+        aspect.ratio=1,
+        axis.text=element_text(colour="black")) +
+  geom_text(NEWDAT, mapping = aes(x=xmax, y=ymax, label=paste0("p=",pval)), size=5,
+            color="blue", hjust=1, vjust=1) +
+  geom_text(NEWDAT, mapping = aes(x=xmin, y=ymax, label=paste0("r²=", r2)), size=5, 
+            color="blue", hjust=0, vjust=1)
+  
+
+
+
+FigS2 <- cowplot::plot_grid(p1, dependencePlot,
+                   labels = c("A", "B"), label_size = 16,
+                   ncol = 1, nrow = 2, rel_heights = c(3.2,1))
+
+FigS2
+
+ggsave("FigureS2.pdf", path = fig_path, width = 10, height = 14)
+ggsave("FigureS2.jpg", path = fig_path, width = 10, height = 14)

Analysis/functions/responseDynamics.R

@@ -0,0 +1,155 @@
+##' Function to Calculate Various Parameters of Response Dynamics
+##' 
+##' The following parameters are calculated: 
+##' Time to 10% of Response Maximum (Minimum), 
+##' Time to 90% of Response Maximum (Minimum),
+##' Time to Response Maximum (Minimum), 
+##' Time to 90% of Response Maximum (Minimum) after Max (Min),
+##' Time to 10% of Response Maximum (Minimum) after Max (Min),
+##' Risetime to Maximum (Minimum),
+##' Falltime from Maximum (Minimum),
+##' Mean Response Maximum (Minimum).
+##' 
+##' @param x Input dataframe with one row per 5 sec trace
+##' @param from Response starting percentage for Risetime calculation,
+##'             defaults to 10 percent of response maximum (or minimum, respectively).
+##' @param to Response ending percentage for Risetime calculation,
+##'           defaults to 90 percent of response maximum (or minimum, respectively).
+##' 
+##' @return New columns with calculated response dynamic parameters
+##' 
+##
+responseDynamics <- function(x, from=0.1, to=0.9)  {
+  
+  ## sub_sub is the time window where the Max and Min is expected
+  ## starting 50ms after valve opening)
+  avo <- 50 ## After Valve Opening
+  
+  sub_sub <- x[(avo+1) :1000]
+  
+  ## win is the window size before and after time of Max (or Min) to calculate the mean of the response
+  win <- 25
+  
+  ## Max is the time pint of maximum response
+  Max <- which(x==max(sub_sub))[1] + avo
+  
+  Max10_1 <- which(x <= max(x * from))
+  Max10_1 <- Max10_1[which(Max10_1 <= Max)]
+  Max10_1 <- Max10_1[length(Max10_1)]
+  
+  Max90_1 <- which(x <= max(x * to))
+  Max90_1 <- Max90_1[which(Max90_1 <= Max)]
+  Max90_1 <- Max90_1[length(Max90_1)]
+  
+  Max10_2 <- which(x <= max(x * from))
+  Max10_2 <- Max10_2[which(Max10_2 >= Max)]
+  Max10_2 <- Max10_2[1]
+  
+  Max90_2 <- which(x <= max(x * to))
+  Max90_2 <- Max90_2[which(Max90_2 >= Max)]
+  Max90_2 <- Max90_2[1]
+  
+  
+  Mean_max <- mean(x[(Max - win + 1) : (Max + win)])
+  
+  
+  ## Min is the time pint of minimum response
+  Min <- which(x==min(sub_sub))[1] + avo
+  
+  Min10_1 <- which(x >= min(x * from))
+  Min10_1 <- Min10_1[which(Min10_1 <= Min)]
+  Min10_1 <- Min10_1[length(Min10_1)]
+  
+  Min90_1 <- which(x >= min(x * to))
+  Min90_1 <- Min90_1[which(Min90_1 <= Min)]
+  Min90_1 <- Min90_1[length(Min90_1)]
+  
+  Min10_2 <- which(x >= min(x * from))
+  Min10_2 <- Min10_2[which(Min10_2 >= Min)]
+  Min10_2 <- Min10_2[1]
+  
+  Min90_2 <- which(x >= min(x * to))
+  Min90_2 <- Min90_2[which(Min90_2 >= Min)]
+  Min90_2 <- Min90_2[1]
+  
+  Mean_min <- mean(x[(Min - win + 1) : (Min + win)])
+  
+  
+  
+  
+  Timeto10Max <- Max10_1
+  Timeto10Min <- Min10_1
+  
+  
+  Max_rise <- Max90_1 - Max10_1
+  Max_fall <- Max10_2 - Max90_2
+  
+  Min_rise <- Min90_1 - Min10_1
+  Min_fall <- Min10_2 - Min90_2
+  
+  
+  if(length(Max10_1) < 1){
+    Max10_1 <- NA
+    Max_rise <- NA
+  }
+  
+  if(length(Max90_1) < 1){
+    Max90_1 <- NA
+    Max_rise <- NA
+  }
+  
+  if(length(Max90_2) < 1){
+    Max90_2 <- NA
+    Max_fall <- NA
+    
+  }
+  
+  if(length(Max10_2) < 1){
+    Max10_2 <- NA
+    Max_fall <- NA
+  }
+  
+  
+  if(length(Min10_1) < 1){
+    Min10_1 <- NA
+    Min_rise <- NA
+  }
+  
+  if(length(Min90_1) < 1){
+    Min90_1 <- NA
+    Min_rise <- NA
+  }
+  
+  if(length(Min90_2) < 1){
+    Min90_2 <- NA
+    Min_fall <- NA
+    
+  }
+  
+  if(length(Min10_2) < 1){
+    Min10_2 <- NA
+    Min_fall <- NA
+  }
+  
+  
+  
+  df <- c(Max10_1,
+          Max90_1,
+          Max, 
+          Max90_2,
+          Max10_2,
+          Max_rise,
+          Max_fall,
+          Min10_1,
+          Min90_1,
+          Min, 
+          Min90_2,
+          Min10_2,
+          Min_rise,
+          Min_fall,
+          Mean_max,
+          Mean_min)
+  
+  return(df)
+  
+}

Analysis/functions/utils.R

@@ -0,0 +1,4 @@
+
+# create the standard error function
+se <- function(x, ...) sd(x)/sqrt(length(x))
+

README.md

@@ -1,3 +1,42 @@
-# Stonefly_EAG
 
-Electroantennogram recordings from stonefly antennae
+## Data sets and analysis scripts underlying the manuscript 'Rapid evolution of olfactory degradation in recently flightless insects'
+
+**Hypothesis**
+
+Fast-moving animals need fast-acting sensory systems. Flying insects have exceptionally fast senses.
+For example, flighted insects can track the temporal structure of odour plumes at rates above 100 Hz, 
+and it has been hypothesized that this fast smelling capability is an adaptation for flying. 
+We test this hypothesis by comparing the olfactory acuity of sympatric flighted versus flightless 
+lineages within a wing-polymorphic stonefly species. 
+
+
+**Data**
+
+We did electroantennogram recordings from sympatric wing-reduced and full-winged stoneflies.
+We always measured 4 stonefly antennae simultanously (2 full-winged and 2 wing-reduced).
+As stimuli, we presented a set of different odorants with varying valve opening durations.
+Each odorant - pulse duration combination was presented 10 times and we calculated the median response
+trace over the ten odorant pulses of the same type.
+
+**Dataset 1 & 01_Analysis_Dataset1.R**
+
+Dataset 1 resembles the data for Fig 1, Fig 2, and Fig S1.
+
+Median traces of electroantennogram recordings from wing-reduced and full-winged stoneflies
+from five different sampling locations in the South Island of New Zealand. As comparison,
+we also recorded honeybee antennae. Recordings with the photoionisation device (PID) show the
+dynamics of odorant concentration changes.
+
+
+**Dataset 2 & 02_Analysis_Dataset2.R**
+
+Dataset 2 resembles the data for Fig S2.
+
+To test whether the signal strength of live antennae is dependent on their physical properties,
+we utilized the negative signals in dead antennae induced by propionic acid (physical property)
+and compared them to the response strength to 2-heptanone (because this odorant induced the
+strongest responses) of the same antennae, but when still alive.
+Dataset 2 contains electroantennogram recordings of nine additional stonefly antennae.
+Each of the recorded antennae was recorded twice with the same stimulus protocol. In between the 
+two protocols, the antennae were killed with hot water vapour.
+