Transcriptome dynamics predict thermotolerance in Caenorhabditis elegans

Background The detrimental effects of a short bout of stress can persist, and potentially turn lethal, long after the return to normal conditions. Thermotolerance, which is the capacity of an organism to withstand relatively extreme temperatures, is influenced by the response during stress exposure, as well as the recovery process afterwards. While heat-shock response mechanisms have been studied intensively, predicting thermal tolerance remains a challenge. Results Here, we use the nematode Caenorhabditis elegans to measure transcriptional resilience to heat stress and predict thermotolerance. Using high dimensionality reduction techniques in combination with genome-wide gene expression profiles collected in three high resolution time-series during control, heat stress and recovery conditions, we infer a quantitative scale capturing the extent of stress-induced transcriptome dynamics in a single value. This scale provides a basis for evaluating transcriptome resilience, defined here as the ability to depart from stress-expression dynamics during recovery. Independent replication across multiple highly divergent genotypes reveals that the transcriptional resilience parameter measured after a spike in temperature is quantitatively linked to long-term survival after heat stress. Conclusion Our findings imply that thermotolerance is an intrinsic property that pre-determines long term outcome of stress and can be predicted by the transcriptional resilience parameter. Inferring the transcriptional resilience parameters of higher organisms could aid in evaluating rehabilitation strategies after stresses such as disease and trauma.

relatively extreme temperatures, is influenced by the response during stress exposure, as well as the recovery 23 process afterwards. While heat-shock response mechanisms have been studied intensively, predicting 24 thermal tolerance remains a challenge. 25 Results: Here, we use the nematode Caenorhabditis elegans to measure transcriptional resilience to heat 26 stress and predict thermotolerance. Using high dimensionality reduction techniques in combination with 27 genome-wide gene expression profiles collected in three high resolution time-series during control, heat 28 stress and recovery conditions, we infer a quantitative scale capturing the extent of stress-induced 29 transcriptome dynamics in a single value. This scale provides a basis for evaluating transcriptome resilience, 30 defined here as the ability to depart from stress-expression dynamics during recovery. Independent 31 replication across multiple highly divergent genotypes reveals that the transcriptional resilience parameter 32 measured after a spike in temperature is quantitatively linked to long-term survival after heat stress. 33 Conclusion: Our findings imply that thermotolerance is an intrinsic property that pre-determines long term 34 outcome of stress and can be predicted by the transcriptional resilience parameter. Inferring the 35 transcriptional resilience parameters of higher organisms could aid in evaluating rehabilitation strategies 36 after stresses such as disease and trauma. heat stress). B, A subset of samples from heat-stress and control treatments were used for the inference of 101 the heat-stress axis, H, describing the gene expression dynamics during heat stress. C, Projection of the data 102 on this axis describes the dynamics of the response to heat stress. Notably, this is true also for the recovery 103 data that was not used to infer axis H. D, Projection of transcriptome data of the recovery process after 2, 3, 104 6 4, and 6 hours of heat-stress shows a decrease in recovery dynamics. E, Axis H also describes the 105 transcriptional heat-stress response for strains other than N2. 106 107 Heat-stress axis H reflects exposure duration, as well as recovery from heat stress 108 By projecting gene-expression profiles on the heat-stress axis H, each sample can be associated with 109 a value h. While only 18 samples were used to infer the axis, all 71 samples from all three time-series align 110 along the axis according to treatment and exposure duration ( showing that h is a quantitative measure of the transcriptional stress response. The value of h increased with 112 increasing heat-stress duration (Fig. 1C, red) until h started to saturate after long exposure (>4 hours). The 113 unperturbed worms had a constant value of h (Fig. 1C, blue), showing that we were able to successfully 114 remove the signal caused by developmental differences on gene expression. Strikingly, even though samples 115 collected during recovery were not used to determine the axis H, the gene expression during recovery from 116 a 2-hour heat stress was also well-explained with samples returning to the level of h typical of unperturbed 117 worms within about four hours (Fig. 1C, green). We concluded that h quantitatively reflects exposure 118 duration, as well as the time elapsed since the end of exposure. Note that although samples returned to the 119 pre-stress treatment level of h after recovery, this does not imply that recovered C. elegans populations are 120 transcriptionally indistinguishable from unperturbed ones (see SI section S2 and figure S7b). Therefore, 121 recovery was defined and measured here by the ability to depart from stress response dynamics. 122 So far, the results have shown that the transcriptional recovery process after a mild stress can be 123 followed over time using the heat-shock axis H. To exclude the possibility that H only captured time since 124 the end of the heat stress without biological meaning towards phenotypic recovery or resilience, we 125 expanded the dataset to include four additional time-series tracking the transcriptome recovery for 4 hours 126 following four different heat-stress intensities (2, 3, 4, and 6 hours at 35°C). The long-term effect of these 127 stress intensities on survival, reproduction, and mobility have been shown to range from mild after short (2 128 hour) exposure to 100% mortality within 24 hours after 6 hours at 35°C [8].

Heat-stress axis H reflects the average heat-stress response across multiple genotypes. 154
Next, we tested whether heat-stress axis H can also reflect the change in gene expression for 155 different genotypes. We used expression profiles of the strain CB4856 (Hawaii), which is genetically 156 distinct from N2, as well as 54 recombinant inbred lines (RILs) [9], which are genetic mosaics derived from 157 a cross between CB4856 and N2 [15,16]. We found that the heat-stress axis H successfully recapitulates 158 the average dynamic transcriptional response and resilience of this genetically diverse set of lines (Fig. 1E  159 and SI section S3). The robustness of the pattern across genotypes reflects the high degree of conservation 160 in transcriptional resilience. It should be noted that the RILs were not used here for the genetic mapping of 161 traits [17], but rather as a genotypic library. 162 163

Variation in stress resilience across genotypes is captured in a genetic heat-stress axis (GH) 164
We have shown that the heat-stress axis H, inferred using solely the isogenic strain N2, describes 165 the average conserved stress response of a library of highly divergent genotypes. On the other hand, there 166 is large natural variation in long-term effects of heat-stress exposure across genotypes, for example marked 167 by differences in the stressed transcriptome [9], survival rates [3, 18], and reproductive rates [3]. 168 Considering that variation is genotype dependent, it implies a difference in transcriptional resilience during 169 and/or after stress. Next, we ask if a single axis could also capture the natural variation in heat-stress 170 response across genotypes. Since genotypes differ in more traits than their transcriptional response to stress, 171 such as developmental timing and size, we needed to isolate stress-induced variation in expression levels 172 from other intrinsic differences in the transcriptome between genotypes. For this purpose, we used gene 173 expression data of RILs collected before and after two hours of heat stress [9]. Analogous to our approach 174 above in inferring the heat-stress axis H for N2 by removing developmental differences, we corrected the 175 heat-stress response of the RILs for their intrinsic gene expression differences in unperturbed conditions 176 (see SI section S3). We inferred a genetic heat-stress axis (GH) that isolates and describes the variability 177 across strains in their stress response. 178 10 The strength of relationship between the genetic axis GH and the environmental heat-stress axis H 179 measures the proportion of the variation of heat-stress response across RILs that is due to timing differences. 180 We found a positive correlation between the two axes (Spearman rho = 0.36, p = 0.01) implying that 181 different strains respond as if they were exposed to the heat stress for different durations. This was confirmed 182 by analysing a second set of heat-stressed gene expression profiles from a separate alternative panel of 183 inbred lines [19, 20](Introgression Lines, ILs; Spearman rho = 0.44, p = 8⋅10 -4 ) (Fig. 3). These results show 184 that the genetic differences also lead to difference in the timing or magnitude of the transcriptional response.

Transcriptional resilience on a short timescale is predictive of the variability in thermotolerance on a 202 longer timescale 203
Heat stress affects gene expression dynamics and resilience in the short term in a predictable way, which is 204 recapitulated by axes H and GH. On the other hand, in the long run, heat stress also affects developmental 205 speed, aging, behaviour, and vitality -for instance by drastically reducing lifespan. We set out to explore 206 how the variability in gene expression dynamics following heat stress on a short timescale is predictive of 207 variability in thermotolerance measured on a longer timescale. Thermotolerance in C. elegans can be 208 recorded by its survival rates. Therefore, in a parallel experiment, we collected lifespan data of over 200 209 different RILs and ILs with and without exposure to heat stress. While two hours at 35 o C are sufficient to 210 induce a strong transcriptional response, previous experiments have shown that overall lifespan is not 211 necessarily shortened at this intensity [8]. Therefore, we increased the exposure duration to four hours at 212 35 o C for lifespan measurements as this duration is known to affect lifespan [3, 8], allowing us to make a 213 better estimate of difference in thermotolerance across genotypes As expected, both RILs and ILs show high 214 variability in their lifespan after heat stress and in control conditions. The average lifespan following a heat 215 stress is always lower than what was found for the same strain when unperturbed ( Fig. 4 and Figure S14). 216 Next, we compared the effect of heat stress on the lifespan of different RILs with the difference in 217 transcriptional resilience, measured by projecting the recovery data of the RILs on the genetic heat-stress 218 axis GH. Figure 4 shows that the ability of different strains to recover from heat stress is predictive of 219 thermotolerance (Spearman rho = -0.41, p = 0.02; Fig 4). In order to test the robustness of this result, we 220 also performed the same analysis on ILs, which are genetically mostly derived from one strain (N2) and 221 were not used to infer the axis. In this case, we also found a significant correlation (Spearman rho = -0.46, 222 p = 10 -3 ), implying that the connection between the ability to recover and lifespan was robust across different 223 inbred line panels. The projection of the heat-stress data onto GH (which is related to the speed at which 224 worms react to heat stress) was not robustly correlated with lifespan (see SI section S5), showing that 225 resilience measured based on recovery data was more directly linked to tolerance. 226 systemic modus operandi based on using genome-wide gene expression profiles. We conclude that a 238 relatively simple axis can measure stress resilience of a dynamic transcriptome in a single quantitative 239 variable and describes the capacity of an organism to recover from heat stress. Our findings show that natural 240 variation in transcriptome resilience after mild stress exposure is predictive of thermotolerance across a 241 14 diverse set of genotypes in C. elegans. The results imply that thermotolerance is an intrinsic trait that largely 242 pre-determines long-term effects of heat-stress exposure. Operationalizing the concept of resilience in 243 higher organisms, like mammals, has been difficult because it includes a range of many different phenotypic 244 traits [21]. Our approach represents a novel way in understanding resilience in a living system, and we show 245 how the inherent complexity of stress recovery can be exploited to predict the chance of survival. We 246 anticipate that our finding will accelerate progress in the study of resilience of complex living systems, added to each sample. The samples were then incubated for 10 minutes at 65°C and 1000 rpm in a 310 Thermomixer (Eppendorf, Hamburg, Germany) before cooling on ice for 1 minute. At this point, the 311 samples were pipetted into the cartridges resuming with the standard protocol. 312

Sample preparation and scanning 313
For cDNA synthesis, labelling and the hybridization reaction, the 'Two-Color Microarray-Based Gene 314 Expression Analysis; Low Input Quick Amp Labeling' -protocol, version 6.0 from Agilent (Agilent 315 Technologies, Santa Clara, CA, USA) was followed, starting at step 5. The Agilent C. elegans (V2) Gene 316 Expression Microarray 4X44K slides were used in combination with an Agilent High Resolution C Scanner 317 using the recommended settings. Data was extracted with the Agilent Feature Extraction Software (version 318 10.7.1.1) following the manufacturers' guidelines. 319

Data normalization 320
Microarray data were normalized using a within array normalization using a standard function of the R 321 package limma (using "loess" method) [24]. 322 The code needed to replicate all the results presented here can be found at 323 https://github.com/jacopogrilli/resiliencevitality.git 324