Important Announcement
PubHTML5 Scheduled Server Maintenance on (GMT) Sunday, June 26th, 2:00 am - 8:00 am.
PubHTML5 site will be inoperative during the times indicated!

Home Explore Science


Published by pochitaem2021, 2022-07-07 15:01:45

Description: Science


Read the Text Version

RESEARCH | REPORT Fig. 3. cis- and trans-regulatory mutants have AAbsolute Log2 fold change Log 2 fold changeB GPD2 expression C 300 distinct effects on expression of genes (% Wild type) downstream of TDH3. (A) Violin plots showing 0.81 0.50 0.48 0.44 0.26 2 200 absolute log2 fold changes in the cis-regulatory 6 1 mutants for the 140 genes identified as downstream 0 of TDH3, as they were significantly differentially 4 expressed (DE) in the TDH3 null mutant. Median 2 100 absolute log2 fold changes are shown above each −1 0 50 100 150 TDH3 expression plot and indicated with black dots. (B) Log2 fold 0 20 50 85 135 −2 0 50 100 (% Wild type) changes in cis-regulatory mutants are shown for the 0 TDH3 expression TDH3 expression same 140 genes downstream of TDH3 connected by (% Wild type) (% Wild type) line segments. Genes whose expression was not D E significantly linearly correlated with TDH3 expres- Number of downstream genes DE sion are shown in gray. (C) Expression of TDH3 and 140 genes significantly DE in TDH3 null GCR1162 GPD2 is shown for the cis-regulatory mutants GCR1339 RAP1238 RAP154 GCR1281 ADE5 CAF40 ADE6 FTR1 GCR1037 TYE7 MRN1 NAM7 HXK2 CYC8 SSN2 BRE2 SDS23 ADE4 PRE7 ATP23 WWM1 ADE2 RIM8 TUP1 GCR1241 CIA2 FRA1 IRA2 MOD5 RAP1484 RAP1357 CCC2 TRA1 NAR1 TDH3 expression Z-score 100 1.0 (circles) and the WT strain (triangle). The best fit linear regression line and 95% CI (gray shaded 50 0.5 area) are shown. (D) Heatmap of the 140 genes downstream of TDH3 showing log2 fold changes in 0 0.0 all cis- and trans-regulatory mutants; genes are rows and mutants—named after the gene bearing the mutation in that mutant—are columns. Color Trans−Rreguullaattoorryymmuutatannt tstsratrianin intensity is scaled by row (by gene) and represents z scores. Mutants are hierarchically clustered as F (Name(Gdeanfetebregaerinneg bmeuataritniognm) utation) shown by the dendrogram. (E) The number of downstream genes that are also significantly differentially expressed in each trans-regulatory Pleiotropic Pleiotropic effects mutant is shown as a column relating to the effects left y axis. The expression level of TDH3 in that Regulatory mutant strain mutant is shown as a green point relating to the (Named after gene bearing mutation) right y axis. (F) Schematic shows that trans-regulatory G H mutants can have pleiotropic effects on genes downstream of TDH3 not mediated by their effect 600 Pleiotropic expression effect (residual) on TDH3 as well as pleiotropic effects on genes in GPD2 expression (% Wild type) parallel to TDH3. (G) Expression of GPD2 and TDH3 is 400 Pleiotropic shown for the trans-regulatory mutants, with the 200 expression expression and linear regression from cis-regulatory effect mutants from (C) included in orange. Red lines delineate 95% prediction intervals for the cis-regulatory mutant relationship between TDH3 and GPD2. Effects 0 of trans-regulatory mutants on GPD2 that are not explained by their effect on TDH3 (pleiotropic 0 50 100 150 TDH3 expression (% Wild type) expression effects, example illustrated by black line) Genes (n = 122) are colored blue. (H) For each of the 122 genes downstream of TDH3 with a significant linear relationship to TDH3 expression in the cis-regulatory mutants (x axis), the pleiotropic expression effect (y axis), as illustrated in (G), is shown. For each gene, each point represents a different trans-regulatory mutant. Genes are ordered on the x axis by median residual (black points). Blue points indicate mutants with significant pleiotropic expression effects, whereas gray points indicate nonsignificant pleiotropic expression effects. As shown in Fig. 1 trans-regulatory mutants these genes, 49 (35%) were underexpressed from nonmetabolic functions of TDH3p in are not only assumed to have effects on expres- in the null mutant relative to the wild-type sion of genes downstream of the focal gene strain and 91 (65%) were overexpressed (data processes such as transcriptional silencing and similar to that of cis-regulatory mutations, but S3). This gene set was significantly enriched for rDNA recombination (25). are also assumed have additional pleiotropic genes encoding proteins involved in glycolytic effects on expression of other genes in parallel. processes (fig. S4), suggesting that many ex- The median absolute log2 fold expression To test this assumption, we sought to sepa- pression changes observed in the TDH3 null changes observed for this set of 140 genes rately analyze the effects of trans-regulatory mutant might be due to a homeostatic re- downstream of TDH3 decreased monotoni- mutants on expression of genes downstream sponse of the cells to maintain metabolism in cally as TDH3 expression approached wild of and in parallel to TDH3. To identify the the absence of the TDH3p enzymatic activity type and was smallest in the cis-regulatory set of genes downstream of TDH3, we iden- involved in glycolysis and gluconeogenesis mutant overexpressing TDH3 (Fig. 3A). Out tified genes whose expression was signifi- (24). These downstream genes were also en- of these 140 genes, 122 (87%) showed a sig- cantly altered in the TDH3 null mutant. We riched for genes associated with the gene on- nificant linear relationship with TDH3 ex- found 140 such genes (10% FDR, P value = tology terms DNA biosynthesis, integration, pression in the five cis-regulatory mutants 0.002, data S3), excluding TDH3 itself. Of and transposition (fig. S4), which might result (10% FDR, P value = 0.09; fig. S5A), with 44 positively correlated and 78 negatively cor- related (Fig. 3B and fig. S5B). For example, Vande Zande et al., Science 377, 105–109 (2022) 1 July 2022 3 of 5

RESEARCH | REPORT Fig. 4. Larger pleiotropic expression effects of A Z-score B trans-regulatory mutants correlate with nega- tive fitness effects. (A) A heatmap shows log2 fold Number of r2 = 0.95 changes for all genes not downstream of TDH3 as downstream genes DE estimated by DESeq2, in which rows are genes and 75 columns are mutants. Color intensity is scaled by row (by gene) and corresponds to z scores. 50 Hierarchical clustering of mutants is shown by the dendrogram above. Three of the four trans- 25 regulatory mutants bearing mutations in the adenine biosynthesis pathway cluster together and are marked All genes not downstream of TDH3 0 by a blue dot and line. The cis-regulatory mutant 0 1000 2000 3000 overexpressing TDH3 clusters with IRA2 (green), whereas the cis-regulatory mutants with reduced Genes DE in parallel to TDH3 TDH3 expression cluster together (orange). A large cluster of trans-regulatory mutants with small CRelative fitness IRA2 effects across the genome are marked by a purple 1.1 NAM7 dot and line. (B) For all trans-regulatory mutants, 0.9 the number of downstream genes differentially expressed is positively correlated with the number 0.7 of genes differentially expressed in parallel to TDH3 in that mutant. (C) For both cis- (orange) and 0.5 trans- (blue) regulatory mutants, the total number of differentially expressed genes is a strong predictor 0.3 R2 = 0.63 of relative fitness. A linear model is shown with gray shading representing 95% CI and two outliers, 0 1000 2000 3000 IRA2 and NAM7 are indicated. (D) For all trans- regulatory mutants with fitness data (i.e., excluding Number of DEGs the two flocculant strains), the pleiotropic fitness effect, as defined in Fig. 2B, is plotted vs the number D of genes differentially expressed in parallel to TDH3 in that trans-regulatory mutant. A linear regression is 0.0 shown in black with 95% CI in gray. Pleiotropic fitness effect −0.2 Regulatory mutant strains −0.4 (Named after gene bearing mutation) R2 = 0.61 −0.6 0 1000 2000 3000 Genes DE in parallel to TDH3 the GPD2 gene—which encodes an enzyme TDH3 that are not explained by their effect on to the pleiotropic effects of trans-regulatory two steps away from TDH3 in the metabolic TDH3 (Fig. 3F). mutants in parallel to TDH3. A heatmap of network—increased linearly when TDH3 ex- expression with hierarchical clustering for pression was decreased by cis-regulatory mu- To further explore this possibility, we used the 5806 genes not classified as downstream tations (Fig. 3C). the linear models fitted to the cis-regulatory of TDH3 shows these other effects (Fig. 4A). mutant data to predict the change in expres- Of the trans-regulatory mutants, 14 showed A heatmap showing expression with hierar- sion expected for each downstream gene as a minimal effects on expression of these genes chical clustering for these 140 genes downstream result of the effect of the trans-regulatory mu- (purple cluster in Fig. 4A). The four hypo- of TDH3 in the 40 cis- and trans-regulatory tant on TDH3 expression alone. Deviations morphic cis-regulatory mutants clustered to- mutants also visually showed the correlation from these expectations indicate pleiotropic gether (orange cluster in Fig. 4A) and showed with TDH3 expression in the cis-regulatory effects of trans-regulatory mutants on the additional effects that appeared to scale with mutants (orange cluster in Fig. 3D). The cis- expression of genes downstream of TDH3. TDH3 expression level (consistent with fig. regulatory mutant overexpressing TDH3 (135% For example, 6 of the 35 trans-regulatory mu- S5), suggesting that there might be more genes TDH3) had opposing effects on expression of tants showed evidence of a pleiotropic effect downstream of TDH3 than were identified as these genes that caused it to cluster separately, on expression of GPD2, as indicated by a change differentially expressed in the TDH3 null mu- with expression most similar to two trans- in GPD2 expression further than one standard tant with the statistical thresholds used. The regulatory mutants (bearing mutations in error outside of the 95% prediction interval trans-regulatory mutants with larger effects NAR1 and IRA2) that also caused overexpres- for the expression change expected based on tended to cluster according to related pheno- sion of TDH3 (green cluster in Fig. 3D). The cis-regulatory mutants (Fig. 3G). Such pleiotropic type or function, such as the two flocculant other 33 trans-regulatory mutants had more effects were observed for at least one trans- strains (bearing mutations in CYC8 and SSN2), distinct patterns of expression for these genes regulatory mutant for 111 of the 122 genes genes involved in adenine biosynthesis (ADE4, (Fig. 3D), with some mutants (e.g., RAP1484 downstream of TDH3, with the magnitude of ADE5, and ADE6), and large-impact mutations and IRA2) showing significant changes in ex- the pleiotropic effects (measured as residuals in GCR1. Finally, the two RAP1 mutants with the pression for many genes despite minimal from the gene-specific regression models based largest impacts on TDH3 expression (RAP154 effects on TDH3 expression (Fig. 3E). These on the cis-regulatory mutants) varying among and RAP1238) showed the most different ex- observations suggest that the pleiotropic ef- trans-regulatory mutants and genes (Fig. 3H). pression patterns from the other mutants and fects of trans-regulatory mutants include ef- each other. fects on expression of genes downstream of These pleiotropic effects on expression of genes downstream of TDH3 are in addition Vande Zande et al., Science 377, 105–109 (2022) 1 July 2022 4 of 5

RESEARCH | REPORT Comparing the impacts of trans-regulatory regulatory mutations but also provide a more 25. A. E. Ringel et al., PLOS Genet. 9, e1003871 (2013). mutations on gene expression downstream of 26. D. E. Featherstone, K. Broadie, BioEssays 24, 267–274 and in parallel to TDH3 showed that mutants nuanced view of these differences at the level affecting expression of many genes in parallel (2002). to TDH3 tended to also affect expression of of specific mutations and show that the pleio- 27. P. V. Zande, M. S. Hill, P. J. Wittkopp, Zenodo (2022); many genes downstream of TDH3 (Fig. 4B). This observation indicates that some trans- tropic effects of trans-regulatory mutations regulatory mutants had larger or smaller over- all effects on the transcriptome. However, the might often affect expression of genes both in ACKNOWLEDGMENTS impact of a mutation on TDH3 expression was not a strong predictor for this overall effect: parallel to and downstream of the focal gene. We thank B. Metzger, F. Duveau, M. Siddiq, H. Ertl, A. Redgrave, Some mutants with minimal effects on TDH3 and other members of the Wittkopp laboratory for helpful expression altered expression of thousands of Studies such as this one that empirically de- discussions and feedback on drafts of this manuscript. We also other genes whereas other mutants changed thank B.Z. He (University of Iowa) for sharing the protocol used for TDH3 expression up to 30% but altered ex- termine and compare the properties of different RNA-seq. The University of Michigan Advanced Genomics Core, pression of <100 other genes (Fig. 2E). The Center for Statistical Consultation and Research, and High- overall number of genes whose expression types of mutations affecting gene expression Performance Computing Cluster provided services used to conduct was affected by a trans-regulatory mutation this work. Funding: This work was funded by National Institutes was a strong predictor of the relative fitness are critical for understanding how regulatory of Health grants R35GM118073 and R01GM108826 to P.J.W. of the mutant [Fig. 4C; consistent with (26)], and T32GM07544 to P.V.Z. Author contributions: P.V.Z. and although two trans-regulatory mutants (IRA2 systems evolve. P.J.W. conceived and designed the study. P.V.Z. performed all and NAM7) were notable exceptions (Fig. 4C). experiments, data collection, and data analysis. M.S.H. contributed Directly comparing the pleiotropic effects of REFERENCES AND NOTES to the design of the data analysis, especially in the analysis of trans-regulatory mutations inferred for fitness previously published gene deletion data. P.V.Z. and P.J.W. wrote and expression of genes in parallel to TDH3 1. D. L. Stern, V. Orgogozo, Evolution 62, 2155–2177 (2008). the manuscript, and M.S.H. provided comments on later drafts. showed a very similar relationship (Fig. 4D) 2. J. D. Coolon, C. J. McManus, K. R. Stevenson, B. R. Graveley, P.V.Z. prepared all figures, code, and supplemental files, with input because the cis-regulatory mutants had much from P.J.W. Competing interests: Authors declare that they smaller effects on both fitness and expression P. J. Wittkopp, Genome Res. 24, 797–808 (2014). have no competing interests. Data and materials availability: of other genes than trans-regulatory mutants 3. B. P. H. Metzger, P. J. Wittkopp, J. D. Coolon, Genome Biol. Evol. RNA-seq data is available at GEO accession GSE175398. Code used with similar effects on TDH3 expression (Fig. for data analysis, as well as supporting data files, are available 2, B and E). 9, 843–854 (2017). from GitHub at 4. D. Gokhman et al., Nat. Genet. 53, 467–476 (2021). with the version of record at time of publication available at In summary, we developed a framework to 5. S. A. Signor, S. V. Nuzhdin, Trends Genet. 34, 532–544 (2018). Zenodo (27). All other data is included with the manuscript or decompose the effects of trans-regulatory mu- 6. M. S. Hill, P. Vande Zande, P. J. Wittkopp, Nat. Rev. Genet. 22, supplementary material. Mutant strains of S. cerevisiae available tations attributable to their effect on expres- upon request. License information: Copyright © 2022 the sion of a focal gene from their effect on other 203–215 (2021). authors, some rights reserved; exclusive licensee American genes in the genome. These data support the 7. G. A. Wray, Nat. Rev. Genet. 8, 206–216 (2007). Association for the Advancement of Science. No claim to original widely held but rarely tested assumptions that 8. P. J. Wittkopp, B. K. Haerum, A. G. Clark, Nat. Genet. 40, US government works. trans-regulatory mutations tend to be more science-licenses-journal-article-reuse pleiotropic and more deleterious than cis- 346–350 (2008). 9. B. Schaefke et al., Mol. Biol. Evol. 30, 2121–2133 (2013). SUPPLEMENTARY MATERIALS 10. M. Kimura, T. Ohta, Proc. Natl. Acad. Sci. U.S.A. 71, 2848–2852 (1974). Materials and Methods 11. A. B. Paaby, M. V. Rockman, Trends Genet. 29, 66–73 Figs. S1 to S9 References (28–33) (2013). MDAR Reproducibility Checklist 12. J. Zhang, G. P. Wagner, Trends Genet. 29, 383–384 (2013). Data S1 to S3 13. A. B. Paaby, M. V. Rockman, Trends Genet. 29, 384 (2013). 14. B. P. H. Metzger, D. C. Yuan, J. D. Gruber, F. Duveau, View/request a protocol for this paper from Bio-protocol. P. J. Wittkopp, Nature 521, 344–347 (2015). Submitted 28 May 2021; resubmitted 7 February 2022 15. B. P. H. Metzger et al., Mol. Biol. Evol. 33, 1131–1146 (2016). Accepted 20 May 2022 16. F. Duveau et al., eLife 10, e67806 (2021). 10.1126/science.abj7185 17. F. Duveau, W. Toubiana, P. J. Wittkopp, Mol. Biol. Evol. 34, 2908–2912 (2017). 18. F. Duveau et al., eLife 7, e37272 (2018). 19. A. L. Sezmis, M. E. Malerba, D. J. Marshall, M. J. McDonald, J. Mol. Evol. 86, 111–117 (2018). 20. S. Venkataram et al., Cell 166, 1585–1596.e22 (2016). 21. Y. Li et al., Curr. Biol. 28, 515–525.e6 (2018). 22. P. Kemmeren et al., Cell 157, 740–752 (2014). 23. C. J. Maclean et al., Mol. Biol. Evol. 34, 2486–2502 (2017). 24. L. McAlister, M. J. Holland, J. Biol. Chem. 260, 15019–15027 (1985). Vande Zande et al., Science 377, 105–109 (2022) 1 July 2022 5 of 5

RESEARCH BIOMEDICINE neural signals are well defined in select ana- tomical regions, (ii) nerves carrying aberrant Soft, bioresorbable coolers for reversible conduction neural signals are already isolated, and (iii) a block of peripheral nerves need for opioid therapy exists after operation (Fig. 1A). Pain management after amputations, Jonathan T. Reeder1,2,3†‡*, Zhaoqian Xie4,5†, Quansan Yang3,6†, Min-Ho Seo2,3,7†, Ying Yan8, nerve grafts, or spinal decompression surgeries Yujun Deng9, Katherine R. Jinkins3, Siddharth R. Krishnan2,3, Claire Liu3,10, Shannon McKay10, represent examples. Here, the relevant nerves Emily Patnaude10, Alexandra Johnson10, Zichen Zhao4,5, Moon Joo Kim11§, Yameng Xu12, Ivy Huang2,3, are already isolated and identified; thus, the Raudel Avila6, Christopher Felicelli13, Emily Ray14, Xu Guo4,5, Wilson Z. Ray8,14, application of the cuff would be straightfor- Yonggang Huang2,3,6,15, Matthew R. MacEwan8,14, John A. Rogers2,3,6,10,16,17,18* ward to integrate into the clinical workflow. In this scheme, implantation of a bioresorb- Implantable devices capable of targeted and reversible blocking of peripheral nerve activity may provide able cooler around the nerve that innervates alternatives to opioids for treating pain. Local cooling represents an attractive means for on-demand damaged tissue enables reversible elimination elimination of pain signals, but traditional technologies are limited by rigid, bulky form factors; imprecise of neural activity and pain signals through cooling; and requirements for extraction surgeries. Here, we introduce soft, bioresorbable, microfluidic the focused application of cooling power (Fig. devices that enable delivery of focused, minimally invasive cooling power at arbitrary depths in 1B). Construction from water-soluble materials living tissues with real-time temperature feedback control. Construction with water-soluble, biocompatible leads naturally to dissolution of the cooling materials leads to dissolution and bioresorption as a mechanism to eliminate unnecessary device system after the completion of the healing load and risk to the patient without additional surgeries. Multiweek in vivo trials demonstrate the process and obviates the need for an extraction ability to rapidly and precisely cool peripheral nerves to provide local, on-demand analgesia in rat models surgery (Fig. 1C), in a manner conceptually for neuropathic pain. similar to that of other recently reported classes of bioresorbable sensors and therapeutic T he high therapeutic efficacy of opioids avoid side effects associated with opioids and systems to monitor and accelerate wound has led to their widespread use despite other analgesics (2, 3). The controlled input healing or recovery processes (18, 19). of electrical (4), pharmacological (5), optical increasing rates of addiction and opioid- (6), mechanical (7), or thermal (8, 9) stimuli We present concepts and device designs for to neural tissue can lead to local and rever- soft, bioresorbable peripheral nerve-cooling related deaths due to overdose (1). The sible neural blocking. and temperature-sensing units that are engi- neered to reversibly block pain signals with a increasing societal burden caused by Of particular interest for the work here is targeted cooling stimulus over a finite time that temporal measures of metabolic, electro- period matched to patient needs. The technol- opiate misuse motivates the development of genic, and ionic activity in neural tissue all ogy consists of a hybrid microfluidic and elec- exhibit a negative temperature dependence (10). tronic system for cooling and simultaneously localized, nonopioid, and nonaddictive pain- Local cooling of peripheral nerves decreases measuring the temperature of a peripheral conduction velocity and signal amplitude of nerve. The elastomeric nature of the micro- management techniques. Miniaturized im- neural activity (11). Blocking of transmission fluidic system and the serpentine shapes of of compound action potentials in mammalian the electrical interconnects yield soft, stretch- plantable devices that eliminate pain signals nerves typically occurs below 15°C (12), but able mechanics at the device level (Fig. 1D) this threshold can be temporarily increased to with effective moduli not substantially higher locally in peripheral nerves suggest a poten- near room temperature by a brief heating than those of peripheral nerves [rat sciatic period preceding the cooling period (13, 14). nerve elastic modulus is 0.6 MPa (20)]. These tial role for engineering-based treatments that Cooling applied to peripheral nerves is a electronic and microfluidic systems termi- promising approach for blocking pain sig- nate in a cuff structure with a diameter 1Knight Campus for Accelerating Scientific Impact, University nals because it is nonaddictive, is rapidly matched to the rat sciatic nerve (1.5 mm) to of Oregon, Eugene, OR, USA. 2Department of Materials reversible, can be applied locally, avoids any provide an intimate mechanical and thermal Science and Engineering, Northwestern University, Evanston, onset response, and allows for simultaneous interface to the nerve, without the need for IL, USA. 3Querrey Simpson Institute for Bioelectronics, electrical interrogation of the blocked nerve sutures (Fig. 1E). The curled geometry of the Northwestern University, Evanston, IL, USA. 4State Key (15). Analgesic nerve cooling requires spatio- cuff and its elastic nature enables manual Laboratory of Structural Analysis for Industrial Equipment, temporally precise control of temperature to unrolling and soft clasping of the nerve. An Department of Engineering Mechanics, Dalian University of maximize desired outcomes and to minimize essential defining characteristic of this Technology, Dalian, China. 5Ningbo Institute of Dalian the chance of cooling-induced tissue damage system is that it is constructed entirely with University of Technology, Ningbo, China. 6Department of (16, 17). Current approaches for nerve cooling water-soluble constituent materials that con- Mechanical Engineering, Northwestern University, Evanston, rely on rigid, bulky systems that prevent the trollably dissolve to biocompatible end pro- IL, USA. 7School of Biomedical Convergence Engineering, use of local cooling as a practical approach ducts in the biofluids that are contained in College of Information and Biomedical Engineering, Pusan for peripheral nerve pain management. subcutaneous tissues. Figure 1F shows devices National University, Busan, Republic of Korea. 8Department wrapped around a silicone phantom nerve and of Neurological Surgery, Washington University School of An implantable device that provides on- submerged in phosphate-buffered saline (PBS) Medicine, St. Louis, MO, USA. 9State Key Laboratory of demand local analgesia over a defined time- (pH 7.4) at 75°C, as an accelerated aging test. The Mechanical System and Vibration, Shanghai Jiao Tong line followed by subsequent dissolution and results show that the materials largely dissolve University, Shanghai, China. 10Department of Biomedical bioresorption would represent a qualitative within 20 days and that elimination of residues Engineering, Northwestern University, Evanston, IL, USA. advance in pain-management techniques. We occurs after 50 days under these conditions. 11Department of Chemical Engineering, Northwestern specifically envision clinical use cases for non- University, Evanston, IL, USA. 12The Institute of Materials opioid management of postoperative acute pain An illustration of the layers of the hybrid Science and Engineering, Washington University, St. Louis, signals in peripheral nerves where (i) aberrant microfluidic-electronic device is shown in Fig. 2A. MO, USA. 13Department of Pathology, Northwestern A bioresorbable elastomer, poly(octanediol cit- University, Evanston, IL, USA. 14Department of Biomedical rate) (POC), forms the microfluidic system. POC Engineering, Washington University, St. Louis, MO, USA. 15Departments of Civil and Environmental Engineering, Northwestern University, Evanston, IL, USA. 16Department of Chemistry, Northwestern University, Evanston, IL, USA. 17Department of Electrical Engineering and Computer Science, Northwestern University, Evanston, IL, USA. 18Department of Neurological Surgery, Feinberg School of Medicine, Northwestern University, Evanston, IL, USA. *Corresponding author. Email: [email protected] (J.T.R.); [email protected] (J.A.R.) †These authors contributed equally to this work. ‡Present address: Science Corporation, Alameda, CA, USA. §Present address: LG Innotek, Seoul, South Korea. Reeder et al., Science 377, 109–115 (2022) 1 July 2022 1 of 7

RESEARCH | REPORT Fig. 1. Soft, bioresorbable, evaporative microfluidic coolers for an nerve-cooling system that features elastomeric interconnects and a terminal on-demand nerve block. (A) Transmission of postoperative acute pain signals cuff structure. (E) A soft, curled, nerve-clasping structure provides secure through peripheral nerves. (B) Local cooling of peripheral nerves provides attachment to nerves without sutures. (F) Construction from water-soluble an on-demand nerve block. (C) Treatment termination and cooler bioresorption materials enables device dissolution and subsequent bioresorption after completion of the healing process. (D) A soft, bioresorbable (PBS at 75°C, pH 7.4). exhibits an elastic modulus of 2.8 MPa (21), system includes transcutaneous colinear inter- the microfluidic system, with the temperature- controllable rates of degradation by means of connects that deliver liquid coolant [perfluoro- sensing element at the distal end of the device. surface erosion (22), and a demonstrated com- pentane (PFP)] and dry N2 to a serpentine The simultaneous initiation of PFP and N2 patibility with nerves (23). Details of the steps evaporation chamber (Fig. 2B) in a completely flows into this structure prompts evaporation for fabrication are provided in fig. S1 and the sealed system that provides fluidic access at of PFP at the microfluidic junction between the the ends. The electronic layer lies coplanar with PFP and N2 channels and along the serpentine supplementary materials. The microfluidic Reeder et al., Science 377, 109–115 (2022) 1 July 2022 2 of 7

RESEARCH | REPORT Fig. 2. Evaporative microfluidic cooling and temperature sensing system. (A) Exploded device render PFP molar flow ratios (XPFP = 0.1), the PFP showing layers of the microfluidic and electronic systems. (B) Schematic illustration of the microfluidic and fully evaporates after passing through three electronic systems design. (C) A flow of dry N2 prompts the complete evaporation of PFP within three serpentines, with marginal liquid PFP build- serpentine turns at low molar ratio (XPFP = 0.1) inside the serpentine microfluidic channel. (D) Increasing up at the corners of the microchannels (Fig. the PFP molar fraction (XPFP = 0.5) extends the evaporation throughout the length of the serpentine 2C). At high molar flow rates (XPFP = 0.5), PFP microfluidic channel. (E) Evaporative microfluidic cooling prompts reduction of the temperature of the device proceeds through annular flow and passes surface to −20°C in ambient conditions, as shown by thermal imaging. (F) A bioresorbable electronic along the sidewalls of the microchannels (Fig. system for providing nerve-temperature feedback consists of a magnesium resistance-based temperature 2D). This phase change prompts the temper- sensor on a cellulose acetate (CA) substrate. ature of a device in a planar, uncurled config- uration to drop to −20°C within 2 min after chamber. PFP, which boils near room temper- ultrasound contrast agent (25), and in ther- initializing flow in ambient, room-temperature ature (28° to 30°C), is bioinert and compatible apeutic hypothermia (26). The mass flow rates conditions (Fig. 2E). The cooled area of the with nonfluorinated elastomers. PFP is clin- of the PFP and N2 and the geometry of the device is confined predominately to the ser- ically approved as a propellant in pressurized evaporation chamber govern the magnitude pentine evaporation chamber, as governed metered-dose inhalers (24), as an intravenous by the microfluidic channel design and fluid and localization of the cooling effect. At low flow rates. A serpentine magnesium trace with a width and length of 25 mm and 72 mm, respec- tively, provides temperature feedback through the temperature coefficient of resistance of Mg (Fig. 2F). Details for the temperature-sensing system are provided in fig. S2 and the supple- mentary materials. Figure 3 summarizes quantitative results of measurements of the efficacy of nerve cooling of a phantom nerve structure and an inte- grated thermocouple inside a 37°C hydrogel tissue mimic (fig. S3). An enabling property of PFP is that it can be superheated beyond its boiling point, to temperatures as high as 60°C (27), because of the energy barrier re- quired for homogeneous nucleation (28). This feature prevents premature boiling of the PFP as it passes along the smooth interior of the microfluidic channels to the evaporation cham- ber. The temperatures of the PFP and N2 flows equilibrate at 37°C within 90 mm of entering the 37°C environment (fig. S4). A condenser placed distal to the cooling cuff enables recap- ture of 90% of the evaporated PFP (fig. S5). The delivery of N2 in a pressure-driven mode mitigates transient pressure spikes that can occur with multiphase microfluidic flows (fig. S6). Holding the N2 pressure constant at values ranging from 7 to 90 kPa and sweeping the PFP flow rates from 0 to 900 ml/min yields data that reveal the dependence of the N2 flow rate on PFP flow rate (Fig. 3A). Experiments based on systematically sweeping the PFP and N2 flow rates over the same range as in Fig. 3A reveal the effect of molar flow rates on resultant nerve temperature (Fig. 3B). The minimum nerve temperature is −1.4°C, which is achieved at a molar PFP fraction of 0.13 and a N2 pressure of 90 kPa (Fig. 3C). Additional details regarding conversion to molar flow rates are provided in fig. S7. Figure 3D demon- strates the effect of PFP flow rate on nerve- temperature cooling rate. The maximum cooling rate of 3°C/s occurs with the 300 ml/min case. Arbitrary cooling and rewarming profiles, as governed by the thermal properties of the surrounding media, follow from controlled increments or decrements of the PFP flow rate (Fig. 3E). Experiments indicate sustained Reeder et al., Science 377, 109–115 (2022) 1 July 2022 3 of 7

RESEARCH | REPORT Fig. 3. In vitro studies of nerve-cooling efficacy. (A) The N2 flow rate nerve over a 20 min period. (F) Evaporative microfluidic cooling enables precise depends on N2 pressure (PN2) and PFP flow rate. QN2, N2 flow rate; QPFP, and stable nerve cooling for more than 15 min. (G) The effect of a convective PFP flow rate. (B) Phantom-nerve temperature depends on PFP and N2 flow environment on the ability to cool a phantom nerve, as shown by a comparison of rates. (C) A PFP molar fraction of 0.13 produces the lowest nerve temperature 37°C water and hydrogel baths. (H) Physiologically relevant rat sciatic nerve blood flow rates induce an increase in nerve temperature during cooling from 3° to 5°C in a (−1.4°C). Qtotal, Total flow rate. (D) Phantom-nerve cooling rates greater than phantom nerve. Error bars represent standard deviation of three trials. (I) POC 3°C/s are prompted by initiation of a 300 ml/min flow of PFP. (E) Systematic microfluidics provide precise nerve cooling for 21 days in vitro (PBS, 37°C, pH 7.4). reduction in PFP flow rate from 300 ml/min enables controlled rewarming of the and consistent nerve cooling to 3.0°C for a convective thermal transfer. Comparisons of (Fig. 3H), the temperature of the nerve increases 15-min interval in Fig. 3F [minimum tem- measurements in a conduction-only environ- by 2.0°C (from 3.5° to 5.5°C). Experimental perature (Tmin) = 2.1°C; maximum temper- ment (hydrogel, 37°C) to those that include details are provided in fig. S8. A thermofluidic ature (Tmax) = 3.5°C]. convective effects (water, 37°C) highlight these model (fig. S9) predicts the resultant changes effects (Fig. 3G). The perfusion of blood through in nerve temperature with and without nerve Thermal transfer in nerves occurs mainly the targeted nerve presents an additional source blood flow and shows similar results to ex- through conduction, though subcutaneous of heat flux. For a nerve blood flow of 50 ml/min perimental results (fig. S10). Details for the biofluid and nerve blood flow contribute to Reeder et al., Science 377, 109–115 (2022) 1 July 2022 4 of 7

RESEARCH | REPORT Fig. 4. Cooling localization. (A to C) Illustrations of the experimental setup for quantifying cooling localization (A) radially, (B) longitudinally, and (C) across the surface of the nerve. (D to F) A thermochromic (TC) hydrogel visually indicates the 10°C thermocline for a plane (D) perpendicular to the nerve, (E) parallel to the nerve, and (F) above a flat, uncurled cooler. (G to I) Simulations of the cases shown in (A) to (C), respectively, starting at the bottom of the TC hydrogel (z = 0). thermofluidic model are provided in the sup- tal study of temperature gradients in radial thermocline extends over nearly the entire plementary materials. Data captured for daily (Fig. 4A) and longitudinal (Fig. 4B) views serpentine evaporation chamber in the flat device. cooling for 180 s over 21 consecutive days in- along the nerve and across the surface of an Images of flow rate sweep experiments for the dicate stable operation in 37°C PBS, with a uncurled, planar device (Fig. 4C). Figure S11 cases shown in Fig. 4, E and F, are in figs. S12 standard deviation in cooling temperature of illustrates the experimental setup for quan- and S13, respectively. Thermal three-dimensional 0.6°C (Fig. 3I). tifying the 10°C thermocline in similar views finite element analysis confirms that the cooling of curled devices and a top-down view of a effect is largely confined radially (Fig. 4G) and Localization of the cooling effect to a pre- flat device. The 10°C thermocline is contained longitudinally (Fig. 4H) within the extent of the defined site without the need for insulating within the cuff (Fig. 4D) except for a region that cooling cuff and above a flat cooler (Fig. 4I). The layers represents a key capability of the evap- extends downward ~500 mm from the exterior temperature of the vapor remains confined in- orative microfluidic cooling approach intro- of the cuff and does not extend past the edge side the microfluidic channel, as indicated by duced here. Nerve coolers embedded in a of the device longitudinally along the nerve the cold region that extends radially down and thermochromic tissue mimic in three different (Fig. 4E). Figure 4F demonstrates that this to the right in the z = 0 mm plane for Fig. 4G. configurations serve as models for experimen- Reeder et al., Science 377, 109–115 (2022) 1 July 2022 5 of 7

RESEARCH | REPORT Fig. 5. Cooling-induced nerve block and analgesia. (A) Amplitude of muscular persistent touch sensitivity of the paw and serves as a model for neuropathic contraction in the tibialis anterior, as measured by EMG, diminishes with pain. (E) Image of a freely moving rat in a test environment for quantifying reduced temperature, and signal latency increases and subsequently recovers the mechanical nociceptive threshold. (F) Cooling the sciatic nerve to after rewarming. (B) Amplitude of CNAP diminishes with reduced temperature, 10°C increases the mechanical nociceptive threshold by a factor of seven in and signal latency increases and subsequently recovers upon rewarming. animals that received both a SNI and cuff, indicating a significant (C) Illustration of the bioresorbable nerve cooler interfacing with the sciatic cooling-induced analgesic effect. Changes in mechanical sensitivity of the nerve of a rat. (D) Illustration of the subcutaneous routing path of the contralateral for the same animals are insignificant. Error bars represent bioresorbable device and nonbioresorbable interconnects. A SNI generates standard error of the mean. ns, not significant; *P < 0.05. The low heat capacity of the PFP and N2 ex- Examples of in vivo cooling profiles are provided in the mechanical sensitivity threshold (1.6 to haust gas mixture yields minimal cooling along in fig. S17. 11.5 g), consistent with a significant cooling- the exterior of the outlet channel, as supported by induced analgesic effect (Fig. 5F). Changes in experiments and simulations (figs. S14 and S15). Experiments based on cooling of sciatic the mechanical sensitivity threshold in the nerves in a rat model of neuropathic pain as- contralateral side during cooling of the SNI- Acute animal trials demonstrate the capa- sociated with spared nerve injury (SNI) elu- treated nerve are statistically insignificant. bility of evaporative microfluidic coolers to cidate capabilities for microfluidic evaporative Histologic analyses after 1, 2, 3, and 6 months reversibly eliminate evoked nerve signals. The coolers to block pain in freely moving animals indicate a close proximity of the cuff to the soft, curled structure defines an intimate ther- (29). In devices constructed for multiday ex- nerve and provide evidence of biocompatibility mal and mechanical interface to the rat sciatic periments, transcutaneous connections to a and bioresorption, as described in the sup- nerve without sutures and without mechani- bioresorbable microfluidic evaporative cooler plementary materials (figs. S21 to S25 and cally induced damage (fig. S16). Electromyo- and temperature sensor mounted to the sciatic table S1). Previous in vivo studies of POC (21), graphy (EMG) of the tibialis anterior muscle nerve (Fig. 5C) route subcutaneously along the Mg (30), SiO2 (31, 32), and cellulose acetate indicates a 92% reduction in EMG magnitude spine to a headcap (Fig. 5D). Device details are (33) provide additional strong evidence of and 64% increase in signal latency of neuro- provided in figs. S18 and S19. An integrated biocompatibility and associated bioresorp- muscular activity during cooling from 31° to 5°C connecter mounted inside a titanium headcap tion processes. over a period of 8 min (Fig. 5A). The amplitude enables reversible fluidic and electronic con- and latency return to 108 and 100% of their ini- nection to an awake, freely moving animal The results presented here demonstrate that tial values, respectively, after rewarming over a (Fig. 5E). Mechanical nociceptive sensitivity spatiotemporally precise cooling enabled by period of 3 min. Electrical recording from the tests in two control animals (SNI only) over soft, bioresorbable evaporative microfluidics sciatic nerve distal to the cooling cuff provides a 3 weeks after the SNI show an expected in- provides for reversible elimination of local measure of compound nerve action potential crease in the mechanical sensitivity threshold peripheral nerve activity. Experiments indi- (CNAP) evoked through a single stimulation of the SNI side as compared with that of the cate capabilities in on-demand analgesia for pulse. Cooling from 33° to 4°C over a period of contralateral side, which persists for more than the management of neuropathic pain in 15 min prompts a decrease in signal amplitude 3 weeks (fig. S20). Studies using three addi- freely moving animal models. These concepts, of 77% and increase in latency of 97% (Fig. 5B). tional animals that received both the SNI materials, and device designs establish the Amplitude and latency return to within 101 and and the cooling cuff show that cooling of the engineering foundations for a class of implant- 97% of their initial values, respectively, after sub- SNI-treated nerve from 37° to 10°C after 3 weeks able cooling systems capable of targeted neu- sequent rewarming over a period of 3 min. of implantation leads to a sevenfold increase ral blocking with relevance across a range of Reeder et al., Science 377, 109–115 (2022) 1 July 2022 6 of 7

RESEARCH | REPORT clinical applications, including targeted, on- 20. G. H. Borschel, K. F. Kia, W. M. Kuzon Jr., R. G. Dennis, which is partially supported by the Soft and Hybrid J. Surg. Res. 114, 133–139 (2003). Nanotechnology Experimental (SHyNE) Resource (NSF ECCS- demand, nonopioid pain management. 1542205), the Materials Research Science and Engineering Center 21. J. Yang, A. R. Webb, G. A. Ameer, Adv. Mater. 16, 511–516 (DMR-1720139), the State of Illinois, and Northwestern University. REFERENCES AND NOTES (2004). Histology services were provided by the Northwestern University Mouse Histology and Phenotyping Laboratory, which is supported 1. J. M. Hah, B. T. Bateman, J. Ratliff, C. Curtin, E. Sun, Anesth. 22. R. T. Tran, J. Yang, G. A. Ameer, Annu. Rev. Mater. Res. 45, by National Cancer Institute grant P30-CA060553 awarded to Analg. 125, 1733–1740 (2017). 277–310 (2015). the Robert H. Lurie Comprehensive Cancer Center. Author contributions: Conceptualization: J.T.R., J.A.R.; Data curation: 2. K. Yu, X. Niu, B. He, Adv. Funct. Mater. 30, 1908999 23. R. T. Tran et al., J. Biomed. Mater. Res. A 102, 2793–2804 J.T.R., Z.X.; Funding acquisition: W.Z.R., Y.H., M.R.M., J.A.R.; (2020). (2014). Investigation: J.T.R., Z.X., Q.Y., M.-H.S., Y.Y., Y.D., K.R.J., C.L., S.M., E.P., A.J., Z.Z., M.J.K., Y.X., I.H., R.A., C.F., X.G.; Methodology: 3. T. R. Deer et al., Bioelectron. Med. 6, 9 (2020). 24. P. G. A. Rogueda, Drug Dev. Ind. Pharm. 29, 39–49 (2003). J.T.R., Z.X., Q.Y., M.-H.S., Y.Y., S.R.K., Y.D., Y.H., C.L., C.F., W.Z.R., 4. R. Melzack, P. D. Wall, Science 150, 971–979 (1965). 25. P. Grayburn, Clin. Cardiol. 20, I12–I18 (1997). Y.H., M.R.M., J.A.R.; Project administration: J.T.R., Z.X., Y.Y., 5. M. J. Cousins, D. B. Carr, T. T. Horlocker, P. O. Bridenbaugh, Eds., 26. M. Castrén et al., Circulation 122, 729–736 (2010). K.R.J.; Visualization: J.T.R.; Writing – original draft: J.T.R., Z.X., E.R., 27. O. D. Kripfgans, J. B. Fowlkes, D. L. Miller, O. P. Eldevik, J.A.R.; Writing – review & editing: J.T.R., Z.X., Q.Y., S.R.K., J.A.R. Cousins & Bridenbaugh’s Neural Blockade in Clinical Anesthesia and Competing interests: One or more provisional patents are being Pain Medicine (Lippincott Williams & Wilkins, 2009). P. L. Carson, Ultrasound Med. Biol. 26, 1177–1189 (2000). filed on this work. Data and materials availability: All data are 6. S. M. Iyer et al., Nat. Biotechnol. 32, 274–278 (2014). 28. P. A. Mountford, M. A. Borden, Adv. Colloid Interface Sci. 237, available in the main text or the supplementary materials. License 7. C. Rabut et al., Neuron 108, 93–110 (2020). information: Copyright © 2022 the authors, some rights reserved; 8. V. B. Brooks, in Reviews of Physiology, Biochemistry and 15–27 (2016). exclusive licensee American Association for the Advancement of Pharmacology, Volume 95, R. H. Adrian et al., Eds. (Springer, 29. I. Decosterd, C. J. Woolf, Pain 87, 149–158 (2000). Science. No claim to original US government works. https://www. 1983), pp. 1–109. 30. X. Gu, Y. Zheng, Y. Cheng, S. Zhong, T. Xi, Biomaterials 30, 9. E. H. Lothet et al., Sci. Rep. 7, 3275 (2017). 10. R. Janssen, Neurosci. Biobehav. Rev. 16, 399–413 484–498 (2009). SUPPLEMENTARY MATERIALS (1992). 31. S.-W. Hwang et al., Science 337, 1640–1644 (2012). 11. S. B. Rutkove, Muscle Nerve 24, 867–882 (2001). 32. S.-K. Kang et al., Nature 530, 71–76 (2016). 12. P. Borgdorff, P. G. A. Versteeg, Eur. J. Appl. Physiol. 33. M. N. Nosar et al., Cellulose 23, 3239–3248 (2016). Materials and Methods Occup. Physiol. 53, 175–179 (1984). Supplementary Text 13. Z. Zhang et al., J. Neurophysiol. 115, 1436–1445 (2016). ACKNOWLEDGMENTS Figs. S1 to S29 14. T. Morgan et al., J. Neurophysiol. 123, 2173–2179 Table S1 (2020). Funding: This work was funded by the Phil and Penny Knight References (34–52) 15. D. M. Ackermann, E. L. Foldes, N. Bhadra, K. L. Kilgore, Campus for Accelerating Scientific Impact (J.T.R.); the Querrey J. Neurosci. Methods 193, 72–76 (2010). Simpson Institute for Bioelectronics (J.A.R.); National Science Submitted 9 August 2021; accepted 13 May 2022 16. J. Jia, M. Pollock, Muscle Nerve 22, 1644–1652 (1999). Foundation grant CMMI1635443 (Y.H.); National Natural Science 10.1126/science.abl8532 17. J. Jia, M. Pollock, J. Jia, Brain 121, 989–1001 (1998). Foundation of China grant 12072057 (Z.X.); LiaoNing Revitalization 18. J. Shin et al., Nat. Biomed. Eng. 3, 37–46 (2019). Talents Program grant XLYC2007196 (Z.X.); Dalian Outstanding 19. J. Koo et al., Nat. Med. 24, 1830–1836 (2018). Young Talents in Science and Technology grant 2021RJ06 (Z.X.) Fundamental Research Funds for the Central Universities grant DUT20RC(3)032 (Z.X.); and National Research Foundation of Korea grant NRF-2021R1A5A1032937 (M.-H.S.). This work used the Northwestern University Micro/Nano Fabrication Facility (NUFAB), Reeder et al., Science 377, 109–115 (2022) 1 July 2022 7 of 7

RESEARCH PA L E O C L I M AT E supported by micropaleontological and sed- imentological observations of open water sur- Tectonic degassing drove global temperature trends rounding much of the Antarctic during the since 20 Ma MCO (5), with vegetation occupying at least coastal areas of that continent during this Timothy D. Herbert1*†, Colleen A. Dalton1†, Zhonghui Liu2, Andrea Salazar3, time (8). A warm high-latitude Southern Ocean Weimin Si1, Douglas S. Wilson4 is also consistent with clumped isotope results on bottom-dwelling foraminifera that indicate The Miocene Climatic Optimum (MCO) from ~17 to 14 million years ago (Ma) represents an enigmatic MCO deep ocean temperatures ~10°C warmer reversal in Cenozoic cooling. A synthesis of marine paleotemperature records shows that the MCO was a than at present (16). local maximum in global sea surface temperature superimposed on a period from at least 19 Ma to 10 Ma, during which global temperatures were on the order of 10°C warmer than at present. Our high- It is thus evident that Miocene warmth was resolution global reconstruction of ocean crustal production, a proxy for tectonic degassing of carbon, global, long-lived, and peaked with the retreat suggests that crustal production rates were ~35% higher than modern rates until ~14 Ma, when of ice over much of Antarctica. Global cooling production began to decline steeply along with global temperatures. The magnitude and timing of occurred with the reoccupation of much of the the inferred changes in tectonic degassing can account for the majority of long-term ice sheet and global Antarctic by ice after the MCO, accelerating temperature evolution since 20 Ma. after ~ 9.5 Ma (Fig. 1). T he climatic journey from global warmth Sea Drilling Project Site 594 (9) spliced to a Here, we investigated the possibility that to the great Pleistocene ice ages began continuous GDGT (glycerol dialkyl glycerol global climate since 20 Ma was paced and many tens of millions of years ago, in tetraether)–based estimate at Integrated synchronized by changes in a slow, but pow- the late Eocene (~40 Ma) when global Ocean Drilling Program Site 1352 over the erful, deep Earth process: the rate of degassing temperatures began to descend from the MCO (10). of CO2 controlled by changes in the rate of strikingly warm conditions of the Mesozoic oceanic plate creation and destruction. Our hy- and Early Cenozoic (1). This path was far from This synthesis of Miocene ocean temperatures pothesis builds on seminal work by Berner et al. monotonic, however. The climate system reached indicates that the MCO was a local thermal (17), who proposed that on time scales longer a tipping point at the Eocene/Oligocene bound- maximum superimposed on a strikingly warm than the residence time of carbon in the ocean- ary, when a substantial icecap grew on Antarctica background climate. In comparison to the pre- atmosphere-biosphere system (>100,000 years), and global temperatures declined. Episodic ad- sent day, temperatures preceding and postdat- vances and retreats of this Southern Hemisphere ing the MCO were high for millions of years: A SST anomaly (oC) Benthic foraminiferal 18O 0 icecap occurred for most of the Oligocene and global alkenone-based reconstruction at 10 Ma 1 Miocene (2, 3). However, an important rever- estimated an area-weighted ocean warming 20 2 sal of the climatic trend occurred during the of +6°C relative to the present day (9); Fig. 1 3 Miocene Climatic Optimum (MCO), when the suggests that temperatures prior to the MCO 15 4 East Antarctic Ice Sheet (EAIS) may have may have been similar. Estimating the global 5 largely disappeared (4, 5). temperature anomaly during the MCO itself NH HL is difficult, because the temperatures were so Modern baseline The MCO was originally recognized by a warm that the alkenone proxy became satu- 10 SH substantial excursion to lighter values in the rated over a band of ± ~40° latitude around benthic oxygen isotope record, which could the equator, implying temperatures in excess ML NH reflect a rise in deep sea temperatures or a of ~29°C. However, by relying on regional tropics decrease in global ice volume (1, 2, 6, 7). Al- Northern and Southern Hemisphere stacks though the partitioning of the two components where paleotemperatures remain below alke- 5 remains challenging, it is clear that both warm- none saturation, we arrive at an estimate of a ing and deglaciation occurred on and around global mean surface air temperature (SAT) 0 the Antarctic continent (4, 5, 8). In Fig. 1, we anomaly of ~ +12° to 19°C using a scaling of provide a continuous global estimate of marine ~1.7 times (11) the global sea surface temper- 0 5 10 15 20 temperatures through the MCO by moving to ature (SST) anomaly of +7.25° to 11.5°C (fig. S1). higher latitudes where the pattern of temper- The warmth we estimate is on the high side Age (Ma) ature change is clear, although temperature of a recent estimate of MCO DSAT (+11.5°C) changes are certainly amplified relative to a compiled from heterogeneous SST proxies, Fig. 1. Regionally averaged marine sea surface global mean ocean temperature. The curves which includes tropical estimates of perhaps temperature anomalies relative to the present rely almost entirely on the alkenone proxy, with questionable reliability (12). day, compared to the evolution of oxygen iso- a single exception in the Southern Hemisphere topic values of bottom-dwelling foraminifera. between an alkenone-based record at Deep That the MCO represents a local maximum Data are from (7). Abbreviations: NH HL, Northern in global temperatures is also confirmed by Hemisphere high latitude (45° to 69°N); SH, 1DEEPS, Brown University, Providence, RI 02912, USA. high-resolution Mg/Ca records showing clear Southern Hemisphere (20° to 49°S); ML NH, mid- 2Department of Earth Sciences, University of Hong Kong, warming into the MCO that parallels the latitudes of the Northern Hemisphere (29° to 43°N); Hong Kong, China. 3Department of Earth and Planetary marine d18O curve at two tropical locations Tropics, all tropical sites (24°N to 2°S). All Science, Harvard University, Cambridge, MA, USA. (13) and pronounced cooling at the end of paleotemperature determinations were made 4Department of Earth Science, University of California, the MCO at South Pacific ODP Site 1171 (14), with the alkenone proxy, with the exception of Santa Barbara, CA, USA. supported by similar cooling in a GDGT-based Site 1352 (part of the Southern Hemisphere *Corresponding author. Email: [email protected] record from the Arabian Sea (15). The link of stack), where we rely on GDGT-based paleother- †These authors contributed equally to this work. inferred warming to deglaciation in Antarctica is mometry (data S1 and S2). Marine temperatures evolve essentially synchronously in both hemi- spheres and reach a local maximum at the time of the MCO. The large cooling from ~9.5 to 6 Ma is not reflected in the benthic record, presumably because there was little growth of continental ice during this period. Herbert et al., Science 377, 116–119 (2022) 1 July 2022 1 of 4

RESEARCH | REPORT the inventory of CO2 in the atmosphere would careful treatment of three distinct sources of poles, as we do, but are focused on a longer be controlled by a plate tectonic degassing error: in the rotation parameters that describe time period and typically consider only two to source term, buffered by the negative feed- seafloor spreading, in the endpoints of each four intervals within the past 20 Ma (36, 39), back of silicate weathering (18, 19). Most sub- mid-ocean ridge, and in the ages of magnetic in contrast to our use of 10 intervals (table S3). sequent treatments of the long-term carbon polarity reversals. Furthermore, relative to the CK95 time scale cycle have instead focused on removal rates of (44) used by many earlier studies, the astro- carbon as the forcing function, often explicitly The globally integrated crustal production nomical polarity time scale used here (40) on the assumption that tectonic degassing has curve reveals large changes since ~20 Ma (24) importantly shortens the interval from C5C to not changed appreciably (19–23). However, in- (Fig. 2A). Total crustal production decreased C6 (~16 to 19 Ma). The smaller errors in the vestigating the tectonic forcing of climate in from at least 3.3 km2/year prior to 14 Ma to astronomical dating of magnetic ages also re- the past 20 million years offers several key no more than 2.7 km2/year after 6 Ma. Ridges duce uncertainties in crust production rate advantages over studies focused on longer time in the eastern Pacific dominate total produc- and allow greater confidence to be assigned periods: The motions of tectonic plates are well tion (Fig. 2, B and C), but crust production to the rate changes we measured (data S3). known because of the preservation of young has slowed along most ridges since 20 Ma. seafloor, magnetic polarity reversals are dated Only the Pacific-Antarctic ridge substantially The similarity of the new crust production with high accuracy and small uncertainties, increased production; ridges in the eastern rates to the broad evolution of ocean temper- and large and well-dated transitions occurred Pacific reduced production by 25 to 50%, and atures motivated us to evaluate the plausibility in global climate. those in the Atlantic slowed by 10 to 30%. of two end-member explanations for cooling since the Miocene, one focused on changes in We derived a global synthesis of ocean crust Previous studies of ocean crust production continental weatherability (CO2 sink) and the production rate, which we propose is a proxy differ in their estimate of the trend since the other on CO2 degassing (CO2 source). The for degassing rates over time (24). This assump- mid-Miocene, ranging from increasing (36) to weatherability argument posits a long-term tion does not specify the proportion of degass- constant (37) to decreasing (38, 39). Among increase in the proportion of chemical weather- ing that occurs at mid-ocean ridges (25), in the subset of studies that identified declining ing in relation to continental breakdown (19, 22) metamorphism in orogenic belts (26, 27), or by ridge flux, our result has a larger amplitude, or to the emergence of maritime continents release in subduction zones (28–30), as faster greater level of detail, and better-defined un- with easily weathered lithologies in the tropical crust production necessarily implies faster lith- certainty. Three main factors may explain these Indo-Pacific in the Plio-Pleistocene (23). osphere destruction (and the converse). Our improvements: an updated reconstruction of simple model neglects factors that are likely the complex and incomplete record of spread- To explore the alternative (source) scenario, to be important in detail, such as the carbon ing in the eastern Pacific since 25 Ma, the we constructed a carbon cycle mass balance content of sediments and rocks fed into oro- inclusion of ridge spreading histories with forced by the new crust production spreading genic and subduction zones, the dip of the higher temporal resolution, and our use of curve, which is used as a proxy for tectonic car- subducting plate, and other factors that might astronomically dated magnetic polarity ages bon release (supplementary materials). Because influence the efficiency of tectonic degassing (40, 41). Important pieces of the record of nearly all plate pairs display a pattern similar to independently of the total rate of marine relative motion of the Cocos, Nazca, and Pacific that of the global ensemble (Fig. 2A), we made crustal recycling (28, 31). plates are obscured by large microplates, poor the simplification that tectonic degassing scales low-latitude magnetic geometry, and hotspot linearly to the global ocean crustal production Crustal production rates are calculated as tracks. Moreover, some studies (36, 37, 42) function. Two adjustable parameters can be the product of spreading rate and ridge length, derive spreading rates from gridded estimates optimized to satisfy the twin constraints of CO2 assuming that crustal thickness is constant in of seafloor age grid, early versions of which degassing rates and global temperatures: (i) the space and time. Compilations of ocean crustal have been shown to contain some errors in Earth system sensitivity that relates atmospheric thickness have shown that it is largely inde- the eastern Pacific for ages of 5 to 25 Ma (43). CO2 to long-term equilibrium global temper- pendent of spreading rate except at ultraslow Although we compiled the eastern Pacific ature (ESS), and (ii) the sensitivity of the silicate spreading ridges where melt supply is low spreading record in more detail than previ- weathering negative feedback to temperature (32). We took advantage of recent reconstruc- ous authors, we caution that subdividing the (Ea), which is believed to keep the carbon sys- tions of ocean ridge spreading histories at interval from 22 to 15 Ma remains difficult tem in long-term balance. There is good reason high temporal resolution (33–35) as well as because of the limited record of the Nazca- to suspect that ESS is larger than a recent new determinations of spreading rates in the Pacific and Cocos-Nazca plate pairs. Other estimate (45) of 2.6° to 3.9°C per doubling eastern Pacific. Central to our analysis is a studies derive spreading rates from rotation of CO2 [referred to as equilibrium climate Fig. 2. Ocean crust production rates with 95% confidence interval. See data S3. (A) Global total. (B and C) By plate boundary. Note the different vertical scales for (B) and (C). See table S2 for plate names. Herbert et al., Science 377, 116–119 (2022) 1 July 2022 2 of 4

RESEARCH | REPORT Fig. 3. Results of carbon mass balance modeling. (A) Global degassing rate driven by ocean crustal production reconstruction. (B) Calculated changes in global surface air temperature (SAT). (C) Corre- sponding implied atmospheric CO2 levels. Thick pink lines in (B) and (C) result from using central crustal production estimate [thick line in (A)]. Central carbon cycle parameters are Ea = 4 kcal/mol and ESS = 6.5°C per doubling of CO2. Upper and lower bounds in (B) and (C) vary Ea from 3.5 to 4.5 kcal/mol and ESS from 5° to 8°C. Gray/ black symbols in (C) show pCO2 estimates from boron isotope analyses of planktonic foraminifera using different ocean carbon system parameters (49); blue and red symbols are based on carbon isotope values of alkenones (48, 50). sensitivity (ECS or “Charney” climate sensi- observed globally in the younger half of the although a gap remains between our preferred tivity)] and also to consider the possibility Miocene (14.5 to 6 Ma). The level of detail in CO2 curve (Fig. 3C, heavy pink line) and some that ESS may increase nonlinearly under the crustal production curve from ~14 to 20 Ma proxy estimates. Last, our finding of large varia- high CO2 (11, 46). We explored the effects is limited by time resolution in tectonic con- tions in seafloor spreading rate over the Neogene of varying ESS between 5° and 8°C per CO2 straints (rotation points at 13.739, 15.974, and implies that interpretations of geochemical sig- doubling, or approximately two times the 18.748 Ma) and by somewhat higher uncertain- nals commonly used in paleoceanography (Mg/Ca, 66th percentile bounds of the ECS (45). On ties in the astronomical calibration of the mag- 87Sr/87Sr) may need to be reinterpreted in light tectonic time scales, ocean total dissolved in- netic polarity time scale. Degassing rates may of changing hydrothermal fluxes driven by organic carbon and alkalinity are primarily have changed very little since ~6 Ma (Fig. 2). If changes in seafloor spreading rates (19, 51). set by degassing, chemical weathering, and our proposed linkage between seafloor spread- calcium concentration of seawater; therefore, ing and tectonic degassing is correct, it carries Forcing by a short-lived and relatively small it is not necessary to know the absolute values of several major implications. First, the similarity input of CO2 from Columbia River basalt mag- these quantities a priori to estimate equilib- in timing between crustal production changes matism (52) appears insufficient as a first- rium pCO2 (supplementary materials). and paleotemperatures suggests a geologically order explanation for the long duration of a efficient (million-year time scale) coupling be- strikingly warm Miocene climate, although it The temperature dependence of weathering tween tectonic processes and CO2 release to may have contributed to the apex during the sensitivity is directly constrained by our esti- the Earth surface; time lags between tectonic MCO. In contrast, the hypothesis of long-term, mate of peak MCO warmth and degassing. forcing and climate response are not apparent large variations in tectonic degassing provides Here, Ea represents an effective weathering at the resolution of our analysis. Second, it sug- a potent, persistent driver. The evolution of dependence on temperature in an Arrhenius gests that many proxy-based reconstructions of temperature, inferred CO2 levels, and benthic formulation; it is a simple parameterization Miocene levels of CO2 may be too low (Fig. 3C). d18O values (Figs. 1 and 3) suggests that polar of the complex factors that control silicate The large range in plausible CO2 levels in our glaciation exhibits strongly nonlinear thresh- weathering, including relief, the insulating reconstruction derive from the wide permissi- old sensitivity to CO2 and temperature: The effect of soil formation (transport limitation), ble range of ESS. However, it is very difficult large retreat of the East Antarctic Ice Sheet and rainfall. The values chosen here, although to reconcile the combination of peak Miocene during the MCO may have required only a considerably lower than those derived from warmth and high degassing with an atmospheric modest rise in CO2 from its generally high pure silicate mineral activation energies, are CO2 inventory much lower than 850 ppm (12) early Miocene level, and the onset of North- consistent with previous studies of global (see Fig. 3). In this regard, we note that the ern Hemisphere ice ages at ~2.7 Ma occurred weathering rates (supplementary materials). trend of more recent CO2 proxy estimates has under tectonic CO2 forcing similar to today. been to revise reconstructed Miocene levels These results are consistent with two very The model satisfies the observation of gen- of CO2 higher than earlier estimates (47–50), different CO2/temperature thresholds for glacia- erally warm early to mid-Miocene conditions, tion in the Southern and Northern hemispheres the timing of the MCO, and the steep cooling Herbert et al., Science 377, 116–119 (2022) 1 July 2022 3 of 4

RESEARCH | REPORT proposed by (53) and especially imply a high REFERENCES AND NOTES 41. A. J. Drury et al., Earth Planet. Sci. Lett. 475, 254–266 (2017). sensitivity of East Antarctic ice volume to 42. J.-P. Cogné, E. Humler, Geochem. Geophys. Geosyst. 7, Q03011 global temperature at a “set point” of ~ +10°C 1. J. Zachos, M. Pagani, L. Sloan, E. Thomas, K. Billups, Science (Fig. 3). 292, 686–693 (2001). (2006). 43. D. B. Rowley, Lithosphere 11, 21–43 (2019). We do not rule out the possibility that the 2. B. P. Flower, J. P. Kennett, Paleoceanography 10, 1095–1112 44. S. C. Cande, D. V. Kent, J. Geophys. Res. 97, 13917–13951 global carbon cycle was also affected by a tran- (1995). sition from weaker silicate “weatherability” in (1992). the Miocene relative to the Plio-Pleistocene 3. D. Liebrand et al., Proc. Natl. Acad. Sci. U.S.A. 114, 3867–3872 45. S. Sherwood et al., Rev. Geophys. 58, e2019RG000678 (22, 23). However, the proposed low weather- (2017). ability cannot serve as the primary explana- (2020). tion for the warm early to mid-Miocene: In the 4. R. Levy, D. Harwood, F. Florindo, SMS Science Team, 46. A. Farnsworth et al., Geophys. Res. Lett. 46, 9880–9889 case of the scenario evaluated by Park et al. Proc. Natl. Acad. Sci. U.S.A. 113, 3453–3458 (2016). (23), a reduced MCO weatherability accounts (2019). for about one-fourth of the observed temper- 5. F. Sangiorgi et al., Nat. Commun. 9, 317 (2018). 47. S. M. Sosdian et al., Earth Planet. Sci. Lett. 498, 362–376 ature anomaly (fig. S3). A 30% higher tectonic 6. A. Holbourn, W. Kuhnt, K. G. D. Kochhann, N. Andersen, CO2 flux rate will, however, generate enough (2018). warming to be consistent with our empirical K. J. Sebastian Meier, Geology 43, 123–126 (2015). 48. H. M. Stoll et al., Quat. Sci. Rev. 208, 1–20 (2019). SST and SAT estimates (Fig. 3). We readily 7. D. De Vleeschouwer, M. Vahlenkamp, M. Crucifix, H. Pälike, 49. J. W. B. Rae et al., Annu. Rev. Earth Planet. Sci. 49, 609–641 admit that our temperature and CO2 model reconstructions are imperfect, because our as- Geology 45, 375–378 (2017). (2021). sumptions that CO2 degassing rate is in 1:1 8. S. Warny et al., Geology 37, 955–958 (2009). 50. T. Tanner, I. Hernández-Almeida, A. J. Drury, J. Guitián, proportion to ocean crustal production, and 9. T. D. Herbert et al., Nat. Geosci. 9, 843–847 (2016). that weatherability and ESS remain fixed for 10. J. Jiang, thesis, University of Hong Kong (2019). H. Stoll, Paleoceanogr. Paleoclimatol. 35, e2020PA003925 the past 20 million years, are unlikely to apply 11. D. J. Lunt et al., Nat. Geosci. 3, 60–64 (2009). (2020). strictly. In this regard, reduced weatherability 12. N. J. Burls et al., Paleoceanogr. Paleoclimatol. 36, 51. J. A. Higgins, D. P. Schrag, Earth Planet. Sci. Lett. 416, 73–81 during the mid-Miocene (22) may have en- (2015). hanced large excursions in atmospheric CO2 e2020PA004054 (2021). 52. J. Kasbohm, B. Schoene, Sci. Adv. 4, eaat8223 (2018). and temperature by reducing the sensitivity of 13. S. M. Sosdian, C. H. Lear, Paleoceanogr. Paleoclimatol. 35, 53. R. M. Deconto et al., Nature 455, 652–656 (2008). the silicate weathering sink in a warm climate 54. S. Brune, S. E. Williams, R. D. Müller, Nat. Geosci. 10, 941–946 in comparison to the present day. e2020PA003920 (2020). (2017). 14. A. E. Shevenell, J. P. Kennett, D. W. Lea, Science 305, 55. D. Archer, A. Winguth, D. Lea, N. Mahowald, Rev. Geophys. 38, Our findings suggest that the coupled evolu- 159–189 (2000). tion of CO2 and climate over the Neogene can 1766–1770 (2004). 56. A. Holbourn, W. Kuhnt, M. Schulz, J.-A. Flores, N. Andersen, be understood as a combination of exogenous, 15. G. Zhuang, M. Pagani, Y. G. Zhang, Geology 45, 655–658 (2017). Earth Planet. Sci. Lett. 261, 534–550 (2007). slow forcing by the tectonic CO2 source, in 16. S. E. Modestou, T. J. Leutert, A. Fernandez, C. H. Lear, combination with more rapid but substantial ACKNOWLEDGMENTS repartitioning of the total surface carbon in- A. N. Meckler, Paleoceanogr. Paleoclimatol. 35, ventory among atmosphere, biosphere, and e2020PA003927 (2020). We thank A. Lee, C. Carchedi, and J. Smith for assistance with ocean paced by orbital cycles. Our study does seafloor magnetic anomaly studies. This research used samples not negate the rich possibility of other tectonic 17. R. A. Berner, A. C. Lasaga, R. M. Garrels, Am. J. Sci. 283, and/or data provided by the International Ocean Discovery forcings of the carbon cycle (19–23, 29, 54), but 641–683 (1983). Program (IODP) and its predecessors. IODP is sponsored by it suggests that the dynamism in ocean crustal NSF and participating countries under management of Joint creation/destruction as a carbon cycle driver 18. J. C. Walker, P. Hays, J. F. Kasting, J. Geophys. Res. 86, Oceanographic Institutions (JOI) Inc. Funding: Supported by NSF has been underestimated since the publica- 9776–9782 (1981). grants 1635127 (T.D.H., W.S., C.A.D.) and 1545859 (T.D.H.), the tion of (17). Riding on top of tectonic forcing, Doherty Fund for Oceanography at Brown University (T.D.H., W.S.), cyclical sequestration and release of carbon 19. L. R. Kump, M. A. Arthur, in Tectonic Uplift and Climate Change Hong Kong Research Grants Council grant 17305019 (Z.L.), from the deep ocean (55) may explain high- (Springer, 1997), pp. 399–426. and the Leadership Alliance (A.S.). Author contributions: frequency orbital-scale variability evident within Conceptualization, T.D.H., C.A.D.; methodology, T.D.H., C.A.D., the MCO, as demonstrated by cyclic varia- 20. M. Raymo, W. F. Ruddiman, Nature 359, 117–122 (1992). Z.L., A.S., W.S., D.W.; writing–original draft, T.D.H., C.A.D., Z.L., tions in foraminiferal d18O-, d13C-, and Mg/ 21. G. Li, H. Elderfield, Geochim. Cosmochim. Acta 103, 11–25 A.S., W.S., D.S.W.; writing–reviewing and editing, T.D.H., C.A.D., Ca-based SST estimates (4, 6, 13, 14, 56); pro- W.S., D.S.W. Competing interests: The authors declare no cesses driving the oceanic carbon reservoir (2013). competing interests. Data and materials availability: The data became dominant during the Pleistocene ice 22. J. K. Caves Rugenstein, D. E. Ibarra, F. von Blanckenburg, supporting the regional SST trends in Fig. 1 include previously ages. published (9, 57–66) data available at the PANGAEA data Nature 571, 99–102 (2019). repository ( Data tables providing new TEX86- 23. Y. Park et al., Proc. Natl. Acad. Sci. U.S.A. 117, 25319–25326 based estimates of SST and bulk stable isotopes at Site U1352, the age model used for Site U1352, and the regional SST stacks (2020). presented in Fig. 1 are also archived at the PANGAEA data 24. C. A. Dalton, D. S. Wilson, T. D. Herbert, Geophys. Res. Lett. 49, repository. License information: Copyright © 2022 the authors, some rights reserved; exclusive licensee American Association e2022GL097937 (2022). for the Advancement of Science. No claim to original 25. A. E. Saal, E. H. Hauri, C. H. Langmuir, M. R. Perfit, Nature 419, US government works. journal-article-reuse 451–455 (2002). 26. J. A. Becker, M. J. Bickle, A. Galy, T. J. Holland, Earth Planet. SUPPLEMENTARY MATERIALS Sci. Lett. 265, 616–629 (2008). 27. A. Skelton, Geology 39, 43–46 (2011). Supplementary Text 28. C.-T. A. Lee et al., Geosphere 9, 21–36 (2013). Figs. S1 and S2 29. N. R. McKenzie et al., Science 352, 444–447 (2016). Tables S1 and S2 30. P. B. Kelemen, C. E. Manning, Proc. Natl. Acad. Sci. U.S.A. 112, Data S1 to S4 Matlab Codes for Carbon Cycle Calculations E3997–E4006 (2015). References (57–86) 31. T. Plank, C. E. Manning, Nature 574, 343–352 (2019). 32. R. S. White, D. McKenzie, R. K. O’Nions, J. Geophys. Res. 97, Submitted 16 July 2021; accepted 3 May 2022 10.1126/science.abl4353 19683–19715 (1992). 33. S. Merkouriev, C. DeMets, Geophys. J. Int. 173, 1064–1083 (2008). 34. C. DeMets, S. Merkouriev, Geophys. J. Int. 207, 741–773 (2016). 35. C. DeMets, S. Merkouriev, S. Jade, Geophys. J. Int. 220, 1149–1171 (2019). 36. C. J. Rowan, D. B. Rowley, Geophys. J. Int. 208, 1173–1183 (2017). 37. D. B. Rowley, Geol. Soc. Am. Bull. 114, 927–933 (2002). 38. C. P. Conrad, C. Lithgow-Bertelloni, Geology 35, 29–32 (2007). 39. R. D. Müller et al., Tectonics 38, 1884–1907 (2019). 40. F. J. Hilgen, L. J. Lourens, J. A. van Dam, in The Geologic Time Scale 2012, F. M. Gradstein, J. G. Ogg, M. Schmitz, G. Ogg, Eds. (Elsevier, 2012), pp. 923–978. Herbert et al., Science 377, 116–119 (2022) 1 July 2022 4 of 4

WORKING LIFE By Kasper Bonnesen Stepping out of my comfort zone “D o you want to go to Sunday brunch?” It was a question I’d never dared ask an office mate, but I was feeling desperate. I’d relocated from Denmark to the United States 2 months earlier for a research abroad experience during my Ph.D. I hadn’t made any friends by that point so I decided to ask another Ph.D. student in my office whether they’d meet up outside work. He looked at me, smiled, and then uttered “Yes.” It was a little thing, but it was a victory. I had finally conquered my shyness and dared to step out of my comfort zone. My university encourages Ph.D. felt I had to learn everything from students to spend time abroad, scratch—where do I get my grocer- so I planned from the beginning ies? How do I get to work? If my to work at a U.S. university for sister was busy, who could I spend 6 months under a collaborator time with? It didn’t help that stu- of my adviser’s. I was excited to dents were largely away from cam- travel, improve my English, and pus for the first 2 months—first expand my scientific network. But because classes weren’t in session I was also afraid that my shy, in- and later because Atlanta was hit troverted self would keep me from by a wave of COVID-19. getting to know people, leading to My sister helped me come up a level of sadness and loneliness with a plan. We decided that, that would eventually force me once the COVID-19 situation to abandon the adventure and go eased, I would go to campus ev- back to Denmark. ery day. There, I would try to It had happened twice before. engage in conversations with my The first time was immediately colleagues—for instance, by greet- after medical school when I took “Even little interactions ing people I passed in the hallway a job at a small hospital in Green- at work helped me feel part or by asking other students how land. I had always wanted to go their weekends went while wait- to the island, and the challenge of a community.” ing for the coffee to brew. Every of living in a small, isolated com- weekend, I would also plan out- munity and doing my best to help ings with at least one person, people appealed to me. But I had barely put my feet on whether it be going out for brunch, visiting a museum, Greenlandic soil before something felt wrong. Despite traveling to another city, or going on a hike. working with open and friendly colleagues, I never felt The advice made a huge difference—I suddenly felt I belonged and was lonely and sad. Two weeks after my much less lonely. Even little interactions at work helped start date, I told the hospital manager I needed to go back me feel part of a community. Gradually, I built up a to Denmark. The second time was during the pandemic. small group of friends—from both within and outside the I moved to Copenhagen because I’d always wanted to live office—whom I could call on to do things with, including there and because my Ph.D. adviser was fine with me work- the Ph.D. student I had invited to brunch. By the end of ing remotely. That relocation only lasted 3 months—again my time in Atlanta, I felt so comfortable that I actually because of loneliness. I missed my friends and family who had mixed feelings about going home. lived near my university, which is less than an hour’s drive I returned to Denmark in May, feeling fulfilled by my from my hometown. time abroad. I was able to travel to new places, meet inter- When my departure date neared for my research abroad esting people, and learn new professional skills. I also came experience in Atlanta, I vowed that this time would be to appreciate the benefits of stepping outside my comfort different. I wanted to make more of an effort to get to zone. I encourage every young scientist to do a research ILLUSTRATION: ROBERT NEUBECKER know people and feel comfortable in the community, and stay abroad—and to have the courage to open up to people I thought it would be easier because my outgoing younger while you’re there. You never know what you’ll learn. j sister happened to be doing her Ph.D. in the same city. Still, the initial few months were difficult. Finding myself Kasper Bonnesen is a Ph.D. student at Aarhus University. Do you have an in a foreign country without my habitual life terrified me. I interesting career story to share? Send it to [email protected]. 122 1 JULY 2022 • VOL 377 ISSUE 6601 SCIENCE

Like this book? You can publish your book online for free in a few minutes!
Create your own flipbook