Gold-nanofève surface-enhanced Raman spectroscopy visualizes hypotaurine as a robust anti-oxidant consumed in cancer survival

Gold deposition with diagonal angle towards boehmite-based nanostructure creates random arrays of horse-bean-shaped nanostructures named gold-nanofève (GNF). GNF generates many electromagnetic hotspots as surface-enhanced Raman spectroscopy (SERS) excitation sources, and enables large-area visualization of molecular vibration fingerprints of metabolites in human cancer xenografts in livers of immunodeficient mice with sufficient sensitivity and uniformity. Differential screening of GNF-SERS signals in tumours and those in parenchyma demarcated tumour boundaries in liver tissues. Furthermore, GNF-SERS combined with quantum chemical calculation identified cysteine-derived glutathione and hypotaurine (HT) as tumour-dominant and parenchyma-dominant metabolites, respectively. CD44 knockdown in cancer diminished glutathione, but not HT in tumours. Mechanisms whereby tumours sustained HT under CD44-knockdown conditions include upregulation of PHGDH, PSAT1 and PSPH that drove glycolysis-dependent activation of serine/glycine-cleavage systems to provide one-methyl group for HT synthesis. HT was rapidly converted into taurine in cancer cells, suggesting that HT is a robust anti-oxidant for their survival under glutathione-suppressed conditions.

4. The manuscript also requires careful editing by a native speaker of English to correct grammatical errors.
5. The authors are over-reaching in the last paragraph of the Discussion regarding clinical applications, particularly considering the data is primarily derived from a single colon carcinoma cell line.
Reviewer #2 (Remarks to the Author): Shiota and her/his coworkers developed beautiful SERS model to characterize intrinsic Raman spectra. This work has high potential to be developed as a wide-field tip-enhanced Raman spectroscopy (TERS) and/or intrinsic Raman endoscope. This paper is also interesting as the authors demonstrated a Raman version of metabolic PET imaging, not only to distinguish tumor/normal tissue boundary via intrinsic Raman spectra, but to analyze metabolism of tumor/parenchymal tissue. This is a risky-as all the intrinsic Raman paper does, but a beautiful challenge. I recommend this article to publish in Nature communications after the revision.
(1) Although they tried very hard to reach the solid discussion across the paper, the article is too long to publish in Nature communications. It is almost 6,000 words with 10 figures. Judging from the contents, there are lots of ways to polish the overall manuscript to be suitable for the publication.
(2) Figure 1c, Figure 2a (in particular) should be magnified to clarify the morphology of the SERS structure.
(3) Could you provide a fair comparison of the enhancement factor between the single nanofeve structure and nanosphere junction which is obtained from zero-angle metal deposition, which was also discussed in your previous paper, ACS Nano, 8, 5622-5632 (2014). In addition, what is the population of nanofeve junction rather than single nanofeve structure across the surface, which will definitely affect the overall enhancement factor of the film? (4) The nanofeve described here is so much similar to the nanorice, which is one of the best single particle SERS design for strong local electromagnetic enhancement. In this regard, please add the related reference of "Nanorice: A Hybrid Plasmonic Nanostructure. Nano Lett. 2006, 6, 827-832." (5) Is there a motivation for the authors to choose liver tumor model? Is that because there are more sufficient metabolites in liver than other organs so that you could build up intrinsic Raman libraries of metabolites easier?

Reply to Reviewers' comments:
General notes for the revised version of manuscript (NCOMMS-17-21247A) The authors thank 3 reviewers and Editorial Boards for giving us the opportunity to submit the revised version of our manuscript (NCOMMS-17-21247A).
The minor change of the title (～in cancer survival) resulted from correction by a native English speaker.
As reviewers #1 and #3 pointed out that conclusions of the original version (17-21247) are derived from the analyses of a single cell line, and that the data collected from xenograft transplantation of human colon cancer to immunodeficient mice hamper their feedback to apply SERS imaging to clinical cancer samples which display more complicated pathologic features. We were unable to accomplish a similar study using clinical cancer tissues, just because sufficient time to pass IRB paperworks within 3 months. However, in the revised version, we performed new sets of animal experiments to improve a quality of the study by using the murine model of Ink4a/Arf-null These data were newly added in Fig. 10 and Suppl. Fig. 15 in the revised version (17-21247A). In this model, distribution of GFP-positive cancer cells does not necessarily exhibit "solid" shapes but display anatomically amorphous main tumour with multiple daughter micro-tumours surrounding the main one that coincide with infiltration of inflammatory cells and/or gliosis. During the limited time for revision. Using this model, we were able to build up "accumulative lists of Raman shifts (ALR in Fig. 4 and Suppl. Fig. 3)" and a single mouse was used for practicing automated extraction of the tumour boundaries based on the ALR-database. Since this model displayed complicated tumour boundaries including irregular borders with reactive gliosis and inflammatory responses surrounding the main tumour, we asked a national-licensed experienced pathologist to carefully extract the tumour regions that did not coincided with gliosis and inflammatory responses to build up the database of T-dominant and P-dominant Raman shifts". As seen in Fig. 10 and Table   S3, SERS imaging and region-specific analyses of the spectral features clearly showed that the technology enables us to clearly distinguish the tumour regions. Such novel data showed that GNF-SERS technology is applicable to the extraction of tumour boundaries even in the regions displaying more complicated features of malignancy, and thus serve as a reply to the Reviewer 3. At the same time, according to the reviewers' and Editorials' suggestions, we softened overarching description of the clinical application of our GNF-SERS technology in the last portion of Discussion.
Immediately after receiving the editorial comments on October 23, 2017, we did our best to submit a proposal for clinical research in which we attempted to use surgical post-operative human samples of carcinoma to apply our SERS technology under collaboration with National Cancer Center Hospital in Japan. While we are still waiting for the answer from IRB of the Institute, it seems impossible to have a permission before the date of the deadline (January 23, 2018).
I appreciated if the Editorial Boards understand our difficulty to perform clinical studies using human pathological samples which usually require much time to be approved. Instead, what we did the best for revision within 3 months is to provide the new datasets of the current mouse glioblastoma model, by which we believe usefulness of the current GNF-SERS technology has been empowered.
According to the Reviewer 2's suggestion, we transferred the major portions of mass spectroscopic data into Suppl. Information. These portions were just transferred to Suppl.
Information without changing the sentences that were underlined with broken lines. On the other hand, all revised portions which were newly added in the revised version were marked by "solid underlines" in the revised manuscript. Suppl. Fig. 2, Suppl. Fig. 4b, Fig. 10 and Suppl. Fig. 15 constitute major portions that were newly added according to the Reviewers' suggetsions.
Because of considerable contribution to accomplishment of new data from in vivo tumours with a perfect penetrance on orthotopic inplantation of cancer stem cells in syngenic mice, I would like to add the following co-authors: Dr. Oltea Sampetrean (oltea@a6.keio.jp) and Dr. Hideaki Nagashima (032m2068@gmail.com) who developed and performed important experiments of the syngeneic cancer stem cell model of glioblastoma, and Dr. Nobuyoshi Hiraoka ( nhiraoka@ncc.go.jp ) who is a professional pathologist who evaluated pathology of mouse glioblastoma-bearing brain tissues. In addition, Dr. Takayuki Morikawa (morikawa150401@gmail.com) contributed to preparation of frozen brain tissues, and Dr. Keiyo Takubo (keiyot@gmail.com) contributed to preparation of the revised manuscript as an expert of stem cell biology. All other authors agreed with adding these important contributors in the revised version.
We asked a professional English writer to correct English. Such corrections were not underlined. Otherwise, all revised protions were underlined in the text, so that the Referees can easily recognize all revised portions of the manuscript.

Reviewer #1 (Remarks to the Author):
Q1: The authors use novel techniques to identify tumor-specific metabolites. Their invention of surface-enhanced Raman spectroscopy using gold nanofeve substrates appears to be a useful addition to the techniques that can be applied to study tumor metabolism in vivo.
1. My major concern is that most of the data (especially in vivo) and conclusions are derived from the analysis of a single cell line, HCT116. The authors should not present conclusions in general terms but refer specifically to the cell line(s) from which the data were derived. Furthermore, using the same lysate samples for CE-MS metabolomics, we determined phospho-serine (P-serine) and its mass-labeled derivatives: As seen in Fig. 7 in the revised version, the newly added data clearly indicated that 13 C 3 -labeled 3-PG in the glycolytic pathway is converted to 13 C 3 -P-serine with statistical significance, while (as shown in the original manuscript) serine is detectable mainly as 13 C 1 -and 13 C 3 -labeled serine with statistical significance versus 13 C 2 -serine. These results strengthen a hypothesis that the C 1group transfer mechanism plays a role in generating 13 C 1 -serine as a source of hypotaurine/taurine in HCT116 cells. The discussion was briefly added in the revised version (Figs 7 and 8 were revised according to the Reviewer 1). PSPH protein expression and the data of p-serine was inserted in the text with being underlined. Regarding Discussion that included the application to human pathology, we deleted overarching description in the original version, and clarified unresolved problems to challenge for clinical application using human samples (page 12, See our reply A1).

Reviewer #2 (Remarks to the Author):
Shiota and her coworkers developed beautiful SERS model to characterize intrinsic Raman spectra. This work has high potential to be developed as a wide-field tip-enhanced Raman spectroscopy (TERS) and/or intrinsic Raman endoscope. This paper is also interesting as the authors demonstrated a Raman version of metabolic PET imaging, not only to distinguish tumor/normal tissue boundary via intrinsic Raman spectra, but to analyze metabolism of tumor/parenchymal tissue. This is a risky-as all the intrinsic Raman paper does, but a beautiful challenge. I recommend this article to publish in Nature communications after the revision.

Q1:
Although they tried very hard to reach the solid discussion across the paper, the article is too long to publish in Nature communications. It is almost 6,000 words with 10 figures.
Judging from the contents, there are lots of ways to polish the overall manuscript to be suitable for the publication.

Q3:
Could you provide a fair comparison of the enhancement factor between the single nanofeve structure and nanosphere junction which is obtained from zero-angle metal deposition, which was also discussed in your previous paper, ACS Nano, 8, 5622-5632 (2014).
In addition, what is the population of nanofeve junction rather than single nanofeve structure across the surface, which will definitely affect the overall enhancement factor of the film? A3: Thank you very much for an intriguing question. According to the Referee's suggestion, we attempted to provide a fair composition of the enhancement factor (G values in Fig. 2 and Suppl. Fig. 2) between single nanofeve structures and nanosphere junction. The density of Au-nanoparticles corresponds to the "gap distance" between the two adjacent nanoparticles. Therefore, in the revised version, we examined differences in the G values as a function of the gap distance (also called "junction" in the other literature including Ref. 20 cited in the revised version) between GNF and GNC. As seen in Suppl. Fig. 2, FDTD simulation revealed that the local field enhancement by two adjacent Au-nanoparticles of GNC are diminished by increasing the gap distance over 10 nm (typical gap distance of GNF in Fig.2a), suggesting that the density of GNC nano-particles rate-limits the enhancement. In contrast, the enhancement by GNF nanoparticles does not regress and stay in a plateau level so far in a distance less than 200 nm of the gap distance (Suppl. Figs. 2a-c). As shown in a high-magnification SEM (Fig. 1d in the revised version, the approximate distance between the adjacent Au-nanoparticles in GNF is in a range between 50~100 nm. Collectively, these results suggest that the anisotropy of Au-nanoparticles rather than effects of the gap of the Au-nanoparticles, which is also called "junction", determine SERS enhancement in GNF substrates, being in good agreement with previous studies (Ref. 20 cited in the revised version).

Q4:
The nanofeve described here is so much similar to the nanorice, which is one of the best single particle SERS design for strong local electromagnetic enhancement. In this regard, please add the related reference of "Nanorice: A Hybrid Plasmonic Nanostructure. Nano Lett. 2006, 6, 827-832." A4: The authors thank this question from Reviewer #2. We added brief comments with citing the reference of this intriguing publication (Reference #20 in the revised version).
Eventually, the citation of this article with FDTD simulation to compare GNC and GNF allowed us to emphasize the importance of morphologic anisotropy than the gap distance ("junction") of Au-nanoparticles as mechanisms for enhancing the electromagnetic fields (Suppl. Fig. 2 in the revised version). However, in the revised version, we attempted to examine if SERS imaging is useful to extract the borders between tumours and parenchyma in other organs such as brain.

Q5. Is there a motivation for the authors to choose liver tumor model
While the addition of these datasets was based on a criticism raised by other referees, this question raised by the Reviewer 2 led us to examine whether the algorithms developed in this study (Fig. 4 and Suppl. Fig. 3) is useful to extract the boundaries of tumours which are developing in other organs such as brain. To collect such data on T-dominant and Pdominant fingerprints of SERS signals, we applied the model of murine Ink4a/Arf-null neural stem/progenitor cells expressing H-RasV12 and GFP which forms glioblastoma-like tumors.
The difference spectra between the ipsilateral glioma regions and the contralateral normal regions (striatum, Suppl. Fig 15a and Fig 10 in the revised version) indicated T-dominant and P-dominant SERS peaks, suggesting the presence of metabolic fingerprints to discriminate the two regions.
To extract the boundaries between tumours and parenchyma, the data of Raman peaks were not accumulated, but processed directly at the smallest detector resolution (~ 2 cm -1 ). The protocol of this kind benefits precise extraction of the specific Raman shifts (Suppl. Tab. S1) to distinguish tumour-dominant signals from parenchyma-dominant ones.
This process was necessary to improve the accuracy rates of tumour boundary extraction as described later in this section. As a result, GNF-SERS analyses yielded 16 T-dominant and 10 P-dominant peaks (Suppl. Fig. 15a and Table S3) in the revised version. Furthermore, The GNF-SERS imaging for accumulative data (central peaks +/-10 cm-1 ) revealed the two major Raman peaks at 485 cm -1 and 726 cm -1 The peak at 726 cm -1 appeared to be derived from metabolites with adenine ring (e.g. IMP and AMP. See Reference 15 from our lab.). Such a hypothesis was in good agreement with the vacuum-type imaging MS data showing distribution of AMP and IMP in the serial tissue section (Fig. 10g in the revised version).
As the reviewer mentioned, livers seem to possess larger number of P-dominant marker peaks than brain. Furthermore, in case of brain analyses, we must further consider intra-organ spatial heterogeneity of metabolites. Because of this, the striatum region of the contralateral hemisphere was carefully chosen for the control region versus glioblastoma occurring in striatum of the ipsilateral hemisphere. These lines of the data were added in the revised version (Page 10, last para. with Fig 10 and Suppl Fig 15). Thank you for your question on this important issue. Fig. 3d, what is the 480 cm-1 Raman shift responsible for?

Q6 In
A6. Thank you for asking us the very important question. In the original version of our manuscript, we were unable to identify molecules responsible for the SERS peak at 480 cm -1 . The question from the other reviewers led us to perform additional experiments using an orthotopic syngeneic mouse model of glioblastoma, and the analyses of difference GNF-SERS spectra (Suppl . Fig 15a) gave us a clue to answer to this question ! Interestingly, likewise in the liver xenograft cancer model, we observed the SERS peak at 485 cm -1 as a Tdominant peak (Suppl Fig 15a). Careful surveillance of previous Raman reports (Janz GJm et al. 1976 Inorg Chem, reference 47) showed that S-S stretches of polysulfide compounds yield the peaks at around 460~480 cm-1. Based on the conventional analyses (Ref. 47) of crystals of Na 2 S x compounds which are commercially available, we collected the new data showing Na 2 S 3 and Na 2 S 4 yields robust Raman signals with multiple and broad peaks at around 450~485 cm-1, which coincided with the peak at < 300 cm-1, indicating the Au-S stretch (Suppl Fig 15c). Use of GNF-SERS for examining these sodium polysulfides in distilled water (Suppl Fig 15d) markedly enhanced both Au-S and S-S stretch signals. Although the concentrations of Na 2 S x , which were used in the experiments in vitro, seemed high as compared with the physiological concentrations (Few methods have been available to determine the exact contents of these metabolites), these results raised a possibility that endogenous polysulfide compounds serve as candidates of metabolites yielding the peak at 480~485 cm -1 .
While there are only limited methods to determine specific polysulfide compounds, overall amounts of H 2 S (the compound yielding no peak at 480 cm -1 : Suppl Fig 4a) and polysulfides were able to be determined by monobromobimane-assisted derivatization of Scontaining compounds (References 39 and 48) using LC-MSMS. We thus conducted new experiments to measure these metabolites, and the result showed that the tumour-bearing brain tissues displayed significant elevation of H 2 S/polysulfides than the tumour-free brain tissues (Suppl Fig 15e). This result is consistent with a notion that glioblastoma generates endogenous polysulfides, which result in generation of T-dominant SERS peaks at around 480 cm -1 . Fig. 3b indicated the peaks at 480 cm -1 occurs in colon cancer xenografts as well as the liver parenchyma. Since the liver constitutes another source of endogenous polysulfides, according to previous articles showed (References 52 and 67), the presence of SERS peaks in the liver (and tumours) at 480 cm -1 does not seem unreasonable.
Q7：When the Raman shift, 968 cm-1 is considered as mainly hypotaurine (HT), could you provide a discussion why beta-carotene and retinol palmitate and retinol can't be also considered? It exhibits lower signal intensity, but not negligible.
A7: Thank you for your careful review and comments. The Reviewer 2's question led us to re-examine the "optimal" peak for hypotaurine which includes the shoulder of retinol/retinol palmitate and beta-carotene at 968 cm -1 . First, we measured amounts of retinol palmitate, retinol and beta-carotene in the control mouse liver (Suppl. Info. Page 3, 1 st para). The data showed that retinol palmitate constituted a major metabolite, the concentration of which was ~ 0.9 mM, which retinol and beta-carotene were undetectable.
Thus, in new experiments in vitro, we compared SERS signals of hypotaurine and retinol palmitate at 1 mmol/L in methanol solution (Suppl. Fig 4b). By shifting the summation band to higher wave number towards 978 cm -1 , we were able to extract HT signal with minimizing contamination of signals derived from retinol palmitate. Accordingly, HT data were reanalyzed by a new criterion of the summation of the band for wave numbers (978 ± 8cm -1 in Suppl. Figs. 4b-c in the revised version). The description of the corrected wave number was underlined throughout the text and Figs and Suppl. Info.

Q8. I do love the MS analysis and DFT calculation, however, it would be better to make MS
analysis discussion more concise.
A8. Thank you for your question: in the revised version, we moved most of the datasets of MS analyses towards Suppl. Info. All portions which were moved to Suppl. Information were identical to those in the original version, and underlined by broken lines.

Q9
Why there is an increase of GS and HT signal not only for shCD44, but also with shControl compared to the signals obtained from normal liver control (Fig.3 and 7)? A9. Thank you for an important question. According to the Reviewer's suggestion in Q7, we re-assessed the optimal SERS band for HT at 978 cm -1 (Fig. 7). Under these circumstances, there were no differences of HT signals among the normal, shControl-and shCD44-bearing livers.
As the Reviewer pointed out, the GS signal increased in tumour-bearing livers (both parenchyma (P) and tumours (T)). We previously reported that GSH elevation results from CD44-mediated stimulation of cysteine incorporation into cancer cells (Reference 11 in the revised version). While detailed mechanisms were unknown, we previously reported that tumour metastases caused regenerative responses of the liver parenchyma which secondarily stimulate GSH elevation in the parenchyma (Reference 12). These lines of description were briefly added in the revised version.

Reviewer #3 (Remarks to the Author):
This is a good but not excellent piece of work reporting on the use of large-area nanostructured substrates for surface-enhanced Raman screening detection of metabolites in tumor and normal tissues. It is based in the author's recent work in showing that beanshaped gold nanostructures can be produced by angled deposition, and such shaped nanostructures show excellent properties for surface plasmonic enhancement. The current work is to apply this type of substrates for potentially distinguishing normal and tumor tissues by measuring their metabolites and other small molecules such as glutathione and hypotaurine. One major advantage is that this is a "label-free" approach, and the tissues can be measured without the use of injected agents or cellular/tissue stains. The authors have carried out detailed and careful studies, but the overall approach has two major problems that will likely limit the utility of this strategy. First, metabolites and small molecules are often not a good indicator of tumor development or malignancy, so they are not reliable markers for differentiating tumors from healthy tissues. Second, there is considerable ambiguity because the reported spectral differences are not distinct enough for evaluating highly heterogeneous clinical tissue specimens. For these reasons, I don't believe that this work is suitable for publication in a Nature series journal.
Thank you for giving us critical comments. As the Reviewer pointed out, small molecules are often not reliable markers for differentiating tumours from healthy tissues. In the revised version, we therefore deleted a description in Discussion which suggested clinical application of the technique, but added some problems (e.g. effects of vascular ligation on alterations in small molecular metabolites in tumours and surrounding tissues), when the technique is applied for clinical samples. But, on the other hand, we would emphasize that the tumour borders can be extracted by examining 28 (6 are Tumour-dominant and 22 are Parenchyma-dominant. See Table S1 in the original version) SERS peaks, but not by determining only 2 metabolites such as GS and HT.
Another point raised by the Referee is "considerable ambiguity", because the reported spectral differences are not distinct enough for evaluating highly heterogeneous clinical tissue specimens. We attempted to tackle this problem by using clinical samples and submitted protocols of clinical study to institutional IRB, but unfortunately, it seemed to take more than 3 months because of a slow process of paper works. Instead, to overcome this problem, we performed new animal experiments in the revised version using the model of murine Ink4a/Arf-null neural stem/progenitor cells expressing H-RasV12 and GFP (forming glioblastoma-like tumors with a 100% penetrance upon orthotopic implantation in syngeneic mice. The glioblastoma model enabled us to examine more complex pathological features analogous to human glioblastoma with vascular involvement, inflammatory responses and gliosis in the surrounding brain tissues (Refs 45~47) SERS imaging and regionspecific analyses of the spectral features revealed several tumour-dominant and parenchyma-dominant metabolites, and the current method enabled us to distinguish the tumour regions from non-tumour intact regions even in this orthotopic syngeneic mouse glioblastoma model which showed complicated pathology. These data were added in Fig. 10 and Suppl. Fig. 15 in the revised version (17-21247A).