Add files using upload-large-folder tool

.gitattributes

@@ -825,3 +825,117 @@ preprint/preprint__385eba12651ca14ca40446725b2cdb543d0ea1431d5e9b1326fccd6ab5c05
 preprint/preprint__3804f4d067d29fa6c47e21129b1f36471a862895cea382b50ff004a7b0975fb7/preprint__3804f4d067d29fa6c47e21129b1f36471a862895cea382b50ff004a7b0975fb7_layouts.pdf filter=lfs diff=lfs merge=lfs -text
 preprint/preprint__39174209d0233158e33c46ef361d060c6a2a584d09d7e6ec7d58818d7e86fec6/preprint__39174209d0233158e33c46ef361d060c6a2a584d09d7e6ec7d58818d7e86fec6_layouts.pdf filter=lfs diff=lfs merge=lfs -text
 preprint/preprint__382891b67562d7dea62f7b2296235e31ef97c54879b0cd210b88fe2bc6b0bb0e/preprint__382891b67562d7dea62f7b2296235e31ef97c54879b0cd210b88fe2bc6b0bb0e_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__392108643a203326a80265726f6bb7237d7c55a374fc63cbf41bb0b6dcaef2c9/preprint__392108643a203326a80265726f6bb7237d7c55a374fc63cbf41bb0b6dcaef2c9_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3d4c909b33a892d3ce9290a5a247fa2f402a1dea034ca1ceef8ddfd146967e63/preprint__3d4c909b33a892d3ce9290a5a247fa2f402a1dea034ca1ceef8ddfd146967e63_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3cec3e9a73bee766e2b674ba16e1fe22921da920c2b05e83ca8fdaace6f48c0b/preprint__3cec3e9a73bee766e2b674ba16e1fe22921da920c2b05e83ca8fdaace6f48c0b_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3d887a9aa691dba3a0f8ab2dd8af12b6712c746fc5c50d92e352d907ed63d557/preprint__3d887a9aa691dba3a0f8ab2dd8af12b6712c746fc5c50d92e352d907ed63d557_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3d66c399a2dbcfa2ece0132fa58c977bdb5645afd48433f18cd6dc1d49d0d239/preprint__3d66c399a2dbcfa2ece0132fa58c977bdb5645afd48433f18cd6dc1d49d0d239_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3c980a9d836d45c440e1873c28630b0774f4ecaf65d2aaa6b0209006cb01f784/preprint__3c980a9d836d45c440e1873c28630b0774f4ecaf65d2aaa6b0209006cb01f784_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3d36d9b0dee867b947fc48a74a7ac6867bcfe4bcb3fb08b3a93c5d08a93a8bea/preprint__3d36d9b0dee867b947fc48a74a7ac6867bcfe4bcb3fb08b3a93c5d08a93a8bea_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3d75317a7a35d815ee7778714a67b43c1f44538a4dd2a60c0da17c0038e767be/preprint__3d75317a7a35d815ee7778714a67b43c1f44538a4dd2a60c0da17c0038e767be_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3daff2a4adcc2d34e5bbae85728b06976b614edf99b302743e3b1dab823f27d7/preprint__3daff2a4adcc2d34e5bbae85728b06976b614edf99b302743e3b1dab823f27d7_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3da6b0c6f58f3d1aa79ef5cbe3a46097c68ac96896dc9df0cc937e7ab7e57220/preprint__3da6b0c6f58f3d1aa79ef5cbe3a46097c68ac96896dc9df0cc937e7ab7e57220_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3db73133208fbdb7959c6aefdf42a73c95e4108483437cc4dd7e9e5241c96b10/preprint__3db73133208fbdb7959c6aefdf42a73c95e4108483437cc4dd7e9e5241c96b10_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3dace3599649bd24eba189c0f7a92111e9f3c702b68b15ce03bcadff93ee02f3/preprint__3dace3599649bd24eba189c0f7a92111e9f3c702b68b15ce03bcadff93ee02f3_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3d26fceb5002e445f20f5fe4b456eadaa6e05a92d63b36444d1c9993ef3a53b0/preprint__3d26fceb5002e445f20f5fe4b456eadaa6e05a92d63b36444d1c9993ef3a53b0_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3da6f7c1dbe65017857cce6485ddcabd264ffece0167ad76290b885845fabdf6/preprint__3da6f7c1dbe65017857cce6485ddcabd264ffece0167ad76290b885845fabdf6_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3de0b315b8b2058df5ebd9726503225df489f1898a62be7e6a8076ff55f35456/preprint__3de0b315b8b2058df5ebd9726503225df489f1898a62be7e6a8076ff55f35456_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3dcfaaa5ae4f03a4b4058598377f9b07db935a618ab3a5d2e98cdd736ce0ffcb/preprint__3dcfaaa5ae4f03a4b4058598377f9b07db935a618ab3a5d2e98cdd736ce0ffcb_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3e48a2fc9760fc035af2cdb1928d07602eea786f8d3f3ac1174a2ce0a8f9ba07/preprint__3e48a2fc9760fc035af2cdb1928d07602eea786f8d3f3ac1174a2ce0a8f9ba07_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3dedbf7ca8dc430a97f77d4627ef8ab0f30941392a786d5ef019b641c43efd5e/preprint__3dedbf7ca8dc430a97f77d4627ef8ab0f30941392a786d5ef019b641c43efd5e_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3f0dba21f43c7defa4497945e5fbabdda25213dc0be5fbf6b30029f1dd145cf3/preprint__3f0dba21f43c7defa4497945e5fbabdda25213dc0be5fbf6b30029f1dd145cf3_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3e6529cc94944971bf23bb398aa126357fece4f7d2280ff48ab1abbb1b9bad41/preprint__3e6529cc94944971bf23bb398aa126357fece4f7d2280ff48ab1abbb1b9bad41_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3e471a92cf652ac3bf640908fdee5b75876a14cd71deba1e8cd13351491ff978/preprint__3e471a92cf652ac3bf640908fdee5b75876a14cd71deba1e8cd13351491ff978_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3e9c3359b0dbb9cebdc84cd7c3e91539ac216ffb4da3ccd80d3cfaacff69aa6e/preprint__3e9c3359b0dbb9cebdc84cd7c3e91539ac216ffb4da3ccd80d3cfaacff69aa6e_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3f08bc92f01efb62ff7bd7a06ac71cf9065d15926b57a874ddd26ccfee9d223a/preprint__3f08bc92f01efb62ff7bd7a06ac71cf9065d15926b57a874ddd26ccfee9d223a_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3e3ce8942455f521a3ff83699282360fedd22f778ac961a6aa218ba2ec60208e/preprint__3e3ce8942455f521a3ff83699282360fedd22f778ac961a6aa218ba2ec60208e_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3f8a40d254876ef32383ce75948d98733d1d2374f1d3ae3de8792bb4f90a2a0d/preprint__3f8a40d254876ef32383ce75948d98733d1d2374f1d3ae3de8792bb4f90a2a0d_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3fac8dc00d4a5720b8defbea6aa58f18f4a0e8b95ae84bf55ac07421aaa0dbcf/preprint__3fac8dc00d4a5720b8defbea6aa58f18f4a0e8b95ae84bf55ac07421aaa0dbcf_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3f7b1c1452f75398fee29b3c6eece66dbcb6e1cd8ed4cd8f399ecf1d72dbb7c0/preprint__3f7b1c1452f75398fee29b3c6eece66dbcb6e1cd8ed4cd8f399ecf1d72dbb7c0_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3ef01892f39ea120df6b13c6fd17e18e296c05b71bf7c7fea1e661282b66363d/preprint__3ef01892f39ea120df6b13c6fd17e18e296c05b71bf7c7fea1e661282b66363d_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3b9c8d82c935aec234f18b48aea21b339ec4e0a67b2824cb25647c4a489e57ba/preprint__3b9c8d82c935aec234f18b48aea21b339ec4e0a67b2824cb25647c4a489e57ba_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3b33893b9eead68bb9de00948dc96940e8118e56f193e989fbdd6045cee7b551/preprint__3b33893b9eead68bb9de00948dc96940e8118e56f193e989fbdd6045cee7b551_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3b4a323d4faa60d098f8ad3666a03cd4e164672a457b181247197da1ee488d27/preprint__3b4a323d4faa60d098f8ad3666a03cd4e164672a457b181247197da1ee488d27_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3af957e8464cf085e9d891b494f06806e3d28dff979e07f8a37c51be0c0cedfa/preprint__3af957e8464cf085e9d891b494f06806e3d28dff979e07f8a37c51be0c0cedfa_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3bd81a72867b860657fc4c78e4994b403bcfb886c91a39bae86f6b3e29f4efd9/preprint__3bd81a72867b860657fc4c78e4994b403bcfb886c91a39bae86f6b3e29f4efd9_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3b59f892148313cc345f347b0bcc36270236e59453d1dc5cd968a79ae24c9841/preprint__3b59f892148313cc345f347b0bcc36270236e59453d1dc5cd968a79ae24c9841_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3bcda0b7009957061396c838ecf4c9f0125f75a6c9f74307b291ead881e1e3ec/preprint__3bcda0b7009957061396c838ecf4c9f0125f75a6c9f74307b291ead881e1e3ec_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3b660e01869a8b7aa521c97af0f2e41877441c649387c2eb05095addb3bc8c5c/preprint__3b660e01869a8b7aa521c97af0f2e41877441c649387c2eb05095addb3bc8c5c_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3bff87d1b34964d77e9b614b539205db568c98abb73c66f53d850637fc2e60a7/preprint__3bff87d1b34964d77e9b614b539205db568c98abb73c66f53d850637fc2e60a7_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3be9d54e25f4f4792beb16ccc8be383a7b29531ad3fcf4aba301337fb6c4be78/preprint__3be9d54e25f4f4792beb16ccc8be383a7b29531ad3fcf4aba301337fb6c4be78_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3bfe8562b7f491e8a7e27faa44870885872f1e420ab4e8b487231927aa575f7e/preprint__3bfe8562b7f491e8a7e27faa44870885872f1e420ab4e8b487231927aa575f7e_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3c312e4f6747ae3f2651de29c4701b28262d04b354478f6feaf8c05edab94d5a/preprint__3c312e4f6747ae3f2651de29c4701b28262d04b354478f6feaf8c05edab94d5a_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3be079717797521ad7cf08b4a529b0cba1f7360c373382086462b1c0e8e948bc/preprint__3be079717797521ad7cf08b4a529b0cba1f7360c373382086462b1c0e8e948bc_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3b483e628a9da69e0d3f5841a3bbe788eef5cc4a8983e564bb34ca45b54094c7/preprint__3b483e628a9da69e0d3f5841a3bbe788eef5cc4a8983e564bb34ca45b54094c7_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3c44fee1be76e59e0d12156ff42dc84e2e7aa008789150907c4922634a4cff10/preprint__3c44fee1be76e59e0d12156ff42dc84e2e7aa008789150907c4922634a4cff10_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3c84502afdca112bad7382e379d882c282b6b5436b7bb77a15033fb25dd043fc/preprint__3c84502afdca112bad7382e379d882c282b6b5436b7bb77a15033fb25dd043fc_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3c30fc18e852ee3aa258f2095bd0e59ef5e9263212e635a56b7a776499b09348/preprint__3c30fc18e852ee3aa258f2095bd0e59ef5e9263212e635a56b7a776499b09348_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3c8d3ff33612502f64b1d8e43d4ce53d1a5a201655a1da1b5d2071cd0bc324e0/preprint__3c8d3ff33612502f64b1d8e43d4ce53d1a5a201655a1da1b5d2071cd0bc324e0_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3c82f547af63d8508927d78b797dbef02af4d2097b22a5ea778f5f6ff39d40d0/preprint__3c82f547af63d8508927d78b797dbef02af4d2097b22a5ea778f5f6ff39d40d0_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3cb23c6a81abc42ad5755b944a2bfbdf675c861ece974ded9e15eaa6fbd1dedd/preprint__3cb23c6a81abc42ad5755b944a2bfbdf675c861ece974ded9e15eaa6fbd1dedd_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3cd3a5055f7cbc61a085aa8307b13bb54dc2b085c9b4d3a36f239e87940a2e73/preprint__3cd3a5055f7cbc61a085aa8307b13bb54dc2b085c9b4d3a36f239e87940a2e73_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3ccf5418df2113c24931c29a01151dd63c4aafa92c30ef2654f34e64b6a2e44a/preprint__3ccf5418df2113c24931c29a01151dd63c4aafa92c30ef2654f34e64b6a2e44a_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3e4f3705a15f1b72ec039c6456d7d9afe52a94f124902f9877eef3985cd61c88/preprint__3e4f3705a15f1b72ec039c6456d7d9afe52a94f124902f9877eef3985cd61c88_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3fa8310f63bb8662939272aa3593e5ef2f2dc9d8fe3656d2324ddc70ec3ea9cd/preprint__3fa8310f63bb8662939272aa3593e5ef2f2dc9d8fe3656d2324ddc70ec3ea9cd_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3ffe0c14c09e593a961ddc489f2292609d69adb16b874700f8ce7de89502bc17/preprint__3ffe0c14c09e593a961ddc489f2292609d69adb16b874700f8ce7de89502bc17_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3e822736679c95d43c9a8744b5255a1986d25d972b0a0e66d6e62eb68534d9b1/preprint__3e822736679c95d43c9a8744b5255a1986d25d972b0a0e66d6e62eb68534d9b1_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__401052def592a707f39c58fa1a6dc997aca95a132e500e929448ba2631e34938/preprint__401052def592a707f39c58fa1a6dc997aca95a132e500e929448ba2631e34938_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3fef3e66f681031be3a99ffbb2944c6ba64960af251f01bfd83241eb03f9ed3e/preprint__3fef3e66f681031be3a99ffbb2944c6ba64960af251f01bfd83241eb03f9ed3e_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3fc72f1e9a17e2bb00988df2a93fef6b38f66719228bf0615d75dea379eea95e/preprint__3fc72f1e9a17e2bb00988df2a93fef6b38f66719228bf0615d75dea379eea95e_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__4007ec3d4e97f862feea6fb4f231b5e29bded5acc59f7f589451a804a7834ac5/preprint__4007ec3d4e97f862feea6fb4f231b5e29bded5acc59f7f589451a804a7834ac5_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__3feab0de8a89ff018e4282ce265b5dbcb28c6b11ea0b470f895a56f39af284a3/preprint__3feab0de8a89ff018e4282ce265b5dbcb28c6b11ea0b470f895a56f39af284a3_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__4042950bc0bc46a5d6a21966eef630879380f4f0c96061789b12fe96518c827e/preprint__4042950bc0bc46a5d6a21966eef630879380f4f0c96061789b12fe96518c827e_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__403ec6da2694449fcfc2b5d1e73e688bc305580ce773f280daa9a1a3c7387963/preprint__403ec6da2694449fcfc2b5d1e73e688bc305580ce773f280daa9a1a3c7387963_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__40251bce42a39adfddfeab4efd14ee10f570c3f236d375df176ade5ad73983dc/preprint__40251bce42a39adfddfeab4efd14ee10f570c3f236d375df176ade5ad73983dc_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__4006797545e18ce576ac06bd9ebce3a2c176fc474fc9745c9ded7858f22f8316/preprint__4006797545e18ce576ac06bd9ebce3a2c176fc474fc9745c9ded7858f22f8316_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__40458895b1986fd180c56085277e1a7fb3576371bf213d2e79c0b9519572c441/preprint__40458895b1986fd180c56085277e1a7fb3576371bf213d2e79c0b9519572c441_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__408bff0163f9fd9040d50cf11154e3c49abffd76eca1b3213580a857a729f58c/preprint__408bff0163f9fd9040d50cf11154e3c49abffd76eca1b3213580a857a729f58c_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__40813a865f675034383df6d42b0566aa6691f1c5b6bb1ef742593c4696a611b0/preprint__40813a865f675034383df6d42b0566aa6691f1c5b6bb1ef742593c4696a611b0_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__40303ce51cba15f9bc54f0a3a26d50387cbce88ae565c3df97df7aa39fcb35bc/preprint__40303ce51cba15f9bc54f0a3a26d50387cbce88ae565c3df97df7aa39fcb35bc_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__40ca971b1db095a25d09d69e308ad56d4a7956b856a8a0c6cfa86e7d225d7630/preprint__40ca971b1db095a25d09d69e308ad56d4a7956b856a8a0c6cfa86e7d225d7630_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__40fe8134f0c115c81aacebf6a22447d2991307a01ea47736fdf29205c60ae2cb/preprint__40fe8134f0c115c81aacebf6a22447d2991307a01ea47736fdf29205c60ae2cb_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__40dc4ebd7a61f6f4ab76218238d2bc278f6e34db09daf7c104505796a153297f/preprint__40dc4ebd7a61f6f4ab76218238d2bc278f6e34db09daf7c104505796a153297f_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__4024343884f52d45cde9702eaeba5cac1c92c5db02b5936c1d2ad1fa6b37eca7/preprint__4024343884f52d45cde9702eaeba5cac1c92c5db02b5936c1d2ad1fa6b37eca7_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__40b8c8c18cb79229e69203114f5df6c28d9b991c6dcdc6797664f5d55714faaf/preprint__40b8c8c18cb79229e69203114f5df6c28d9b991c6dcdc6797664f5d55714faaf_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__40dd48c2038698367c3078c58c8629fcd6b3fa8603918e646d0c05da9009685c/preprint__40dd48c2038698367c3078c58c8629fcd6b3fa8603918e646d0c05da9009685c_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__413d403d33d701e2de74e57f7015922ff20ebbc6cdf3506db1f6078098c231e6/preprint__413d403d33d701e2de74e57f7015922ff20ebbc6cdf3506db1f6078098c231e6_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__415ad37dc70afb1f55f617b8fc3fd3f8fbec446f19dd1a78191d6d302b0516dc/preprint__415ad37dc70afb1f55f617b8fc3fd3f8fbec446f19dd1a78191d6d302b0516dc_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__40933eae2bd1a05a737eb27ed7403f2aed7bde62314b2b7225c4accbac207c48/preprint__40933eae2bd1a05a737eb27ed7403f2aed7bde62314b2b7225c4accbac207c48_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__405a7cd58813aacfa43f90fe3fbb57ee49bf754e7b6e8d36b426c3e215c3bce0/preprint__405a7cd58813aacfa43f90fe3fbb57ee49bf754e7b6e8d36b426c3e215c3bce0_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__411b58522f3f4c2aea78ebeda083a7bbeff2fb49e757dc238124fcc7c6580abc/preprint__411b58522f3f4c2aea78ebeda083a7bbeff2fb49e757dc238124fcc7c6580abc_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__411f68b0072d1982d41fa9ef4552289f25b219c2817e825365f72fd748f7b0cb/preprint__411f68b0072d1982d41fa9ef4552289f25b219c2817e825365f72fd748f7b0cb_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__41393f2f83f62809f436ab9d0aa8169ea4b417a252c50a631ef48724c1a6dd14/preprint__41393f2f83f62809f436ab9d0aa8169ea4b417a252c50a631ef48724c1a6dd14_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__41aa10c555038d95b50df5ed6d0e642149771b01f23c660f9cfd134323f39c51/preprint__41aa10c555038d95b50df5ed6d0e642149771b01f23c660f9cfd134323f39c51_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__41f73bb80f7f4860b41cedbe7ceb9226dfa693d3d58be9b65f6123b0558820c4/preprint__41f73bb80f7f4860b41cedbe7ceb9226dfa693d3d58be9b65f6123b0558820c4_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__42151a82c0281305539beb0a5a086a26c865e127f2c665b3801e9dc9d28b53f2/preprint__42151a82c0281305539beb0a5a086a26c865e127f2c665b3801e9dc9d28b53f2_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__41fb1a86f69cec4959ccaa38486a6a3e79227b326beb0a65cf327ec871396ccf/preprint__41fb1a86f69cec4959ccaa38486a6a3e79227b326beb0a65cf327ec871396ccf_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__4244a34dfb55a6cd608059035d42b932ec3f8a4f2a9d25b0367057b5f6547fe8/preprint__4244a34dfb55a6cd608059035d42b932ec3f8a4f2a9d25b0367057b5f6547fe8_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__41f876c27c0e1382b538f6f6956a01ddcd533eea25edced44d573a2da694537a/preprint__41f876c27c0e1382b538f6f6956a01ddcd533eea25edced44d573a2da694537a_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__423f716c7bee5b7e2aadcb37b9787c0560a003bd0a3ef89ceb22c4eb562faa96/preprint__423f716c7bee5b7e2aadcb37b9787c0560a003bd0a3ef89ceb22c4eb562faa96_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__4228ec7d6481bacf2560445f50fe7bcf8f7676305226d1aaf248f6d5fb553c25/preprint__4228ec7d6481bacf2560445f50fe7bcf8f7676305226d1aaf248f6d5fb553c25_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__42236cdce814b6f16452541b829274fa16205f7d595121945ec587de3c00c67c/preprint__42236cdce814b6f16452541b829274fa16205f7d595121945ec587de3c00c67c_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__422eb454ab109a207e99c88e9eba2e272d57ac88a694913805829cddff0fe530/preprint__422eb454ab109a207e99c88e9eba2e272d57ac88a694913805829cddff0fe530_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__42e78634e2d28d5b1674fd88701cd63406dcdb39a52d86f577116a7e6cb4c4d5/preprint__42e78634e2d28d5b1674fd88701cd63406dcdb39a52d86f577116a7e6cb4c4d5_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__42b24df073d0df35238c69cc51af1604a05a01628c1be1ed9acaba5415bba72b/preprint__42b24df073d0df35238c69cc51af1604a05a01628c1be1ed9acaba5415bba72b_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__42e00d20e9d549cfc52e3f47786978426d2bd90cf868b9b2017627c51f0f960a/preprint__42e00d20e9d549cfc52e3f47786978426d2bd90cf868b9b2017627c51f0f960a_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__42d4699c4e65ed95775b859ea7b95f01d6cc639156a8bd344e894404ab1add48/preprint__42d4699c4e65ed95775b859ea7b95f01d6cc639156a8bd344e894404ab1add48_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__4302c12f3068b25952571899019868b7bfdd86ab8da45890222281471ac5f9e3/preprint__4302c12f3068b25952571899019868b7bfdd86ab8da45890222281471ac5f9e3_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__433a982c7f92e80ecabf78601dad5208f2632f6c8bc07d7d182169b8d0461ae3/preprint__433a982c7f92e80ecabf78601dad5208f2632f6c8bc07d7d182169b8d0461ae3_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__441f9407302393914095b7edc3ad47804b8a172b40d660d04f980b77da7e41d6/preprint__441f9407302393914095b7edc3ad47804b8a172b40d660d04f980b77da7e41d6_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__43f57920d0cc1b4b5cc7738208cf1e5082ada58cf4830c9ce4fa16eabc64f49e/preprint__43f57920d0cc1b4b5cc7738208cf1e5082ada58cf4830c9ce4fa16eabc64f49e_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__436406321e00a52550b472893c19a8554e472787fd2aa808f22288ed54eb9a65/preprint__436406321e00a52550b472893c19a8554e472787fd2aa808f22288ed54eb9a65_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__43865603d5a26c4d69a5cc2b37b48b2437fdf8a2a31357deb4840ab90c8a6924/preprint__43865603d5a26c4d69a5cc2b37b48b2437fdf8a2a31357deb4840ab90c8a6924_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__43470755cca0263d20ba27a2e315ebda8bd3f51b485e75fb05673e70392cb22a/preprint__43470755cca0263d20ba27a2e315ebda8bd3f51b485e75fb05673e70392cb22a_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__4347ad13967a0fb2d71346bb2c587edc0d41f311204b855d620ba628a8cbde27/preprint__4347ad13967a0fb2d71346bb2c587edc0d41f311204b855d620ba628a8cbde27_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__43bb2eb4af46fc261e5879f2334298e35bfae4fe061d0255df4dae97fb2b4fb6/preprint__43bb2eb4af46fc261e5879f2334298e35bfae4fe061d0255df4dae97fb2b4fb6_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__43d05b5532f3de5b25ca7acf67ecece0ce1dad6a611a6b1a0acd967fded10394/preprint__43d05b5532f3de5b25ca7acf67ecece0ce1dad6a611a6b1a0acd967fded10394_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__43fb5128bebda86efbec026bfbec7f5523ac49c2619d58b77b26e2f2828414ac/preprint__43fb5128bebda86efbec026bfbec7f5523ac49c2619d58b77b26e2f2828414ac_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__443a52426249a8ff8ecc991ca463004118766906ed804812d4cb606632655b20/preprint__443a52426249a8ff8ecc991ca463004118766906ed804812d4cb606632655b20_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__441124b1c3fab64b922c49edf7921f08fbd012f369724bded85496a42ec56c78/preprint__441124b1c3fab64b922c49edf7921f08fbd012f369724bded85496a42ec56c78_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__44369edda93280f645159c8e6434e5b6ddec924a744aa0eb53df797854b332ca/preprint__44369edda93280f645159c8e6434e5b6ddec924a744aa0eb53df797854b332ca_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__43a438ad55c12c7c430994081e89ce1899b856e6563e09e4569576d8deeaaae8/preprint__43a438ad55c12c7c430994081e89ce1899b856e6563e09e4569576d8deeaaae8_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__44a7509242cdaba16c462fd4674041ec91a2babb941823b30d1b4d1fc3bd2b08/preprint__44a7509242cdaba16c462fd4674041ec91a2babb941823b30d1b4d1fc3bd2b08_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__4480d0108b1ac86fbd7de14d6ed11178e827bbb675fa636a1e0079b7242e2256/preprint__4480d0108b1ac86fbd7de14d6ed11178e827bbb675fa636a1e0079b7242e2256_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__446a6d1639de149e2cdadfff5059c93994fb6bf575933900ed819a058513438a/preprint__446a6d1639de149e2cdadfff5059c93994fb6bf575933900ed819a058513438a_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__44bd8119853e2fcc7fb044a9f959e97bc3b0440d36ba5953911b7e794c249100/preprint__44bd8119853e2fcc7fb044a9f959e97bc3b0440d36ba5953911b7e794c249100_layouts.pdf filter=lfs diff=lfs merge=lfs -text
+preprint/preprint__447ea8ea6e006d78bc64133f35a1e45dcbd8cb9ac5fc966fd3285782d3d6b1a8/preprint__447ea8ea6e006d78bc64133f35a1e45dcbd8cb9ac5fc966fd3285782d3d6b1a8_layouts.pdf filter=lfs diff=lfs merge=lfs -text

peer_reviews/supplementary_0_Peer Review file__9a456fb9be35f764c17c5b6789e7c12d7f00d83d968976ea125d37cd9aec4eeb/images_list.json

@@ -0,0 +1,32 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Figure S7. Sensitivity of sea-level contributions with linear Weertman law to various mesh resolutions.",
+    "footnote": [],
+    "bbox": [
+      [
+        297,
+        92,
+        684,
+        275
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2:",
+    "footnote": [],
+    "bbox": [
+      [
+        131,
+        336,
+        844,
+        530
+      ]
+    ],
+    "page_idx": 21
+  }
+]

peer_reviews/supplementary_0_Peer Review file__9a456fb9be35f764c17c5b6789e7c12d7f00d83d968976ea125d37cd9aec4eeb/supplementary_0_Peer Review file__9a456fb9be35f764c17c5b6789e7c12d7f00d83d968976ea125d37cd9aec4eeb.mmd

@@ -0,0 +1,724 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+# Subglacial Water Amplifies Antarctic Contributions to Sea-Level Rise  
+
+Corresponding Author: Dr Chen Zhao  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+1. What is the element sizes in the Model and how dependent is the SLE for th e different scenarios on the mesh size.In general you should in a few lines give basic information on model parameters  
+
+2. is there a correlation between the bed topography and the distribution of effective pressures und inverted basal drag. If so you should point it out  
+
+3. I would like to point to a publication where although in a different manner the same problem is adressed and maybe you can use it to further complete your argument.  
+
+Goeller et al.:A balanced water layer concept for subglacial hydrology in large-scale ice sheet models, The Cryosphere, 7, 1095- 1106,2013 doi:10.51947c-7-1095-2013  
+
+4. Finally - from a person which is color blind - you would make life much easier for about \(10\%\) of your readers if you would use different line styles in addition to colors  
+
+Reviewer #2  
+
+(Remarks to the Author) Subglacial Water Amplifies Antarctic Contributions to Sea-Level Rise  
+
+The present manuscript investigates the effect of different variants for computing the basal pressure below the Antarctic ice sheet used in the basal sliding law. Basal sliding is a key component in ice sheet modeling that is hard to constrain due to a lack of direct observations and it has a strong influence on the predicted sea level rise and timing of tipping points. Using the ice sheet model Elmer/lce, the authors conduct a number of different experiments each with a different treatment for computing the effective pressure N (ice overburden pressure minus subglacial water pressure) and investigate the effect under several climate forcing scenarios. They show that different Antarctic sectors react differently to the sliding relation chosen, depending mostly on the topography and dominating driving stress regime.  
+
+This study is original in that it focuses on comparing the effects of different sliding relations within the same ice sheet model. Subglacial hydrology is an important missing piece in ice sheet modeling (and especially relevant for sea level rise predictions), and while the paper does not examine the validity of them, it investigates the implications of the different approaches that are used in the wild.  
+
+## REMARKS  
+
+Discussion of sliding laws  
+
+In the introduction, the authors compare the two overarching classes of sliding laws: classical Weertman and regularized Coulomb and state that models often ignore or crudely implement the effective pressure N "by adopting the Weertman
+
+<--- Page Split --->
+
+relation as the commonplace approach" (L. 39f). While this is generally correct, I would prefer a more nuanced statement because the current formulation can imply that most models have no treatment of effective pressure at all. From skimming over Seroussi et al. (2020), I noticed that out of the 13 used models, only 4 to 6 seem to use classical Weertman without considering basal pressure (unfortunately, not all of them clearly state their sliding relation). Maybe the variety in the approaches can be mentioned in the paper. The following paragraph on HAF (L. 45ff) describes one simple approach for including N, but I think it could be better integrated with the previous part.  
+
+Sliding laws used in ISMIP6: AWI_ PISM: School/Coulomb DOE_MALI: linear basal friction law with spatially varying basal friction coefficient ILTS_PIK_SICOPOLIS: Weertman- Budd- type sliding law with sub- melt sliding (Sato and Greve, 2012) and subglacial hydrology IMAU_IMAUICE: ?? probably School/Coulomb? citing Bueler Brown 2009 JPL_ISSM: ?? probably coulomb- like, citing Morlighem et al. 2010 LCSE_GRISLI: power law basal friction (Weertman) NCAR_CISM: basal sliding is similar to that of School (2005), combining power law and Coulomb behavior PIK_PISM: Mohr- Coulomb UCJPL_ISSM: ??? ULB_FETISH: power law basal sliding UTAS_ELMERICE: linear sliding relation VUB_AISMPALEO: Weertman, but with height above bouancy VUW_PISM: probably School/Coulomb  
+
+## 2. Explaining the experiments in the main text  
+
+The Method section does a great job of explaining the different experiments, but it would be helpful for the reader to get a short overview already in the main text to facilitate the understanding of the results. It would be enough to briefly explain the different treatments of N and possibly further explain the difference between HAF and POC (they seem similar in that they both have the goal of enforcing floating condition at the grounding line).  
+
+I would also appreciate a comment on why these particular treatments were chosen; are they representative of what is used in current models?  
+
+It would be good to lead the reader along from simple implementations to more complex/realistic and explain why you chose the GlaDS model out of a variety of subglacial hydrology models (see e.g. De Fleurian et al. 2018). I am a little bit concerned with the validity of the inversion result for future projections when the basal conditions (especially at the grounding line) have changed. This affects mostly the experiments where N is treated implicitly (as part of the inversion). Are they still a good baseline after e.g. a substantial retreat of the GL which would imply a major reconfiguration of the hydraulic system?  
+
+3. Structure of results and headlines  
+
+I believe the presentation of the results could be enhanced for better readability and comprehension. While the beginning that focuses on the implicit handling of N does a good job of describing the difference between LW and RC_IN_HAF, the following parts could be arranged in a more organized way, as I experienced them hard to follow. One idea could be to adapt something like the following structure (very similar to the current one):  
+
+Keep the first part on implicit handling of N: LW vs RC_IN_HAF as it is (L. 59 to L. 80). Then have a second part that covers the explicit handling of N: RC_en_GlaDS vs RC_en_POC vs RC_en_GlaDS_HAF. Here I would suggest to start with RC_en_POC as the most simple approach and then introduce GlaDS and the extended GlaDS (RC_en_GlaDS_HAF).  
+
+With this continental view established, continue with the Tipping points and the regional analysis. Not much change is needed, but these sections should be clearly identifiable and might deserve their own headings, also because the current in- between headings ("Subglacial effective pressure must be included in sliding relations" and "Subglacial effective pressure controls ice fluxes") are very general and do not help much to navigate the text. The final part before the discussion ("Increased sensitivity to effective pressure near the grounding line") is in a good place.  
+
+At some point, Kazmierczak et al. (2022) should also be referenced, as they also investigate the influence of different sliding relations for Antarctica.  
+
+## Remarks by lines  
+
+L. 29: also frictional heating due to internal deformation of ice  
+
+L. 45f: Can't the effect of N be also approximated in the sliding coefficients for Weertman sliding laws? see also concern 1.  
+
+L. 48: more realistic than what?  
+
+L. 56: Why SSA, what is the implication?  
+
+L. 57: Why is SSA used here? Why not hybrid/higher order? What are the implications?  
+
+L. 60: Why is RC_IN_HAF more realistic? is combined effects of seawater intrusion and meltwater production is also considered in the eN variants? is this only about RC vs LW? more explanation is needed.  
+
+L. 68: Replace "near mass balance" with "near equilibrium". I don't understand the second part of the sentence regarding
+
+<--- Page Split --->
+
+observed mass loss.  
+
+L.71: "For WAIS..." Sentence is unclear. Maybe "For the WAIS the experiment using the RC relation and high emission scenario..."  
+
+L. 73: produce.  
+
+L. 74: "However, the RC..." -> "However, using the RC..."  
+
+L. 75: "than LW" -> "than using the LW relation" The whole paragraph seems to be missing words.  
+
+L. 82: This sentence is not very clear, seems obvious.  
+
+L 99ff: Isn't the addition of HAF to the GlaDS N a little bit of double counting, because the effect of the ocean is already included in the GlaDS model? Though it is only appropriate at the time where N was computed in GlaDS.  
+
+L. 106: Explain coulomb limit  
+
+L. 115- 122: Can the experiments using POC produce meaningful results for basins with reverse bed slope? Won't POC on the retrograde bed will lead to higher basal pressure in the inland than at the grounding line?  
+
+L. 153: How valid is the distribution of N computed by GlaDS for present day conditions after more than a century? Even with the HAF scaling, the whole hydraulic regime might have shifted.  
+
+L. 163: What does "constrained future surface mass budgets estimates" mean? Improvement of the predictions of SMB?  
+
+Figure 1: good colors, bigger fonts, no box around legend  
+
+Figure 2: caption: "sea-level contributions between linear Weertman law" -> "sea-level contributions in experiment using linear Weertman law".  
+
+Figure 4: caption: "with a Hydrology model" -> "from GlaDS" Does channel area mean cross section of individual channels? why is a diverging colormap used? What is special about 0.5 MPa? small inset figures in G- H and D- Dp are hardly readable.  
+
+Figure 5: Why are the lines dashed?  
+
+## References  
+
+DE FLEURIAN B, WERDER MA, BEYER S, et al. SHMIP The subglacial hydrology model intercomparison Project. Journal of Glaciology. 2018;64(248):897- 916. doi:10.1017/jog.2018.78  
+
+Kazmierczak, E., Sun, S., Coulon, V., and Pattyn, F.: Subglacial hydrology modulates basal sliding response of the Antarctic ice sheet to climate forcing, The Cryosphere, 16, 4537- 4552, https://doi.org/10.5194/tc- 16- 4537- 2022, 2022.  
+
+Bueler, E., and J. Brown (2009), Shallow shelf approximation as a "sliding law" in a thermomechanically coupled ice sheet model, J. Geophys. Res., 114, F03008, doi:10.1029/2008JF001179.  
+
+Morlighem, M., E. Rignot, H. Seroussi, E. Larour, H. Ben Dhia, and D. Aubry (2010), Spatial patterns of basal drag inferred using control methods from a full- Stokes and simpler models for Pine Island Glacier, West Antarctica, Geophys. Res. Lett., 37, L14502, doi:10.1029/2010GL043853.  
+
+Reviewer #3  
+
+(Remarks to the Author) Review of "Subglacial Water Amplifies Antarctic Contributions to Sea- Level Rise", for Nature Communications, by Zhao, Gladstone, Zwinger, Gillet- Chaulet, Wang, Caillet, Mathiot, Saraste, Jager, Galton- Fenzi, Christoffersen and King  
+
+This manuscript describes the importance of incorporating subglacial hydrology parameterizations into whole Antarctic ice sheet models, by varying parameterization of basal water effective pressure in a basal slip relation. The authors conclude
+
+<--- Page Split --->
+
+that effective pressure is required in the slip relationship, and that ice flux is sensitive to effective pressure model, especially near the grounding line.  
+
+We believe the topic of this manuscript is important and can be of interest to a wide range of readers. However, we believe that this manuscript can be significantly improved by considering the comments listed below.  
+
+For now, the manuscript is "expert- facing", i.e., it is assumed that the readers have a substantial understanding of the background and methods of this study. The manuscript leaps over many (essential) details that could have been helpful for the readers to follow along.  
+
+- For an important study that can be of interest to the general public, especially in the journal of Nature Communication, we argue this is not appropriate. We recommend the authors to provide more detailed elaborations on the background, methods, findings, and implications of this study. A few examples can be found in the comments below.  
+
+- In the section "Subglacial effective pressure must be included in sliding relations"; its implied the observed mass gain of the linear Weertman models demonstrate the need for subglacial effective pressure. The authors need to make a stronger case - ie demonstrate that mass gain is not plausible over these time frames.  
+
+- We recommend the authors to elaborate on the differences between model setups. We appreciate the balance between the amount of work presented in this manuscript and the space limitations. But the current manuscript makes it really difficult to follow the different model setups, especially considering the number of runs that are being presented and discussed.  
+
+- It is not clear why/how including different subglacial hydrology components can change the AIS SLR projection by such a large degree. What is the physical process behind the differences? We recommend the authors to elaborate on this and make it more clear.  
+
+- Line 47-52: HAF is a very important concept for this study. However, the authors didn't provide any explanation/elaboration on the concept, other than providing one reference. (This is also true for some other important concepts throughout the manuscript.) This approach makes it difficult for non-expert readers to understand the work and follow the logic flow. We recommend providing a more detailed explanation of the concept, and/or maybe even a conceptual diagram.  
+
+- It is not clear what assumption is made about the subglacial meltwater budget. Does the model run only require a certain and finite amount of subglacial water? Or do the authors assume an "unlimited" supply of subglacial water generated by basal melting upstream? And does a limited subglacial meltwater budget change the model output?  
+
+\* The authors need to address the limitation of the study in the Discussion section. We understand that the authors need to consider the space limitations per journal guideline, but we strongly recommend the authors to reconsider the balance between (1) further elaboration and discussion of the study findings, (2) connection with other studies, (3) broader implications, and (4) limitation of the study, for the Discussion section.  
+
+minor comments:  
+
+\* This manuscript uses a lot of acronyms - are they really necessary? Like, "GL" for "grounding line"? Having too many acronyms can be counterproductive. Also, it may help to spell out the run names in Figure 1.  
+
+\* We noticed that no doi/web- link is provided for the references. We encourage the authors to consider including such information.  
+
+Unsure of what is meant by the second and third authors contributed equally.  
+
+## Reviewer #4  
+
+(Remarks to the Author)  
+
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+The paper is the result of a numerical study investigating the effect of basal pressure under the Antarctic ice sheet and the role of the pressurized water layer in basal slip. While for a direct prognostic application there is lack of data, the paper defines the range of solutions for Ice Sheet/iceshelf dynamics depending on pressure in particular at he icesheet- iceshelf boundary.  
+
+The authors have responded to all reviewers comments and i am satisfied with the answers to my specific comments.
+
+<--- Page Split --->
+
+## Reviewer #2  
+
+(Remarks to the Author) The authors have addressed the reviewers' concerns effectively. I'm pleased with the improved paper structure and extended description. I have only one minor comment:  
+
+I appreciate the more detailed description of sliding laws used in the ISMIP/ABUMIP experiments, but I would like to point out that the PISM model uses a Mohr- Coulomb type sliding law and not a Weertman type. It is specifically mentioned in the manual that Weertman should not be used with PISM: https://www.pism.io/docs/manual/modeling- choices/dynamics/weertman.html and also in the model description it is pretty clear:  
+
+Seroussi et al. 2020 PIK_PISM  
+
+We apply a power law for sliding with a Mohr- Coulomb criterion relating the yield stress to parameterized till material properties and the effective pressure of the overlaying ice on the saturated till (Bueler and van Pelt, 2015).  
+
+Seroussi et al. 2024: PIK_PISM  
+
+A generalized power law (Schoof & Hindmarsh, 2010) is applied to parameterize basal sliding. The basal friction coefficient depends on the effective pressure and till friction angle, that is parameterized using a heuristic, piecewise linear function of the bed elevation (Martin et al., 2011).  
+
+I know from personal communication that AWI_PISM uses the same setup in that regard.  
+
+This does of course not change the overall conclusion of the authors that Weertman type sliding laws are dominant in the benchmarks.  
+
+## Reviewer #3  
+
+(Remarks to the Author)  
+
+This work illustrates the need for improved knowledge of the evolution of subglacial hydrology, and we recommend publication.  
+
+The revised manuscript reads much better with acronym expansion and with more detailed concept explanations and discussions. Our only minor follow- up suggestion is to explicitly state in the Method section that the subglacial water budget is held constant because (1) the model requires such assumption to reach steady state, and (2) different assumptions on this do not cause significant variation in the modeling results.  
+
+## Reviewer #4  
+
+(Remarks to the Author)  
+
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.
+
+<--- Page Split --->
+
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+## Response to Reviewers  
+
+Many thanks to the reviewers for their valuable comments. Our responses are highlighted in blue, and the line numbers refer to the revised manuscript.  
+
+Reviewer #1 (Remarks to the Author):  
+
+1. What is the element sizes in the Model and how dependent is the SLE for the different scenarios on the mesh size. In general you should in a few lines give basic information on model parameters.  
+
+Thanks for the comments. We have included the element size on the Method Setup Section in the original manuscript "The low-computation cost of SSA model allows a high-resolution mesh of 1 km in fast-flowing regions and near the GL and a coarser resolution (25 km) further inland."  
+
+We have done sensitivity tests to different mesh resolutions (500 m, 1 km, 2 km) near the grounding line with the linear Weertman relation (LW). See the figure below. It suggests that higher resolution would predict less mass loss than coarser resolution, which is consistent with similar sensitivity tests done on Wilkes Subglacial Basin (Wang et al. 2024) and an idealised domain (Seroussi et al., 2018). It can be seen that the 1 km mesh used in this study suggests a more consistent result with 500 m mesh compared with the 2 km mesh. Previous studies (Wang et al., 2024; Seroussi et al., 2018) indicate that the Regularised Coulomb sliding relation with effective pressure included is less sensitive to the mesh resolution compared with LW. So we think the sensitivity of sea-level rise projections to the mesh size is little compared with the dependence on the different treatment of effective pressure.  
+
+To make it clearer and include the information above, we added the figure below into the supplementary material and a few sentences in the Methods section (Line 426- 431).  
+
+"The low-computation cost of SSA model allows a high-resolution mesh of 1 km in fast-flowing regions and near the grounding line, with a coarser resolution (25 km) further inland. This represents the highest resolution used in continental Antarctic ice sheet models from ISMIP6 studies \(^{8,9}\) . Sensitivity analysis using the LW sliding relation (Fig. S7) shows consistency between model outputs generated with 1 km and 500 m meshes. Given the projected ice mass loss using the Weertman sliding relation is more grid- size sensitive than the Coulomb relation used here \(^{10,11}\) , the mesh resolution artefact in this study is acceptable."
+
+<--- Page Split --->
+![](images/Figure_unknown_0.jpg)
+
+<center>Figure S7. Sensitivity of sea-level contributions with linear Weertman law to various mesh resolutions. </center>  
+
+2. is there a correlation between the bed topography and the distribution of effective pressures und inverted basal drag. If so you should point it out  
+
+Thank you for your comment. There is indeed a correlation between the bed topography and the distribution of effective pressure and inverted basal drag. Low- lying areas tend to have lower effective pressure and basal drag, while elevated regions exhibit higher values. To include this, we add one sentence in the Methods section (Line 520- 523):  
+
+"Bedrock geometry and ice thickness strongly influences subglacial water flow and channel distribution, as lower- elevation areas, such as valleys, accumulate water due to hydraulic potential gradients shaped by bedrock topography and ice thickness, resulting in higher water pressures and lower effective pressures."  
+
+The correlation between bed topography and inverted basal drag is mediated by how bed geometry influences ice flow resistance, with high basal drag generally associated with elevated or rough areas and low basal drag linked to smooth, low- lying regions. To include this, we add one sentence in the inversion section (Line 440- 441):  
+
+"The inverted basal drag is largely influenced by the bed geometry through affecting the ice flow resistance."  
+
+3. I would like to point to a publication where although in a different manner the same problem is addressed and maybe you can use it to further complete your argument. Goeller et al.: A balanced water layer concept for subglacial hydrology in large-scale ice sheet models, The Cryosphere, 7, 1095-1106, 2013 doi:10.51947tc-7-1095-2013  
+
+Many thanks for recommending this valuable paper. This paper introduces the balanced water layer concept, which provides a simplified representation of subglacial hydrology suitable for large-scale ice sheet models. We have cited this paper in our Introduction:
+
+<--- Page Split --->
+
+"The widespread production of subglacial water beneath the Antarctic Ice Sheet, and the potential intrusion of warm seawater near the grounding line, plays a crucial role in modulating basal sliding and enabling rapid ice flow \(^{7 - 11}\) ." (Line 26- 27)  
+
+4. Finally - from a person which is color blind - you would make life much easier for about \(10\%\) of your readers if you would use different line styles and addition to colors  
+
+Thank you for highlighting this important point. Although Reviewer 2 noted that Figure 1 uses good colors, we have updated all figures to ensure they are more color- blind- friendly by incorporating both improved color schemes and distinct line styles.
+
+<--- Page Split --->
+
+Reviewer #2 (Remarks to the Author): Subglacial Water Amplifies Antarctic Contributions to Sea-Level Rise  
+
+The present manuscript investigates the effect of different variants for computing the basal pressure below the Antarctic ice sheet used in the basal sliding law. Basal sliding is a key component in ice sheet modeling that is hard to constrain due to a lack of direct observations and it has a strong influence on the predicted sea level rise and timing of tipping points.  
+
+Using the Ice sheet model Elmer/Ice, the authors conduct a number of different experiments each with a different treatment for computing the effective pressure N (ice overburden pressure minus subglacial water pressure) and investigate the effect under several climate forcing scenarios. They show that different Antarctic sectors react differently to the sliding relation chosen, depending mostly on the topography and dominating driving stress regime.  
+
+This study is original in that it focuses on comparing the effects of different sliding relations within the same ice sheet model. Subglacial hydrology is an important missing piece in ice sheet modeling (and especially relevant for sea level rise predictions), and while the paper does not examine the validity of them, it investigates the implications of the different approaches that are used in the wild.  
+
+## REMARKS  
+
+Discussion of sliding laws  
+
+In the introduction, the authors compare the two overarching classes of sliding laws: classical Weertman and regularized Coulomb and state that models often ignore or crudely implement the effective pressure N "by adopting the Weertman relation as the commonplace approach" (L. 39f). While this is generally correct, I would prefer a more nuanced statement because the current formulation can imply that most models have no treatment of effective pressure at all. From skimming over Seroussi et al. (2020), I noticed that out of the 13 used models, only 4 to 6 seem to use classical Weertman without considering basal pressure (unfortunately, not all of them clearly state their sliding relation). Maybe the variety in the approaches can be mentioned in the paper.  
+
+Sliding laws used in ISMIP6:  
+
+AWI_PISM: Schoof/Coulomb
+
+<--- Page Split --->
+
+DOE_MALI: linear basal friction law with spatially varying basal friction coefficient ILTS_PIK_SICOPOLIS: Weertman- Budd- type sliding law with sub- melt sliding (Sato and Greve, 2012) and subglacial hydrology\\(PIMAU_IMAUICE: ?? probably Schoof/Coulomb? citing Bueler Brown 2009\\(PL_ISSM: ??? probably coulomb- like, citing Morlighem et al. 2010 LCSE_GRISLI: power law basal friction (Weertman)  
+
+NCAR_CISM: basal sliding is similar to that of Schoof (2005), combining power law and Coulomb behavior\\(PIPIK_PISM: Mohr- Coulomb\\(UCIJPL_ISSM: ???\\(ULB_FETISH: power law basal sliding\\(UTAS_ELMERICE: linear sliding relation\\(VUB_AISMPALEO: Weertman, but with height above bouncy\\(VUW_PISM: probably Schoof/Coulomb  
+
+Thanks for the suggestions and providing the sliding law used in the ISMIP6- 2100 projections. However, I would like to clarify that, based on Seroussi et al. 2020, 13 out of 16 models in ISMIP6- 2100 have used Weerman law.  
+
+To clarify and respond to this comment we have included a table that summarises the sliding laws used in all recent model inter- comparison projects (ISMIP6- 2100, ISMIP6- 2300, and ABUMIP) in Table S1 (Supplementary Material).  
+
+In Line 47- 51, we added the text:  
+
+"Because the actual distribution of effective pressure under the Antarctic Ice Sheet is unknown, model- based sea- level rise projections typically either use Weertman sliding everywhere, which does not include effective pressure (e.g., refs. \(^{24 - 26}\) ), or make simple assumptions about effective pressure when applying RC sliding, (e.g., refs. \(^{27 - 29}\) ). A summary of sliding laws used in three major Model Intercomparison Projects reveals that the majority of Antarctic Ice Sheet models (e.g., 13 out of 16 models in ISMIP6- 2100 \(^{2}\) , 9 out of 16 models in ISMIP6- 2300 \(^{5}\) , and 7 out of 15 models in ABUMIP \(^{30}\) ) employed the Weertman sliding relation."
+
+<--- Page Split --->
+
+
+<table><tr><td colspan="2">ISMIP6-2100e</td><td colspan="2">ISMIP6-2300e</td><td colspan="2">ABUMIP8</td></tr><tr><td>Model Name</td><td>Sliding Law</td><td>Model Name</td><td>Sliding Law</td><td>Model Name</td><td>Sliding Law</td></tr><tr><td>AWI_PISM</td><td>Weertman(m=3)</td><td>-</td><td>-</td><td>AWI_PISMPal</td><td>Coulomb</td></tr><tr><td>-</td><td>-</td><td>UCM_Yelmo</td><td>Regularised Coulomb</td><td>-</td><td>-</td></tr><tr><td>IMAU_IMAUIC1</td><td>Weertman(m=3)</td><td>MAU_UFEMISM</td><td>Regularised Coulomb</td><td>IMAU_ICE</td><td>Coulomb</td></tr><tr><td>IMAU_IMAUIC2</td><td>Weertman(m=3)</td><td>-</td><td>-</td><td>-</td><td>-</td></tr><tr><td>NCAR_CISM</td><td>Weertman(m=3)-Coulomb</td><td>NCAR_CISM1</td><td>Weertman(m=3)-Coulomb</td><td>NCAR_CISM</td><td>Weertman(m=3)-Coulomb</td></tr><tr><td>-</td><td>-</td><td>NCAR_CISM2</td><td>Zoet-Iversen</td><td>-</td><td>-</td></tr><tr><td>-</td><td>-</td><td>NORCE_CISM</td><td>Zoet-Iversen</td><td>-</td><td>-</td></tr><tr><td>DOE_MALI</td><td>Weertman(m=1)</td><td>DOE_MALI</td><td>Weertman(m=3)</td><td>DOE_MALI</td><td>Weertman(m=1)</td></tr><tr><td>ILTS_PIK_SICOPOLIS</td><td>Weertman(m=3)-Budd</td><td>ILTS_PIK_SICOPOLIS</td><td>Weertman(m=3)-Bud</td><td>ILTS_PIK_SICOPOLIS</td><td>Weertman(m=3)</td></tr><tr><td>JPL_ISSM</td><td>Budd(m=1)</td><td>UCSD_ISSM</td><td>Weertman(m=3)-Bud</td><td>JPL_ISSM</td><td>Weertman(m=1)</td></tr><tr><td>UCIPLP_ISSM</td><td>Weertman(m=3)</td><td>-</td><td>-</td><td>-</td><td>-</td></tr><tr><td>PIK_PISM1</td><td>Weertman(m=3)</td><td>PIK_PISM</td><td>Weertman(m=3)</td><td>PIK_PISM</td><td>Coulomb</td></tr><tr><td>PIK_PISM2</td><td>Weertman(m=3)</td><td>-</td><td>-</td><td>-</td><td>-</td></tr><tr><td>ULB_FETISH_16km</td><td>Weertman(m=2)</td><td>ULB_FETISH</td><td>Weertman(m=3)</td><td>ULB_FETISH</td><td>Weertman(m=2)</td></tr><tr><td>ULB_FETISH_32km</td><td>Weertman(m=2)</td><td>-</td><td>-</td><td>-</td><td>-</td></tr><tr><td>UTAS_ELMERICE</td><td>Weertman(m=1)</td><td>UTAS_ELMERICE</td><td>Weertman(m=1)</td><td>-</td><td>-</td></tr><tr><td>-</td><td>-</td><td>IGE_Emerfce</td><td>Weertman(m=1)</td><td>IGE_Emerfce</td><td>Weertman(m=3)</td></tr><tr><td>VUB_AISMPALEO</td><td>Weertman(m=3)</td><td>VUW_PISM1</td><td>Weertman(m=3)</td><td>-</td><td>-</td></tr><tr><td>VUW_PISM</td><td>Weertman(m=3)</td><td>VUW_PISM2</td><td>Weertman(m=3)</td><td>-</td><td>-</td></tr><tr><td>-</td><td>-</td><td>-</td><td>-</td><td>ARC_PISM1</td><td>Coulomb</td></tr><tr><td>-</td><td>-</td><td>-</td><td>-</td><td>ARC_PISM2</td><td>Coulomb</td></tr><tr><td>LcSE_GRISLI</td><td>Weertman(m=3)</td><td>LCSE_GRISLI</td><td>Weertman(m=3)</td><td>LCSE_ GRISLI</td><td>Coulomb</td></tr><tr><td>-</td><td>-</td><td>LCSE_GRISLI2</td><td>Weertman(m=3)</td><td>CPOM_BISCLES</td><td>Weertman(m=3)-Coulomb</td></tr><tr><td>-</td><td>-</td><td>-</td><td>-</td><td>PSU_PSU3D1</td><td>Weertman(m=2)</td></tr><tr><td>-</td><td>-</td><td>-</td><td>-</td><td>PSU_PSU3D2</td><td>Weertman(m=2)</td></tr></table>
+
+Table S1. Sliding law used in ISMIP6-2100 Antarctic, ISMIP6-2300 Antarctic7, and ABUMIP8.
+
+The following paragraph on HAF (L. 45ff) describes one simple approach for including N, but I think it could be better integrated with the previous part.
+
+To better illustrate the differences in how effective pressure is treated across experiments,we have reorganized the Introduction as suggested by Reviewer 2. We now introduce the explicit representation of effective pressure first (Line 52-59), followed by the implicit representation (Line 60-63). Additionally, we have added a separate paragraph (Line 64-73) specifically discussing HAF-scaling. To further clarify the experimental design, we have also moved Table 1 from the Methods section to the Introduction.
+
+# 2. Explaining the experiments in the main text
+
+The Method section does a great job of explaining the different experiments, but it would be helpful for the reader to get a short overview already in the main text to facilitate the understanding of the results. It would be enough to briefly explain the different treatments of N and possibly further explain the difference between HAF and POC (they seem similar in that they both have the goal of enforcing floating condition at the grounding line).
+
+Thanks for the helpful suggestions. As mentioned in our response to the comment above,we reorganized the Introduction to better explain the difference between experiments
+
+<--- Page Split --->
+
+and added a summary paragraph at the end (Line 74- 81). A modified Table 1 is moved to the Introduction to help better interpret the differences.  
+
+To explain the differences between HAF and POC, we added two sentences at Line 70- 73:  
+
+"Both the HAF- scaling and perfect ocean connection assumption enforce the floating condition at the grounding line. However, HAF- scaling, applied only near the grounding line, adjusts basal resistance based on ice elevation relative to flotation, potentially under- or overestimating effective pressure. In contrast, the perfect ocean connection assumes uniform hydraulic connectivity across the domain, systematically overestimating effective pressure by neglecting hydraulic potential gradients."  
+
+I would also appreciate a comment on why these particular treatments were chosen; are they representative of what is used in current models?  
+
+In the reorganized Introduction section, we now provide a clearer and more logical explanation for why these specific treatments were chosen, with appropriate citations as evidence for their use in current models. To be consistent with the logical structure of the Results section suggested by Reviewer 2, we begin with the explicit representation of effective pressure, employing widely recognized approaches: the perfect ocean connection assumption, which is commonly used to enforce the floating condition at the grounding line (Line 52- 55), and an alternative approach that derives effective pressure using a subglacial hydrology model (GlaDS) to account for spatial variability (Line 56- 59). We then include the implicit representation of effective pressure, as used in Pine Island Glacier simulations by Joughin et al. (2019). This approach, chosen for its computational efficiency and ability to reproduce observed ice flow speeds through spatially optimized parameters, remains a practical and widely adopted method for large- scale ice sheet simulations (Line 60- 63). Then a separate paragraph (Line 64- 73) about Height Above Flotation scaling (HAF- scaling) is added to manually scale the basal resistance or the effective pressure from a prescribed HAF value down to zero at the grounding line. The last paragraph is a summary of experiment design in this study with a modified Table 1.  
+
+It would be good to lead the reader along from simple implementations to more complex/realistic and explain why you chose the GlaDS model out of a variety of subglacial hydrology models (see e.g. De Fleurian et al. 2018).  
+
+Thank you for your comment. We use the GlaDS model because it was designed for glacial applications (Werder et al., 2013) and has been successfully utilized in previous ice sheet studies (Gagliardini et al., 2018, Dow et al. 2022, Pelle et al. 2024, etc). Introducing an alternative model would be impractical and time- consuming.
+
+<--- Page Split --->
+
+GlaDS is in fact particularly well- suited for our study as it represents a continuum approach that combines distributed and channelized drainage systems, making it ideal for capturing the diverse hydrological processes that occur beneath glaciers and ice sheets. Its ability to simulate both water- filled channels and sheet- like flows aligns well with the goals of our research, particularly in exploring how bed topography influences effective pressure and basal drag.  
+
+While other hydrology models exist, such as those focused exclusively on either distributed or channelized systems, GlaDS offers a balance of physical realism and computational feasibility, making it a robust choice for studying subglacial hydrology in coupled ice sheet modeling frameworks. This combination of practical implementation within Elmer/Ice and its theoretical strengths guided our decision to use GlaDS in this study.  
+
+To address this comment in the manuscript, we include a brief explanation in the Methods section (Line 511- 515):  
+
+"The GlaDS model was selected for this study as it combines a distributed water sheet and channelized drainage system, providing a good representation of subglacial hydrology that is particularly suited to investigating how bed topography influences effective pressure and basal drag \(^{19 - 22}\) . It is fully implemented within Elmer/Ice, allowing for seamless spin- up of the hydrology model with a restart from the ice sheet model."  
+
+I am a little bit concerned with the validity of the inversion result for future projections when the basal conditions (especially at the grounding line) have changed. This affects mostly the experiments where N is treated implicitly (as part of the inversion). Are they still a good baseline after e.g. a substantial retreat of the GL which would imply a major reconfiguration of the hydraulic system?  
+
+We share the reviewer's concern, which is precisely what this study aims to emphasize. We agree that basal drag should vary over time, especially near the grounding line.  
+
+As the reviewer pointed out, in experiment with an implicit effective pressure (RC_in), assuming effective pressure into a constant parameter fails to account for the temporal variability of effective pressure in fast- flowing regions, which is why we use HAF- scaling to manually reduce basal drag near the grounding line in RC_in_HAF. Similarly, in the experiment with simulated effective pressure from a hydrology model (RC_eN_GlaDS), we assumed a constant effective pressure, neglecting changes in basal drag over time. To address this, we conducted a separate experiment, RC_eN_GlaDS_HAF, by applying HAF- scaling to the simulated effective pressure.
+
+<--- Page Split --->
+
+Our results show that ice dynamics are highly sensitive to how temporally varying effective pressure is treated, particularly near the grounding line. This underlines the importance of using a coupled ice sheet- hydrology model, which would allow for a more realistic representation of the evolving hydraulic system, particularly under future projections when significant changes to basal conditions are expected. This has been well addressed in the Discussion (Line 212- 219).  
+
+## 3. Structure of results and headlines  
+
+I believe the presentation of the results could be enhanced for better readability and comprehension. While the beginning that focuses on the implicit handling of N does a good job of describing the difference between LW and RC_in_HAF, the following parts could be arranged in a more organized way, as I experienced them hard to follow. One idea could be to adapt something like the following structure (very similar to the current one):  
+
+Keep the first part on implicit handling of N: LW vs RC_in_HAF as it is (L. 59 to L. 80). Then have a second part that covers the explicit handling of N: RC_en_GlaDS vs RC_en_POC vs RC_en_GlaDS_HAF. Here I would suggest to start with RC_en_POC as the most simple approach and then introduce GlaDS and the extended GlaDS (RC_en_GlaDS_HAF).  
+
+With this continental view established, continue with the Tipping points and the regional analysis.  
+
+Not much change is needed, but these sections should be clearly identifiable and might deserve their own headings, also because the current in- between headings ("Subglacial effective pressure must be included in sliding relations" and "Subglacial effective pressure controls ice fluxes") are very general and do not help much to navigate the text. The final part before the discussion ("Increased sensitivity to effective pressure near the grounding line") is in a good place.  
+
+Many thanks for the suggestions, which are super helpful. We have reorganised the Results section as recommended:  
+
+## Subglacial effective pressure must be included in sliding relations  
+
+This section compares the experiments using linear Weertman and Regularised Coulomb sliding relations and indicates that effective pressure is necessary in sliding relations.  
+
+## Subglacial effective pressure controls ice fluxes  
+
+This section indicates the influence of different treatments of effective pressure on the projected mass loss. To maintain consistency with the structure of the Introduction section and adhere to Reviewer 2's suggestions, we presented the results using three bullet points:
+
+<--- Page Split --->
+
+- Implicit representation of effective pressure N- Explicit representation of effective pressure N- Basin-scale response  
+
+## Increased sensitivity to effective pressure near the grounding line  
+
+This section emphasizes the critical role of effective pressure near the grounding line in affecting the low- drag area and projected ice mass loss.  
+
+We believe that the original subtitles effectively convey the key messages when supplemented by these key bullets. Please see the modified main text for detailed changes.  
+
+At some point, Kazmierczak et al. (2022) should also be referenced, as they also investigate the influence of different sliding relations for Antarctica.  
+
+Thanks for sharing this paper. We have cited this paper in our Introduction (Line 26- 27) and Discussion (Line 213- 215), especially about the finding that decreased effective pressure near the grounding line amplifies the ice sheet's sensitivity to climatic forcing, particularly for a given sliding relation power.  
+
+Remarks by lines  
+
+L. 29: also frictional heating due to internal deformation of ice  
+
+Thanks. Added.  
+
+L. 45f: Can't the effect of N be also approximated in the sliding coefficients for Weertman sliding laws? see also concern 1.  
+
+The short answer is "no", except for the present- day state of the ice sheet. The effect of effective pressure N can be viewed as being implicitly represented in the spatial pattern of the Weertman coefficient, but this varies by many orders of magnitude because the Weertman sliding laws do not explicitly capture anything about the role of subglacial hydrology in sliding nor its evolution over time; all the relevant physics are hidden in one spatially tuned parameter. So we have no confidence in the contribution of N to projected evolution of the ice sheet when using Weertman sliding.  
+
+L. 48: more realistic than what?  
+
+Here we mean the HAF- scaling is more realistic than using a constant coefficient in RC_in or a constant simulated effective pressure in RC_en_GlaDS. This is because the
+
+<--- Page Split --->
+
+effective pressure is expected to vary with time, especially in regions where the ice approaches flotation with a moving grounding line. To address this and incorporate other related comments regarding the HAF- scaling, we have rewritten this entire paragraph as follows (Line 64- 73):  
+
+"A further assumption about the effective pressure and its impact on sliding can be imposed using on "height above floatation" (HAF) scaling<sup>21</sup>. HAF is defined as the height of the upper ice surface above sea level minus the height required for the ice overburden pressure to match the ocean water pressure at the bed. This assumption scales the basal resistance or the effective pressure from a prescribed HAF value down to zero at the grounding line, reflecting the influence of warm seawater intrusion beneath grounded ice<sup>41</sup>. However, the inland extent of this effect is poorly constrained due to limited observations, and this approach neglects significant spatial and temporal variability driven by evolving glacier geometry<sup>42,43</sup>. Both the HAF- scaling and perfect ocean connection assumption enforce the floating condition at the grounding line. However, HAF- scaling, applied only near the grounding line, adjusts basal resistance based on ice elevation relative to flotation, potentially under- or overestimating effective pressure. In contrast, the perfect ocean connection assumes uniform hydraulic connectivity across the domain, systematically overestimating effective pressure by neglecting hydraulic potential gradients."  
+
+## L. 56: Why SSA, what is the implication?  
+
+The SSA simplifies the full- Stokes equations by neglecting vertical shear stresses, focusing only on the horizontal stress balance. This reduction significantly decreases the computational cost, making SSA suitable for large- scale and long- term simulations, i.e. century- scale Antarctic Ice Sheet projections. The SSA is well- suited to simulate the behavior of fast- flowing ice streams, outlet glaciers, and floating ice shelves, which are critical regions influencing ice discharge from Antarctica. However, the neglect of vertical shear can lead to inaccuracies in simulating inland ice dynamics and interactions with the grounding zone, which might underestimate or oversimplify the response of grounded ice to changes in basal conditions or surface mass balance.  
+
+To address this limitation, we expanded the Discussion section to provide a more detailed explanation of the study's limitations (Line \*\*\*):  
+
+"Our simulations explore the sensitivity to basal sliding paramerisations under various climate scenarios. However, limitations in our approach highlight areas for improvement. The use of SSA assumes vertical hydrostatic equilibrium and neglects vertical shear stresses, which restricts its ability to fully capture the ice dynamics at the grounding line. This is particularly relevant in regions with pronounced bedrock
+
+<--- Page Split --->
+
+slopes, where non- hydrostatic effects and deviatoric (bridging) stresses—ignored by SSA become significant. While a fully resolved full- Stokes model, combined with a contact problem formulation for the lower ice surface, would be needed to consistently resolve these dynamics \(^{51}\) , we argue that the sliding relation itself is less sensitive to the choice of stress representation in the ice- flow model. For instance, a similar sensitivity to driving stress has been found between an SSA model and one incorporating simple vertical shear, particularly under a regularized Coulomb law or a Weertman sliding parameterisation \(^{21}\) .  
+
+Nevertheless, unresolved processes may further alter projected sea- level contributions. Given the importance of the grounding line region to our results, we expect further changes when considering three- dimensional shear fields, hydrology- ocean- bedrock- sediment interactions, iceberg calving, and uncertainties in predicted future surface mass balance. While these processes are unlikely to negate the effects we identify—given that effective pressure is largely isolated from external forcings—they underscore the complexity of grounding line dynamics and the need for urgent model development to advance the fidelity of sea- level projections."  
+
+L. 57: Why is SSA used here? Why not hybrid/higher order? What are the implications?  
+
+See my response to L.56 above.  
+
+L. 60: Why is RC_in_HAF more realistic? is combined effects of seawater intrusion and meltwater production is also considered in the eN variants? is this only about RC vs LW? more explanation is needed.  
+
+For the \(1^{\text{st}}\) and \(3^{\text{rd}}\) question here, Yes, the comparison here is only about RC vs LW. We suggest that RC relation is more physically realistic because it can capture both Weertman and Coulomb regimes, provides a unified framework for modelling sliding across diverse bed types as explained in the modified Introduction (Line 43- 46). The HAF- scaling in RC_in_HAF can well represent the combined effects of seawater intrusion and meltwater production near the grounding line as explained in Line 66- 68. However, considering the restructured Introduction and Results sections, we deleted this sentence here and added a similar sentence in the Introduction (Line 77- 80):  
+
+"For comparison against the more physically justifiable RC sliding parameterisation, we also apply the linear Weertman sliding relation (hereafter "LW"), which, despite its tendency to underestimate mass loss \(^{21}\) , is still widely used for its simplicity in both numerical and computational aspects \(^{2,5,26,30}\) ."
+
+<--- Page Split --->
+
+For the second question, while we agree that RC_eN_POC also considers the combined effects of seawater intrusion and meltwater production, it is important to note that RC_eN_GlaDS uses a constant effective pressure simulated from the GlaDS hydrology model. To account for a temporally evolving effective pressure, we applied HAF- scaling to the simulated effective pressure in RC_eN_GlaDS_HAF. We have further clarified the HAF scaling, along with its similarities and differences with POC, in the revised Introduction (Line 70- 73).  
+
+L. 68: Replace "near mass balance" with "near equilibrium". I don't understand the second part of the sentence regarding observed mass loss.  
+
+Modified.  
+
+The second part of the sentence refers to how the RC relation suggests a near- zero mass loss by 2300, which aligns more closely with the observed mass balance trends in recent decades. We have revised the sentence for greater clarity, as follows (Line 91- 93):  
+
+"Under the two low emission scenarios, the RC relation suggests near equilibrium by 2300, which aligns more closely with observed mass loss in recent decades 44, ..."  
+
+L.71: "For WAIS...": Sentence is unclear. Maybe "For the WAIS the experiment using the RC relation and high emission scenario..."  
+
+Modified.  
+
+L. 73: produce.  
+
+Modified.  
+
+L. 74: "However, the RC..." -> "However, using the RC..."  
+
+Modified.  
+
+L. 75: "than LW" -> "than using the LW relation". The whole paragraph seems to be missing words.  
+
+Thanks for the suggestions. The whole paragraph has been modified accordingly (Line 95- 105).  
+
+L. 82: This sentence is not very clear, seems obvious.  
+
+We modified the sentence to increase the clarity but we're not sure what the reviewer is requesting here.
+
+<--- Page Split --->
+
+"Our simulations reveal the treatment of effective pressure significantly affects the spatial tuning of sliding coefficients, which, in turn, impacts the evolution of basal shear stress" (Line 111- 112)  
+
+L 99ff: Isn't the addition of HAF to the GlaDS N a little bit of double counting, because the effect of the ocean is already included in the GlaDS model? Though it is only appropriate at the time where N was computed in GlaDS.  
+
+No, as we mentioned in response to an earlier comment, RC_eN_GlaDS uses a constant effective pressure simulated by the GlaDS hydrology model. While the GlaDS model does account for ocean effects, the constant simulated effective pressure becomes unsuitable for regions experiencing grounding line retreat. To address this time- varying effect, we apply HAF- scaling to the simulated effective pressure.  
+
+## L. 106: Explain coulomb limit  
+
+Thanks for your comment. The Coulomb limit defines the maximum basal shear stress under a Coulomb sliding regime, where sliding is primarily controlled by effective pressure and the sliding coefficient. We decide to move all the sentences related with Coulomb limit into a separate paragraph in the subsection "Increased sensitivity to effective pressure near the grounding line" (Line 199- 210). In the modified Methods section, we further explained the shift between Weertman and Coulomb regime under different assumptions of effective pressure (Line 475, 492).  
+
+We modified the sentence here into: "While this parameterisation ensures the Coulomb regime dominates near the grounding line, ..." (Line 131- 132)  
+
+L. 115- 122: Can the experiments using POC produce meaningful results for basins with reverse bed slope? Won't POC on the retrograde bed will lead to higher basal pressure in the inland than at the grounding line?  
+
+No, the POC assumption does not produce meaningful results, as it consistently underestimates water pressure inland. We included it in our study because it is one of the commonly used assumptions (see modified Introduction, Line 52- 55, Line 72- 73), not because we believe it to be a good assumption.  
+
+Yes, on a retrograde bed, POC will lead to higher basal water pressures inland compared to the grounding line, but these pressures are still insufficient. Given that basal friction and geothermal heat provide a basal water source nearly everywhere under the ice, and basal water sinks are rare, there is always a hydraulic gradient driving water toward the grounding line. In other words, subglacial water pressure is expected to build up sufficiently to drive flow toward the grounding line under normal conditions.
+
+<--- Page Split --->
+
+L. 153: How valid is the distribution of N computed by GlaDS for present day conditions after more than a century? Even with the HAF scaling, the whole hydraulic regime might have shifted.  
+
+The distribution of N computed by GlaDS for present- day conditions is based on the initial geometry and the friction heating calculated from the inverted basal shear stress. We acknowledge that over more than a century, changes in the ice sheet and surrounding environment—such as variations in oceanic conditions, climate, and ice dynamics—could have altered the hydraulic regime, especially near the moving grounding line. While the HAF- scaling helps account for some of these time- varying effects, it is true that the entire hydraulic regime may have shifted, particularly in regions where significant changes in the ice flow or grounding line dynamics have occurred. However, in this study, we're not aiming to provide an accurate representation of the evolving hydrology over time, which can only be achieved by applying a coupled ice sheet- hydrology model. We have made it clear at the start of the Methods section (Line 414- 416):  
+
+"Our simulations do not capture a fully coupled ice- hydrology system, as such a model is not yet available for continental- scale applications. Instead, our simulations aim to determine whether a coupling is indeed necessary, rather than predetermining what might be considered a valid future projection of ice mass change."  
+
+Moreover, we emphasized the limitation of HAF- scaling and the importance of having a coupled ice- hydrology model in the modified Discussion (Line 221- 229):  
+
+"Among these RC experiments, we place greater confidence in the results from experiment RC_en_GlaDS_HAF, as it uses a more physically realistic subglacial hydrology system where HAF- scaling adjusts effective pressure as the grounding line retreats. However, while the HAF- scaling accounts for some of these time- varying effects, the entire hydraulic regime may have shifted, particularly in regions experiencing significant changes in ice flow or grounding line dynamics<sup>12, 39</sup>. Moreover, the GlaDS model, originally designed to simulate subglacial hydrology on hard bedrock, may predict more efficient drainage systems than those occurring on a soft bed, potentially leading to lower modelled subglacial water pressures<sup>12,37,39</sup>. This highlights the importance of accurately predicting subglacial water pressure and developing a more comprehensive ice sheet- hydrology model that can account for the temporal co- evolution of ice sheet dynamics and the subglacial hydrological system."
+
+<--- Page Split --->
+
+L. 163: What does "constrained future surface mass budgets estimates" mean? Improvement of the predictions of SMB?  
+
+Yes, by "constrained future surface mass budget estimates," we mean the uncertainties in predicting the future surface mass balance (SMB). To clarify, we have revised the sentence to: "...and uncertainties in predicted future surface mass balance." (Line 241)  
+
+Figure 1: good colors, bigger fonts, no box around legend  
+
+Thanks for the suggestions. Based on comments from Reviewer 1, we have modified Figure 1 to be more colour- blind friendly with bigger fonts and no box around legend.  
+
+![](images/Figure_2.jpg)
+
+<center>Figure 2: </center>  
+
+caption: "sea- level contributions between linear Weertman law" -> "sea- level contributions in experiment using linear Weertman law".  
+
+Modified.  
+
+Figure 4:  
+
+caption: "with a Hydrology model" -> "from GlaDS"  
+
+Modified.  
+
+Does channel area mean cross section of individual channels?  
+
+Yes, the channel area refers to the cross section of each channel. To clarify this, we have updated the caption of Fig. 4 to read: "The background image shows the simulated effective pressure and channel area (cross- sectional area of each channel) generated from GlaDS."
+
+<--- Page Split --->
+
+why is a diverging colormap used? What is special about 0.5 MPa?  
+
+The diverging colormap and the value of 0.5 MPa do not have any special significance. We chose this color scale to emphasize regions of low effective pressure, such as the Siple Coast and Lake Vostok.  
+
+small inset figures in G- H and D- Dp are hardly readable.  
+
+We have zoomed in the inset figures for G- H and D- Dp.  
+
+Figure 5:  
+
+Why are the lines dashed?  
+
+There is no special meaning here. We have modified them into solid lines.  
+
+## References  
+
+DE FLEURIAN B, WERDER MA, BEYER S, et al. SHMIP The subglacial hydrology model intercomparison Project. Journal of Glaciology. 2018;64(248):897- 916. doi:10.1017/jog.2018.78  
+
+Kazmierczak, E., Sun, S., Coulon, V., and Pattyn, F.: Subglacial hydrology modulates basal sliding response of the Antarctic ice sheet to climate forcing, The Cryosphere, 16, 4537- 4552, https://doi.org/10.5194/tc- 16- 4537- 2022, 2022.  
+
+Bueler, E., and J. Brown (2009), Shallow shelf approximation as a "sliding law" in a thermomechanically coupled ice sheet model, J. Geophys. Res., 114, F03008, doi:10.1029/2008JF001179.  
+
+Morlighem, M., E. Rignot, H. Seroussi, E. Larour, H. Ben Dhia, and D. Aubry (2010), Spatial patterns of basal drag inferred using control methods from a full- Stokes and simpler models for Pine Island Glacier, West Antarctica, Geophys. Res. Lett., 37, L14502, doi:10.1029/2010GL043853.
+
+<--- Page Split --->
+
+Reviewer #3 (Remarks to the Author):  
+
+Review of "Subglacial Water Amplifies Antarctic Contributions to Sea- Level Rise", for Nature Communications, by Zhao, Gladstone, Zwinger, Gillet- Chaulet, Wang, Caillet, Mathiot, Saraste, Jager, Galton- Fenzi, Christoffersen and King  
+
+This manuscript describes the importance of incorporating subglacial hydrology parameterizations into whole Antarctic ice sheet models, by varying parameterization of basal water effective pressure in a basal slip relation. The authors conclude that effective pressure is required in the slip relationship, and that ice flux is sensitive to effective pressure model, especially near the grounding line.  
+
+We believe the topic of this manuscript is important and can be of interest to a wide range of readers. However, we believe that this manuscript can be significantly improved by considering the comments listed below.  
+
+For now, the manuscript is "expert- facing", i.e., it is assumed that the readers have a substantial understanding of the background and methods of this study. The manuscript leaps over many (essential) details that could have been helpful for the readers to follow along.  
+
+We have gone through the manuscript line by line with a view to providing clarity for non- experts. This is most noticeable in the modified Introduction. See also our specific responses to further comments below.  
+
+- For an important study that can be of interest to the general public, especially in the journal of Nature Communication, we argue this is not appropriate. We recommend the authors to provide more detailed elaborations on the background, methods, findings, and implications of this study. A few examples can be found in the comments below. Thanks for the comments. We have further elaborated on these sections. See our specific responses to further comments below.  
+
+- In the section "Subglacial effective pressure must be included in sliding relations"; its implied the observed mass gain of the linear Weertman models demonstrate the need for subglacial effective pressure. The authors need to make a stronger case - ie demonstrate that mass gain is not plausible over these time frames.  
+
+The comparison between linear Weertman (LW) and regularised Coulomb (RC) sliding relations suggests that inclusion of effective pressure in the sliding relation largely enhances the basal sliding and projected mass loss. Whether the mass gain simulated
+
+<--- Page Split --->
+
+in LW is plausible is not the focus here. We have clarified this in the modified Methods section (Line 414- 416) that we do not aim to predetermine what might be considered a valid future projection in terms of mass change. We have also made minor modifications to the Introduction section (Line 77- 80) to make it clear that the existing literature already provides strong evidence for favouring RC over LW. The existence of a wide spread hydrological network under the Antarctic Ice Sheet can lubricate the ice base and thus lead to increased ice velocities, which should be considered in the basal sliding relations. Our aim in the current study is not to convince the reader that RC is better than LW (this has already been done), but rather to show just how large an impact this choice can have. • We recommend the authors to elaborate on the differences between model setups. We appreciate the balance between the amount of work presented in this manuscript and the space limitations. But the current manuscript makes it really difficult to follow the different model setups, especially considering the number of runs that are being presented and discussed.  
+
+Thanks for the suggestions. We have provided a more accessible conceptual description of the different approaches in the modified Introduction section, i.e. the different assumptions implied by the different versions of sliding laws. We have also moved the modified Table 1 to the Introduction to help better interpret the differences. See our response to Reviewer 2 for more details.  
+
+• It is not clear why/how including different subglacial hydrology components can change the AIS SLR projection by such a large degree. What is the physical process behind the differences? We recommend the authors to elaborate on this and make it more clear.  
+
+Thanks for the comment. We have added a description of the most important physical processes at the start of the subsection "Increased sensitivity to effective pressure near the grounding line" (Line 174- 182). The key concept, which we now believe comes across clearly to the non- expert, is that of a low friction "ice plain" immediately upstream of the grounding line. This section has also been expanded to include a discussion of how assumptions about effective pressure influence the extent of this low- friction "ice plain" region (Line 262- 273), which can be widely different depending on the bedrock and ice geometry.  
+
+• Line 47- 52: HAF is a very important concept for this study. However, the authors didn't provide any explanation/elaboration on the concept, other than providing one reference. (This is also true for some other important concepts throughout the manuscript.) This approach makes it difficult for non- expert readers to understand the work and follow the logic flow. We recommend providing a more detailed explanation of the concept, and/or maybe even a conceptual diagram.
+
+<--- Page Split --->
+
+Thanks for the comment. We have provided an accessible description of the concept of HAF in the modified Introduction (Line 65- 68):  
+
+"HAF is defined as the height of the upper ice surface above sea level minus the height required for the ice overburden pressure to match the ocean water pressure at the bed. This assumption scales the basal resistance or the effective pressure from a prescribed HAF value down to zero at the grounding line, reflecting the influence of warm seawater intrusion beneath grounded ice \(^{41}\) ."  
+
+We have also provided a more detailed explanation for other concepts like the Coulomb limit (Line 201- 202)  
+
+- It is not clear what assumption is made about the subglacial meltwater budget. Does the model run only require a certain and finite amount of subglacial water? Or do the authors assume an "unlimited" supply of subglacial water generated by basal melting upstream? And does a limited subglacial meltwater budget change the model output?  
+
+Thanks for the comments. We have expanded our description of the hydrology model in the Methods section (Line 511- 519). The hydrology model spin- up does make specific assumptions about the basal water source and the other parameters on which basal water pressure depends, most important of which are parameters describing conductivity of both basal channels and the distributed basal water system. A brief description is given in the Methods section (Line 515- 516), and the citation to the recent paper (Zhang et al., 2024) is the most directly relevant as their use of GlaDS is almost identical to the current study.  
+
+The ice dynamic simulations have no direct dependence on water budget; instead, they depend on basal water pressure. This pressure is influenced by various factors, particularly the conductivities mentioned above, with basal water production being driven solely by frictional heating. Consequently, the only simulations where assumptions about the basal water budget are directly relevant are those utilizing the spatial distribution of basal water from the hydrology model. Even in those cases, the dependence remains weak due to the influence of other factors.  
+
+To answer the question literally (though the fact that our hydrology model is in steady state probably makes this clarification unnecessary), the subglacial water supply is effectively unlimited, but any other answer would result in no hydrology at all at steady state.  
+
+The feedback between ice dynamics and basal water generation, through friction heating due to sliding, may be important for future evolution of the coupled system, but this is beyond the scope of the current study.
+
+<--- Page Split --->
+
+\* The authors need to address the limitation of the study in the Discussion section. We understand that the authors need to consider the space limitations per journal guideline, but we strongly recommend the authors to reconsider the balance between (1) further elaboration and discussion of the study findings, (2) connection with other studies, (3) broader implications, and (4) limitation of the study, for the Discussion section.  
+
+Thanks for your comments. We have carefully considered your suggestions and revised the Discussion section, which we hope will satisfy this request. Specifically:  
+
+1. Elaboration and discussion of findings: We have added a paragraph emphasizing the critical role of subglacial water pressure and the significant influence of a small region near the grounding line on multi-century ice sheet mass loss. This discussion highlights the connection between effective pressure and sliding relations, as well as the significance of model resolution in capturing small-scale transitions near the grounding line (Lines 212-219).  
+
+2. Connection with other studies: We have incorporated relevant citations in the Discussion to place our findings in the context of previous research. For example, we cite Kazmierczak et al. (2022) to emphasize the influence of different sliding relations for Antarctica (Lines 213-215), suggested by Reviewer 2.  
+
+3. Broader implications: In the final paragraph of the Discussion (Lines 245-251), we elaborate on the implications of our findings for understanding ice sheet dynamics and their influence on sea-level projections. This paragraph emphasizes the importance of integrating realistic subglacial processes into future models to improve predictions.  
+
+4. Study limitations: In response to your comments and those of Reviewer 2, we have added a paragraph discussing the limitations of using the Shallow Shelf Approximation (SSA) in this study (Line 230-238). Additionally, the original submission already addressed other limitations, such as hydrology-ocean-bedrock-sediment interactions, iceberg calving, and uncertainties in future surface mass balance projections (Line 239-241).  
+
+minor comments:  
+
+\* This manuscript uses a lot of acronyms - are they really necessary? Like, "GL" for "grounding line"? Having too many acronyms can be counterproductive. Also, it may help to spell out the run names in Figure 1.  
+
+Thanks for the comments. We have removed the acronyms including GL (grounding line), N (effective pressure), SLR (sea-level rise), AIS (Antarctic Ice Sheet), WAIS (West
+
+<--- Page Split --->
+
+Antarctic Ice Sheet), EAlS (East Antarctic Ice Sheet). We now only keep three acronyms: HAF (height above flotation), LW (linear Weertman), RC (regularised Coulomb).  
+
+We have moved the modified Table 1 from Methods to Introduction, which can well explain the difference across these experiments in Figure 1.  
+
+\* We noticed that no doi/web- link is provided for the references. We encourage the authors to consider including such information.  
+
+Our manuscript is available in LaTeX format, which includes the DOI and web links in the references. If the manuscript is accepted, we hope the editing team can assist with ensuring the DOI and web links are properly included in the final references.  
+
+Unsure of what is meant by the second and third authors contributed equally.  
+
+We mean that the second and third authors contributed to this paper equally. To reduce the confusion, we removed this statement.  
+
+Reviewer #4 (Remarks to the Author):  
+
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+Many thanks for your contribution to provide the valuable comments.
+
+<--- Page Split --->
+
+## References:  
+
+Dow, C.F., Ross, N., Jeofry, H. et al. Antarctic basal environment shaped by high- pressure flow through a subglacial river system. Nat. Geosci. 15, 892- 898 (2022). https://doi.org/10.1038/s41561- 022- 01059- 1  
+
+GAGLIARDINI, O. and WERDER, M.A. (2018) 'Influence of increasing surface melt over decadal timescales on land- terminating Greenland- type outlet glaciers', Journal of Glaciology, 64(247), pp. 700- 710. doi:10.1017/jog.2018.59.  
+
+Joughin, I., Smith, B. E., & Schoof, C. G. (2019). Regularized Coulomb friction laws for ice sheet sliding: Application to Pine Island Glacier, Antarctica. Geophysical Research Letters, 46, 4764- 4771. https://doi.org/10.1029/2019GL082526  
+
+Pelle, T., Greenbaum, J. S., Ehrenfeucht, S., Dow, C. F., & McCormack, F. S. (2024). Subglacial discharge accelerates dynamic retreat of aurora subglacial basin outlet glaciers, East Antarctica, over the 21st century. Journal of Geophysical Research: Earth Surface, 129, e2023JF007513. https://doi.org/10.1029/2023JF007513  
+
+Seroussi, H. and Morlighem, M.: Representation of basal melting at the grounding line in ice flow models, The Cryosphere, 12, 3085- 3096, https://doi.org/10.5194/tc- 12- 3085- 2018, 2018.  
+
+Seroussi, H., Pelle, T., Lipscomb, W. H., Abe- Ouchi, A., Albrecht, T., Alvarez- Solas, J., et al. (2024). Evolution of the Antarctic Ice Sheet over the next three centuries from an ISMIP6 model ensemble. Earth's Future, 12, e2024EF004561. https://doi.org/10.1029/2024EF004561  
+
+Wang, Y., Zhao, C., Gladstone, R., Zwinger, T., Galton- Fenzi, B. K., and Christoffersen, P.: Sensitivity of the future evolution of the Wilkes Subglacial Basin ice sheet to grounding- line melt parameterizations, The Cryosphere, 18, 5117- 5137, https://doi.org/10.5194/tc- 18- 5117- 2024, 2024.  
+
+Weertman, J. (1972). General theory of ice- sheet flow. Journal of Glaciology, 11(62), 157- 170. DOI: 10.1017/S002214300002302X  
+
+Werder, M. A., I. J. Hewitt, C. G. Schoof, and G. E. Flowers (2013), Modeling channelized and distributed subglacial drainage in two dimensions, J. Geophys. Res. Earth Surf., 118, doi:10.1002/jgrf.20146.  
+
+Yufang Zhang, John C Moore, Liyun Zhao, Mauro A Werder, Rupert Gladstone, and Michael Wolovick. The role of hydraulic conductivity in the Pine Island Glacier's subglacial water distribution. Science of The Total Environment, 927:172144, 2024.
+
+<--- Page Split --->
+
+## Response to Reviewers  
+
+Many thanks to the reviewers for their valuable comments. Our responses are highlighted in blue, and the line numbers refer to the revised manuscript.  
+
+Reviewer #1 (Remarks to the Author):  
+
+The paper is the result of a numerical study investigating the effect of basal pressure under the Antarctic ice sheet and the role of the pressurized water layer in basal slip. While for a direct prognostic application there is lack of data, the paper defines the range of solutions for Ice Sheet/iceshelf dynamics depending on pressure in particular at he icesheet- iceshelf boundary.  
+
+The authors have responded to all reviewers comments and i am satisfied with the answers to my specific comments.  
+
+Thanks again for your comments.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors have addressed the reviewers' concerns effectively. I'm pleased with the improved paper structure and extended description. I have only one minor comment:  
+
+I appreciate the more detailed description of sliding laws used in the ISMIP/ABUMIP experiments, but I would like to point out that the PISM model uses a Mohr- Coulomb type sliding law and not a Weertman type. It is specifically mentioned in the manual that Weertman should not be used with PISM: https://www.pism.io/docs/manual/modeling- choices/dynamics/weertman.html and also in the model description it is pretty clear:  
+
+Seroussi et al. 2020 PIK_PISM  
+
+We apply a power law for sliding with a Mohr- Coulomb criterion relating the yield stress to parameterized till material properties and the effective pressure of the overlaying ice on the saturated till (Bueler and van Pelt, 2015).  
+
+Seroussi et al. 2024: PIK_PISM  
+
+A generalized power law (Schoof & Hindmarsh, 2010) is applied to parameterize basal sliding. The basal friction coefficient depends on the effective pressure and till friction angle, that is parameterized using a heuristic, piecewise linear function of the bed elevation (Martin et al., 2011).
+
+<--- Page Split --->
+
+I know from personal communication that AWI_PISM uses the same setup in that regard.  
+
+This does of course not change the overall conclusion of the authors that Weertman type sliding laws are dominant in the benchmarks.  
+
+Many thanks for the correction and providing the detailed information. I have modified the Table S1 in the supplementary material accordingly. See the modified table below.  
+
+Table S1. Sliding law used in ISMIP6-2100 Antarctic, ISMIP6-2300 Antarctic, and ABUMIP   
+
+<table><tr><td colspan="3">ISMIP6-2100</td><td colspan="3">ISMIP6-2300</td><td colspan="3">ABUMIP8</td></tr><tr><td>Model Name</td><td>Sliding Law</td><td>Model Name</td><td>Sliding Law</td><td>Model Name</td><td>Sliding Law</td><td></td><td></td><td></td></tr><tr><td>AWI_PISM</td><td>Mohr-Coulomb (m=3)</td><td>-</td><td>-</td><td>AWI_PISM</td><td>Coulomb</td><td></td><td></td><td></td></tr><tr><td>-</td><td>-</td><td>UCM_Yelmo</td><td>Regularized Coulomb</td><td>-</td><td>-</td><td></td><td></td><td></td></tr><tr><td>IMAU_IMAUIICE1</td><td>Weertman(m=3)</td><td>MAU_UFFEMISM</td><td>Regularized Coulomb</td><td>IMAU_ICE</td><td>Coulomb</td><td></td><td></td><td></td></tr><tr><td>IMAU_IMAUIICE2</td><td>Weertman(m=3)</td><td>-</td><td>-</td><td>-</td><td>-</td><td></td><td></td><td></td></tr><tr><td>NCAR_CISM</td><td>Weertman(m=3)-Coulomb</td><td>NCAR_CISM1</td><td>Weertman(m=3)-Coulomb</td><td>NCAR_CISM</td><td>Weertman(m=3)-Coulomb</td><td></td><td></td><td></td></tr><tr><td>-</td><td>-</td><td>NCAR_CISM2</td><td>Zot-Iversen</td><td>-</td><td>-</td><td></td><td></td><td></td></tr><tr><td>-</td><td>-</td><td>NORCE_CISM</td><td>Zot-Iversen</td><td>-</td><td>-</td><td></td><td></td><td></td></tr><tr><td>DOE_MALI</td><td>Weertman(m=1)</td><td>DOE_MALI</td><td>Weertman(m=3)</td><td>DOE_MALI</td><td>Weertman(m=1)</td><td></td><td></td><td></td></tr><tr><td>ILTS_PIK_SICOPOLIS</td><td>Weertman(m=3)-Budd</td><td>ILTS_PIK_SICOPOLIS</td><td>Weertman(m=3)-Bud</td><td>ILTS_PIK_SICOPOLIS</td><td>Weertman(m=3)</td><td></td><td></td><td></td></tr><tr><td>JPL_ISSM</td><td>Budd(m=1)</td><td>UCSD_ISSM</td><td>Weertman(m=3)-Bud</td><td>JPL_ISSM</td><td>Weertman(m=1)</td><td></td><td></td><td></td></tr><tr><td>UCIPL_ISSM</td><td>Weertman(m=3)</td><td>-</td><td>-</td><td>-</td><td>-</td><td></td><td></td><td></td></tr><tr><td>PIK_PISM1</td><td>Mohr-Coulomb(m=3)</td><td>PIK_PISM</td><td>Mohr-Coulomb (m=3)</td><td>PIK_PISM</td><td>Coulomb</td><td></td><td></td><td></td></tr><tr><td>PIK_PISM2</td><td>Mohr-Coulomb (m=3)</td><td>-</td><td>-</td><td>-</td><td>-</td><td></td><td></td><td></td></tr><tr><td>ULB_FETISH_16km</td><td>Weertman(m=2)</td><td>ULB_FETISH</td><td>Weertman(m=3)</td><td>ULB_FETISH</td><td>Weertman(m=2)</td><td></td><td></td><td></td></tr><tr><td>ULB_FETISH_32km</td><td>Weertman(m=2)</td><td>-</td><td>-</td><td>-</td><td>-</td><td></td><td></td><td></td></tr><tr><td>UTAS_ELMERICE</td><td>Weertman(m=1)</td><td>UTAS_ELMERICE</td><td>Weertman(m=1)</td><td>-</td><td>-</td><td></td><td></td><td></td></tr><tr><td>-</td><td>-</td><td>IGE_Emerice</td><td>Weertman(m=1)</td><td>IGE_Emerice</td><td>Weertman(m=3)</td><td></td><td></td><td></td></tr><tr><td>VUB_AISMIPALEO</td><td>Weertman(m=3)</td><td>-</td><td>-</td><td>-</td><td>-</td><td></td><td></td><td></td></tr><tr><td>VUW_PISM</td><td>Weertman(m=3)</td><td>VUW_PISM1</td><td>Weertman(m=3)</td><td>-</td><td>-</td><td></td><td></td><td></td></tr><tr><td>-</td><td>-</td><td>VUW_PISM2</td><td>Weertman(m=3)</td><td>-</td><td>-</td><td></td><td></td><td></td></tr><tr><td>-</td><td>-</td><td>-</td><td>-</td><td>ARC_PISM1</td><td>Coulomb</td><td></td><td></td><td></td></tr><tr><td>-</td><td>-</td><td>-</td><td>ARC_PISM2</td><td>Coulomb</td><td></td><td></td><td></td><td></td></tr><tr><td>LCS_E_GRISLI</td><td>Weertman(m=3)</td><td>LCS_E_GRISLI</td><td>Weertman(m=3)</td><td>LCSE_GRISLI</td><td>Coulomb</td><td></td><td></td><td></td></tr><tr><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td><td></td><td></td><td></td></tr><tr><td>-</td><td>-</td><td>-</td><td>CPOM_BISICLES</td><td>Weertman(m=3)-Coulomb</td><td></td><td></td><td></td><td></td></tr><tr><td>-</td><td>-</td><td>-</td><td>PSU_PSU3D1</td><td>Weertman(m=2)</td><td></td><td></td><td></td><td></td></tr><tr><td>-</td><td>-</td><td>-</td><td>PSU_PSU3D2</td><td>Weertman(m=2)</td><td></td><td></td><td></td><td></td></tr></table>  
+
+As the reviewer noted, this does not change our conclusion stated in the Introduction "the majority of Antarctic Ice Sheet models employed the Weertman sliding relation". To ensure statistical accuracy, we have revised this sentence to:  
+
+"A summary of sliding laws used in three major Model Intercomparison Projects (Table S1) reveals that the majority of Antarctic Ice Sheet models (e.g., 10 of 16 in ISMIP6- 2100², 8 of 16 in ISMIP6- 2300⁵, and 7 of 15 in ABUMIP³⁰) employed the Weertman sliding relation." (Line 49- 51)  
+
+Reviewer #3 (Remarks to the Author):  
+
+This work illustrates the need for improved knowledge of the evolution of subglacial hydrology, and we recommend publication.
+
+<--- Page Split --->
+
+The revised manuscript reads much better with acronym expansion and with more detailed concept explanations and discussions. Our only minor follow- up suggestion is to explicitly state in the Method section that the subglacial water budget is held constant because (1) the model requires such assumption to reach steady state, and (2) different assumptions on this do not cause significant variation in the modeling results. Many thanks for the suggestions. We have added this statement in the Methods section: "To ensure that the hydrology system reaches a steady state, the total subglacial water budget is held constant throughout the spin- up period (over approximately 100 model years). This assumption introduces negligible variability in the final results compared to alternative assumptions on subglacial melt supply." (Line 519- 521)  
+
+Reviewer #4 (Remarks to the Author):  
+
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+Many thanks for your contribution to provide the valuable comments.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__9f2277ee7aacb1f4b8c740ce1111b21e6282130d24dd421de8747248239cab4e/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review file__9f2277ee7aacb1f4b8c740ce1111b21e6282130d24dd421de8747248239cab4e/supplementary_0_Peer Review file__9f2277ee7aacb1f4b8c740ce1111b21e6282130d24dd421de8747248239cab4e.mmd

@@ -0,0 +1,232 @@
+
+# Structural molecular and developmental evidence for cell-type diversity in cnidarian mechanosensory neurons  
+
+Corresponding Author: Professor Nagayasu Nakanishi  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+In "Cryptic diversity of cnidarian mechanosensory neurons", Baranyk and co- authors investigate the molecular, anatomical, and development of novel sensory neurons, type II hair cells, in a cnidarian sea anemone Nematostella vectensis and aim to provide insight into the sensory response mediated by this type of cells in cnidarians.  
+
+Using gene KO techniques, this study clearly demonstrated that the function of the polycystin- 1 gene, an ion channel, is essential for mechanosensing in type II hair cells. The main conclusion of this manuscript is that type II hair cells with long cilium are responsible for sensing gentle stimuli in tentacles and that this is mediated by the function of the polycystin- 1 gene product. Additionally, the authors identified the upstream genomic region of polycystin- 1, which contains the binding sequence of pou- IV, drove reporter gene expression specifically in type II hair cells. The authors also used a pou- IV transgenic line to induce reporter gene expression in type I and II hair cells, suggesting that these sensory cells are not in different differentiation states of the identical cell type, but are distinct differentiated cells.  
+
+These observations are new, yet this represents only an incremental advancement of the field of cnidarian neurobiology. In addition to the question about whether it will be of interest to a wider range of researchers, there are also some technical issues, as described below. I am not fully convinced that it fits into the journal Nature Communications.  
+
+Major points  
+
+1) In order to argue for the existence of chemical synapses in the sensory cells analyzed by the authors in this study, it is necessary to show high electron density in the cell-cell junction regions and the accumulation of vesicles that the authors claim are "synaptic vesicles" through 3D reconstruction, rather than relying solely on single-section electrophysiological photographs. Furthermore, quantitative data should be presented to discuss what types of cells form synapse-like structures with hair cells.  
+
+2) It is necessary to confirm that endogenous pkd1 is expressed in type II hair cells. The authors used a transgenic reporter line to examine the expression specificity of pkd1. In this paper, the authors used only the 2.1 kb genomic region upstream of pkd1, but the validity of using only this region has not been demonstrated. In a previous paper (Ozment et al., 2021, Figure 8M), a 3.7 kb upstream region was used, which appears to be expressed more widely than in Supplementary Figure 3C of this manuscript. Whether Pkd1-dependent stimulus responses can be fully attributed to type II hair cells can only be determined if the genomic region used to express the reporter gene completely mimics the expression pattern of endogenous pkd1.  
+
+3) It is puzzling that although the authors mention responses to mechanical stimuli in multiple cnidarian lineages, they do not discuss the extent to which the pou-IV and pkd1 genes they analyzed are universally present in the phylum Cnidaria. With many cnidarian genomes now publicly available, and if single-cell transcriptomic data are available, the presence and expression profiles of genes such as pkd1 should be demonstrated.  
+
+4) Furthermore, the authors need to clarify how their findings relate to mechanosensation in other nonbilaterians and bilaterians. Otherwise, this study will remain a description of something "specialized" about a certain type of mechanosensory cell in a type of sea anemone called Nematostella. For example, what are the genetic characteristics of mechanosensory hair cells expressing Pkd-1? Does the gene signature of the cnidarian Pkd-1-positive cell cluster(s) show any connections to certain sensory cell types in "post-cnidarian" bilaterians and "pre-cnidarian" metazoans such as poriferans, ctenophores, and placozoans?
+
+<--- Page Split --->
+
+## Minor point  
+
+Minor point5) The response to gentle touch is relative to the response to strong touch \((100\%)\) . The authors suggest that the gentle touch insensitivity phenotype of pkd- 1 homozygous mutants is not due to impaired development of the type II hair cell sensory cilia, but this conclusion requires corroboration with quantitative data on the number of type II hair cells in the tentacles of pkd- 1 homozygous mutants.  
+
+## Reviewer #2  
+
+(Remarks to the Author)Cryptic Diversity of cnidarian mechanosensory neurons identified the morphological, functional, molecular, and developmental differences between two distinct types of mechanosensory neurons in Nematostella vectensis. While similar morphological diversity has been described for mechanosensory neurons in other cnidarian species, the developmental, synaptic, and functional differences had not been interrogated at the level provided by this work. The author's demonstrate nicely that the two types of neuron are morphologically distinct, functionally perform different roles, use distinct molecules in their likely mechanosensory mechanism, and that they are not related developmentally. The findings strongly support the conclusion that these two cell types are distinct from one another. The authors then extend their conclusions to suggest that these data support the idea that mechanosensory neurons had diverged into distinct subtypes potentially in the cnidarian- bilaterian common ancestor. Collectively the data are robust and the findings consistent the core of the author's conclusions. I found no major concerns with this study, and only four minor issues. Three of the minor issues are easy to fix, and third requires some expansion of the discussion and a reigning in of the broader implications of mechanosensory diversity in the cnidarian- bilaterian common ancestor.  
+
+## Minor issues:  
+
+Figure 1 - The long cilia of cell type II is really hard to see, even turning the monitor brightness to high. I would suggest adding a panel or enhancing brightness and contract to make the cilium easier to see. It was clear in the supplemental movie, but it should be more obvious in the static images of figure 1, given that most will not download the movie and it is the feature that stood out initially to distinguish these cells.  
+
+Regarding the investigation of lineage relationship between type I and type II cells. The word lineage is loaded and typically refers to proliferative cells. The authors did not test if these two cell types are derived from different stages in the lineage of stem or progenitor cell. They demonstrate that the cells are not different stages of a post- mitotic cell type. The section should reflect this and not use the term lineage that could lead to misinterpretations of the findings.  
+
+I don't understand why the diversity is cryptic? Usually cryptic refers to something that is difficult to observe when comparing two or more things. Here the morphological diversity was already known in other species (see the paper's introduction), and the test that the differences might reflect different functions seems like the obvious next experiment.  
+
+The more complicated issue is whether the data as presented reflect clear support for an already existing diversity of mechanosensory cell types in the cnidarian- bilaterian (or even in just the cnidarian) common ancestor. While the morphological diversity certainly existed in cnidarians these animals have been evolving for a long time, and an alteration of cilia length seems like a trait that could easily be selected for in different contexts multiple times. The molecular comparisons between the two types of mechanosensory cells across cnidarians or even cnidarians and bilaterians was not done. Without that it doesn't seem possible to refute the hypothesis that a single ancestral mechanosensory cell radiated in both lineages. The authors either need to enhance the discussion to put their results into context that refutes the single cell type radiating hypothesis, or reframe the discussion to point out how this study poises future efforts to better test the competing hypotheses.  
+
+## Reviewer #3  
+
+(Remarks to the Author)In the article "Cryptic diversity of cnidarian mechanosensory neurons" the authors uncovered a previously unrecognized mechanosensory neuron - type II hair cell - in the tentacular ectoderm of the sea anemone Nematostella Previously "only" the type I hair cells was known. The newly identified sensory neuron differs in morphology, mechanotransduction mechanism, and development from the previously known type I hair cell. Using a plethora of different methods (transgenesis, 2d and 3D electron microscopy, behavioral experiments, knockout and photoconversion experiments) the authors convincingly show that the newly identified cell is indeed a mechanosensitive receptor with a distinct mechanotransduction mechanisms and show how gentle touch response vs harsh touch response is sensed in cnidarians.  
+
+This is a very interesting manuscript describing cnidarian mechanosensory neurons. This manuscript will be of great interest to a broad readership (important for the fields of neurobiology, cell biology and evolutionary biology), and I can also see how this study can provide the foundation for future studies in this area.  
+
+Nevertheless, there are concerns which need to be addressed:
+
+<--- Page Split --->
+
+Major concerns:  
+
+1) The authors claim that Type I and Type II hair cells are developmental distinct cell types. The recent single cell atlas of Nematostella developmental stages (Cole et al 2024 Frontiers in zoology [https://link.springer.com/article/10.1186/s12983-024-00529-z] and Steger et al 2022 Cell Reports [https://www.cell.com/cell-reports/fulltext/S2211-1247(22)01202-5? uuid=uid%3A2af2d94a-ebc5-4473-a3bc-c2ccc09012a0] would allow for the direct comparison and surprisingly has not been leveraged in this paper. What is the transcriptional profile of the TRP channel Polycystin1, marker of the Type II hair cell? What is the transcriptional profile of the PO POU-IV transcription factor, marker of the Type I hair cell previously described by the lab (elife paper)?  
+
+2) The synapses (using SBFSEM) look different in Type I hair cells vs Type II hair cells (less vesicles and bigger with clear content in Type I vs more vesicles and darker content in Type II). If indeed the case authors should mention and highlight this in the paper. Moreover, the authors, as they have the SBFSEM data, should mention how many synapses in total have been observed/counted and quantify how many synaptic vesicles are present in the different synapses.  
+
+3) The authors claim to see different patterns of synaptic activity of Type I hair cells vs Type II hair cells. How many synapses have the authors studied for this? How often do they see two-way synapses. More data analysis/quantification is needed.  
+
+4) The authors show a beautiful 3D reconstruction of the Type II hair cell. However, it's unclear where the synapses are (at the end of the protrusions, at the cell body?). I would suggest adding synapse(s) to the 3D model.  
+
+5) Type I hair cells and Type II hair cells differ by their cilia length, however how many cilia have the authors measured remains unclear. Please add this to the manuscript.  
+
+Minor concerns:  
+
+6) In Panel L of Figure 1 the cilia is very hard to see (transparent and blue). Could the authors make it less transparent?  
+
+7) Software and procedure the authors have used to reconstruct type II hair cells form SBFSEM data are not mentioned in the methods part. Please add!  
+
+8) In general, the discussion is very short and should be expanded. For example, if the authors include work that has been done on single cell RNAseq during development to distinguish Type I vs Type II sensory cells, they could expand and discuss this more.  
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+The authors have responded to all requests in this revised manuscript with appropriate data and text. It is now clearer that significant insights into the evolution of mechanosensory cells/systems are provided by the detailed analysis of cnidarian species in this revised version. As I think that significant improvements have made the manuscript more accessible to a wider audience for the value of this excellent work, I respectfully endorse the acceptance of the revised manuscript.  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+The authors adequately addressed concerns. I suggest the paper is accepted  
+
+Reviewer #3  
+
+(Remarks to the Author)  
+
+The authors addressed all the major and minor points raised satisfactorily. I think the paper is an important contribution to the field and I fully support acceptance in its current version.
+
+<--- Page Split --->
+
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author)  
+
+In "Cryptic diversity of cnidarian mechanosensory neurons", Baranyk and co- authors investigate the molecular, anatomical, and development of novel sensory neurons, type II hair cells, in a cnidarian sea anemone Nematostella vectensis and aim to provide insight into the sensory response mediated by this type of cells in cnidarians. Using gene KO techniques, this study clearly demonstrated that the function of the polycystin- 1 gene, an ion channel, is essential for mechanosensing in type II hair cells. The main conclusion of this manuscript is that type II hair cells with long cilium are responsible for sensing gentle stimuli in tentacles and that this is mediated by the function of the polycystin- 1 gene product. Additionally, the authors identified the upstream genomic region of polycystin- 1, which contains the binding sequence of pou- IV, drove reporter gene expression specifically in type II hair cells. The authors also used a pou- IV transgenic line to induce reporter gene expression in type I and II hair cells, suggesting that these sensory cells are not in different differentiation states of the identical cell type, but are distinct differentiated cells.  
+
+These observations are new, yet this represents only an incremental advancement of the field of cnidarian neurobiology. In addition to the question about whether it will be of interest to a wider range of researchers, there are also some technical issues, as described below. I am not fully convinced that it fits into the journal Nature Communications.  
+
+## Major points  
+
+1) In order to argue for the existence of chemical synapses in the sensory cells analyzed by the authors in this study, it is necessary to show high electron density in the cell-cell junction regions and the accumulation of vesicles that the authors claim are "synaptic vesicles" through 3D reconstruction, rather than relying solely on single-section electrophysiological photographs.  
+
+Response: As suggested, we have added supplementary materials showing electron- dense cell-cell junction regions as well as 3D reconstruction of vesicle clustering at a synapse (Supplementary Fig. 2, Supplementary Movies 2-4).  
+
+Furthermore, quantitative data should be presented to discuss what types of cells form synapse- like structures with hair cells.  
+
+Response: Accordingly, we have included in the results section quantitative data on the cell types that form structurally defined synapses with hair cells.  
+
+2) It is necessary to confirm that endogenous pkd1 is expressed in type II hair cells. The authors used a transgenic reporter line to examine the expression specificity of pkd1. In this paper, the authors used only the 2.1 kb genomic region upstream of pkd1, but the validity of using only this region has not been demonstrated. In a previous paper (Ozment et al., 2021, Figure 8M), a 3.7 kb upstream region was used, which appears to be expressed more widely than in Supplementary Figure 3C of this manuscript. Whether Pkd1-dependent stimulus responses can be fully attributed to type II hair cells
+
+<--- Page Split --->
+
+can only be determined if the genomic region used to express the reporter gene completely mimics the expression pattern of endogenous pkd1.  
+
+Response: This is an important point, and we have carried out in situ hybridization to confirm that the transgenic reporter expression indeed recapitulates the expression pattern of endogenous polycystin- 1. We have added the confirmatory data to the revised manuscript (Supplementary Fig. 5).  
+
+3) It is puzzling that although the authors mention responses to mechanical stimuli in multiple cnidarian lineages, they do not discuss the extent to which the pou-IV and pkd1 genes they analyzed are universally present in the phylum Cnidaria. With many cnidarian genomes now publicly available, and if single-cell transcriptomic data are available, the presence and expression profiles of genes such as pkd1 should be demonstrated.  
+
+Response: We have included in the introduction a statement that POU- IV is deeply conserved across animals, and expanded the discussion to integrate recently published single- cell transcriptome data suggesting the possibility of coexpression of pou- iv and polycystin- 1 in a hydrozoan cnidarian and in a placozoan.  
+
+4) Furthermore, the authors need to clarify how their findings relate to mechanosensation in other nonbilaterians and bilaterians. Otherwise, this study will remain a description of something "specialized" about a certain type of mechanosensory cell in a type of sea anemone called Nematostella. For example, what are the genetic characteristics of mechanosensory hair cells expressing Pkd-1? Does the gene signature of the cnidarian Pkd-1-positive cell cluster(s) show any connections to certain sensory cell types in "post-cnidarian" bilaterians and "pre-cnidarian" metazoans such as poriferans, ctenophores, and placozoans?  
+
+Response: Accordingly, we have expanded the discussion in the revised manuscript, explicitly comparing the genetic feature of type II hair cells of Nematostella with that of cell types in Bilateria, Cnidaria, and Placozoa.  
+
+## Minor point  
+
+5) The response to gentle touch is relative to the response to strong touch (100%). The authors suggest that the gentle touch insensitivity phenotype of pkd-1 homozygous mutants is not due to impaired development of the type II hair cell sensory cilia, but this conclusion requires corroboration with quantitative data on the number of type II hair cells in the tentacles of pkd-1 homozygous mutants.  
+
+Response: We agree with the reviewer's comment, and have quantified the linear density of type II hair cells in polycystin- 1 mutants and their wildtype siblings, confirming that type II hair cells of mutants and wildtype siblings do not quantitatively differ. We have included the data in the revised manuscript.
+
+<--- Page Split --->
+
+Reviewer #2 (Remarks to the Author):  
+
+Crypto Diversity of cnidarian mechanosensory neurons identified the morphological, functional, molecular, and developmental differences between two distinct types of mechanosensory neurons in Nematostella vectensis. While similar morphological diversity has been described for mechanosensory neurons in other cnidarian species, the developmental, synaptic, and functional differences had not been interrogated at the level provided by this work. The author's demonstrate nicely that the two types of neuron are morphologically distinct, functionally perform different roles, use distinct molecules in their likely mechanosensory mechanism, and that they are not related developmentally. The findings strongly support the conclusion that these two cell types are distinct from one another. The authors then extend their conclusions to suggest that these data support the idea that mechanosensory neurons had diverged into distinct subtypes potentially in the cnidarian- bilaterian common ancestor. Collectively the data are robust and the findings consistent the core of the author's conclusions. I found no major concerns with this study, and only four minor issues. Three of the minor issues are easy to fix, and third requires some expansion of the discussion and a reigning in of the broader implications of mechanosensory diversity in the cnidarian- bilaterian common ancestor.  
+
+Minor issues:  
+
+Figure 1 - The long cilia of cell type II is really hard to see, even turning the monitor brightness to high. I would suggest adding a panel or enhancing brightness and contract to make the cilium easier to see. It was clear in the supplemental movie, but it should be more obvious in the static images of figure 1, given that most will not download the movie and it is the feature that stood out initially to distinguish these cells.  
+
+Response: Accordingly, we have increased the brightness and contrast to make the cilia of the type II hair cell in Figure 1 more readily visible.  
+
+Regarding the investigation of lineage relationship between type I and type II cells. The word lineage is loaded and typically refers to proliferative cells. The authors did not test if these two cell types are derived from different stages in the lineage of stem or progenitor cell. They demonstrate that the cells are not different stages of a post- mitotic cell type. The section should reflect this and not use the term lineage that could lead to misinterpretations of the findings.  
+
+Response: We agree and therefore avoided the use of the term throughout the revised manuscript; instead we explicitly state what we mean (that type I and type II hair cells do not represent different maturation phases of a post- mitotic cell type).  
+
+I don't understand why the diversity is cryptic? Usually cryptic refers to something that is difficult to observe when comparing two or more things. Here the morphological diversity was already known in other species (see the paper's introduction), and the test that the differences might reflect different functions seems like the obvious next experiment.
+
+<--- Page Split --->
+
+Response: We thank the reviewer for this comment. Although, as the reviewer point out, morphological diversity of hair cells has been documented in medusozoan cnidarians, this is not the case in anthozoan cnidarians. The use of the term 'cryptic' is therefore appropriate in this narrow sense. We do realize, however, that the earlier version of the manuscript generalized the diversity to be cryptic across Cnidaria, and therefore we removed such generalization in the revised manuscript.  
+
+The more complicated issue is whether the data as presented reflect clear support for an already existing diversity of mechanosensory cell types in the cnidarian- bilaterian (or even in just the cnidarian) common ancestor. While the morphological diversity certainly existed in cnidarians these animals have been evolving for a long time, and an alteration of cilia length seems like a trait that could easily be selected for in different contexts multiple times. The molecular comparisons between the two types of mechanosensory cells across cnidarians or even cnidarians and bilaterians was not done. Without that it doesn't seem possible to refute the hypothesis that a single ancestral mechanosensory cell radiated in both lineages. The authors either need to enhance the discussion to put their results into context that refutes the single cell type radiating hypothesis, or reframe the discussion to point out how this study poises future efforts to better test the competing hypotheses.  
+
+Response: We agree that independent diversification of mechanosensory neural cell types in Bilateria and Cnidaria cannot be ruled out, and did not intend to discount this hypothesis in the original version of the manuscript. To better clarify, we have made it explicit in the revised manuscript that further comparative studies of mechanism of development and function of mechanosensory neural cell types are necessary to resolve alternative evolutionary scenarios regarding how mechanosensory neurons diversified in Cnidaria and across major animal lineages.  
+
+Reviewer #3 (Remarks to the Author):  
+
+In the article "Cryptic diversity of cnidarian mechanosensory neurons" the authors uncovered a previously unrecognized mechanosensory neuron - type II hair cell - in the tentacular ectoderm of the sea anemone Nematostella Previously "only" the type I hair cells was known. The newly identified sensory neuron differs in morphology, mechanotransduction mechanism, and development from the previously known type I hair cell. Using a plethora of different methods (transgenesis, 2d and 3D electron microscopy, behavioral experiments, knockout and photoconversion experiments) the authors convincingly show that the newly identified cell is indeed a mechanosensitive receptor with a distinct mechanotransduction mechanisms and show how gentle touch response vs harsh touch response is sensed in cnidarians.  
+
+This is a very interesting manuscript describing cnidarian mechanosensory neurons. This manuscript will be of great interest to a broad readership (important for the fields of neurobiology, cell biology and evolutionary biology), and I can also see how this study can provide the foundation for future studies in this area.
+
+<--- Page Split --->
+
+Nevertheless, there are concerns which need to be addressed:  
+
+Major concerns:  
+
+1) The authors claim that Type I and Type II hair cells are developmental distinct cell types. The recent single cell atlas of Nematostella developmental stages (Cole et al 2024 Frontiers in zoology [https://link.springer.com/article/10.1186/s12983-024-00529-z] and Steger et al 2022 Cell Reports [https://www.cell.com/cell-reports/fulltext/S2211-1247(22)01202-5?uuid=uuid%3A2af2d94a-ebc5-4473-a3bc-c2ccc09012a0] would allow for the direct comparison and surprisingly has not been leveraged in this paper. What is the transcriptional profile of the TRP channel Polycystin1, marker of the Type II hair cell? What is the transcriptional profile of the PO POU-IV transcription factor, marker of the Type I hair cell previously described by the lab (elife paper)?  
+
+Response: Transcriptomic comparison of type I and type II hair cells would certainly be desirable, as it would shed further light on molecular- level differences between the cell types. The problem, however, is that we have yet to identify type I hair cells in the single- cell datasets because type I hair cell- specific marker genes remain unknown. POU- IV indeed is expressed in type I hair cells and is necessary for their development as described in the manuscript. But it is not a specific marker of type I hair cells; it is expressed in other cell types including type II hair cells and cnicocytes. Identification of type I hair cell- specific markers is essential before transcriptomic comparison of type I and type II hair cells can be performed. We therefore think that this is beyond the scope of this study.  
+
+2) The synapses (using SBFSEM) look different in Type I hair cells vs Type II hair cells (less vesicles and bigger with clear content in Type I vs more vesicles and darker content in Type II). If indeed the case authors should mention and highlight this in the paper.  
+
+Response: We are very thankful for the reviewer for this insight. We looked into these features and indeed found significant differences in size and number of synaptic vesicles between type I and type II hair cells. We have added these results to the revised manuscript.  
+
+Moreover, the authors, as they have the SBFSEM data, should mention how many synapses in total have been observed/counted and quantify how many synaptic vesicles are present in the different synapses.  
+
+Response: We have added the requested information to the results section of the revised manuscript.  
+
+3) The authors claim to see different patterns of synaptic activity of Type I hair cells vs Type II hair cells. How many synapses have the authors studied for this? How often do they see two-way synapses. More data analysis/quantification is needed.
+
+<--- Page Split --->
+
+Response: We have added the requested information to the results section of the revised manuscript.  
+
+4) The authors show a beautiful 3D reconstruction of the Type II hair cell. However, it's unclear where the synapses are (at the end of the protrusions, at the cell body?). I would suggest adding synapse(s) to the 3D model.  
+
+Response: To clarify the location of synapses in the type II hair cell, we have included a supplementary movie (4) where a 3D model of a type II hair cell showing synaptic vesicles is rotated around the apical-basal axis.  
+
+5) Type I hair cells and Type II hair cells differ by their cilia length, however how many cilia have the authors measured remains unclear. Please add this to the manuscript.  
+
+Response: We have added the requested information to the results section of the revised manuscript.  
+
+Minor concerns:  
+
+6) In Panel L of Figure 1 the cilia is very hard to see (transparent and blue). Could the authors make it less transparent?  
+
+Response: For Figure 1L, we want to additionally show the entire set of microvilli which would be masked if the cilium is made less transparent. Instead, we have made the cilium of the type II hair cell in the newly added Supplementary Movie 4 ��� showing the side view of the cell – less transparent.  
+
+7) Software and procedure the authors have used to reconstruct type II hair cells form SBFSEM data are not mentioned in the methods part. Please add!  
+
+Response: We have added this information to the methods section of the revised manuscript.  
+
+8) In general, the discussion is very short and should be expanded. For example, if the authors include work that has been done on single cell RNAseq during development to distinguish Type I vs Type II sensory cells, they could expand and discuss this more.  
+
+Response: We have added a discussion about the evolutionary relationship of type II hair cells of the sea anemone with cells in other animals, and included statements about future direction emphasizing the need for more comparative mechanistic data on development and function to elucidate evolutionary histories of mechanosensory neural cell types across animals. As discussed above, we could not perform transcriptome- wide comparison of type I and type II hair cells yet.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__9f2277ee7aacb1f4b8c740ce1111b21e6282130d24dd421de8747248239cab4e/supplementary_0_Peer Review file__9f2277ee7aacb1f4b8c740ce1111b21e6282130d24dd421de8747248239cab4e_det.mmd

@@ -0,0 +1,328 @@
+<|ref|>title<|/ref|><|det|>[[72, 163, 872, 211]]<|/det|>
+# Structural molecular and developmental evidence for cell-type diversity in cnidarian mechanosensory neurons  
+
+<|ref|>text<|/ref|><|det|>[[73, 224, 540, 241]]<|/det|>
+Corresponding Author: Professor Nagayasu Nakanishi  
+
+<|ref|>text<|/ref|><|det|>[[72, 275, 864, 289]]<|/det|>
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+<|ref|>text<|/ref|><|det|>[[73, 327, 144, 340]]<|/det|>
+Version 0:  
+
+<|ref|>text<|/ref|><|det|>[[73, 354, 220, 367]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 380, 160, 393]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 405, 238, 418]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 419, 923, 457]]<|/det|>
+In "Cryptic diversity of cnidarian mechanosensory neurons", Baranyk and co- authors investigate the molecular, anatomical, and development of novel sensory neurons, type II hair cells, in a cnidarian sea anemone Nematostella vectensis and aim to provide insight into the sensory response mediated by this type of cells in cnidarians.  
+
+<|ref|>text<|/ref|><|det|>[[72, 457, 923, 548]]<|/det|>
+Using gene KO techniques, this study clearly demonstrated that the function of the polycystin- 1 gene, an ion channel, is essential for mechanosensing in type II hair cells. The main conclusion of this manuscript is that type II hair cells with long cilium are responsible for sensing gentle stimuli in tentacles and that this is mediated by the function of the polycystin- 1 gene product. Additionally, the authors identified the upstream genomic region of polycystin- 1, which contains the binding sequence of pou- IV, drove reporter gene expression specifically in type II hair cells. The authors also used a pou- IV transgenic line to induce reporter gene expression in type I and II hair cells, suggesting that these sensory cells are not in different differentiation states of the identical cell type, but are distinct differentiated cells.  
+
+<|ref|>text<|/ref|><|det|>[[72, 548, 915, 587]]<|/det|>
+These observations are new, yet this represents only an incremental advancement of the field of cnidarian neurobiology. In addition to the question about whether it will be of interest to a wider range of researchers, there are also some technical issues, as described below. I am not fully convinced that it fits into the journal Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[72, 600, 159, 613]]<|/det|>
+Major points  
+
+<|ref|>text<|/ref|><|det|>[[72, 614, 918, 678]]<|/det|>
+1) In order to argue for the existence of chemical synapses in the sensory cells analyzed by the authors in this study, it is necessary to show high electron density in the cell-cell junction regions and the accumulation of vesicles that the authors claim are "synaptic vesicles" through 3D reconstruction, rather than relying solely on single-section electrophysiological photographs. Furthermore, quantitative data should be presented to discuss what types of cells form synapse-like structures with hair cells.  
+
+<|ref|>text<|/ref|><|det|>[[71, 690, 924, 783]]<|/det|>
+2) It is necessary to confirm that endogenous pkd1 is expressed in type II hair cells. The authors used a transgenic reporter line to examine the expression specificity of pkd1. In this paper, the authors used only the 2.1 kb genomic region upstream of pkd1, but the validity of using only this region has not been demonstrated. In a previous paper (Ozment et al., 2021, Figure 8M), a 3.7 kb upstream region was used, which appears to be expressed more widely than in Supplementary Figure 3C of this manuscript. Whether Pkd1-dependent stimulus responses can be fully attributed to type II hair cells can only be determined if the genomic region used to express the reporter gene completely mimics the expression pattern of endogenous pkd1.  
+
+<|ref|>text<|/ref|><|det|>[[72, 795, 923, 848]]<|/det|>
+3) It is puzzling that although the authors mention responses to mechanical stimuli in multiple cnidarian lineages, they do not discuss the extent to which the pou-IV and pkd1 genes they analyzed are universally present in the phylum Cnidaria. With many cnidarian genomes now publicly available, and if single-cell transcriptomic data are available, the presence and expression profiles of genes such as pkd1 should be demonstrated.  
+
+<|ref|>text<|/ref|><|det|>[[72, 860, 914, 938]]<|/det|>
+4) Furthermore, the authors need to clarify how their findings relate to mechanosensation in other nonbilaterians and bilaterians. Otherwise, this study will remain a description of something "specialized" about a certain type of mechanosensory cell in a type of sea anemone called Nematostella. For example, what are the genetic characteristics of mechanosensory hair cells expressing Pkd-1? Does the gene signature of the cnidarian Pkd-1-positive cell cluster(s) show any connections to certain sensory cell types in "post-cnidarian" bilaterians and "pre-cnidarian" metazoans such as poriferans, ctenophores, and placozoans?
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[72, 61, 152, 74]]<|/det|>
+## Minor point  
+
+<|ref|>text<|/ref|><|det|>[[72, 74, 921, 127]]<|/det|>
+Minor point5) The response to gentle touch is relative to the response to strong touch \((100\%)\) . The authors suggest that the gentle touch insensitivity phenotype of pkd- 1 homozygous mutants is not due to impaired development of the type II hair cell sensory cilia, but this conclusion requires corroboration with quantitative data on the number of type II hair cells in the tentacles of pkd- 1 homozygous mutants.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 163, 161, 176]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[72, 189, 922, 360]]<|/det|>
+(Remarks to the Author)Cryptic Diversity of cnidarian mechanosensory neurons identified the morphological, functional, molecular, and developmental differences between two distinct types of mechanosensory neurons in Nematostella vectensis. While similar morphological diversity has been described for mechanosensory neurons in other cnidarian species, the developmental, synaptic, and functional differences had not been interrogated at the level provided by this work. The author's demonstrate nicely that the two types of neuron are morphologically distinct, functionally perform different roles, use distinct molecules in their likely mechanosensory mechanism, and that they are not related developmentally. The findings strongly support the conclusion that these two cell types are distinct from one another. The authors then extend their conclusions to suggest that these data support the idea that mechanosensory neurons had diverged into distinct subtypes potentially in the cnidarian- bilaterian common ancestor. Collectively the data are robust and the findings consistent the core of the author's conclusions. I found no major concerns with this study, and only four minor issues. Three of the minor issues are easy to fix, and third requires some expansion of the discussion and a reigning in of the broader implications of mechanosensory diversity in the cnidarian- bilaterian common ancestor.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 373, 165, 385]]<|/det|>
+## Minor issues:  
+
+<|ref|>text<|/ref|><|det|>[[72, 385, 920, 438]]<|/det|>
+Figure 1 - The long cilia of cell type II is really hard to see, even turning the monitor brightness to high. I would suggest adding a panel or enhancing brightness and contract to make the cilium easier to see. It was clear in the supplemental movie, but it should be more obvious in the static images of figure 1, given that most will not download the movie and it is the feature that stood out initially to distinguish these cells.  
+
+<|ref|>text<|/ref|><|det|>[[72, 449, 920, 503]]<|/det|>
+Regarding the investigation of lineage relationship between type I and type II cells. The word lineage is loaded and typically refers to proliferative cells. The authors did not test if these two cell types are derived from different stages in the lineage of stem or progenitor cell. They demonstrate that the cells are not different stages of a post- mitotic cell type. The section should reflect this and not use the term lineage that could lead to misinterpretations of the findings.  
+
+<|ref|>text<|/ref|><|det|>[[72, 515, 920, 556]]<|/det|>
+I don't understand why the diversity is cryptic? Usually cryptic refers to something that is difficult to observe when comparing two or more things. Here the morphological diversity was already known in other species (see the paper's introduction), and the test that the differences might reflect different functions seems like the obvious next experiment.  
+
+<|ref|>text<|/ref|><|det|>[[72, 567, 921, 673]]<|/det|>
+The more complicated issue is whether the data as presented reflect clear support for an already existing diversity of mechanosensory cell types in the cnidarian- bilaterian (or even in just the cnidarian) common ancestor. While the morphological diversity certainly existed in cnidarians these animals have been evolving for a long time, and an alteration of cilia length seems like a trait that could easily be selected for in different contexts multiple times. The molecular comparisons between the two types of mechanosensory cells across cnidarians or even cnidarians and bilaterians was not done. Without that it doesn't seem possible to refute the hypothesis that a single ancestral mechanosensory cell radiated in both lineages. The authors either need to enhance the discussion to put their results into context that refutes the single cell type radiating hypothesis, or reframe the discussion to point out how this study poises future efforts to better test the competing hypotheses.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 722, 161, 735]]<|/det|>
+## Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[72, 748, 922, 853]]<|/det|>
+(Remarks to the Author)In the article "Cryptic diversity of cnidarian mechanosensory neurons" the authors uncovered a previously unrecognized mechanosensory neuron - type II hair cell - in the tentacular ectoderm of the sea anemone Nematostella Previously "only" the type I hair cells was known. The newly identified sensory neuron differs in morphology, mechanotransduction mechanism, and development from the previously known type I hair cell. Using a plethora of different methods (transgenesis, 2d and 3D electron microscopy, behavioral experiments, knockout and photoconversion experiments) the authors convincingly show that the newly identified cell is indeed a mechanosensitive receptor with a distinct mechanotransduction mechanisms and show how gentle touch response vs harsh touch response is sensed in cnidarians.  
+
+<|ref|>text<|/ref|><|det|>[[72, 865, 920, 906]]<|/det|>
+This is a very interesting manuscript describing cnidarian mechanosensory neurons. This manuscript will be of great interest to a broad readership (important for the fields of neurobiology, cell biology and evolutionary biology), and I can also see how this study can provide the foundation for future studies in this area.  
+
+<|ref|>text<|/ref|><|det|>[[72, 918, 503, 931]]<|/det|>
+Nevertheless, there are concerns which need to be addressed:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 47, 184, 60]]<|/det|>
+Major concerns:  
+
+<|ref|>text<|/ref|><|det|>[[72, 72, 910, 166]]<|/det|>
+1) The authors claim that Type I and Type II hair cells are developmental distinct cell types. The recent single cell atlas of Nematostella developmental stages (Cole et al 2024 Frontiers in zoology [https://link.springer.com/article/10.1186/s12983-024-00529-z] and Steger et al 2022 Cell Reports [https://www.cell.com/cell-reports/fulltext/S2211-1247(22)01202-5? uuid=uid%3A2af2d94a-ebc5-4473-a3bc-c2ccc09012a0] would allow for the direct comparison and surprisingly has not been leveraged in this paper. What is the transcriptional profile of the TRP channel Polycystin1, marker of the Type II hair cell? What is the transcriptional profile of the PO POU-IV transcription factor, marker of the Type I hair cell previously described by the lab (elife paper)?  
+
+<|ref|>text<|/ref|><|det|>[[72, 177, 913, 231]]<|/det|>
+2) The synapses (using SBFSEM) look different in Type I hair cells vs Type II hair cells (less vesicles and bigger with clear content in Type I vs more vesicles and darker content in Type II). If indeed the case authors should mention and highlight this in the paper. Moreover, the authors, as they have the SBFSEM data, should mention how many synapses in total have been observed/counted and quantify how many synaptic vesicles are present in the different synapses.  
+
+<|ref|>text<|/ref|><|det|>[[70, 243, 920, 271]]<|/det|>
+3) The authors claim to see different patterns of synaptic activity of Type I hair cells vs Type II hair cells. How many synapses have the authors studied for this? How often do they see two-way synapses. More data analysis/quantification is needed.  
+
+<|ref|>text<|/ref|><|det|>[[70, 282, 911, 310]]<|/det|>
+4) The authors show a beautiful 3D reconstruction of the Type II hair cell. However, it's unclear where the synapses are (at the end of the protrusions, at the cell body?). I would suggest adding synapse(s) to the 3D model.  
+
+<|ref|>text<|/ref|><|det|>[[70, 320, 884, 348]]<|/det|>
+5) Type I hair cells and Type II hair cells differ by their cilia length, however how many cilia have the authors measured remains unclear. Please add this to the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[72, 360, 185, 373]]<|/det|>
+Minor concerns:  
+
+<|ref|>text<|/ref|><|det|>[[70, 385, 888, 400]]<|/det|>
+6) In Panel L of Figure 1 the cilia is very hard to see (transparent and blue). Could the authors make it less transparent?  
+
+<|ref|>text<|/ref|><|det|>[[70, 411, 900, 439]]<|/det|>
+7) Software and procedure the authors have used to reconstruct type II hair cells form SBFSEM data are not mentioned in the methods part. Please add!  
+
+<|ref|>text<|/ref|><|det|>[[70, 450, 907, 491]]<|/det|>
+8) In general, the discussion is very short and should be expanded. For example, if the authors include work that has been done on single cell RNAseq during development to distinguish Type I vs Type II sensory cells, they could expand and discuss this more.  
+
+<|ref|>text<|/ref|><|det|>[[72, 618, 144, 631]]<|/det|>
+Version 1:  
+
+<|ref|>text<|/ref|><|det|>[[72, 644, 219, 658]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[72, 670, 160, 684]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[72, 697, 238, 710]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 710, 902, 764]]<|/det|>
+The authors have responded to all requests in this revised manuscript with appropriate data and text. It is now clearer that significant insights into the evolution of mechanosensory cells/systems are provided by the detailed analysis of cnidarian species in this revised version. As I think that significant improvements have made the manuscript more accessible to a wider audience for the value of this excellent work, I respectfully endorse the acceptance of the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[72, 776, 161, 789]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[72, 802, 238, 815]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 815, 602, 829]]<|/det|>
+The authors adequately addressed concerns. I suggest the paper is accepted  
+
+<|ref|>text<|/ref|><|det|>[[72, 840, 161, 853]]<|/det|>
+Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[72, 866, 238, 879]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[70, 880, 923, 907]]<|/det|>
+The authors addressed all the major and minor points raised satisfactorily. I think the paper is an important contribution to the field and I fully support acceptance in its current version.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 45, 916, 99]]<|/det|>
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+<|ref|>text<|/ref|><|det|>[[72, 99, 796, 113]]<|/det|>
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+<|ref|>text<|/ref|><|det|>[[72, 112, 910, 165]]<|/det|>
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+<|ref|>text<|/ref|><|det|>[[72, 165, 618, 179]]<|/det|>
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 88, 440, 106]]<|/det|>
+Reviewer #1 (Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[113, 123, 875, 394]]<|/det|>
+In "Cryptic diversity of cnidarian mechanosensory neurons", Baranyk and co- authors investigate the molecular, anatomical, and development of novel sensory neurons, type II hair cells, in a cnidarian sea anemone Nematostella vectensis and aim to provide insight into the sensory response mediated by this type of cells in cnidarians. Using gene KO techniques, this study clearly demonstrated that the function of the polycystin- 1 gene, an ion channel, is essential for mechanosensing in type II hair cells. The main conclusion of this manuscript is that type II hair cells with long cilium are responsible for sensing gentle stimuli in tentacles and that this is mediated by the function of the polycystin- 1 gene product. Additionally, the authors identified the upstream genomic region of polycystin- 1, which contains the binding sequence of pou- IV, drove reporter gene expression specifically in type II hair cells. The authors also used a pou- IV transgenic line to induce reporter gene expression in type I and II hair cells, suggesting that these sensory cells are not in different differentiation states of the identical cell type, but are distinct differentiated cells.  
+
+<|ref|>text<|/ref|><|det|>[[115, 371, 875, 456]]<|/det|>
+These observations are new, yet this represents only an incremental advancement of the field of cnidarian neurobiology. In addition to the question about whether it will be of interest to a wider range of researchers, there are also some technical issues, as described below. I am not fully convinced that it fits into the journal Nature Communications.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 473, 224, 490]]<|/det|>
+## Major points  
+
+<|ref|>text<|/ref|><|det|>[[115, 491, 882, 578]]<|/det|>
+1) In order to argue for the existence of chemical synapses in the sensory cells analyzed by the authors in this study, it is necessary to show high electron density in the cell-cell junction regions and the accumulation of vesicles that the authors claim are "synaptic vesicles" through 3D reconstruction, rather than relying solely on single-section electrophysiological photographs.  
+
+<|ref|>text<|/ref|><|det|>[[115, 594, 858, 648]]<|/det|>
+Response: As suggested, we have added supplementary materials showing electron- dense cell-cell junction regions as well as 3D reconstruction of vesicle clustering at a synapse (Supplementary Fig. 2, Supplementary Movies 2-4).  
+
+<|ref|>text<|/ref|><|det|>[[115, 664, 869, 700]]<|/det|>
+Furthermore, quantitative data should be presented to discuss what types of cells form synapse- like structures with hair cells.  
+
+<|ref|>text<|/ref|><|det|>[[115, 716, 872, 751]]<|/det|>
+Response: Accordingly, we have included in the results section quantitative data on the cell types that form structurally defined synapses with hair cells.  
+
+<|ref|>text<|/ref|><|det|>[[115, 768, 879, 891]]<|/det|>
+2) It is necessary to confirm that endogenous pkd1 is expressed in type II hair cells. The authors used a transgenic reporter line to examine the expression specificity of pkd1. In this paper, the authors used only the 2.1 kb genomic region upstream of pkd1, but the validity of using only this region has not been demonstrated. In a previous paper (Ozment et al., 2021, Figure 8M), a 3.7 kb upstream region was used, which appears to be expressed more widely than in Supplementary Figure 3C of this manuscript. Whether Pkd1-dependent stimulus responses can be fully attributed to type II hair cells
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 808, 125]]<|/det|>
+can only be determined if the genomic region used to express the reporter gene completely mimics the expression pattern of endogenous pkd1.  
+
+<|ref|>text<|/ref|><|det|>[[115, 141, 850, 212]]<|/det|>
+Response: This is an important point, and we have carried out in situ hybridization to confirm that the transgenic reporter expression indeed recapitulates the expression pattern of endogenous polycystin- 1. We have added the confirmatory data to the revised manuscript (Supplementary Fig. 5).  
+
+<|ref|>text<|/ref|><|det|>[[115, 228, 876, 333]]<|/det|>
+3) It is puzzling that although the authors mention responses to mechanical stimuli in multiple cnidarian lineages, they do not discuss the extent to which the pou-IV and pkd1 genes they analyzed are universally present in the phylum Cnidaria. With many cnidarian genomes now publicly available, and if single-cell transcriptomic data are available, the presence and expression profiles of genes such as pkd1 should be demonstrated.  
+
+<|ref|>text<|/ref|><|det|>[[115, 350, 878, 420]]<|/det|>
+Response: We have included in the introduction a statement that POU- IV is deeply conserved across animals, and expanded the discussion to integrate recently published single- cell transcriptome data suggesting the possibility of coexpression of pou- iv and polycystin- 1 in a hydrozoan cnidarian and in a placozoan.  
+
+<|ref|>text<|/ref|><|det|>[[115, 437, 881, 577]]<|/det|>
+4) Furthermore, the authors need to clarify how their findings relate to mechanosensation in other nonbilaterians and bilaterians. Otherwise, this study will remain a description of something "specialized" about a certain type of mechanosensory cell in a type of sea anemone called Nematostella. For example, what are the genetic characteristics of mechanosensory hair cells expressing Pkd-1? Does the gene signature of the cnidarian Pkd-1-positive cell cluster(s) show any connections to certain sensory cell types in "post-cnidarian" bilaterians and "pre-cnidarian" metazoans such as poriferans, ctenophores, and placozoans?  
+
+<|ref|>text<|/ref|><|det|>[[115, 594, 864, 647]]<|/det|>
+Response: Accordingly, we have expanded the discussion in the revised manuscript, explicitly comparing the genetic feature of type II hair cells of Nematostella with that of cell types in Bilateria, Cnidaria, and Placozoa.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 665, 215, 681]]<|/det|>
+## Minor point  
+
+<|ref|>text<|/ref|><|det|>[[115, 682, 874, 769]]<|/det|>
+5) The response to gentle touch is relative to the response to strong touch (100%). The authors suggest that the gentle touch insensitivity phenotype of pkd-1 homozygous mutants is not due to impaired development of the type II hair cell sensory cilia, but this conclusion requires corroboration with quantitative data on the number of type II hair cells in the tentacles of pkd-1 homozygous mutants.  
+
+<|ref|>text<|/ref|><|det|>[[115, 785, 878, 855]]<|/det|>
+Response: We agree with the reviewer's comment, and have quantified the linear density of type II hair cells in polycystin- 1 mutants and their wildtype siblings, confirming that type II hair cells of mutants and wildtype siblings do not quantitatively differ. We have included the data in the revised manuscript.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 445, 107]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 123, 880, 421]]<|/det|>
+Crypto Diversity of cnidarian mechanosensory neurons identified the morphological, functional, molecular, and developmental differences between two distinct types of mechanosensory neurons in Nematostella vectensis. While similar morphological diversity has been described for mechanosensory neurons in other cnidarian species, the developmental, synaptic, and functional differences had not been interrogated at the level provided by this work. The author's demonstrate nicely that the two types of neuron are morphologically distinct, functionally perform different roles, use distinct molecules in their likely mechanosensory mechanism, and that they are not related developmentally. The findings strongly support the conclusion that these two cell types are distinct from one another. The authors then extend their conclusions to suggest that these data support the idea that mechanosensory neurons had diverged into distinct subtypes potentially in the cnidarian- bilaterian common ancestor. Collectively the data are robust and the findings consistent the core of the author's conclusions. I found no major concerns with this study, and only four minor issues. Three of the minor issues are easy to fix, and third requires some expansion of the discussion and a reigning in of the broader implications of mechanosensory diversity in the cnidarian- bilaterian common ancestor.  
+
+<|ref|>text<|/ref|><|det|>[[115, 439, 232, 455]]<|/det|>
+Minor issues:  
+
+<|ref|>text<|/ref|><|det|>[[115, 456, 883, 543]]<|/det|>
+Figure 1 - The long cilia of cell type II is really hard to see, even turning the monitor brightness to high. I would suggest adding a panel or enhancing brightness and contract to make the cilium easier to see. It was clear in the supplemental movie, but it should be more obvious in the static images of figure 1, given that most will not download the movie and it is the feature that stood out initially to distinguish these cells.  
+
+<|ref|>text<|/ref|><|det|>[[115, 559, 880, 595]]<|/det|>
+Response: Accordingly, we have increased the brightness and contrast to make the cilia of the type II hair cell in Figure 1 more readily visible.  
+
+<|ref|>text<|/ref|><|det|>[[115, 611, 876, 717]]<|/det|>
+Regarding the investigation of lineage relationship between type I and type II cells. The word lineage is loaded and typically refers to proliferative cells. The authors did not test if these two cell types are derived from different stages in the lineage of stem or progenitor cell. They demonstrate that the cells are not different stages of a post- mitotic cell type. The section should reflect this and not use the term lineage that could lead to misinterpretations of the findings.  
+
+<|ref|>text<|/ref|><|det|>[[115, 733, 870, 787]]<|/det|>
+Response: We agree and therefore avoided the use of the term throughout the revised manuscript; instead we explicitly state what we mean (that type I and type II hair cells do not represent different maturation phases of a post- mitotic cell type).  
+
+<|ref|>text<|/ref|><|det|>[[115, 803, 881, 874]]<|/det|>
+I don't understand why the diversity is cryptic? Usually cryptic refers to something that is difficult to observe when comparing two or more things. Here the morphological diversity was already known in other species (see the paper's introduction), and the test that the differences might reflect different functions seems like the obvious next experiment.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 879, 195]]<|/det|>
+Response: We thank the reviewer for this comment. Although, as the reviewer point out, morphological diversity of hair cells has been documented in medusozoan cnidarians, this is not the case in anthozoan cnidarians. The use of the term 'cryptic' is therefore appropriate in this narrow sense. We do realize, however, that the earlier version of the manuscript generalized the diversity to be cryptic across Cnidaria, and therefore we removed such generalization in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[114, 210, 877, 421]]<|/det|>
+The more complicated issue is whether the data as presented reflect clear support for an already existing diversity of mechanosensory cell types in the cnidarian- bilaterian (or even in just the cnidarian) common ancestor. While the morphological diversity certainly existed in cnidarians these animals have been evolving for a long time, and an alteration of cilia length seems like a trait that could easily be selected for in different contexts multiple times. The molecular comparisons between the two types of mechanosensory cells across cnidarians or even cnidarians and bilaterians was not done. Without that it doesn't seem possible to refute the hypothesis that a single ancestral mechanosensory cell radiated in both lineages. The authors either need to enhance the discussion to put their results into context that refutes the single cell type radiating hypothesis, or reframe the discussion to point out how this study poises future efforts to better test the competing hypotheses.  
+
+<|ref|>text<|/ref|><|det|>[[115, 437, 861, 561]]<|/det|>
+Response: We agree that independent diversification of mechanosensory neural cell types in Bilateria and Cnidaria cannot be ruled out, and did not intend to discount this hypothesis in the original version of the manuscript. To better clarify, we have made it explicit in the revised manuscript that further comparative studies of mechanism of development and function of mechanosensory neural cell types are necessary to resolve alternative evolutionary scenarios regarding how mechanosensory neurons diversified in Cnidaria and across major animal lineages.  
+
+<|ref|>text<|/ref|><|det|>[[115, 577, 445, 594]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 611, 880, 787]]<|/det|>
+In the article "Cryptic diversity of cnidarian mechanosensory neurons" the authors uncovered a previously unrecognized mechanosensory neuron - type II hair cell - in the tentacular ectoderm of the sea anemone Nematostella Previously "only" the type I hair cells was known. The newly identified sensory neuron differs in morphology, mechanotransduction mechanism, and development from the previously known type I hair cell. Using a plethora of different methods (transgenesis, 2d and 3D electron microscopy, behavioral experiments, knockout and photoconversion experiments) the authors convincingly show that the newly identified cell is indeed a mechanosensitive receptor with a distinct mechanotransduction mechanisms and show how gentle touch response vs harsh touch response is sensed in cnidarians.  
+
+<|ref|>text<|/ref|><|det|>[[115, 804, 878, 874]]<|/det|>
+This is a very interesting manuscript describing cnidarian mechanosensory neurons. This manuscript will be of great interest to a broad readership (important for the fields of neurobiology, cell biology and evolutionary biology), and I can also see how this study can provide the foundation for future studies in this area.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 660, 107]]<|/det|>
+Nevertheless, there are concerns which need to be addressed:  
+
+<|ref|>text<|/ref|><|det|>[[115, 124, 256, 142]]<|/det|>
+Major concerns:  
+
+<|ref|>text<|/ref|><|det|>[[115, 158, 880, 317]]<|/det|>
+1) The authors claim that Type I and Type II hair cells are developmental distinct cell types. The recent single cell atlas of Nematostella developmental stages (Cole et al 2024 Frontiers in zoology [https://link.springer.com/article/10.1186/s12983-024-00529-z] and Steger et al 2022 Cell Reports [https://www.cell.com/cell-reports/fulltext/S2211-1247(22)01202-5?uuid=uuid%3A2af2d94a-ebc5-4473-a3bc-c2ccc09012a0] would allow for the direct comparison and surprisingly has not been leveraged in this paper. What is the transcriptional profile of the TRP channel Polycystin1, marker of the Type II hair cell? What is the transcriptional profile of the PO POU-IV transcription factor, marker of the Type I hair cell previously described by the lab (elife paper)?  
+
+<|ref|>text<|/ref|><|det|>[[114, 333, 875, 508]]<|/det|>
+Response: Transcriptomic comparison of type I and type II hair cells would certainly be desirable, as it would shed further light on molecular- level differences between the cell types. The problem, however, is that we have yet to identify type I hair cells in the single- cell datasets because type I hair cell- specific marker genes remain unknown. POU- IV indeed is expressed in type I hair cells and is necessary for their development as described in the manuscript. But it is not a specific marker of type I hair cells; it is expressed in other cell types including type II hair cells and cnicocytes. Identification of type I hair cell- specific markers is essential before transcriptomic comparison of type I and type II hair cells can be performed. We therefore think that this is beyond the scope of this study.  
+
+<|ref|>text<|/ref|><|det|>[[115, 524, 866, 595]]<|/det|>
+2) The synapses (using SBFSEM) look different in Type I hair cells vs Type II hair cells (less vesicles and bigger with clear content in Type I vs more vesicles and darker content in Type II). If indeed the case authors should mention and highlight this in the paper.  
+
+<|ref|>text<|/ref|><|det|>[[115, 596, 860, 665]]<|/det|>
+Response: We are very thankful for the reviewer for this insight. We looked into these features and indeed found significant differences in size and number of synaptic vesicles between type I and type II hair cells. We have added these results to the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 681, 881, 734]]<|/det|>
+Moreover, the authors, as they have the SBFSEM data, should mention how many synapses in total have been observed/counted and quantify how many synaptic vesicles are present in the different synapses.  
+
+<|ref|>text<|/ref|><|det|>[[115, 751, 822, 786]]<|/det|>
+Response: We have added the requested information to the results section of the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 803, 868, 856]]<|/det|>
+3) The authors claim to see different patterns of synaptic activity of Type I hair cells vs Type II hair cells. How many synapses have the authors studied for this? How often do they see two-way synapses. More data analysis/quantification is needed.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 825, 125]]<|/det|>
+Response: We have added the requested information to the results section of the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 141, 866, 194]]<|/det|>
+4) The authors show a beautiful 3D reconstruction of the Type II hair cell. However, it's unclear where the synapses are (at the end of the protrusions, at the cell body?). I would suggest adding synapse(s) to the 3D model.  
+
+<|ref|>text<|/ref|><|det|>[[115, 211, 875, 264]]<|/det|>
+Response: To clarify the location of synapses in the type II hair cell, we have included a supplementary movie (4) where a 3D model of a type II hair cell showing synaptic vesicles is rotated around the apical-basal axis.  
+
+<|ref|>text<|/ref|><|det|>[[115, 280, 860, 316]]<|/det|>
+5) Type I hair cells and Type II hair cells differ by their cilia length, however how many cilia have the authors measured remains unclear. Please add this to the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 333, 825, 368]]<|/det|>
+Response: We have added the requested information to the results section of the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 385, 256, 402]]<|/det|>
+Minor concerns:  
+
+<|ref|>text<|/ref|><|det|>[[115, 420, 860, 455]]<|/det|>
+6) In Panel L of Figure 1 the cilia is very hard to see (transparent and blue). Could the authors make it less transparent?  
+
+<|ref|>text<|/ref|><|det|>[[115, 472, 870, 542]]<|/det|>
+Response: For Figure 1L, we want to additionally show the entire set of microvilli which would be masked if the cilium is made less transparent. Instead, we have made the cilium of the type II hair cell in the newly added Supplementary Movie 4 – showing the side view of the cell – less transparent.  
+
+<|ref|>text<|/ref|><|det|>[[115, 558, 861, 594]]<|/det|>
+7) Software and procedure the authors have used to reconstruct type II hair cells form SBFSEM data are not mentioned in the methods part. Please add!  
+
+<|ref|>text<|/ref|><|det|>[[115, 611, 822, 646]]<|/det|>
+Response: We have added this information to the methods section of the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 663, 870, 716]]<|/det|>
+8) In general, the discussion is very short and should be expanded. For example, if the authors include work that has been done on single cell RNAseq during development to distinguish Type I vs Type II sensory cells, they could expand and discuss this more.  
+
+<|ref|>text<|/ref|><|det|>[[115, 732, 880, 838]]<|/det|>
+Response: We have added a discussion about the evolutionary relationship of type II hair cells of the sea anemone with cells in other animals, and included statements about future direction emphasizing the need for more comparative mechanistic data on development and function to elucidate evolutionary histories of mechanosensory neural cell types across animals. As discussed above, we could not perform transcriptome- wide comparison of type I and type II hair cells yet.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__a3b85b67f9321a4530b60ff6cfe1ab0a84067204fc9e97aa5fdcb2b4121daa60/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review file__a3b85b67f9321a4530b60ff6cfe1ab0a84067204fc9e97aa5fdcb2b4121daa60/supplementary_0_Peer Review file__a3b85b67f9321a4530b60ff6cfe1ab0a84067204fc9e97aa5fdcb2b4121daa60.mmd

@@ -0,0 +1,378 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+# Accurate Cross-Species 5mC Detection for Oxford Nanopore Sequencing in Plants with DeepPlant  
+
+Corresponding Author: Dr Chuan- Le Xiao  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+In the paper "DeepPlant: An Accurate Cross- Species 5mC Detection Tool for Oxford Nanopore Sequencing in Plants," He- Xu Chen et al. present a new approach for improving 5- methylcytosine (5mC) detection in plant genomes. The key contributions include generating and publishing new plant datasets using ONT R10.4.1 and WGBS, which are valuable for training models, as well as providing pre- trained models for detecting 5mC in CpG, CHG, and CHH contexts. While the model architecture is similar to existing tools like DeepSignal, the strength of this work lies in the inclusion of diverse datasets and the tool specifically designed for plant epigenomics and the ONT R10.4.1 pore type. The authors compare DeepPlant to the Dorado model, which also detects 5mC in CpG, CHG, and CHH motifs, showing that DeepPlant outperforms Dorado on most metrics, particularly in the CHH context, which is unique to plant genomes. Overall, the paper is well- written and clear, making it easy to follow and understand.  
+
+Concerns:  
+
+A major concern with the evaluation presented in this paper is the inclusion of training data in the evaluation tables. For example, the reported Pearson's correlation values between DeepPlant predictions and BS- seq data at \(30\times\) sequencing depth range from 0.705 to 0.881 across all evaluated datasets. However, the highest correlation value (0.881) is achieved for a species that was included in the training set (R. communis). This undermines the validity of the evaluation, as models often perform significantly better on data they were trained on compared to unseen datasets. The inclusion of training data in the evaluation skews the results, making it difficult to assess the model's true generalization capability across new species. A clear separation between training and evaluation datasets is essential for a reliable assessment of model performance.  
+
+A concern with the current model design is the lack of systematic ablation analysis to evaluate the contributions of its components. The model is composed of three main parts: the first encodes sequence features (including signal length, base qualities, mean, standard deviation, and the base at each position in a k- mer), the second encodes signal features (normalized signal), and the third combines and processes the outputs from the first two parts. While the architecture is well- designed and incorporates diverse feature representations, it's unclear how much each feature and encoder contributes to the final predictions. Performing a systematic ablation study—by removing specific features or encoders—would provide crucial insights into the importance of each component, highlight potential redundancies, and help refine the architecture for improved performance.  
+
+A significant concern is the omission of Glycine max (G. max) from the analysis without any explanation. The authors state: "We then collected tissue samples from six of these species (excluding S. lycopersicum due to its close relation to S. tuberosum) and conducted BS- seq. For better sample diversity, A. thaliana, O. sativa, Glycine max, Vitis vinifera, and Marchantia polymorpha were also added to the analysis." However, while S. miltiorrhiza, S. tuberosum, and R. communis were used for training, and five additional species (A. thaliana, V. vinifera, O. sativa, B. vulgaris, and C. sinensis) were included for evaluation, G. max is not mentioned in the final analysis, nor is its exclusion explained.  
+
+In contrast, the authors explicitly state that P. patens and M. polymorpha were excluded due to possible sample impurities, which is reasonable. However, Figure 1 includes ratios of high- methylation CHH sites for all excluded datasets, including these two species, indicating that they were used in the manuscript. Given their inclusion in the analysis, it would be appropriate for the authors to publish these datasets, regardless of their exclusion from training and evaluation, to support
+
+<--- Page Split --->
+
+transparency and reproducibility. Additionally, since G. max was part of the original sample collection, its data should either be included in the analysis or its omission clearly justified.  
+
+For the read- level evaluation, the authors subsample 100,000 positive and 100,000 negative examples to create a balanced dataset, ensuring an equal number of positive and negative samples. For datasets where there are not enough positive examples to reach this number, they apply SMOTE (Synthetic Minority Over- sampling Technique) to artificially generate additional positive samples. While this approach helps create a balanced dataset for evaluation, it alters the natural distribution of the data, potentially making the evaluation less reflective of real- world conditions, where such balance rarely exists. This distribution shift could bias the results, as the model might perform differently on balanced datasets compared to naturally imbalanced data. To address this, it would be beneficial to include results on the full, unsampled datasets to provide a more realistic and unbiased evaluation of the model's true performance on real- world data.  
+
+The authors state, "To increase the reliability of the results, we repeated this process five times and calculated the average of the five evaluation results." However, they do not report the standard deviation or any other measure of variability to accompany the mean.  
+
+The authors do not specify whether the evaluations were performed on examples successfully called by both Dorado and DeepPlant (i.e., the intersection of valid predictions for both tools) or if the evaluations were conducted independently for each tool. Regardless, the number of evaluated examples should be explicitly reported for clarity and transparency.  
+
+The method for determining methylation status in a site- level evaluation is unclear. The authors state: "We determine the methylation threshold Pth based on the output methylation probability distribution, selecting the probability value within the range of 0.2 to 0.9 with the minor proportion in the probability distribution as Pth. If the methylation probability \(\mathrm{Pm} \geq \mathrm{Pth}\) , the sample is classified as methylated; otherwise, it is classified as non- methylated." However, it is not explained how the threshold is chosen in practical terms, particularly how the "minor proportion" is defined or calculated within the specified range. Greater clarity on the selection process is needed to fully understand this approach.  
+
+Only Dorado is used for comparison in the detection of 5mC in CpG motifs, despite the availability of other tools like DeepMod and Rockfish. While the authors exclude older tools such as Tombo, Megalodon, and DeepSignal- Plant due to their incompatibility with R10.4, this reasoning does not apply to DeepMod and Rockfish. The authors should explain why Dorado was chosen as the sole comparison tool or include comparisons with these additional methods to provide a more thorough evaluation of DeepPlant's performance in CpG detection.  
+
+The paper does not provide information about the running time or memory requirements of DeepPlant. While performance metrics like accuracy and F1 score are essential, practical factors such as computational efficiency and resource usage are equally important. Reporting details on running time and memory consumption would offer valuable insights into the model's scalability and suitability for large- scale or resource- constrained applications.  
+
+In Figure 3B, it is unclear whether the ratio of profiled sites corresponds to Nanopore coverage or DeepPlant coverage. The authors state that "Reads not aligned to the reference or had low alignment quality were filtered out." From this, it can be inferred that Nanopore coverage is likely higher than DeepPlant coverage. Specifying the coverage obtained for Dorado would also be helpful. Furthermore, the quality threshold for filtering low- quality alignments is not explicitly defined in the manuscript. However, in the code (GitHub commit: 67db8ef; Call_Modification.cpp, lines 107- 109), the authors define three constants for filtering, one of which (mapq_thresh_hold) appears to set the mapping quality threshold at 20. More detailed information about the other two heuristics used for filtering is necessary to fully understand the criteria applied during the analysis.  
+
+(Remarks on code availability)  
+
+The installation process described on GitHub is problematic and was unsuccessful. Despite having CUDA 11.8 installed and correctly configured (with 'nvcc - version', 'torch.version.cuda', and 'torch.cuda.is_available()' returning the expected output), an error occurs when running cmake. The first line of the error reads: "Could NOT find CUDA (missing: CUDA_INCLUDE_DIRS CUDA_CUDART_LIBRARY) (found version '11.8')." The installation was performed on a server with the 'cuda/11.8.0' and 'cudnn/8.9.2_cu1' module loaded.  
+
+Additionally, the installation instructions for CUDA are not user- friendly, as they assume the user has root privileges, which may not always be the case. It's also worth noting that running 'pip install torch==2.0.1' without the '- - index- url https://download.pytorch.org/whl/cu118' flag installs PyTorch for CUDA version 11.7. Providing a Dockerfile, a simpler execution option, or clearer setup instructions would improve accessibility and ensure a smoother installation process.  
+
+## Reviewer #2  
+
+(Remarks to the Author)  
+
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+(Remarks on code availability)
+
+<--- Page Split --->
+
+The installation process on GitHub is problematic and fails despite CUDA 11.8 being correctly configured. Errors occur during cmake, and the instructions assume root privileges, which may not be available to all users. Additionally, installing PyTorch without specific flags defaults to CUDA 11.7, causing potential compatibility issues. Providing a Dockerfile, a simpler installation process, or clearer instructions would greatly improve accessibility and reliability.  
+
+## Reviewer #3  
+
+(Remarks to the Author) Major Revision:  
+
+The authors developed a deep learning model to process DNA methylome data generated from ONT R10.4 platform. Particularly, they used mCHH enriched genome in the training to improve the accuracy of mCHH data detection. In general, the new model provides valuable tool for the application of ONT R10.4 in plant epigenome sequencing. I have several concerns:  
+
+1) When the authors processed the BS-seq data, how did they treat with the multiple mapped reads? These reads could be generated from repeat enriched regions, such as centromere. If they discard all multiple mapped reads, it will lead to underestimation of mCHH caption in BS-seq data., such as in figure 3b and 3c.  
+
+2) In figure 3B and 3C, besides BS-seq and Deepplant/Nanopore, the Nanopole data analysis based on Dorodo should also be included to see if there is any improvement using Deepplant then using Dorodo.  
+
+3) All the analysis are statistical results. The authors should show several screenshots of BS-seq, Dorado, Deepplant data in regions where contain mCG, mCHG and mCHH, and the authors could show these three contexts respectively, to demonstrate that the capture of mCHH is more accurate in Deepplant processed dataset.  
+
+(Remarks on code availability)  
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+We appreciate the authors for taking the time to incorporate our suggestions and improve the clarity and completeness of their manuscript. The clear separation of training and evaluation datasets, along with the additional explanations regarding dataset exclusions and public availability, significantly enhance the transparency of the study. The ablation study provides valuable insight into the contribution of individual components, and the inclusion of variability measures strengthens the reliability of the reported results. We also appreciate the clarification on the exclusion of DeepMod2 and the detailed information on DeepPlant's runtime and memory requirements. These thoughtful revisions contribute to a more comprehensive and well- supported analysis.  
+
+# Concern 1:  
+
+A remaining major concern is balanced sampling for read- level evaluation. For evaluation to be truly informative, results should be presented on real- world, naturally imbalanced datasets. Artificially balancing datasets during evaluation does not help users assess a method's practical applicability, as it distorts the class distribution and can inflate performance metrics. While we agree that some metrics, such as accuracy and AUC- ROC, are not suitable for evaluating highly imbalanced datasets, we cannot agree with the claim that metrics like AP, F1- score, precision, and recall "fail to accurately reflect true performance." These metrics are widely used in the machine learning and bioinformatics communities for precisely this purpose—evaluating models in extreme class imbalance scenarios.  
+
+Precision, in particular, can be "unforgiving" when there is a significant imbalance, but this does not make it unrepresentative or biased. Instead, low precision highlights a model's struggle to distinguish true positives from false positives. This is valuable information for users who care about the reliability of positive predictions. Moreover, metrics like average precision (AP) and precision- recall (PR) curves provide an even clearer picture, as they evaluate how well a model ranks positive examples rather than just how it labels them.  
+
+In summary, artificially balancing datasets for evaluation risks misleading users about real- world performance. Instead, presenting results on naturally imbalanced datasets—using well- established metrics like precision, recall, F1- score, and AP—ensures transparency and allows users to make informed decisions about method suitability.  
+
+# Concern 2:  
+
+We appreciate the authors for providing a detailed description of their method for determining methylation status in a site- level evaluation. While the procedural explanation is now clearer, the specific application of this method remains ambiguous. Is it applied to WGBS data, DeepPlant calls, or both?  
+
+Additionally, if this method is applied to DeepPlant, does the term methylation probability refer to the site- level prediction (i.e., the number of methylated Cs divided by the total number of Cs at that site), or does it correspond directly to the model's output? If the threshold is determined using WGBS data, is the same threshold applied to DeepPlant predictions? Furthermore, if this method is applied to DeepPlant but not Dorodo, it would be valuable to evaluate its impact on model performance by applying it to Dorodo or by comparing it with a simpler thresholding approach for DeepPlant predictions.
+
+<--- Page Split --->
+
+Minor comments:  
+
+We thank the authors for clearly separate training and evaluation datasets. However, some figures could still benefit from having clear separation, e.g. Supplementary Fig. 3  
+
+(Remarks on code availability) No additional remarks  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+(Remarks on code availability)  
+
+The instructions for running the tool using Docker are clearly written. However, the run command is missing the repository prefix (image name should be given with 'chenhx26/deepplant:1.1.0' instead of just 'deepplant:1.1.0'), which may cause an error when users attempt to run the image.  
+
+Reviewer #3  
+
+(Remarks to the Author)  
+
+After reviewing the manuscript, I am pleased to suggest that I have no revisions or objections to raise. The content is well- presented, and it contributes valuable insights to the field. Therefore, I recommend proceeding with its acceptance. Thank you for considering my feedback. Should you need any further information or discussion, please do not hesitate to contact me.  
+
+(Remarks on code availability) Accept  
+
+Version 2:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+We acknowledge the authors' efforts in incorporating the suggestions and improving the clarity of the manuscript. The addition of evaluations on datasets that were not sampled is particularly valuable. Furthermore, their explanation of the method for determining methylation status in site- level evaluations is satisfactory. We have no further concerns regarding the manuscript.  
+
+(Remarks on code availability)  
+
+We included this in previous reviews.  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+(Remarks on code availability)
+
+<--- Page Split --->
+
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+In the paper "DeepPlant: An Accurate Cross- Species 5mC Detection Tool for Oxford Nanopore Sequencing in Plants," He- Xu Chen et al. present a new approach for improving 5- methylcytosine (5mC) detection in plant genomes. The key contributions include generating and publishing new plant datasets using ONT R10.4.1 and WGBS, which are valuable for training models, as well as providing pre- trained models for detecting 5mC in CpG, CHG, and CHH contexts. While the model architecture is similar to existing tools like DeepSignal, the strength of this work lies in the inclusion of diverse datasets and the tool specifically designed for plant epigenomics and the ONT R10.4.1 pore type. The authors compare DeepPlant to the Dorado model, which also detects 5mC in CpG, CHG, and CHH motifs, showing that DeepPlant outperforms Dorado on most metrics, particularly in the CHH context, which is unique to plant genomes. Overall, the paper is well- written and clear, making it easy to follow and understand.  
+
+## Concerns:  
+
+A major concern with the evaluation presented in this paper is the inclusion of training data in the evaluation tables. For example, the reported Pearson's correlation values between DeepPlant predictions and BS- seq data at \(30\times\) sequencing depth range from 0.705 to 0.881 across all evaluated datasets. However, the highest correlation value (0.881) is achieved for a species that was included in the training set (R. communis). This undermines the validity of the evaluation, as models often perform significantly better on data they were trained on compared to unseen datasets. The inclusion of training data in the evaluation skews the results, making it difficult to assess the model's true generalization capability across new species. A clear separation between training and evaluation datasets is essential for a reliable assessment of model performance.  
+
+## Response:  
+
+Thank you for the valuable suggestions. We have now clearly separated the training and evaluation datasets in this revision. For fairness, when describing the model evaluation results, we have now excluded the three datasets used for training in the abstract and the main manuscript. To better illustrate the model's generalization capability, we sequenced new BS- seq and nanopore dataset for S. lycopersicum and further compared DeepPlant and the previous state- of- art software Dorado on it. Still, the results support that DeepPlant has advantage on the CHH motif. To be noticed, the correlation coefficients are affected by many factors such as mapping rate, reference genome divergence level, and methylation frequency distribution, making direct comparison of correlation coefficients across datasets less meaningful. After all, the evaluation results showed that the DeepPlant showed even greater advantage compared to Dorado on several test- only datasets than on the training datasets, suggesting better generalization capability across tested datasets. We hope these revisions (in lines 42- 43, 108, and et al.) have made the evaluation conclusion more robust.
+
+<--- Page Split --->
+
+A concern with the current model design is the lack of systematic ablation analysis to evaluate the contributions of its components. The model is composed of three main parts: the first encodes sequence features (including signal length, base qualities, mean, standard deviation, and the base at each position in a k- mer), the second encodes signal features (normalized signal), and the third combines and processes the outputs from the first two parts. While the architecture is well- designed and incorporates diverse feature representations, it's unclear how much each feature and encoder contributes to the final predictions. Performing a systematic ablation study—by removing specific features or encoders—would provide crucial insights into the importance of each component, highlight potential redundancies, and help refine the architecture for improved performance.  
+
+## Response:  
+
+Thank you for the suggestions. Understanding the importance of each component indeed will deepen the insights of this study. Accordingly, we have conducted systematic ablation experiments on DeepPlant to evaluate the contribution of its triple- encoder. By removing each of the encoders and training the model using the same samples and conditions, we quantitatively assessed the resulted models on the O. sativa and A. thaliana datasets. The results, presented in Supplementary Table 3, demonstrated that all three encoders have played a critical role in the model's performance. Among them, the sequence encoder has the greatest impact on the results, while the combine encoder has the least. For instance, the removal of any encoder resulted in at least a \(15\%\) decrease in performance on the A. thaliana dataset, indicating that each encoder is indispensable. We have the revised the manuscript in lines 171- 173.  
+
+A significant concern is the omission of Glycine max (G. max) from the analysis without any explanation. The authors state: "We then collected tissue samples from six of these species (excluding S. lycopersicum due to its close relation to S. tuberosum) and conducted BS- seq. For better sample diversity, A. thaliana, O. sativa, Glycine max, Vitis vinifera, and Marchantia polymorpha were also added to the analysis." However, while S. miltiorrhiza, S. tuberosum, and R. communis were used for training, and five additional species (A. thaliana, V. vinifera, O. sativa, B. vulgaris, and C. sinensis) were included for evaluation, G. max is not mentioned in the final analysis, nor is its exclusion explained.  
+
+## Response:  
+
+Thank you pointing it out. We initially sequenced BS- seq and Pacbio Hifi (for another project) for the G.max var. Huachun No.6, which showed high heterozygosity and abundant nucleotide variances to the reference genome (GCF_000004515.6). The high difference level compared to the reference will affect the quality of BS- seq that uses base conversion for detection of methylation state, therefore we have not sequenced nanopore data for it. We have now explained it in the manuscript (in lines 148- 149) and uploaded the BS- seq data to public database as requested (Data availability statement updated).
+
+<--- Page Split --->
+
+In contrast, the authors explicitly state that P. patens and M. polymorpha were excluded due to possible sample impurities, which is reasonable. However, Figure 1 includes ratios of high- methylation CHH sites for all excluded datasets, including these two species, indicating that they were used in the manuscript. Given their inclusion in the analysis, it would be appropriate for the authors to publish these datasets, regardless of their exclusion from training and evaluation, to support transparency and reproducibility. Additionally, since G. max was part of the original sample collection, its data should either be included in the analysis or its omission clearly justified.  
+
+## Response:  
+
+Thank you for the comments. We have uploaded our BS- seq data for P. patens and M. polymorpha, which are available at http://gsa.big.ac.cn under Project Accession No. PRJCA030666. As mentioned above, we have also uploaded the BS- seq data of G. max.  
+
+For the read- level evaluation, the authors subsample 100,000 positive and 100,000 negative examples to create a balanced dataset, ensuring an equal number of positive and negative samples. For datasets where there are not enough positive examples to reach this number, they apply SMOTE (Synthetic Minority Over- sampling Technique) to artificially generate additional positive samples. While this approach helps create a balanced dataset for evaluation, it alters the natural distribution of the data, potentially making the evaluation less reflective of real- world conditions, where such balance rarely exists. This distribution shift could bias the results, as the model might perform differently on balanced datasets compared to naturally imbalanced data. To address this, it would be beneficial to include results on the full, unsampled datasets to provide a more realistic and unbiased evaluation of the model's true performance on real- world data.  
+
+## Response:  
+
+Thank you for the comments. Following your suggestions, we evaluated the single- molecule performance of DeepPlant and Dorado on three full unbalanced dataset: A. thaliana (positive sample ratio \(0.0506\%\) ), B. vulgaris \((0.1249\%)\) , and O. sativa \((0.033\%)\) . The results are shown in the table below. Due to the extreme scarcity of positive samples, the evaluation metrics AP, F1, Precision, and Accuracy fail to accurately reflect the true performance of the models. This is because even if only a very small proportion of negative samples are misclassified as positive (false positives), their number would significantly exceed the number of true positive samples, leading to a decline in both F1 and Precision metrics. Similarly, the AP value, which is calculated based on the precision- recall curve, is also reduced as a result. Since the vast majority of samples are negative, Accuracy shows a substantial increase. In fact, under such circumstances, Accuracy largely reflects the model's ability to classify negative samples correctly. When the model correctly classifies all negative samples, its Accuracy can exceed \(99.8\%\) , rendering this metric unrepresentative of the model's overall performance. For the reasons, using extreme unbalanced datasets for single- molecule evaluations reduces its also severely biased. Therefore, we prefer to use balanced datasets to reflect the
+
+<--- Page Split --->
+
+single-molecule performance of the models. Additionally, to reflect the real-world sample distribution, we used the full datasets for site-level (methylation frequency) evaluation, which provides some reference value for practical scenarios.  
+
+<table><tr><td></td><td>tools</td><td>AUC</td><td>AP</td><td>F1-score</td><td>Precision</td><td>Accuracy</td><td>Recall</td><td>Specificity</td></tr><tr><td rowspan="2">A. thaliana</td><td>DeepPlant</td><td>0.8668</td><td>0.0073</td><td>0.001373</td><td>0.000687</td><td>0.97023</td><td>0.669626</td><td>0.970239</td></tr><tr><td>Dorado</td><td>0.8492</td><td>0.0024</td><td>0.001231</td><td>0.000616</td><td>0.969593</td><td>0.625466</td><td>0.969604</td></tr><tr><td rowspan="2">B. vulgaris</td><td>DeepPlant</td><td>0.933079</td><td>0.056785</td><td>0.007849</td><td>0.003943</td><td>0.931204</td><td>0.835183</td><td>0.931235</td></tr><tr><td>Dorado</td><td>0.880859</td><td>0.010706</td><td>0.004315</td><td>0.002164</td><td>0.901871</td><td>0.733399</td><td>0.90192</td></tr><tr><td rowspan="2">O. sativa</td><td>DeepPlant</td><td>0.9191</td><td>0.1076</td><td>0.026955</td><td>0.013706</td><td>0.947117</td><td>0.808072</td><td>0.947243</td></tr><tr><td>Dorado</td><td>0.907942</td><td>0.01916</td><td>0.008923</td><td>0.004486</td><td>0.877289</td><td>0.813208</td><td>0.877333</td></tr></table>  
+
+The authors state, "To increase the reliability of the results, we repeated this process five times and calculated the average of the five evaluation results." However, they do not report the standard deviation or any other measure of variability to accompany the mean.  
+
+## Response:  
+
+Thank you for pointing it out. We noticed the lack of rigor in this regard; therefore, we calculated the standard deviation of all single- molecule evaluation results, as shown in revised Supplementary Tables 6 and 7.  
+
+The authors do not specify whether the evaluations were performed on examples successfully called by both Dorado and DeepPlant (i.e., the intersection of valid predictions for both tools) or if the evaluations were conducted independently for each tool. Regardless, the number of evaluated examples should be explicitly reported for clarity and transparency.  
+
+## Response:  
+
+Thank you for pointing it out. We conducted all evaluations for each tool independently. For single- molecule evaluations, we extracted 200,000 samples under the same criteria and repeated the process five times, which has now been described in lines 498- 505. For quantitative evaluations, we additionally provided the number of sites detected for three types of motifs by each method, as shown in Supplementary Table 2.  
+
+The method for determining methylation status in a site- level evaluation is unclear. The authors state: "We determine the methylation threshold Pth based on the output methylation probability distribution, selecting the probability value within the range of 0.2 to 0.9 with the minor proportion in the probability distribution as Pth. If the methylation probability \(\mathrm{Pm} \geq \mathrm{Pth}\) , the sample is classified as methylated; otherwise, it is classified as non- methylated." However, it is not explained how the threshold is chosen in practical
+
+<--- Page Split --->
+
+terms, particularly how the "minor proportion" is defined or calculated within the specified range. Greater clarity on the selection process is needed to fully understand this approach.  
+
+## Response:  
+
+Thank you. We have described the detailed procedure in the Methods section now. The method applied is as follows in lines 510- 518:  
+
+'Divide the range from 0.2 to 0.9 into 70 intervals with a step size of 0.01, using the left endpoint of each interval as the representative value for that interval. For the detection results of a single dataset, group the samples into the corresponding intervals based on their methylation probabilities, count the number of samples in each interval, and calculate their proportion. The value corresponding to the interval with the smallest proportion is selected as the methylation threshold.'  
+
+Only Dorado is used for comparison in the detection of 5mC in CpG motifs, despite the availability of other tools like DeepMod and Rockfish. While the authors exclude older tools such as Tombo, Megalodon, and DeepSignal- Plant due to their incompatibility with R10.4, this reasoning does not apply to DeepMod and Rockfish. The authors should explain why Dorado was chosen as the sole comparison tool or include comparisons with these additional methods to provide a more thorough evaluation of DeepPlant's performance in CpG detection.  
+
+## Response:  
+
+Thank you for the suggestions. Accordingly, we further evaluated the recently published in this revision. For DeepMode2, because its original study (Ahsan et al.,) has carried out multiple assessments showing that it performed not as well as Dorado on R10.4 nanopore data, we have not included it in comparison in this revision. The results have been provided in Supplementary Table 6 and still identify DeepPlant as one of the state- of- art tools on CpG detection. The new benchmark results have also been updated in the manuscript.  
+
+Ahsan, M.U., Gouru, A., Chan, J. et al. A signal processing and deep learning framework for methylation detection using Oxford Nanopore sequencing. Nat Commun 15, 1448 (2024). https://doi.org/10.1038/s41467- 024- 45778- y  
+
+The paper does not provide information about the running time or memory requirements of DeepPlant. While performance metrics like accuracy and F1 score are essential, practical factors such as computational efficiency and resource usage are equally important. Reporting details on running time and memory consumption would offer valuable insights into the model's scalability and suitability for large- scale or resource- constrained applications.  
+
+## Response:
+
+<--- Page Split --->
+
+Benchmarking results are indeed crucial for evaluating the usability of the model. We performed benchmarking of DeepPlant on the entire sequencing dataset of A. thaliana. The benchmarking results, including runtime and memory consumption, are provided in Supplementary Table 9. We have also added a sentence about the benchmark results in lines 339- 341 in the discussion section.  
+
+In Figure 3B, it is unclear whether the ratio of profiled sites corresponds to Nanopore coverage or DeepPlant coverage. The authors state that "Reads not aligned to the reference or had low alignment quality were filtered out." From this, it can be inferred that Nanopore coverage is likely higher than DeepPlant coverage. Specifying the coverage obtained for Dorado would also be helpful. Furthermore, the quality threshold for filtering low- quality alignments is not explicitly defined in the manuscript. However, in the code (GitHub commit: 67db8ef; Call_Modification.cpp, lines 107- 109), the authors define three constants for filtering, one of which (mapq_thresh_hold) appears to set the mapping quality threshold at 20. More detailed information about the other two heuristics used for filtering is necessary to fully understand the criteria applied during the analysis.  
+
+## Response:  
+
+We recalculated the coverage of DeepPlant and Dorado across different genomic regions and revised Fig. 3b to more clearly illustrate the coverage differences between the three methods. The filtering criteria for low- quality alignments are as follows: (1) MAPQ<20; (2) primary alignment length \(< 80\%\) of read length; (3) mapping identity \(< 80\%\) . Reads were filtered if less than \(80\%\) of the reads were aligned to the. The details have now been described in both the Figure legend in lines 737- 739 and in the Methods section in lines 434- 436.  
+
+Reviewer #1 (Remarks on code availability):  
+
+The installation process described on GitHub is problematic and was unsuccessful. Despite having CUDA 11.8 installed and correctly configured (with 'nvcc - version', 'torch.version.cuda', and 'torch.cuda.is_available()' returning the expected output), an error occurs when running cmake. The first line of the error reads: "Could NOT find CUDA (missing: CUDA_INCLUDE_DIRS CUDA_CUDART_LIBRARY) (found version '11.8')." The installation was performed on a server with the 'cuda/11.8.0' and 'cudnn/8.9.2_cu1' module loaded.  
+
+Additionally, the installation instructions for CUDA are not user- friendly, as they assume the user has root privileges, which may not always be the case. It's also worth noting that running 'pip install torch==2.0.1' without the '- index- url https://download.pytorch.org/whl/cu118&#x2019; flag installs PyTorch for
+
+<--- Page Split --->
+
+CUDA version 11.7. Providing a Dockerfile, a simpler execution option, or clearer setup instructions would improve accessibility and ensure a smoother installation process.  
+
+## Response:  
+
+Thank you for pointing it out. In most cases, the linking location of CUDA in a computer is fixed. However, in some cases, the CUDA path configuration might be inconsistent, leading to compilation errors due to missing libraries. To address potential compilation issues, we provide a Dockerfile along with detailed usage instructions to simplify the use of the tool. The usage instructions are available at https://github.com/xiaochuanle/DeepPlant.  
+
+Reviewer #2 (Remarks to the Author):  
+
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+Reviewer #2 (Remarks on code availability):  
+
+The installation process on GitHub is problematic and fails despite CUDA 11.8 being correctly configured. Errors occur during cmake, and the instructions assume root privileges, which may not be available to all users. Additionally, installing PyTorch without specific flags defaults to CUDA 11.7, causing potential compatibility issues. Providing a Dockerfile, a simpler installation process, or clearer instructions would greatly improve accessibility and reliability.  
+
+## Response:  
+
+Thank you for your suggestions. In most cases, the linking location of CUDA in a computer is fixed. However, in some cases, the CUDA path configuration might be inconsistent, leading to compilation errors due to missing libraries. To address potential compilation issues, we provide a Dockerfile along with detailed usage instructions to simplify the use of the tool. The usage instructions are available at https://github.com/xiaochuanle/DeepPlant.  
+
+Reviewer #3 (Remarks to the Author):  
+
+Major Revision:  
+
+The authors developed a deep learning model to process DNA methylome data generated from ONT R10.4 platform. Particularly, they used mCHH enriched genome in the training to improve the accuracy of mCHH data detection. In general, the new model provides
+
+<--- Page Split --->
+
+valuable tool for the application of ONT R10.4 in plant epigenome sequencing. I have several concerns:  
+
+1) When the authors processed the BS-seq data, how did they treat with the multiple mapped reads? These reads could be generated from repeat enriched regions, such as centromere. if they discard all multiple mapped reads, it will lead to underestimation of mCHH caption in BS-seq data., such as in figure 3b and 3c.  
+
+## Response:  
+
+Thanks for bringing it out. We did not discard all multiple mapped reads. We applied the widely applied Bismark pipeline in BS-seq analysis which only uses unique best alignments in methylation frequency quantification. Unique best alignments refer to alignments where the next best alignment does not exist or is not as good (MAPQ>0), indicating that only multiple mapped reads with multiple best alignments were dumped. For the nanopore data, we also used unique best alignments, except that we required a much higher MAPQ threshold (20) for high alignment accuracy. For causing less confusion, related information has been added in lines 404- 405 and 434- 436.  
+
+2) In figure 3B and 3C, besides BS-seq and Deepplant/Nanopore, the Nanopole data analysis based on Dorodo should also be included to see if there is any improvement using Deepplant then using Dorado.  
+
+## Response:  
+
+Thank you for the suggestion. We have added Dorado's methylation analysis results for the relevant regions, as shown in Figure 3B, Figure 3C, and Supplementary Figures 2 and 5. The manuscript has been revised at related locations.  
+
+3) All the analysis are statistical results. The authors should show several screenshots of BS-seq, Dorado, Deepplant data in regions where contain mCG, mCHG and mCHH, and the authors could show these three contexts respectively, to demonstrate that the capture of mCHH is more accurate in Deepplant processed dataset.  
+
+## Response:  
+
+Thank you for the suggestions. We have added methylation analysis results of three methods for other regions, as shown in the Supplementary Figure 2. From these regions, it can be observed that DeepPlant's mCHH analysis results are closer to those of BS-seq, while Dorado's mCHH analysis results in some regions contain a large number of false positives, which should explain the higher correlation coefficient of DeepPlant with BS-seq. The manuscript has been revised accordingly in lines 207- 208.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+We appreciate the authors for taking the time to incorporate our suggestions and improve the clarity and completeness of their manuscript. The clear separation of training and evaluation datasets, along with the additional explanations regarding dataset exclusions and public availability, significantly enhance the transparency of the study. The ablation study provides valuable insight into the contribution of individual components, and the inclusion of variability measures strengthens the reliability of the reported results. We also appreciate the clarification on the exclusion of DeepMod2 and the detailed information on DeepPlant's runtime and memory requirements. These thoughtful revisions contribute to a more comprehensive and well- supported analysis.  
+
+# Concern 1:  
+
+A remaining major concern is balanced sampling for read- level evaluation. For evaluation to be truly informative, results should be presented on real- world, naturally imbalanced datasets. Artificially balancing datasets during evaluation does not help users assess a method's practical applicability, as it distorts the class distribution and can inflate performance metrics.  
+
+While we agree that some metrics, such as accuracy and AUC- ROC, are not suitable for evaluating highly imbalanced datasets, we cannot agree with the claim that metrics like AP, F1- score, precision, and recall "fail to accurately reflect true performance." These metrics are widely used in the machine learning and bioinformatics communities for precisely this purpose—evaluating models in extreme class imbalance scenarios.  
+
+Precision, in particular, can be "unforgiving" when there is a significant imbalance, but this does not make it unrepresentative or biased. Instead, low precision highlights a model's struggle to distinguish true positives from false positives. This is valuable information for users who care about the reliability of positive predictions. Moreover, metrics like average precision (AP) and precision- recall (PR) curves provide an even clearer picture, as they evaluate how well a model ranks positive examples rather than just how it labels them.  
+
+In summary, artificially balancing datasets for evaluation risks misleading users about real- world performance. Instead, presenting results on naturally imbalanced datasets—using well- established metrics like precision, recall, F1- score, and AP—ensures transparency and allows users to make informed decisions about method suitability.  
+
+Thank you very much for your suggestions. In this revision, we have included single- molecule test results on the imbalanced fully methylated/unmethylated CHH sites across the nine datasets. We agree that evaluating single- molecule performance on realistic, imbalanced datasets can provide valuable insights for researchers. However, the selection of high- quality reference control sites, such as fully methylated (100% methylation frequency) and fully unmethylated (0% methylation
+
+<--- Page Split --->
+
+frequency) sites in BS- seq, has been a limiting factor.  
+
+Given the scarcity of fully methylated CHH sites in the analyzed species, the ratio of fully methylated to unmethylated CHH sites (Supplementary Table 7) is much lower than the realistic methylation ratios observed in single- molecule data from the same species (Supplementary Table 2). In such cases, single- molecule performance is more effectively indirectly evaluated through methylation frequency quantification. For example, in A. thaliana, although the overall CHH methylation content is \(3\%\) , the ratio of \(100\%\) methylated to \(0\%\) methylated sites is 1:32,109. Consequently, direct evaluations of DeepPlant and Dorado on the imbalanced fully methylated/unmethylated datasets primarily provided insights into the accuracy of unmethylated site detection, with DeepPlant outperforming Dorado in all cases (Supplementary Table 7).  
+
+Thanks to your valuable suggestions, we have more clearly described the evaluation results and considerations in lines 237- 248.  
+
+## #Concern 2:  
+
+We appreciate the authors for providing a detailed description of their method for determining methylation status in a site- level evaluation. While the procedural explanation is now clearer, the specific application of this method remains ambiguous. Is it applied to WGBS data, DeepPlant calls, or both?  
+
+Additionally, if this method is applied to DeepPlant, does the term methylation probability refer to the site- level prediction (i.e., the number of methylated Cs divided by the total number of Cs at that site), or does it correspond directly to the model's output? If the threshold is determined using WGBS data, is the same threshold applied to DeepPlant predictions?  
+
+Furthermore, if this method is applied to DeepPlant but not Dorado, it would be valuable to evaluate its impact on model performance by applying it to Dorado or by comparing it with a simpler thresholding approach for DeepPlant predictions.  
+
+Thank you for your valuable suggestions. Methylation probability refers to a numerical value between 0 and 1, representing the likelihood that a specific CpG motif in a read is methylated, and is therefore different from site- level methylation frequency which is aggregated from the methylation status of multiple reads. As a result, the threshold selection method is only applicable to DeepPlant and Dorado. Following your advice, we assessed the impact of this threshold selection approach on Dorado, and the results showed that the method could also improve the accuracy of Dorado's site- level methylation frequency quantifications, though it only reduced the advantage of DeepPlant by a small proportion. Relevant results are summarized in Table 1. Additionally, we provided a more detailed explanation of this method, which essentially seeks a balance between false positives and false negatives in the model's predictions, thereby reducing the total number of false positives and false negatives as detailed in Supplementary Fig 2. The manuscript has been revised accordingly in lines 203- 207.
+
+<--- Page Split --->
+
+Minor comments:  
+
+We thank the authors for clearly separate training and evaluation datasets. However, some figures could still benefit from having clear separation, e.g. Supplementary Fig. 3Thank you for pointing out the detail issues in some of the figures. We have separated the datasets in the relevant figures.  
+
+Reviewer #1 (Remarks on code availability):  
+
+No additional remarks  
+
+Reviewer #2 (Remarks to the Author):  
+
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+Reviewer #2 (Remarks on code availability):  
+
+The instructions for running the tool using Docker are clearly written. However, the run command is missing the repository prefix (image name should be given with 'chenhx26/deepplant:1.1.0' instead of just 'deepplant:1.1.0'), which may cause an error when users attempt to run the image.  
+
+Thank you for pointing out the issues in the command. We have made corrections to the relevant commands.  
+
+Reviewer #3 (Remarks to the Author):  
+
+After reviewing the manuscript, I am pleased to suggest that that I have no revisions or objections to raise. The content is well- presented, and it contributes valuable insights to the field. Therefore, I recommend proceeding with its acceptance.  
+
+Thank you for considering my feedback. Should you need any further information or discussion, please do not hesitate to contact me.  
+
+Reviewer #3 (Remarks on code availability):  
+
+Accept
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__a3b85b67f9321a4530b60ff6cfe1ab0a84067204fc9e97aa5fdcb2b4121daa60/supplementary_0_Peer Review file__a3b85b67f9321a4530b60ff6cfe1ab0a84067204fc9e97aa5fdcb2b4121daa60_det.mmd

@@ -0,0 +1,536 @@
+<|ref|>title<|/ref|><|det|>[[73, 50, 296, 78]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[74, 96, 297, 118]]<|/det|>
+Peer Review File  
+
+<|ref|>title<|/ref|><|det|>[[73, 161, 917, 211]]<|/det|>
+# Accurate Cross-Species 5mC Detection for Oxford Nanopore Sequencing in Plants with DeepPlant  
+
+<|ref|>text<|/ref|><|det|>[[73, 224, 430, 240]]<|/det|>
+Corresponding Author: Dr Chuan- Le Xiao  
+
+<|ref|>text<|/ref|><|det|>[[72, 275, 864, 289]]<|/det|>
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+<|ref|>text<|/ref|><|det|>[[73, 327, 144, 340]]<|/det|>
+Version 0:  
+
+<|ref|>text<|/ref|><|det|>[[73, 354, 219, 367]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 380, 160, 393]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 405, 238, 418]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 418, 923, 536]]<|/det|>
+In the paper "DeepPlant: An Accurate Cross- Species 5mC Detection Tool for Oxford Nanopore Sequencing in Plants," He- Xu Chen et al. present a new approach for improving 5- methylcytosine (5mC) detection in plant genomes. The key contributions include generating and publishing new plant datasets using ONT R10.4.1 and WGBS, which are valuable for training models, as well as providing pre- trained models for detecting 5mC in CpG, CHG, and CHH contexts. While the model architecture is similar to existing tools like DeepSignal, the strength of this work lies in the inclusion of diverse datasets and the tool specifically designed for plant epigenomics and the ONT R10.4.1 pore type. The authors compare DeepPlant to the Dorado model, which also detects 5mC in CpG, CHG, and CHH motifs, showing that DeepPlant outperforms Dorado on most metrics, particularly in the CHH context, which is unique to plant genomes. Overall, the paper is well- written and clear, making it easy to follow and understand.  
+
+<|ref|>text<|/ref|><|det|>[[73, 549, 145, 562]]<|/det|>
+Concerns:  
+
+<|ref|>text<|/ref|><|det|>[[72, 574, 925, 666]]<|/det|>
+A major concern with the evaluation presented in this paper is the inclusion of training data in the evaluation tables. For example, the reported Pearson's correlation values between DeepPlant predictions and BS- seq data at \(30\times\) sequencing depth range from 0.705 to 0.881 across all evaluated datasets. However, the highest correlation value (0.881) is achieved for a species that was included in the training set (R. communis). This undermines the validity of the evaluation, as models often perform significantly better on data they were trained on compared to unseen datasets. The inclusion of training data in the evaluation skews the results, making it difficult to assess the model's true generalization capability across new species. A clear separation between training and evaluation datasets is essential for a reliable assessment of model performance.  
+
+<|ref|>text<|/ref|><|det|>[[72, 678, 920, 784]]<|/det|>
+A concern with the current model design is the lack of systematic ablation analysis to evaluate the contributions of its components. The model is composed of three main parts: the first encodes sequence features (including signal length, base qualities, mean, standard deviation, and the base at each position in a k- mer), the second encodes signal features (normalized signal), and the third combines and processes the outputs from the first two parts. While the architecture is well- designed and incorporates diverse feature representations, it's unclear how much each feature and encoder contributes to the final predictions. Performing a systematic ablation study—by removing specific features or encoders—would provide crucial insights into the importance of each component, highlight potential redundancies, and help refine the architecture for improved performance.  
+
+<|ref|>text<|/ref|><|det|>[[72, 795, 912, 875]]<|/det|>
+A significant concern is the omission of Glycine max (G. max) from the analysis without any explanation. The authors state: "We then collected tissue samples from six of these species (excluding S. lycopersicum due to its close relation to S. tuberosum) and conducted BS- seq. For better sample diversity, A. thaliana, O. sativa, Glycine max, Vitis vinifera, and Marchantia polymorpha were also added to the analysis." However, while S. miltiorrhiza, S. tuberosum, and R. communis were used for training, and five additional species (A. thaliana, V. vinifera, O. sativa, B. vulgaris, and C. sinensis) were included for evaluation, G. max is not mentioned in the final analysis, nor is its exclusion explained.  
+
+<|ref|>text<|/ref|><|det|>[[72, 887, 907, 940]]<|/det|>
+In contrast, the authors explicitly state that P. patens and M. polymorpha were excluded due to possible sample impurities, which is reasonable. However, Figure 1 includes ratios of high- methylation CHH sites for all excluded datasets, including these two species, indicating that they were used in the manuscript. Given their inclusion in the analysis, it would be appropriate for the authors to publish these datasets, regardless of their exclusion from training and evaluation, to support
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 46, 916, 74]]<|/det|>
+transparency and reproducibility. Additionally, since G. max was part of the original sample collection, its data should either be included in the analysis or its omission clearly justified.  
+
+<|ref|>text<|/ref|><|det|>[[72, 85, 920, 191]]<|/det|>
+For the read- level evaluation, the authors subsample 100,000 positive and 100,000 negative examples to create a balanced dataset, ensuring an equal number of positive and negative samples. For datasets where there are not enough positive examples to reach this number, they apply SMOTE (Synthetic Minority Over- sampling Technique) to artificially generate additional positive samples. While this approach helps create a balanced dataset for evaluation, it alters the natural distribution of the data, potentially making the evaluation less reflective of real- world conditions, where such balance rarely exists. This distribution shift could bias the results, as the model might perform differently on balanced datasets compared to naturally imbalanced data. To address this, it would be beneficial to include results on the full, unsampled datasets to provide a more realistic and unbiased evaluation of the model's true performance on real- world data.  
+
+<|ref|>text<|/ref|><|det|>[[73, 202, 911, 243]]<|/det|>
+The authors state, "To increase the reliability of the results, we repeated this process five times and calculated the average of the five evaluation results." However, they do not report the standard deviation or any other measure of variability to accompany the mean.  
+
+<|ref|>text<|/ref|><|det|>[[73, 254, 904, 296]]<|/det|>
+The authors do not specify whether the evaluations were performed on examples successfully called by both Dorado and DeepPlant (i.e., the intersection of valid predictions for both tools) or if the evaluations were conducted independently for each tool. Regardless, the number of evaluated examples should be explicitly reported for clarity and transparency.  
+
+<|ref|>text<|/ref|><|det|>[[72, 306, 911, 386]]<|/det|>
+The method for determining methylation status in a site- level evaluation is unclear. The authors state: "We determine the methylation threshold Pth based on the output methylation probability distribution, selecting the probability value within the range of 0.2 to 0.9 with the minor proportion in the probability distribution as Pth. If the methylation probability \(\mathrm{Pm} \geq \mathrm{Pth}\) , the sample is classified as methylated; otherwise, it is classified as non- methylated." However, it is not explained how the threshold is chosen in practical terms, particularly how the "minor proportion" is defined or calculated within the specified range. Greater clarity on the selection process is needed to fully understand this approach.  
+
+<|ref|>text<|/ref|><|det|>[[72, 397, 907, 465]]<|/det|>
+Only Dorado is used for comparison in the detection of 5mC in CpG motifs, despite the availability of other tools like DeepMod and Rockfish. While the authors exclude older tools such as Tombo, Megalodon, and DeepSignal- Plant due to their incompatibility with R10.4, this reasoning does not apply to DeepMod and Rockfish. The authors should explain why Dorado was chosen as the sole comparison tool or include comparisons with these additional methods to provide a more thorough evaluation of DeepPlant's performance in CpG detection.  
+
+<|ref|>text<|/ref|><|det|>[[72, 475, 920, 530]]<|/det|>
+The paper does not provide information about the running time or memory requirements of DeepPlant. While performance metrics like accuracy and F1 score are essential, practical factors such as computational efficiency and resource usage are equally important. Reporting details on running time and memory consumption would offer valuable insights into the model's scalability and suitability for large- scale or resource- constrained applications.  
+
+<|ref|>text<|/ref|><|det|>[[72, 541, 917, 648]]<|/det|>
+In Figure 3B, it is unclear whether the ratio of profiled sites corresponds to Nanopore coverage or DeepPlant coverage. The authors state that "Reads not aligned to the reference or had low alignment quality were filtered out." From this, it can be inferred that Nanopore coverage is likely higher than DeepPlant coverage. Specifying the coverage obtained for Dorado would also be helpful. Furthermore, the quality threshold for filtering low- quality alignments is not explicitly defined in the manuscript. However, in the code (GitHub commit: 67db8ef; Call_Modification.cpp, lines 107- 109), the authors define three constants for filtering, one of which (mapq_thresh_hold) appears to set the mapping quality threshold at 20. More detailed information about the other two heuristics used for filtering is necessary to fully understand the criteria applied during the analysis.  
+
+<|ref|>text<|/ref|><|det|>[[73, 686, 282, 700]]<|/det|>
+(Remarks on code availability)  
+
+<|ref|>text<|/ref|><|det|>[[72, 699, 905, 764]]<|/det|>
+The installation process described on GitHub is problematic and was unsuccessful. Despite having CUDA 11.8 installed and correctly configured (with 'nvcc - version', 'torch.version.cuda', and 'torch.cuda.is_available()' returning the expected output), an error occurs when running cmake. The first line of the error reads: "Could NOT find CUDA (missing: CUDA_INCLUDE_DIRS CUDA_CUDART_LIBRARY) (found version '11.8')." The installation was performed on a server with the 'cuda/11.8.0' and 'cudnn/8.9.2_cu1' module loaded.  
+
+<|ref|>text<|/ref|><|det|>[[72, 775, 911, 829]]<|/det|>
+Additionally, the installation instructions for CUDA are not user- friendly, as they assume the user has root privileges, which may not always be the case. It's also worth noting that running 'pip install torch==2.0.1' without the '- - index- url https://download.pytorch.org/whl/cu118' flag installs PyTorch for CUDA version 11.7. Providing a Dockerfile, a simpler execution option, or clearer setup instructions would improve accessibility and ensure a smoother installation process.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 840, 161, 853]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 866, 237, 879]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 880, 864, 920]]<|/det|>
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+<|ref|>text<|/ref|><|det|>[[73, 932, 282, 946]]<|/det|>
+(Remarks on code availability)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[73, 46, 901, 100]]<|/det|>
+The installation process on GitHub is problematic and fails despite CUDA 11.8 being correctly configured. Errors occur during cmake, and the instructions assume root privileges, which may not be available to all users. Additionally, installing PyTorch without specific flags defaults to CUDA 11.7, causing potential compatibility issues. Providing a Dockerfile, a simpler installation process, or clearer instructions would greatly improve accessibility and reliability.  
+
+<|ref|>sub_title<|/ref|><|det|>[[73, 112, 162, 125]]<|/det|>
+## Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[73, 138, 237, 164]]<|/det|>
+(Remarks to the Author) Major Revision:  
+
+<|ref|>text<|/ref|><|det|>[[73, 164, 916, 216]]<|/det|>
+The authors developed a deep learning model to process DNA methylome data generated from ONT R10.4 platform. Particularly, they used mCHH enriched genome in the training to improve the accuracy of mCHH data detection. In general, the new model provides valuable tool for the application of ONT R10.4 in plant epigenome sequencing. I have several concerns:  
+
+<|ref|>text<|/ref|><|det|>[[73, 216, 912, 257]]<|/det|>
+1) When the authors processed the BS-seq data, how did they treat with the multiple mapped reads? These reads could be generated from repeat enriched regions, such as centromere. If they discard all multiple mapped reads, it will lead to underestimation of mCHH caption in BS-seq data., such as in figure 3b and 3c.  
+
+<|ref|>text<|/ref|><|det|>[[73, 256, 918, 284]]<|/det|>
+2) In figure 3B and 3C, besides BS-seq and Deepplant/Nanopore, the Nanopole data analysis based on Dorodo should also be included to see if there is any improvement using Deepplant then using Dorodo.  
+
+<|ref|>text<|/ref|><|det|>[[73, 283, 920, 322]]<|/det|>
+3) All the analysis are statistical results. The authors should show several screenshots of BS-seq, Dorado, Deepplant data in regions where contain mCG, mCHG and mCHH, and the authors could show these three contexts respectively, to demonstrate that the capture of mCHH is more accurate in Deepplant processed dataset.  
+
+<|ref|>text<|/ref|><|det|>[[74, 334, 283, 348]]<|/det|>
+(Remarks on code availability)  
+
+<|ref|>text<|/ref|><|det|>[[73, 373, 144, 386]]<|/det|>
+Version 1:  
+
+<|ref|>text<|/ref|><|det|>[[73, 399, 218, 412]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 425, 159, 438]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 451, 238, 464]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 464, 911, 555]]<|/det|>
+We appreciate the authors for taking the time to incorporate our suggestions and improve the clarity and completeness of their manuscript. The clear separation of training and evaluation datasets, along with the additional explanations regarding dataset exclusions and public availability, significantly enhance the transparency of the study. The ablation study provides valuable insight into the contribution of individual components, and the inclusion of variability measures strengthens the reliability of the reported results. We also appreciate the clarification on the exclusion of DeepMod2 and the detailed information on DeepPlant's runtime and memory requirements. These thoughtful revisions contribute to a more comprehensive and well- supported analysis.  
+
+<|ref|>text<|/ref|><|det|>[[73, 568, 163, 580]]<|/det|>
+# Concern 1:  
+
+<|ref|>text<|/ref|><|det|>[[72, 592, 920, 688]]<|/det|>
+A remaining major concern is balanced sampling for read- level evaluation. For evaluation to be truly informative, results should be presented on real- world, naturally imbalanced datasets. Artificially balancing datasets during evaluation does not help users assess a method's practical applicability, as it distorts the class distribution and can inflate performance metrics. While we agree that some metrics, such as accuracy and AUC- ROC, are not suitable for evaluating highly imbalanced datasets, we cannot agree with the claim that metrics like AP, F1- score, precision, and recall "fail to accurately reflect true performance." These metrics are widely used in the machine learning and bioinformatics communities for precisely this purpose—evaluating models in extreme class imbalance scenarios.  
+
+<|ref|>text<|/ref|><|det|>[[72, 686, 920, 750]]<|/det|>
+Precision, in particular, can be "unforgiving" when there is a significant imbalance, but this does not make it unrepresentative or biased. Instead, low precision highlights a model's struggle to distinguish true positives from false positives. This is valuable information for users who care about the reliability of positive predictions. Moreover, metrics like average precision (AP) and precision- recall (PR) curves provide an even clearer picture, as they evaluate how well a model ranks positive examples rather than just how it labels them.  
+
+<|ref|>text<|/ref|><|det|>[[72, 750, 920, 790]]<|/det|>
+In summary, artificially balancing datasets for evaluation risks misleading users about real- world performance. Instead, presenting results on naturally imbalanced datasets—using well- established metrics like precision, recall, F1- score, and AP—ensures transparency and allows users to make informed decisions about method suitability.  
+
+<|ref|>text<|/ref|><|det|>[[73, 803, 161, 816]]<|/det|>
+# Concern 2:  
+
+<|ref|>text<|/ref|><|det|>[[72, 828, 900, 868]]<|/det|>
+We appreciate the authors for providing a detailed description of their method for determining methylation status in a site- level evaluation. While the procedural explanation is now clearer, the specific application of this method remains ambiguous. Is it applied to WGBS data, DeepPlant calls, or both?  
+
+<|ref|>text<|/ref|><|det|>[[72, 880, 918, 947]]<|/det|>
+Additionally, if this method is applied to DeepPlant, does the term methylation probability refer to the site- level prediction (i.e., the number of methylated Cs divided by the total number of Cs at that site), or does it correspond directly to the model's output? If the threshold is determined using WGBS data, is the same threshold applied to DeepPlant predictions? Furthermore, if this method is applied to DeepPlant but not Dorodo, it would be valuable to evaluate its impact on model performance by applying it to Dorodo or by comparing it with a simpler thresholding approach for DeepPlant predictions.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 60, 192, 73]]<|/det|>
+Minor comments:  
+
+<|ref|>text<|/ref|><|det|>[[70, 86, 900, 114]]<|/det|>
+We thank the authors for clearly separate training and evaluation datasets. However, some figures could still benefit from having clear separation, e.g. Supplementary Fig. 3  
+
+<|ref|>text<|/ref|><|det|>[[72, 139, 283, 165]]<|/det|>
+(Remarks on code availability) No additional remarks  
+
+<|ref|>text<|/ref|><|det|>[[72, 177, 162, 190]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[72, 204, 238, 216]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 217, 864, 257]]<|/det|>
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+<|ref|>text<|/ref|><|det|>[[72, 270, 283, 282]]<|/det|>
+(Remarks on code availability)  
+
+<|ref|>text<|/ref|><|det|>[[72, 283, 909, 321]]<|/det|>
+The instructions for running the tool using Docker are clearly written. However, the run command is missing the repository prefix (image name should be given with 'chenhx26/deepplant:1.1.0' instead of just 'deepplant:1.1.0'), which may cause an error when users attempt to run the image.  
+
+<|ref|>text<|/ref|><|det|>[[72, 334, 162, 346]]<|/det|>
+Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[72, 361, 238, 372]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 373, 899, 425]]<|/det|>
+After reviewing the manuscript, I am pleased to suggest that I have no revisions or objections to raise. The content is well- presented, and it contributes valuable insights to the field. Therefore, I recommend proceeding with its acceptance. Thank you for considering my feedback. Should you need any further information or discussion, please do not hesitate to contact me.  
+
+<|ref|>text<|/ref|><|det|>[[72, 438, 283, 464]]<|/det|>
+(Remarks on code availability) Accept  
+
+<|ref|>text<|/ref|><|det|>[[72, 476, 144, 489]]<|/det|>
+Version 2:  
+
+<|ref|>text<|/ref|><|det|>[[72, 503, 219, 515]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[72, 529, 160, 541]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[72, 555, 238, 567]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 568, 921, 620]]<|/det|>
+We acknowledge the authors' efforts in incorporating the suggestions and improving the clarity of the manuscript. The addition of evaluations on datasets that were not sampled is particularly valuable. Furthermore, their explanation of the method for determining methylation status in site- level evaluations is satisfactory. We have no further concerns regarding the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[72, 633, 283, 645]]<|/det|>
+(Remarks on code availability)  
+
+<|ref|>text<|/ref|><|det|>[[72, 646, 328, 658]]<|/det|>
+We included this in previous reviews.  
+
+<|ref|>text<|/ref|><|det|>[[72, 671, 162, 684]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[72, 698, 238, 710]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 711, 864, 750]]<|/det|>
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+<|ref|>text<|/ref|><|det|>[[72, 763, 283, 776]]<|/det|>
+(Remarks on code availability)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 45, 916, 99]]<|/det|>
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+<|ref|>text<|/ref|><|det|>[[72, 99, 796, 113]]<|/det|>
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+<|ref|>text<|/ref|><|det|>[[72, 112, 910, 165]]<|/det|>
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+<|ref|>text<|/ref|><|det|>[[72, 165, 618, 179]]<|/det|>
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[150, 104, 445, 119]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[148, 140, 851, 362]]<|/det|>
+In the paper "DeepPlant: An Accurate Cross- Species 5mC Detection Tool for Oxford Nanopore Sequencing in Plants," He- Xu Chen et al. present a new approach for improving 5- methylcytosine (5mC) detection in plant genomes. The key contributions include generating and publishing new plant datasets using ONT R10.4.1 and WGBS, which are valuable for training models, as well as providing pre- trained models for detecting 5mC in CpG, CHG, and CHH contexts. While the model architecture is similar to existing tools like DeepSignal, the strength of this work lies in the inclusion of diverse datasets and the tool specifically designed for plant epigenomics and the ONT R10.4.1 pore type. The authors compare DeepPlant to the Dorado model, which also detects 5mC in CpG, CHG, and CHH motifs, showing that DeepPlant outperforms Dorado on most metrics, particularly in the CHH context, which is unique to plant genomes. Overall, the paper is well- written and clear, making it easy to follow and understand.  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 383, 230, 398]]<|/det|>
+## Concerns:  
+
+<|ref|>text<|/ref|><|det|>[[148, 418, 851, 602]]<|/det|>
+A major concern with the evaluation presented in this paper is the inclusion of training data in the evaluation tables. For example, the reported Pearson's correlation values between DeepPlant predictions and BS- seq data at \(30\times\) sequencing depth range from 0.705 to 0.881 across all evaluated datasets. However, the highest correlation value (0.881) is achieved for a species that was included in the training set (R. communis). This undermines the validity of the evaluation, as models often perform significantly better on data they were trained on compared to unseen datasets. The inclusion of training data in the evaluation skews the results, making it difficult to assess the model's true generalization capability across new species. A clear separation between training and evaluation datasets is essential for a reliable assessment of model performance.  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 623, 234, 638]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[148, 641, 851, 899]]<|/det|>
+Thank you for the valuable suggestions. We have now clearly separated the training and evaluation datasets in this revision. For fairness, when describing the model evaluation results, we have now excluded the three datasets used for training in the abstract and the main manuscript. To better illustrate the model's generalization capability, we sequenced new BS- seq and nanopore dataset for S. lycopersicum and further compared DeepPlant and the previous state- of- art software Dorado on it. Still, the results support that DeepPlant has advantage on the CHH motif. To be noticed, the correlation coefficients are affected by many factors such as mapping rate, reference genome divergence level, and methylation frequency distribution, making direct comparison of correlation coefficients across datasets less meaningful. After all, the evaluation results showed that the DeepPlant showed even greater advantage compared to Dorado on several test- only datasets than on the training datasets, suggesting better generalization capability across tested datasets. We hope these revisions (in lines 42- 43, 108, and et al.) have made the evaluation conclusion more robust.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 103, 848, 306]]<|/det|>
+A concern with the current model design is the lack of systematic ablation analysis to evaluate the contributions of its components. The model is composed of three main parts: the first encodes sequence features (including signal length, base qualities, mean, standard deviation, and the base at each position in a k- mer), the second encodes signal features (normalized signal), and the third combines and processes the outputs from the first two parts. While the architecture is well- designed and incorporates diverse feature representations, it's unclear how much each feature and encoder contributes to the final predictions. Performing a systematic ablation study—by removing specific features or encoders—would provide crucial insights into the importance of each component, highlight potential redundancies, and help refine the architecture for improved performance.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 325, 234, 340]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[147, 343, 852, 545]]<|/det|>
+Thank you for the suggestions. Understanding the importance of each component indeed will deepen the insights of this study. Accordingly, we have conducted systematic ablation experiments on DeepPlant to evaluate the contribution of its triple- encoder. By removing each of the encoders and training the model using the same samples and conditions, we quantitatively assessed the resulted models on the O. sativa and A. thaliana datasets. The results, presented in Supplementary Table 3, demonstrated that all three encoders have played a critical role in the model's performance. Among them, the sequence encoder has the greatest impact on the results, while the combine encoder has the least. For instance, the removal of any encoder resulted in at least a \(15\%\) decrease in performance on the A. thaliana dataset, indicating that each encoder is indispensable. We have the revised the manuscript in lines 171- 173.  
+
+<|ref|>text<|/ref|><|det|>[[147, 564, 844, 712]]<|/det|>
+A significant concern is the omission of Glycine max (G. max) from the analysis without any explanation. The authors state: "We then collected tissue samples from six of these species (excluding S. lycopersicum due to its close relation to S. tuberosum) and conducted BS- seq. For better sample diversity, A. thaliana, O. sativa, Glycine max, Vitis vinifera, and Marchantia polymorpha were also added to the analysis." However, while S. miltiorrhiza, S. tuberosum, and R. communis were used for training, and five additional species (A. thaliana, V. vinifera, O. sativa, B. vulgaris, and C. sinensis) were included for evaluation, G. max is not mentioned in the final analysis, nor is its exclusion explained.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 732, 234, 747]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[148, 750, 850, 878]]<|/det|>
+Thank you pointing it out. We initially sequenced BS- seq and Pacbio Hifi (for another project) for the G.max var. Huachun No.6, which showed high heterozygosity and abundant nucleotide variances to the reference genome (GCF_000004515.6). The high difference level compared to the reference will affect the quality of BS- seq that uses base conversion for detection of methylation state, therefore we have not sequenced nanopore data for it. We have now explained it in the manuscript (in lines 148- 149) and uploaded the BS- seq data to public database as requested (Data availability statement updated).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 102, 850, 250]]<|/det|>
+In contrast, the authors explicitly state that P. patens and M. polymorpha were excluded due to possible sample impurities, which is reasonable. However, Figure 1 includes ratios of high- methylation CHH sites for all excluded datasets, including these two species, indicating that they were used in the manuscript. Given their inclusion in the analysis, it would be appropriate for the authors to publish these datasets, regardless of their exclusion from training and evaluation, to support transparency and reproducibility. Additionally, since G. max was part of the original sample collection, its data should either be included in the analysis or its omission clearly justified.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 272, 234, 286]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[148, 290, 846, 343]]<|/det|>
+Thank you for the comments. We have uploaded our BS- seq data for P. patens and M. polymorpha, which are available at http://gsa.big.ac.cn under Project Accession No. PRJCA030666. As mentioned above, we have also uploaded the BS- seq data of G. max.  
+
+<|ref|>text<|/ref|><|det|>[[147, 378, 850, 581]]<|/det|>
+For the read- level evaluation, the authors subsample 100,000 positive and 100,000 negative examples to create a balanced dataset, ensuring an equal number of positive and negative samples. For datasets where there are not enough positive examples to reach this number, they apply SMOTE (Synthetic Minority Over- sampling Technique) to artificially generate additional positive samples. While this approach helps create a balanced dataset for evaluation, it alters the natural distribution of the data, potentially making the evaluation less reflective of real- world conditions, where such balance rarely exists. This distribution shift could bias the results, as the model might perform differently on balanced datasets compared to naturally imbalanced data. To address this, it would be beneficial to include results on the full, unsampled datasets to provide a more realistic and unbiased evaluation of the model's true performance on real- world data.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 601, 234, 615]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[147, 618, 850, 913]]<|/det|>
+Thank you for the comments. Following your suggestions, we evaluated the single- molecule performance of DeepPlant and Dorado on three full unbalanced dataset: A. thaliana (positive sample ratio \(0.0506\%\) ), B. vulgaris \((0.1249\%)\) , and O. sativa \((0.033\%)\) . The results are shown in the table below. Due to the extreme scarcity of positive samples, the evaluation metrics AP, F1, Precision, and Accuracy fail to accurately reflect the true performance of the models. This is because even if only a very small proportion of negative samples are misclassified as positive (false positives), their number would significantly exceed the number of true positive samples, leading to a decline in both F1 and Precision metrics. Similarly, the AP value, which is calculated based on the precision- recall curve, is also reduced as a result. Since the vast majority of samples are negative, Accuracy shows a substantial increase. In fact, under such circumstances, Accuracy largely reflects the model's ability to classify negative samples correctly. When the model correctly classifies all negative samples, its Accuracy can exceed \(99.8\%\) , rendering this metric unrepresentative of the model's overall performance. For the reasons, using extreme unbalanced datasets for single- molecule evaluations reduces its also severely biased. Therefore, we prefer to use balanced datasets to reflect the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 86, 840, 138]]<|/det|>
+single-molecule performance of the models. Additionally, to reflect the real-world sample distribution, we used the full datasets for site-level (methylation frequency) evaluation, which provides some reference value for practical scenarios.  
+
+<|ref|>table<|/ref|><|det|>[[55, 157, 941, 293]]<|/det|>
+
+<table><tr><td></td><td>tools</td><td>AUC</td><td>AP</td><td>F1-score</td><td>Precision</td><td>Accuracy</td><td>Recall</td><td>Specificity</td></tr><tr><td rowspan="2">A. thaliana</td><td>DeepPlant</td><td>0.8668</td><td>0.0073</td><td>0.001373</td><td>0.000687</td><td>0.97023</td><td>0.669626</td><td>0.970239</td></tr><tr><td>Dorado</td><td>0.8492</td><td>0.0024</td><td>0.001231</td><td>0.000616</td><td>0.969593</td><td>0.625466</td><td>0.969604</td></tr><tr><td rowspan="2">B. vulgaris</td><td>DeepPlant</td><td>0.933079</td><td>0.056785</td><td>0.007849</td><td>0.003943</td><td>0.931204</td><td>0.835183</td><td>0.931235</td></tr><tr><td>Dorado</td><td>0.880859</td><td>0.010706</td><td>0.004315</td><td>0.002164</td><td>0.901871</td><td>0.733399</td><td>0.90192</td></tr><tr><td rowspan="2">O. sativa</td><td>DeepPlant</td><td>0.9191</td><td>0.1076</td><td>0.026955</td><td>0.013706</td><td>0.947117</td><td>0.808072</td><td>0.947243</td></tr><tr><td>Dorado</td><td>0.907942</td><td>0.01916</td><td>0.008923</td><td>0.004486</td><td>0.877289</td><td>0.813208</td><td>0.877333</td></tr></table>  
+
+<|ref|>text<|/ref|><|det|>[[148, 361, 844, 415]]<|/det|>
+The authors state, "To increase the reliability of the results, we repeated this process five times and calculated the average of the five evaluation results." However, they do not report the standard deviation or any other measure of variability to accompany the mean.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 434, 235, 449]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[148, 453, 826, 506]]<|/det|>
+Thank you for pointing it out. We noticed the lack of rigor in this regard; therefore, we calculated the standard deviation of all single- molecule evaluation results, as shown in revised Supplementary Tables 6 and 7.  
+
+<|ref|>text<|/ref|><|det|>[[148, 543, 853, 635]]<|/det|>
+The authors do not specify whether the evaluations were performed on examples successfully called by both Dorado and DeepPlant (i.e., the intersection of valid predictions for both tools) or if the evaluations were conducted independently for each tool. Regardless, the number of evaluated examples should be explicitly reported for clarity and transparency.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 654, 235, 668]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[148, 672, 850, 762]]<|/det|>
+Thank you for pointing it out. We conducted all evaluations for each tool independently. For single- molecule evaluations, we extracted 200,000 samples under the same criteria and repeated the process five times, which has now been described in lines 498- 505. For quantitative evaluations, we additionally provided the number of sites detected for three types of motifs by each method, as shown in Supplementary Table 2.  
+
+<|ref|>text<|/ref|><|det|>[[148, 799, 840, 908]]<|/det|>
+The method for determining methylation status in a site- level evaluation is unclear. The authors state: "We determine the methylation threshold Pth based on the output methylation probability distribution, selecting the probability value within the range of 0.2 to 0.9 with the minor proportion in the probability distribution as Pth. If the methylation probability \(\mathrm{Pm} \geq \mathrm{Pth}\) , the sample is classified as methylated; otherwise, it is classified as non- methylated." However, it is not explained how the threshold is chosen in practical
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[148, 85, 844, 139]]<|/det|>
+terms, particularly how the "minor proportion" is defined or calculated within the specified range. Greater clarity on the selection process is needed to fully understand this approach.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 159, 234, 173]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[148, 177, 844, 220]]<|/det|>
+Thank you. We have described the detailed procedure in the Methods section now. The method applied is as follows in lines 510- 518:  
+
+<|ref|>text<|/ref|><|det|>[[148, 213, 850, 323]]<|/det|>
+'Divide the range from 0.2 to 0.9 into 70 intervals with a step size of 0.01, using the left endpoint of each interval as the representative value for that interval. For the detection results of a single dataset, group the samples into the corresponding intervals based on their methylation probabilities, count the number of samples in each interval, and calculate their proportion. The value corresponding to the interval with the smallest proportion is selected as the methylation threshold.'  
+
+<|ref|>text<|/ref|><|det|>[[148, 359, 848, 488]]<|/det|>
+Only Dorado is used for comparison in the detection of 5mC in CpG motifs, despite the availability of other tools like DeepMod and Rockfish. While the authors exclude older tools such as Tombo, Megalodon, and DeepSignal- Plant due to their incompatibility with R10.4, this reasoning does not apply to DeepMod and Rockfish. The authors should explain why Dorado was chosen as the sole comparison tool or include comparisons with these additional methods to provide a more thorough evaluation of DeepPlant's performance in CpG detection.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 508, 234, 522]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[148, 525, 851, 653]]<|/det|>
+Thank you for the suggestions. Accordingly, we further evaluated the recently published in this revision. For DeepMode2, because its original study (Ahsan et al.,) has carried out multiple assessments showing that it performed not as well as Dorado on R10.4 nanopore data, we have not included it in comparison in this revision. The results have been provided in Supplementary Table 6 and still identify DeepPlant as one of the state- of- art tools on CpG detection. The new benchmark results have also been updated in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[148, 671, 845, 725]]<|/det|>
+Ahsan, M.U., Gouru, A., Chan, J. et al. A signal processing and deep learning framework for methylation detection using Oxford Nanopore sequencing. Nat Commun 15, 1448 (2024). https://doi.org/10.1038/s41467- 024- 45778- y  
+
+<|ref|>text<|/ref|><|det|>[[148, 761, 839, 871]]<|/det|>
+The paper does not provide information about the running time or memory requirements of DeepPlant. While performance metrics like accuracy and F1 score are essential, practical factors such as computational efficiency and resource usage are equally important. Reporting details on running time and memory consumption would offer valuable insights into the model's scalability and suitability for large- scale or resource- constrained applications.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 891, 234, 905]]<|/det|>
+## Response:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[148, 85, 843, 176]]<|/det|>
+Benchmarking results are indeed crucial for evaluating the usability of the model. We performed benchmarking of DeepPlant on the entire sequencing dataset of A. thaliana. The benchmarking results, including runtime and memory consumption, are provided in Supplementary Table 9. We have also added a sentence about the benchmark results in lines 339- 341 in the discussion section.  
+
+<|ref|>text<|/ref|><|det|>[[147, 213, 850, 397]]<|/det|>
+In Figure 3B, it is unclear whether the ratio of profiled sites corresponds to Nanopore coverage or DeepPlant coverage. The authors state that "Reads not aligned to the reference or had low alignment quality were filtered out." From this, it can be inferred that Nanopore coverage is likely higher than DeepPlant coverage. Specifying the coverage obtained for Dorado would also be helpful. Furthermore, the quality threshold for filtering low- quality alignments is not explicitly defined in the manuscript. However, in the code (GitHub commit: 67db8ef; Call_Modification.cpp, lines 107- 109), the authors define three constants for filtering, one of which (mapq_thresh_hold) appears to set the mapping quality threshold at 20. More detailed information about the other two heuristics used for filtering is necessary to fully understand the criteria applied during the analysis.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 435, 234, 450]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[147, 453, 850, 580]]<|/det|>
+We recalculated the coverage of DeepPlant and Dorado across different genomic regions and revised Fig. 3b to more clearly illustrate the coverage differences between the three methods. The filtering criteria for low- quality alignments are as follows: (1) MAPQ<20; (2) primary alignment length \(< 80\%\) of read length; (3) mapping identity \(< 80\%\) . Reads were filtered if less than \(80\%\) of the reads were aligned to the. The details have now been described in both the Figure legend in lines 737- 739 and in the Methods section in lines 434- 436.  
+
+<|ref|>text<|/ref|><|det|>[[148, 637, 495, 653]]<|/det|>
+Reviewer #1 (Remarks on code availability):  
+
+<|ref|>text<|/ref|><|det|>[[147, 673, 850, 802]]<|/det|>
+The installation process described on GitHub is problematic and was unsuccessful. Despite having CUDA 11.8 installed and correctly configured (with 'nvcc - version', 'torch.version.cuda', and 'torch.cuda.is_available()' returning the expected output), an error occurs when running cmake. The first line of the error reads: "Could NOT find CUDA (missing: CUDA_INCLUDE_DIRS CUDA_CUDART_LIBRARY) (found version '11.8')." The installation was performed on a server with the 'cuda/11.8.0' and 'cudnn/8.9.2_cu1' module loaded.  
+
+<|ref|>text<|/ref|><|det|>[[147, 822, 848, 895]]<|/det|>
+Additionally, the installation instructions for CUDA are not user- friendly, as they assume the user has root privileges, which may not always be the case. It's also worth noting that running 'pip install torch==2.0.1' without the '- index- url https://download.pytorch.org/whl/cu118&#x2019; flag installs PyTorch for
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[149, 85, 832, 120]]<|/det|>
+CUDA version 11.7. Providing a Dockerfile, a simpler execution option, or clearer setup instructions would improve accessibility and ensure a smoother installation process.  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 142, 235, 156]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[148, 160, 847, 270]]<|/det|>
+Thank you for pointing it out. In most cases, the linking location of CUDA in a computer is fixed. However, in some cases, the CUDA path configuration might be inconsistent, leading to compilation errors due to missing libraries. To address potential compilation issues, we provide a Dockerfile along with detailed usage instructions to simplify the use of the tool. The usage instructions are available at https://github.com/xiaochuanle/DeepPlant.  
+
+<|ref|>text<|/ref|><|det|>[[149, 325, 446, 340]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[149, 361, 850, 434]]<|/det|>
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+<|ref|>text<|/ref|><|det|>[[149, 454, 494, 470]]<|/det|>
+Reviewer #2 (Remarks on code availability):  
+
+<|ref|>text<|/ref|><|det|>[[148, 491, 850, 600]]<|/det|>
+The installation process on GitHub is problematic and fails despite CUDA 11.8 being correctly configured. Errors occur during cmake, and the instructions assume root privileges, which may not be available to all users. Additionally, installing PyTorch without specific flags defaults to CUDA 11.7, causing potential compatibility issues. Providing a Dockerfile, a simpler installation process, or clearer instructions would greatly improve accessibility and reliability.  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 620, 235, 634]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[148, 637, 833, 747]]<|/det|>
+Thank you for your suggestions. In most cases, the linking location of CUDA in a computer is fixed. However, in some cases, the CUDA path configuration might be inconsistent, leading to compilation errors due to missing libraries. To address potential compilation issues, we provide a Dockerfile along with detailed usage instructions to simplify the use of the tool. The usage instructions are available at https://github.com/xiaochuanle/DeepPlant.  
+
+<|ref|>text<|/ref|><|det|>[[149, 803, 446, 818]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[149, 840, 282, 855]]<|/det|>
+Major Revision:  
+
+<|ref|>text<|/ref|><|det|>[[149, 858, 848, 911]]<|/det|>
+The authors developed a deep learning model to process DNA methylome data generated from ONT R10.4 platform. Particularly, they used mCHH enriched genome in the training to improve the accuracy of mCHH data detection. In general, the new model provides
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[148, 85, 816, 120]]<|/det|>
+valuable tool for the application of ONT R10.4 in plant epigenome sequencing. I have several concerns:  
+
+<|ref|>text<|/ref|><|det|>[[148, 123, 835, 195]]<|/det|>
+1) When the authors processed the BS-seq data, how did they treat with the multiple mapped reads? These reads could be generated from repeat enriched regions, such as centromere. if they discard all multiple mapped reads, it will lead to underestimation of mCHH caption in BS-seq data., such as in figure 3b and 3c.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 215, 234, 230]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[148, 233, 837, 378]]<|/det|>
+Thanks for bringing it out. We did not discard all multiple mapped reads. We applied the widely applied Bismark pipeline in BS-seq analysis which only uses unique best alignments in methylation frequency quantification. Unique best alignments refer to alignments where the next best alignment does not exist or is not as good (MAPQ>0), indicating that only multiple mapped reads with multiple best alignments were dumped. For the nanopore data, we also used unique best alignments, except that we required a much higher MAPQ threshold (20) for high alignment accuracy. For causing less confusion, related information has been added in lines 404- 405 and 434- 436.  
+
+<|ref|>text<|/ref|><|det|>[[148, 398, 816, 453]]<|/det|>
+2) In figure 3B and 3C, besides BS-seq and Deepplant/Nanopore, the Nanopole data analysis based on Dorodo should also be included to see if there is any improvement using Deepplant then using Dorado.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 474, 234, 488]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[148, 492, 850, 545]]<|/det|>
+Thank you for the suggestion. We have added Dorado's methylation analysis results for the relevant regions, as shown in Figure 3B, Figure 3C, and Supplementary Figures 2 and 5. The manuscript has been revised at related locations.  
+
+<|ref|>text<|/ref|><|det|>[[148, 564, 850, 636]]<|/det|>
+3) All the analysis are statistical results. The authors should show several screenshots of BS-seq, Dorado, Deepplant data in regions where contain mCG, mCHG and mCHH, and the authors could show these three contexts respectively, to demonstrate that the capture of mCHH is more accurate in Deepplant processed dataset.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 656, 234, 670]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[148, 673, 853, 782]]<|/det|>
+Thank you for the suggestions. We have added methylation analysis results of three methods for other regions, as shown in the Supplementary Figure 2. From these regions, it can be observed that DeepPlant's mCHH analysis results are closer to those of BS-seq, while Dorado's mCHH analysis results in some regions contain a large number of false positives, which should explain the higher correlation coefficient of DeepPlant with BS-seq. The manuscript has been revised accordingly in lines 207- 208.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[149, 85, 459, 101]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[148, 122, 850, 308]]<|/det|>
+We appreciate the authors for taking the time to incorporate our suggestions and improve the clarity and completeness of their manuscript. The clear separation of training and evaluation datasets, along with the additional explanations regarding dataset exclusions and public availability, significantly enhance the transparency of the study. The ablation study provides valuable insight into the contribution of individual components, and the inclusion of variability measures strengthens the reliability of the reported results. We also appreciate the clarification on the exclusion of DeepMod2 and the detailed information on DeepPlant's runtime and memory requirements. These thoughtful revisions contribute to a more comprehensive and well- supported analysis.  
+
+<|ref|>text<|/ref|><|det|>[[148, 327, 253, 343]]<|/det|>
+# Concern 1:  
+
+<|ref|>text<|/ref|><|det|>[[148, 363, 849, 454]]<|/det|>
+A remaining major concern is balanced sampling for read- level evaluation. For evaluation to be truly informative, results should be presented on real- world, naturally imbalanced datasets. Artificially balancing datasets during evaluation does not help users assess a method's practical applicability, as it distorts the class distribution and can inflate performance metrics.  
+
+<|ref|>text<|/ref|><|det|>[[148, 455, 849, 567]]<|/det|>
+While we agree that some metrics, such as accuracy and AUC- ROC, are not suitable for evaluating highly imbalanced datasets, we cannot agree with the claim that metrics like AP, F1- score, precision, and recall "fail to accurately reflect true performance." These metrics are widely used in the machine learning and bioinformatics communities for precisely this purpose—evaluating models in extreme class imbalance scenarios.  
+
+<|ref|>text<|/ref|><|det|>[[148, 568, 849, 696]]<|/det|>
+Precision, in particular, can be "unforgiving" when there is a significant imbalance, but this does not make it unrepresentative or biased. Instead, low precision highlights a model's struggle to distinguish true positives from false positives. This is valuable information for users who care about the reliability of positive predictions. Moreover, metrics like average precision (AP) and precision- recall (PR) curves provide an even clearer picture, as they evaluate how well a model ranks positive examples rather than just how it labels them.  
+
+<|ref|>text<|/ref|><|det|>[[148, 697, 849, 789]]<|/det|>
+In summary, artificially balancing datasets for evaluation risks misleading users about real- world performance. Instead, presenting results on naturally imbalanced datasets—using well- established metrics like precision, recall, F1- score, and AP—ensures transparency and allows users to make informed decisions about method suitability.  
+
+<|ref|>text<|/ref|><|det|>[[148, 790, 849, 900]]<|/det|>
+Thank you very much for your suggestions. In this revision, we have included single- molecule test results on the imbalanced fully methylated/unmethylated CHH sites across the nine datasets. We agree that evaluating single- molecule performance on realistic, imbalanced datasets can provide valuable insights for researchers. However, the selection of high- quality reference control sites, such as fully methylated (100% methylation frequency) and fully unmethylated (0% methylation
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[149, 85, 580, 102]]<|/det|>
+frequency) sites in BS- seq, has been a limiting factor.  
+
+<|ref|>text<|/ref|><|det|>[[148, 104, 850, 306]]<|/det|>
+Given the scarcity of fully methylated CHH sites in the analyzed species, the ratio of fully methylated to unmethylated CHH sites (Supplementary Table 7) is much lower than the realistic methylation ratios observed in single- molecule data from the same species (Supplementary Table 2). In such cases, single- molecule performance is more effectively indirectly evaluated through methylation frequency quantification. For example, in A. thaliana, although the overall CHH methylation content is \(3\%\) , the ratio of \(100\%\) methylated to \(0\%\) methylated sites is 1:32,109. Consequently, direct evaluations of DeepPlant and Dorado on the imbalanced fully methylated/unmethylated datasets primarily provided insights into the accuracy of unmethylated site detection, with DeepPlant outperforming Dorado in all cases (Supplementary Table 7).  
+
+<|ref|>text<|/ref|><|det|>[[149, 308, 848, 343]]<|/det|>
+Thanks to your valuable suggestions, we have more clearly described the evaluation results and considerations in lines 237- 248.  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 363, 249, 379]]<|/det|>
+## #Concern 2:  
+
+<|ref|>text<|/ref|><|det|>[[149, 400, 849, 473]]<|/det|>
+We appreciate the authors for providing a detailed description of their method for determining methylation status in a site- level evaluation. While the procedural explanation is now clearer, the specific application of this method remains ambiguous. Is it applied to WGBS data, DeepPlant calls, or both?  
+
+<|ref|>text<|/ref|><|det|>[[148, 493, 849, 585]]<|/det|>
+Additionally, if this method is applied to DeepPlant, does the term methylation probability refer to the site- level prediction (i.e., the number of methylated Cs divided by the total number of Cs at that site), or does it correspond directly to the model's output? If the threshold is determined using WGBS data, is the same threshold applied to DeepPlant predictions?  
+
+<|ref|>text<|/ref|><|det|>[[149, 586, 849, 641]]<|/det|>
+Furthermore, if this method is applied to DeepPlant but not Dorado, it would be valuable to evaluate its impact on model performance by applying it to Dorado or by comparing it with a simpler thresholding approach for DeepPlant predictions.  
+
+<|ref|>text<|/ref|><|det|>[[148, 642, 850, 899]]<|/det|>
+Thank you for your valuable suggestions. Methylation probability refers to a numerical value between 0 and 1, representing the likelihood that a specific CpG motif in a read is methylated, and is therefore different from site- level methylation frequency which is aggregated from the methylation status of multiple reads. As a result, the threshold selection method is only applicable to DeepPlant and Dorado. Following your advice, we assessed the impact of this threshold selection approach on Dorado, and the results showed that the method could also improve the accuracy of Dorado's site- level methylation frequency quantifications, though it only reduced the advantage of DeepPlant by a small proportion. Relevant results are summarized in Table 1. Additionally, we provided a more detailed explanation of this method, which essentially seeks a balance between false positives and false negatives in the model's predictions, thereby reducing the total number of false positives and false negatives as detailed in Supplementary Fig 2. The manuscript has been revised accordingly in lines 203- 207.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[148, 105, 293, 120]]<|/det|>
+Minor comments:  
+
+<|ref|>text<|/ref|><|det|>[[147, 141, 850, 232]]<|/det|>
+We thank the authors for clearly separate training and evaluation datasets. However, some figures could still benefit from having clear separation, e.g. Supplementary Fig. 3Thank you for pointing out the detail issues in some of the figures. We have separated the datasets in the relevant figures.  
+
+<|ref|>text<|/ref|><|det|>[[148, 252, 512, 269]]<|/det|>
+Reviewer #1 (Remarks on code availability):  
+
+<|ref|>text<|/ref|><|det|>[[148, 290, 330, 306]]<|/det|>
+No additional remarks  
+
+<|ref|>text<|/ref|><|det|>[[148, 344, 459, 361]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[148, 382, 850, 454]]<|/det|>
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+<|ref|>text<|/ref|><|det|>[[148, 475, 512, 491]]<|/det|>
+Reviewer #2 (Remarks on code availability):  
+
+<|ref|>text<|/ref|><|det|>[[148, 512, 850, 585]]<|/det|>
+The instructions for running the tool using Docker are clearly written. However, the run command is missing the repository prefix (image name should be given with 'chenhx26/deepplant:1.1.0' instead of just 'deepplant:1.1.0'), which may cause an error when users attempt to run the image.  
+
+<|ref|>text<|/ref|><|det|>[[148, 587, 848, 621]]<|/det|>
+Thank you for pointing out the issues in the command. We have made corrections to the relevant commands.  
+
+<|ref|>text<|/ref|><|det|>[[148, 660, 459, 677]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[148, 698, 850, 770]]<|/det|>
+After reviewing the manuscript, I am pleased to suggest that that I have no revisions or objections to raise. The content is well- presented, and it contributes valuable insights to the field. Therefore, I recommend proceeding with its acceptance.  
+
+<|ref|>text<|/ref|><|det|>[[148, 752, 850, 787]]<|/det|>
+Thank you for considering my feedback. Should you need any further information or discussion, please do not hesitate to contact me.  
+
+<|ref|>text<|/ref|><|det|>[[148, 808, 512, 824]]<|/det|>
+Reviewer #3 (Remarks on code availability):  
+
+<|ref|>text<|/ref|><|det|>[[148, 846, 207, 861]]<|/det|>
+Accept
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__a58c0301e8ee191db7719ea6d64710681f6fbb8cf59d314c19fdca9c4b2c2302/images_list.json

@@ -0,0 +1,153 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Fig. 2 The generalization ability of VITAP based on VMR-MSL38 database. The VMR-MSL38 database was divided into a training set comprising 70% of the data and a test set comprising 30%. Sequences in the test set were sliced into genome fragments of varying lengths (1-kb, 5-kb, 10-kb, 20-",
+    "footnote": [],
+    "bbox": [
+      [
+        211,
+        384,
+        800,
+        844
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 Benchmarking performance metrics of VITAP compared to vConTACT2, CAT, VPF-Class, PhaGCN2, and geNomad. a) Benchmarking performance metrics based on VMR-MSL37, including vConTACT2, CAT, PhaGCN2, geNomad and VITAP. Each boxplot represents the results of five fragment subsets with different lengths (1-kb, 5-kb, 10-kb, 20-kb, and 30-kb), which were generated from viral reference genomes. In total, 3,705 viral reference genomes were sliced to generate 191,596, 40,425, 21,857, 13,622, and 8,844 fragments with 1-kb, 5-kb, 10-kb, 20-kb, and 30-kb length, respectively. The line and fork inside each box represent the median and mean, respectively. The standard deviation of each pipeline is labelled above the box. b)",
+    "footnote": [],
+    "bbox": [
+      [
+        200,
+        278,
+        810,
+        732
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4 The benchmarking of VITAP compared to other pipelines. a) On the aspect of family-level taxonomic assignments on different viral phyla, the modified-F1-score-based benchmarking performance of VITAP compared to vConTACT2, CAT, PhaGCN2, and geNomad. Based on the VMR-MSL37 database or the viral RefSeq209 database, viral RefSeqs were added after January 2022 as \"new\" viral genomes to perform taxonomic assignments. Sequences with varying lengths (1-kb, 5-kb, 10-kb, 20-kb, and 30-kb) from 16 viral phyla were subjected to taxonomic assignment using six different pipelines, which were used in classification effectiveness comparison. Each bar represents the modified F1 score of the family level taxonomic assignment result from different pipelines. The",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        215,
+        850,
+        521
+      ]
+    ],
+    "page_idx": 18
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Fig. S2 The number of taxonomic assignments on deep-sea viromes by vConTACT2, CAT, VPF-Class, PhaGCN2, geNomad, VITAP. The taxonomic databases used by these pipelines include VMR-MS L37/RefSeq209/IMGVR4-hybrid database (VITAP), VMR-MSL37 (vConTACT2), RefSeq209 (CAT), and pipeline-integrated databases (VPF-Class, PhaGCN2, and geNomad). These bars describe the numbers of family-level, genus-level, and all annotated results.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_7.jpg",
+    "caption": "Fig. 7 (re-performed used by hybrid database) The viral taxonomy of deep-sea viromes. a) The realm-level taxonomy of four deep-sea virome datasets. The pie chart and bar chart indicate the number of viral operated units (vOTUs) contained in each viral phylum, which are represented by distinct viral symbols; b) The family level taxonomy of four deep-sea virome datasets. The bubble chart displays the top ten assigned viral families in each dataset. Viral family names and their respective phylum symbols are annotated on the left side of the heatmap; c) The bar graph represents the density of each phylum level taxonomic unit in each dataset based on sequencing depth (giga base, Gb). Different datasets are indicated by bars of varying colors.; d) The bar graph represents the density of top ten family level taxonomic units in each dataset based on sequencing depth (giga base, Gb). Different datasets are indicated by bars of varying colors. Corresponding phylum symbols",
+    "footnote": [],
+    "bbox": [
+      [
+        184,
+        92,
+        775,
+        686
+      ]
+    ],
+    "page_idx": 21
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": ">Figure",
+    "footnote": [],
+    "bbox": [
+      [
+        207,
+        110,
+        820,
+        691
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Fig. 5 The robustness of VITAP's taxonomic assignments for short sequences within different viral phyla, and across whole viral genomes. This result focuses on the taxonomic assignment efficiency of short sequences that were successfully assigned taxonomic units by VITAP, without considering sequences that were not assigned, as the annotation ratio of VITAP has already been thoroughly evaluated in previous results. a) The comparison of the taxonomic assignment efficiency at the family and genus level for 1-kb and 5-kb short sequences, which derived from different viral phyla. For 1-kb sequences, VITAP can produce acceptable family level taxonomic assignment results; For 5-kb sequences, VITAP",
+    "footnote": [],
+    "bbox": [
+      [
+        222,
+        155,
+        797,
+        740
+      ]
+    ],
+    "page_idx": 24
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 29
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3",
+    "footnote": [],
+    "bbox": [
+      [
+        175,
+        87,
+        870,
+        770
+      ]
+    ],
+    "page_idx": 35
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4",
+    "footnote": [],
+    "bbox": [
+      [
+        175,
+        80,
+        875,
+        603
+      ]
+    ],
+    "page_idx": 52
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Figure S2",
+    "footnote": [],
+    "bbox": [
+      [
+        200,
+        260,
+        820,
+        808
+      ]
+    ],
+    "page_idx": 53
+  }
+]

peer_reviews/supplementary_0_Peer Review file__a67efe6a3db9cdb08824acfcefc7ccfcf97b085f73a6f42f1c9d733274c4bf70/images_list.json

@@ -0,0 +1,17 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Figure SM6. Probabilistic EF realizations plotted against their average values using our approach (blue markers) and the approach by Platt et al. (2024) (red markers).\"",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        87,
+        870,
+        402
+      ]
+    ],
+    "page_idx": 28
+  }
+]

peer_reviews/supplementary_0_Peer Review file__aac9d4784990993e86fdeaac6a1fa75dd22448c7eea2daca4690c8309be907e5/images_list.json

@@ -0,0 +1,63 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Fig. R1. The electron tunneling from graphene to silicon based on the field direction.",
+    "footnote": [],
+    "bbox": [
+      [
+        567,
+        543,
+        886,
+        757
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Extended_Data_Figure_1.jpg",
+    "caption": "Extended Data Fig. 1. The measured cross-correlation \\(\\mathrm{I_E}\\) current modulation driven by a field of chirped few-cycle pulse.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Extended_Data_Figure_1.jpg",
+    "caption": "Extended Data Fig. 1. The measured cross-correlation \\(\\mathrm{I_E}\\) current modulation driven by a field of chirped few-cycle pulse.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Extended_Data_Figure_1.jpg",
+    "caption": "Fig. R1: The retrieved waveforms of the two pulses used to obtain the cross-correlation current modulation in Extended data Fig. 1: (a) pulse 1 (6.5 fs) (b) pulse 2 (chirped pulse). (c) The calculated cross-correlation of the two pulse is plotted in red in comparison with the measured current modulation presented in the Extended Data Fig. 1.",
+    "footnote": [],
+    "bbox": [
+      [
+        98,
+        541,
+        839,
+        696
+      ]
+    ],
+    "page_idx": 20
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Fig. S3: The measured modulated current using circular polarized pulse.",
+    "footnote": [],
+    "bbox": [
+      [
+        275,
+        762,
+        664,
+        900
+      ]
+    ],
+    "page_idx": 21
+  }
+]

peer_reviews/supplementary_0_Peer Review file__aac9d4784990993e86fdeaac6a1fa75dd22448c7eea2daca4690c8309be907e5/supplementary_0_Peer Review file__aac9d4784990993e86fdeaac6a1fa75dd22448c7eea2daca4690c8309be907e5.mmd

@@ -0,0 +1,734 @@
+
+# nature portfolio  
+
+# Peer Review File  
+
+# Light-induced quantum tunnelling current in graphene  
+
+Corresponding Author: Professor Mohammed Hassan  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Attachments originally included by the reviewers as part of their assessment can be found at the end of this file.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+Ultrafast switching of the current in the graphene transistor is very interesting. The theory including an absorbing potential is also interesting. However, there are several concerns (see below) that must be addressed before publication can be recommended.  
+
+First, the origin of the current is interpreted as "quantum tunneling between two graphene sides" in the main text. On the other hand, it is written in the theory part as "tunneling from the conduction band of graphene to silicon". These two physics are completely different, and I can only agree with the latter. How large is the Si gap between the two graphene? Scale bar(s) should be added to Fig. 1. If the gap is very large (In Ref. 55, the gap is only 5 nm thick and the tunneling picture is correct), how can electrons tunnel over a large distance within the ultrashort time scale?  
+
+Second, Fig. S1 shows that there is a 90- nm- thick SiO2 layer between graphene and silicon. Why can the authors make the graphene- silicon- graphene transistor rather than the graphene- SiO2- silicon- SiO2- graphene transistor? This point must be made clearer because it is closely related to all the results and interpretations.  
+
+Third, the authors discuss two types of currents, I_E and I_p. The main claim of this article is the observation of I_E by the use of the graphene transistor. However, if my understanding is correct, the contribution of I_p is still much stronger (Extended Data 1) and the observation of I_E requires the excitation with the two pulses even with the special transistor. Moreover, I_E is not important in the results in Figs. 3- 4 and Table 1- 2 (the logic gates). These points should be explained more clearly and carefully. In addition, it should be described how the background signal was removed from the raw data to obtain the results in Fig. 2.  
+
+Fourth, it is written in the theory part that "the presence of the graphene- silicon junction breaks the symmetry" but the experimental sample (Fig. 1a) holds a symmetry due to the two junctions with opposite directions. Can the theory still explain the generation of current with the two junctions?  
+
+Fifth, as far as I understand from Ref. 54, the current produced in Ref. 54 (and probably in some others as well) is not from I_p. The currents are originated from the quantum interference occurring within a cycle and induced by the classical electric field (E), not photons (p). The interference prevents I_E from averaging out to zero. Authors should check the article again and modify their definitions of I_E and I_p, if necessary. On page 5, there is a sentence on the origin of currents in this work, "drifting of the excited carriers ...following the driver laser field". If this sentence explains the physics of I_E, the currents in Ref. 54 should be I_E, in my opinion. In addition, there is a question how is this drift picture related to the tunneling?  
+
+Below are minor comments:  
+
+(1) The authors should compare the achieved switching time (630 as) to the other works on the petahertz/light-wave electronics. In addition, "room temperature" and "ambient pressure" are emphasized. How are the relevant experiments done before? 
+(2) Regarding "cross-correlation" on page 5, is it the first-order cross-correlation or the second-order? 
+(3) Did the authors confirm the laser-polarization-angle dependence? There must be no I_E current when the polarization is parallel to the Si stripe. 
+(4) Authors should add an explanation as to why the Raman spectrum in the supplement shows that there is no graphene
+
+<--- Page Split --->
+
+oxide.  
+
+oxide.(5) Is the result in the inset of Fig. 2b measured or calculated? If this is calculated, it should be clearly stated that the number was obtained by calculation.(6) It should be stated that the theory considers only two electronic bands.(7) Label should be added to the horizontal axis of Fig. 1b.  
+
+## Reviewer #2  
+
+(Remarks to the Author)  
+
+The manuscript presents very interesting experimental results on the measurement of ultrafast current a cross a Gr- Si- Gr junction upon photoexcitation. When the few cycle laser pulses are incident the junction reports a several hundred nano ampere level of current. There is an oscillation on the measured current that corresponds to the quarter of the period of the driving laser field. The maximum current is defined as ON and the minimum current is defined as OFF. So, there is a 630 as of delay between the ON peak and OFF dip, as shown in figure 2. They also present IV curve in figure 3, where the current shows an increase with respect to the applied higher peak intensity of the laser in the form of its field strength.  
+
+Through a cross- correlation method they claim to also separate the instantaneous field induced current (IE) from the photo- induced current (IP) although I find the methodology is not clearly presented.  
+
+The IV curve is described as the device performing like a transistor although unfortunately I do not find any specific features of the IV curves of a transistor such as an increase followed by a saturation of the current after certain amount of voltage. Also, the cross- correlation scheme could be clearer if they sketch some pictures of the setup showing two pulses.  
+
+If the measured current is the tunneling current across the interface the calculations should also try to calculate the tunneling across Si- Ge interface. It seems to me that the calculated current densities are intra- band current densities on graphene. Presented calculation results do not seem to support the main claim of the manuscript although the intra- band current density maps on graphene look pretty.  
+
+If the current is a field- driven current, I would expect it has strong directionality with respect to the polarization direction of the laser. Have they observed anything like that?  
+
+Also, the photon energy is above the band- gap of silicon so I am not sure how the photo- excited carriers are not dominating. If the photon- energy of the light is smaller than the indirect band- gap of silicon then it would be easier to understand.  
+
+In overall, in the present manuscript, while I find interesting experimental observations, I feel that the manuscript lack a clear interpretation of the data, and I do not find any similarities with IV curves of a typical transistor.  
+
+## Reviewer #3  
+
+(Remarks to the Author)  
+
+Key results  
+
+A) The photoresponse of the planar-graphene tunneling phototransistor designed in this study is a result from the tunneling of photoexcited carriers between the graphene layers of a Gr-Si-Gr transistor  
+
+B) Two components contributing to the total photoresponse are the direct tunneling of photoexcited charge carriers and the field-induced tunneling current which is carrier motion driven directly by the light field itself. The latter is attosecond scale response.  
+
+C) There is a notable photoconductivity enhancement at higher light field strength  
+
+D) This phototransistor is at the technology readiness level for developing attosecond and lightwave quantum optoelectronics  
+
+Comments on A)  
+
+Based on the current profile in Fig 1C and also the geometry of the device being graphene separated by a gap, it is reasonable to believe the current is due to tunneling photoexcited carriers. This device response is also consistent with other non- photoexcited lateral graphene tunneling devices (J. H. Yang et al., "Geometrically Enhanced Graphene Tunneling Diode with Lateral Nano- Scale Gap," IEEE Electron Device Letters, vol. 40, no. 11, pp. 1840- 1843, Nov. 2019, doi: 10.1109/LED.2019.2940818). However, clarification on device configuration would be helpful.  
+
+It would be helpful to the reader if the schematic in Fig S1a reflects the actual geometry of the device with the tunneling junction in the middle along with it the dimensions of this junction. The classification of this being a graphene- silicon- graphene photoresistor is also unclear, since the methods described in the text for creating the devices involve removing the middle section of the graphene and SiO2 which means the Si is not in contact with the graphene in the lateral direction. Base on this description it would make the device a Graphene/vacuum/air- Graphene device. This may or may not affect the simulations outcomes.  
+
+It would also be helpful to the reader to have the focused laser spot size and location described along with the tunneling junction dimensions. The size of the laser spot and also its location is important in understanding the origin of the tunneling current. For example, if the laser is focused such that it is only on one side of the graphene junction, then one might expect to only have field- induced current in one direction, so either it is always positive or always negative. This would help with visualizing the behavior of the field induced current.
+
+<--- Page Split --->
+
+## Comments on B)  
+
+Comments on B)The tunneling configuration describe in this manuscript is expected to capture the field- induced current generated in graphene, since the junction would capture any charges moving into the junction and not away from it thus breaking the symmetry of the field- induced motion of the carriers. Moreover, this field- induced current has been observed previously in graphene under similar field intensities as such one would expect a field- induced current to contribute to the total photoresponse in the configuration as well. A petahertz responsive device with a simple easily reproducible geometry is indeed instrumental in studying sub- femtosecond dynamics.  
+
+On the other hand, more clarification maybe needed when describing the origins of the oscillating behavior of the field- induced current. The method in which the field- induced current was extracted was by removing the photo- induced current. This was done so by applying an external voltage across the contact of the device such that the total current is zero when the laser pulses are not overlapping. However, the magnitude of the photo- induced current could potentially vary when the laser pulses overlap due to the interference of the pulses which mean the photo- induced current may not be completely removed by applying a fixed voltage across the device. This would lead to a photo- induced current which also varies in time- delay of the pulses as the average number of excited carriers changes when the pulse interfere constructively or destructively. Thus, this current would also modulate similarly to the cross- correlation profile of the pulses. An explanation of the potential changes in the photo- induce current, or lack thereof, would be helpful to further solidify the conclusions of the field induced current.  
+
+## Comments on C)  
+
+Comments on C)Increase of the photoconductivity with intensity is expected due to excitation of carriers into the conduction band, this is also confirmed with the model calculations which seems reasonable. This gives a good metric for determining the efficiency of the logic circuit since it is expected to operate at high field intensities.  
+
+## Comments on D)  
+
+Comments on D)The technological readiness of this device would be clearer when the comments for the extraction of the field- induced current is addressed. The fabrication of the device also appears to be able to be scaled up given that a high- power scanning laser could produce the tunneling junction required. It would be helpful to give examples of methods to distinguish the ultrafast- intra- band current response from the slower overall inter- band current. As in how would one suppress the IE signal which changes sign to achieve an ultrafast logic gate states as described in table 1 and 2.  
+
+## Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+The authors adequately replied to many of my comments. I now understand physics better. The directional current in Fig. 1b is produced due to the unequal laser intensities on the two graphene- silicon interfaces. The offset of \(75~\mathrm{nA}\) in Fig. 2b is canceled by an external DC voltage, and therefore the direction of the measured current ( \(29~\mathrm{nA}\) amplitude) could be reversed. On the other hand, there are two remaining points which should be made clearer before publication.  
+
+1. The structure of the sample is still not clear to me. The authors removed graphene and SiO2 and exposed the silicon substrate. How can it make a graphene-silicon interface? From a three-dimensional perspective, isn't there at least \(90~\mathrm{nm}\) distance (=SiO2 thickness) between graphene and silicon?  
+
+2. Regarding the 'first-order' cross-correlation, I mentioned 'cross-correlation' at a different place, which is in the sentence 'In our calculation, ... cross-correlation of the laser fields'. The question was asked because it was a bit of a surprise to me that the observed result in Fig. 2b can well be approximated by the first-order cross-correlation of the laser fields, meaning that the switching speed of the device is solely determined by the cycle period time of the laser field.  
+
+The authors excluded the contribution of optical interference (reply to reviewer #3) but the revised manuscript states that 7.5 nA (= 10% of the \(75~\mathrm{nA}\) ) of the \(29~\mathrm{nA}\) can come from the optical interference. On this point, the authors need to prove more carefully that the dominant effect is still the tuning, not the optical interference. My specific comments are  
+
+(2-a) An explanation should be added as to how and where the \(10\%\) power modulation was measured. Since two noncollinear beams can produce delay-dependent interference fringes in space, the local intensities on the graphene- silicon interfaces can be modulated by more than \(10\%\) . Since the current is driven by unequal local intensities, the interference fringes may cause a strong influence. Did the authors exclude this possibility, too?  
+
+(2-b) Is the noncollinear geometry considered in the theory with the dipole approximation? If not, the theory gives a strong optical interference effect and cannot support the interpretation. Additional explanation is needed on this point.  
+
+## Reviewer #3  
+
+[Editorial Note: The attachment is displayed at the end of the file]  
+
+(Remarks to the Author)The follow up comments below should be addressed before publication can be recommended.
+
+<--- Page Split --->
+
+The main claims of the manuscript is restated below for reference, follow up comments follows.  
+
+A) The photoresponse of the planar-graphene tunneling phototransistor designed in this study is a result from the tunneling of photoexcited carriers between the graphene layers of a Gr-Si-Gr transistor.  
+
+B) Two components contributing to the total photoresponse are the direct tunneling of photoexcited charge carriers and the field-induced tunneling current which is carrier motion driven directly by the light field itself. The latter is attosecond scale response. 
+C) There is a notable photoconductivity enhancement at higher light field strength 
+D) This phototransistor is at the technology readiness level for developing attosecond and lightwave quantum optoelectronics.  
+
+Follow up Comments on A)  
+
+The added scale for device in the revised manuscript is helpful. However the overall geometry of the device is still unclear. I am unsure as to why the Graphene would be in contact with the silicon since the graphene should be ontop of the SiO2 from the fabrication process. A cross section schematic of the structure would be helpful here (see attached example). As for the Raman spectroscopy, while the silicon and graphene peaks are shown, that could potentially mean both the graphene and silicon are within the beamspot of the laser, but they could be vertically displaced.  
+
+Follow up Comments on B)  
+
+Since the light is exciting both graphene and silicon, is there a difference between the carriers tunneling from silicon to graphene vs graphene tunneling to silicon?  
+
+Also, in reference to the polarization testing mentioned by reviewer #1. It is not immediately clear why the circular polarization does not produce the same variation with since you can consider it as two orthogonal linear polarizations (one parallel to the stripe, and another perpendicular to the stripe)  
+
+Lastly the linear polarization dependence is a strong experimental evidence for this affect being field driven and should be included in the main text.  
+
+Version 2:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+The authors have adequately responded to my last comments. But the new data, Extended Data 1, is difficult to understand. If my understanding is correct, the two pulses are nearly identical. If so, the same signals must be observed at plus and minus delay times because the two pulses are indistinguishable regardless of the chirp and the nonlinearity involved. If the two pulses are significantly unequal, the new result would not strongly support the claim. Additional explanation(s) should be added.  
+
+Reviewer #3  
+
+(Remarks to the Author)  
+
+The comments on two of the main claims in the manuscripts was addressed in the previous rounds of comments. However more clarification is needed before publication to support the following two claims listed below (restated from the first round of comments for reference).  
+
+A) The photoresponse of the planar-graphene tunneling phototransistor designed in this study is a result from the tunneling of photoexcited carriers between the graphene layers of a Gr-Si-Gr transistor. 
+B) Two components contributing to the total photoresponse are the direct tunneling of photoexcited charge carriers and the field-induced tunneling current which is carrier motion driven directly by the light field itself. The latter is attosecond scale response.  
+
+(Follow up comments on A)  
+
+Both reviewer #1 and myself asked for clarification of the structure, which the authors has provided. While the provided explanation could explain the formation of the structure it is only a speculation on the melting of the Si. Perhaps an AFM measurement of the surface can show this effect.
+
+<--- Page Split --->
+
+(Follow up comments on B)  
+
+A clarification on the asymmetric time delayed data using a chirped pulse, and a demonstration of the time- delay photocurrent data with different polarization is needed to support this effect being field driven.  
+
+A more clear explanation on why an asymmetric current modulation implies a field driven effect is needed. This asymmetry suggests the charge carriers cares about which of the two identical pulses arrives first, which intuitively even with a field driven affect should not occur.  
+
+Data showing the diminished tunneling current with respect to excitation by a pulse with linear- polarization parallel or semi parallel to the channel as mentioned in a previous rounds of comments should be included in the manuscript for this would be one of the hallmarks of this effect being field driven.  
+
+Version 3:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+Regarding the new information that a combination of chirped and un- chirped pulses was used for Extended Data 1, I cannot follow why and how the asymmetry in the data is connected to their conclusion of "minor effect from the optical interference". Optical interference can give asymmetric data because one of the pulses is asymmetric in time. A clear explanation should be added before publication.  
+
+In addition, the authors should describe in the method section how the pulse was chirped.  
+
+If the asymmetry in data cannot exclude the possibility of optical interference, the measurement result with a power meter (10% modulation) is the only direct evidence. To strengthen the authors' claim, I suggest adding a description of how this number (10%) is translated into the amount of modulation of I. If the modulation of I. E is proportional to the laser intensity, I E should then be modulated by 10% due to optical interference. However, if I E varies nonlinearly with laser intensity, optical interference causes stronger modulation.  
+
+Lastly, although the authors replied to referee #3 and me previously that the connection between graphene and Si was confirmed by Raman measurements, I cannot find any proof that the SiO2 layer was removed. There is no Raman peaks assigned to SiO2 in Fig. S2. The authors should add an explanation before publication. If the removal of the SiO2 layer is a hypothetical level, it should be expressed as such.  
+
+Reviewer #3  
+
+(Remarks to the Author) The authors have adequately answered all my questions.
+
+<--- Page Split --->
+
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+Response to reviewers' comments - NCOMMS- 24- 50952  
+
+The referees' comments are shown in bold and Italics, followed by the corresponding response in blue and the change in the revised version.  
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+Ultrafast switching of the current in the graphene transistor is very interesting. The theory including an absorbing potential is also interesting. However, there are several concerns (see below) that must be addressed before publication can be recommended.  
+
+We would like to thank the reviewer for finding our work interesting. We appreciate the opportunity to address all the reviewer's concerns.  
+
+1- First, the origin of the current is interpreted as "quantum tunneling between two graphene sides" in the main text. On the other hand, it is written in the theory part as "tunneling from the conduction band of graphene to silicon". These two physics are completely different, and I can only agree with the latter. How large is the Si gap between the two graphene? Scale bar(s) should be added to Fig. 1. If the gap is very large (In Ref. 55, the gap is only 5 nm thick and the tunneling picture is correct.), how can electrons tunnel over a large distance within the ultrashort time scale?  
+
+Thank you for the reviewer's comment! The Si gap is estimated to be \(4.3 \mu \mathrm{m}\) , and a scale bar has been added to the revised version of Fig. 1. We agree with the reviewer's suggestion that the tunneling is taking place in the interfaces between graphene and Si. In this studied system we have two Gr- Si interface. To clarify this point we, and we have revised the manuscript accordingly:  
+
+Page 3:  
+
+\(\Rightarrow\) "In our transistor, the current flows based on quantum tunnelling between the graphene and the silicon junction."  
+
+Page 4:  
+
+\(\Rightarrow\) "Accordingly, the generated \(I_{L}\) flows in our device by quantum tunnelling of the carriers from the graphene to silicon."  
+
+Page 5:  
+
+\(\Rightarrow\) "It is noteworthy that the modulation of the \(I_{E}\) oscillates between negative and positive values. This indicates that the \(I_{E}\) flows alternatively from the two graphene sides to the silicon junction every half- cycle of the driver field (depending on the driver field direction) causing the switching in subfemtosecond time window."  
+
+Page 8:  
+
+\(\Rightarrow\) "The current flow is based on the tunnelling of electrons between the graphene to the Si Junction."
+
+<--- Page Split --->
+
+2- Second, Fig. S1 shows that there is a 90- nm- thick \(\mathrm{SiO_2}\) layer between graphene and silicon. Why can the authors make the graphene- silicon- graphene transistor rather than the graphene- \(\mathrm{SiO_2}\) - silicon- \(\mathrm{SiO_2}\) - graphene transistor? This point must be made clearer because it is closely related to all the results and interpretations.  
+
+The \(\mathrm{SiO_2}\) has a large band gap of \(9\mathrm{eV}\) , which prevents electron tunneling (our pump pulse is centered at \(1.65\mathrm{eV}\) ). As a result, the graphene- \(\mathrm{SiO_2}\) - silicon- \(\mathrm{SiO_2}\) - graphene transistor doesn't allow photo- induced current to flow and, therefore, cannot be measured by our setup. We have clarified this point in the revised supplementary information (SI), page 2, thanks to the reviewer's suggestion.  
+
+\(\Rightarrow\) To generate the Gr- Si- Gr channel, a high- power laser beam (300 mW) is focused into the centre of the graphene channel. Then, we created a thin layer (junction) of Si by depleting the thin layers of graphene and \(\mathrm{SiO_2}\) . If the \(\mathrm{SiO_2}\) layer (see Fig. S1) — which is an isolator with high bandgap (9 eV)— is not totally depleted, the photo- induced current won't tunnel, flow, and generate a measurable current in our transistor setup.  
+
+3- Third, the authors discuss two types of currents, \(I_E\) and \(I_P\) . The main claim of this article is the observation of \(I_E\) by the use of the graphene transistor. However, if my understanding is correct, the contribution of \(I_P\) is still much stronger (Extended Data 1) and the observation of \(I_E\) requires the excitation with the two pulses even with the special transistor. Moreover, \(I_E\) is not important in the results in Figs. 3- 4 and Table 1- 2 (the logic gates). These points should be explained more clearly and carefully. In addition, it should be described how the background signal was removed from the raw data to obtain the results in Fig. 2.  
+
+We thank the reviewer for the comment! We would like to clarify that in our experiment setup the main contribution of \(I_P\) signal was compensated (canceled out) by applying external voltage with the same value but different direction. To avoid any confusion we considered the reviewer's comment in the revised version of the manuscript in many positions. i.e. Page 5  
+
+\(\Rightarrow\) "Our setup's capability enabled the compensation of the \(I_P\) current (generated by the interband dynamics)—which has the main contribution in \(I_L\) —by applying an external voltage ( \(V_{ext}\) ) value until the output measured current is zero amperes."  
+
+It is also mentioned in the paragraph discussing the results in Figure 3- 4, in page 6  
+
+\(\Rightarrow\) "The photo- induced \(I_P\) current signal has the main contribution to the \(I_L\) current." in page 7  
+
+\(\Rightarrow\) "Next, we studied the effect of the light- induced current and carrier excitation on the resistivity and photoconductivity, mainly driven by \(I_P\) , of our phototransistor \(^{59}\) ."  
+
+The explanation of the "how is the background signal was removed" is explained in page 5.  
+
+\(\Rightarrow\) "Accordingly, we modified our setup by splitting the input laser beam into two beams using a beamsplitter (Fig. 2a); each beam power has been set to have a similar estimated low field strength of \(\sim 0.85\mathrm{V / nm}\) . Then, we recorded the current as a function of the time delay between the two pulses. The recorded current, when the two pulses are not overlapping in
+
+<--- Page Split --->
+
+time, is 75 nA. Our setup's capability enabled the compensation (cancelling out) of the \(I_{\mathrm{P}}\) current signal (generated by the interband dynamics)—which has the main contribution in \(I_{\mathrm{L}}\) —by applying an external voltage \((V_{\mathrm{ext}})\) value until the output measured current is zero amperes."  
+
+Thanks to the reviewer's comment and suggestion! we show a case of using \(I_{\mathrm{E}}\) in demonstrating the logic gate NOT. See page 7- 8 and table 4 in the revised version of the manuscript.  
+
+\(\Rightarrow\) "Furthermore, the petahertz logic gate can be demonstrated by using \(I_{\mathrm{E}}\) . In this case, the delay between two laser pulses \(\tau\) is the input signal and the total \(I_{\mathrm{E}}\) is the output signal (see Fig. 2b). When the delay between the two pulses \(\tau = 360\) as, the output \(I_{\mathrm{E}}\) signal \(= 0\) . Alternatively, when the delay between the two pulses \(\tau = 0\) as, the output \(I_{\mathrm{E}}\) has 29 nA (see Fig. 2 a) which present the status 1, demonstrating the NOT gate as summarized in table 4, open the door for establishing ultrafast optical computers."  
+
+4- Fourth, it is written in the theory part that "the presence of the graphene-silicon junction breaks the symmetry" but the experimental sample (Fig. 1a) holds a symmetry due to the two junctions with opposite directions. Can the theory still explain the generation of current with the two junctions?  
+
+First, we would like to clarify the following: by "the presence of the graphene- silicon junction breaks the symmetry," we are referring to the disruption of the structural symmetry of the graphene layer caused by the junction. However, the symmetry breaking those results in the propagation of the light- induced current arises from the asymmetric illumination of the laser beam on both sides of the graphene (see the revised Fig. 1Sa).  
+
+Theoretically, we modeled the tunneling of electrons by a simple CAP placed in a graphene unit cell located near the junction. Importantly, within this formalism the tunneling depends on the strength of the applied vector potential (see Eq. (8) of the main text). Accordingly, if two junctions with identical but reversed in direction CAPs are included in the simulations, we still will see the generated current. This is because the applied vector potential with a certain local strength will push or pull certain number of electrons through the Gr- Si junction on one side and will do the same but with decreased/increased probability through another Si- Gr junction where the local strength of the vector potential is lower/higher. This is further explained with an illustration in the next point (Point 5).  
+
+Thanks to the reviewer comment! We clarify this in the revised manuscript in page 4 we added  
+
+\(\Rightarrow\) Moreover, the asymmetric of this obtain IV curve (around Vext=0 in Fig. 1c) suggests that the illumination of the laser beam of the two graphene sides is uneven (see Fig. S1a), which cause the symmetry breaking and explains the flow of the light- induced current in our setup.  
+
+5- Fifth, as far as I understand from Ref. 54, the current produced in Ref. 54 (and probably in some others as well) is not from Ip. The currents are originated from the quantum interference occurring within a cycle and induced by the classical electric field (E), not photons (p). The interference prevents \(I_{\mathrm{E}}\) from averaging out to zero. Authors should check the article again and modify their definitions of \(I_{\mathrm{E}}\) and \(I_{\mathrm{p}}\) if necessary. On page 5, there is a sentence on the origin of currents in this work, "drifting of the excited carriers ...following the driver laser field". If this sentence explains the physics of \(I_{\mathrm{E}}\) , the currents in Ref. 54
+
+<--- Page Split --->
+
+should be \(I_{\mathrm{E}}\) , in my opinion. In addition, there is a question how is this drift picture related to the tunneling?  
+
+We would like to thank the reviewer for this insightful comment. As we understand it, the light- induced current results presented in Ref. 54 (Fig. 2), which illustrates how the generated current varies as a function of field strength and CEP. According to the discussion in Ref. 54, "The CEP- dependent conduction- band population can be estimated by treating the electron dynamics in a fully coherent manner when the carrier decay timescales are longer than the optical cycle, which is satisfied here. The residual current is obtained from the conduction- band population distribution and the electron velocity given by the slopes of the bands, with the assumptions of a ballistic carrier lifetime of 40 fs and a diffusive decay length of 350 nm, consistent with previous literature12- 14 (see Supplementary Methods)." Based on this, the primary change in current is attributed to the conduction- band population distribution after the laser pulse.  
+
+We completely agree with this explanation; however, we note that it does not fully confirm that the CEP dependence is exclusively related to the generated induced electric field ( \(\mathrm{I}_{\mathrm{E}}\) ), as the authors did not provide results from an experiment that would exclude changes in the peak intensity of the pulse when the CEP is changing. Given the fact that the pulse duration is very short (5.4 fs), a change of \(\pi\) in the CEP would induce an intensity value by approximately 20- 30%. Therefore, the observed change in the generated current could also be related to the intensity dependence.  
+
+Importantly, we would like to mention that in a follow- up work (Ref. 53), the authors of Ref. 54 did indeed present results showing field- induced currents due to interference occurring within a cycle, which breaks the symmetry in graphene, which could provide further context for understanding the phenomenon.  
+
+Regarding the drifting, we believe that the charge carriers are first drifted in the conduction band by the driving field (vector potential), generating an intraband current. These carriers then tunnel  
+
+from graphene into silicon. Depending on the field direction, the carriers tunnel from graphene 1 to silicon or from graphene 2 (depends on the field direction) to silicon (see illustration below in Fig. R1. ).  
+
+Additionally, the measurements in Fig. 1c (where the obtained IV curve is not symmetric around zero) and Fig. 2b (where the average \(+I_{\mathrm{E}}\) is greater than the average \(- I_{\mathrm{E}}\) ) suggest that the laser illumination on both graphene sides is asymmetric. This asymmetry is responsible for breaking the symmetry of the system, which, together with the tunneling process, leads to measure the induced electric ( \(\mathrm{I}_{\mathrm{E}}\) ) current signal.  
+
+![](images/Figure_unknown_0.jpg)
+
+<center>Fig. R1. The electron tunneling from graphene to silicon based on the field direction. </center>  
+
+Hence, we considered the reviewer's suggestion and revised our explanation to be more accurate in page 5 in our revised manuscript  
+
+\(\Rightarrow\) "We attributed this measured current oscillation to the drifting and tunnelling of the excited carriers within the conduction band (intraband current) of graphene following the driver laser field".
+
+<--- Page Split --->
+
+## Below are minor comments:  
+
+(1) The authors should compare the achieved switching time (630 as) to the other works on the petahertz/light-wave electronics. In addition, "room temperature" and "ambient pressure" are emphasized. How are the relevant experiments done before?  
+
+For example, our work can be compared to one of the most recognized in the petahertz/light- wave electronic work reported in Nature 605, 251- 255, (2022). The demonstrated switching from 0 A to 4 pA is \(\pi /2\) of the driver pulses which is estimated to be 1.4 fs, while in our work the switching speed is 0.630 fs and with a contrast from 0 to \(29\mathrm{nA}\) . In addition, the work demonstrated in Nature 605, 251- 255, (2022) is done at room temperature (similar to our presented work) but under vacuum conditions \((1\times 10^{- 8}\mathrm{Pa})\) , while our work is done at atmospheric pressure, as mentioned in the manuscript conclusion paragraph (page 8):  
+
+In this work, we demonstrate the light- induced quantum tunnelling current in a Gr- Si- Gr phototransistor. The current flow is based on the tunnelling of electrons between the graphene to the Si Junction. This current has more than three orders of magnitude better efficiency than the typical graphene transistor53,54. Moreover, this high efficiency led to generating a decent light- induced current amplitude at low pumping laser power. Hence, this Gr- Si- Gr transistor can operate in ambient conditions (normal pressure and temperature conditions) in analogy to the typical graphene phototransistor, which operates in vacuum to avoid the oxidation of graphene and the degradation of the transistor when illuminated with a high- intense laser beam. Furthermore, the presented current tunnelling mechanism in the Gr- Si- Gr transistor gate the laser field- induced current signal; thus, it subsists after the pulse, which is not possible in a symmetric graphene transistor. Hence, this ultrafast current—which has a sub- femtosecond switching time—can be logged, demonstrating the petahertz current switching speed in our transistor.  
+
+(2) Regarding "cross-correlation" on page 5, is it the first-order cross-correlation or the second-order?  
+
+It is a first- order cross- correlation since out work is done under a field strength drive one- photon excitation as shown in Fig. 3b (now Fig. 3c). This information is now included in the revised version of the manuscript, Page 5  
+
+"The average of three first-order cross-correlation current measurements is shown in Fig. 2a (black dots connected with red lines)."  
+
+(3) Did the authors confirm the laser-polarization-angle dependence? There must be no \(I_{E}\) current when the polarization is parallel to the Si stripe.  
+
+We appreciate the reviewer's suggestion and have conducted the proposed test. Specifically, when we changed the polarization to be parallel to the Si stripe, we observed a significantly reduced IE; however, it did not vanish entirely. This behavior can be attributed to the 6.5- fs pump pulse used in our experiment, which is generated via the nonlinear propagation of multicycle pulses in a hollow- core fiber (HCF), as described in the Methods section. Consequently, the output beam from the HCF does not exhibit perfectly pure linear polarization (our measurements indicate a polarization ratio of \(85\%\) to \(15\%\) ).  
+
+To address this, we introduced a polarizer to improve the beam's linear polarization. However, the polarizer (10 mm thick) introduced temporal broadening of the pulse to approximately 200
+
+<--- Page Split --->
+
+femtoseconds, rendering the results inconclusive. This is because the observed signal disappearance could either result from the extended pulse duration or the change in polarization. As an alternative, we employed a thin quarter- wave plate to generate a circularly polarized beam. In this case, the current oscillation \(\mathrm{I_E}\) signal disappeared then reappeared again when switching back to a linearly polarized beam.  
+
+### (4) Authors should add an explanation as to why the Raman spectrum in the supplement shows that there is no graphene oxide.  
+
+The ionization energy of graphene is \(4.5 - 5\mathrm{eV}\) and for oxygen is \(13.6\mathrm{eV}\) so the laser intensity at the focus \((50\mu \mathrm{m})\) doesn't have enough intensity to ionize graphene and oxygen through multiphoton excitation to induce the oxidization reaction. Hence, the Raman spectrum show no graphene oxide. This explanation in added to the revised version of the SI.  
+
+### (5) Is the result in the inset of Fig. 2b measured or calculated? If this is calculated, it should be clearly stated that the number was obtained by calculation.  
+
+The inset is a zoom- in from the measured data on Fig. 2b (now Fig. 2c in the revised version). Thanks to the reviewer! this now included in Fig. 2 caption to avoid any confusion for the reader in the revised version of the manuscript.  
+
+### (6) It should be stated that the theory considers only two electronic bands.  
+
+The reviewer is considered and included in the revised version of the manuscript page 9  
+
+"The light- induced population transfer dynamics in graphene can be obtained by solving the semiconductor Bloch equation considers only two electronic bands".  
+
+### (7) Label should be added to the horizontal axis of Fig. 1b.  
+
+Thanks to the reviewer! The note is considered in the revised version of Fig. 1.  
+
+Finally, we would like to express our gratitude for the reviewer's comments and suggestions, which have greatly helped us improve both the manuscript and the discussion of our results. We hope we have adequately addressed all the raised questions, and we sincerely appreciate the reviewer's recommendation for publishing our manuscript in Nature Communications.
+
+<--- Page Split --->
+
+## Reviewer #2 (Remarks to the Author):  
+
+The manuscript presents very interesting experimental results on the measurement of ultrafast current across a Gr- Si- Gr junction upon photoexcitation. When the few cycle laser pulses are incident, the junction reports a several hundred nano ampere level of current. There is an oscillation on the measured current that corresponds to the quarter of the period of the driving laser field. The maximum current is defined as ON and the minimum current is defined as OFF. So, there is a 630 as of delay between the ON peak and OFF dip, as shown in figure 2. They also present IV curve in figure 3, where the current shows an increase with respect to the applied higher peak intensity of the laser in the form of its field strength.  
+
+We thank the reviewer for finding our experimental results very interesting. We are happy to address all the points raised by the reviewer to further improve our presentation.  
+
+1- Through a cross- correlation method they claim to also separate the instantaneous field induced current \((I_{E})\) from the photo- induced current \((I_{P})\) although I find the methodology is not clearly presented.  
+
+We have revised the explanation of the related paragraph in the revised version, see page 5  
+
+\(\Rightarrow\) "Hence, to measure \(I_{E}\) in real- time, we opted to perform a cross- correlation current measurement between two current signals generated by two pump laser pulses. Accordingly, we modified our setup by splitting the input laser beam into two beams using a beamsplitter (see Methods and Fig. S.3c); each beam power has been set to have a similar estimated low field strength of \(\sim 0.85 \mathrm{V / nm}\) . Then, we recorded the current as a function of the time delay between the two pulses. The recorded current, when the two pulses are not overlapping in time, is \(75 \mathrm{nA}\) . Our setup's capability enabled the compensation of the \(I_{P}\) current (generated by the interband dynamics)—which has the main contribution in \(I_{L}\) —by applying an external voltage \((V_{\mathrm{ext}})\) value until the output measured current is zero amperes. The average of three first- order cross- correlation current measurements is shown in Fig. 2a (black dots connected with red lines). A minor contribution of the current amplitude oscillation (Fig. 2a) potentially originated from the optical interference, since we observed only a \(10\%\) oscillation in the power between the two pulses at the temporal overlap. Please note the two beams aren't collinearly propagating, and they incident on the sample with small angles \((< 5^{\circ})\) (Fig. S2c), which minimises the optical interference effect. Furthermore, the absolute measured \(I_{E}\) current amplitude signal in real- time (plotted in Fig. 2b) switches from \(29 \mathrm{nA}\) (ON status) to \(< 1 \mathrm{nA}\) (OFF status) in 630 attoseconds (see the inset of Fig. 2b), demonstrating the attosecond current switching in our phototransistor."  
+
+We will happily consider any further specific suggestion from the reviewer to clearly more the cross- correlation measurements if it is needed.  
+
+2- The IV curve is described as the device performing like a transistor although unfortunately, I do not find any specific features of the IV curves of a transistor such as an increase followed by a saturation of the current after certain amount of voltage. Also, the cross- correlation scheme could be clearer if they sketch some pictures of the setup showing two pulses.
+
+<--- Page Split --->
+
+We thank the reviewer for their helpful comment! We would like to clarify that the IV curve presented in Fig. 1c is used to illustrate the tunneling effect, as previously reported in Fig. 3A of Science 335, 947- 950, (2012) ("Field- Effect Tunneling Transistor Based on Vertical Graphene Heterostructures"). Our IV measurement in Fig. 1c shows that the current (I) increases linearly with voltage (V), saturates in the range of - 0.6 to \(0\mathrm{mV}\) , and then resumes a linear increase as the voltage rises further.  
+
+We are grateful for the reviewer's insightful suggestions, which have guided important revisions. Specifically:  
+
+- Fig. 2 has been updated by moving Fig. S3c from the Supplementary Material to become Fig. 2a, as recommended.- The IV curve, previously included in Extended Data 1, has been relocated to Fig. 1b in the revised manuscript to provide a clearer representation of the typical behavior of a Graphene Field-Effect Transistor, as shown here.- Fig. 3 has been revised to better illustrate the experimental setup, improving clarity in accordance with the reviewer's advice.  
+
+We believe these updates significantly enhance the presentation of our results and appreciate the reviewer's constructive feedback.  
+
+3- If the measured current is the tunneling current across the interface the calculations should also try to calculate the tunneling across Si- Ge interface. It seems to me that the calculated current densities are intra- band current densities on graphene. Presented calculation results do not seem to support the main claim of the manuscript although the intra- band current density maps on graphene look pretty.  
+
+The calculated currents are, indeed, arising from the intra- band dynamics of electrons in graphene. However, in pure symmetric graphene one can only observe instantaneous current but not tunneling current which we believe is measured in our experiment. To include the possibility for a tunneling current to occur in our system, we use the CAP which mimics the Si- Gr interface (see the last paragraph of the Methods section). Accordingly, we do, although phenomenologically, simulate the Si- Gr interface and observe the tunneling current in our simulations (see Extended Data 2b). Without this interface, the intra- band current in graphene is found to be zero as expected in simulation shown in Extended Data 3b.  
+
+## 4- If the current is a field-driven current, I would expect it has strong directionality with respect to the polarization direction of the laser. Have they observed anything like that?  
+
+Yes, we observed that measured current \(\mathrm{I}_{\mathrm{E}}\) current is sensitive to the polarization direction change. When we changed the polarization we observed a significantly reduced \(\mathrm{I}_{\mathrm{E}}\) ; however, it did not vanish entirely because our pump doesn't have perfect pure linear polarization (our measurements indicate a polarization ratio of \(85\%\) to \(15\%\) ), due to the nonlinear propagation nature for generating the pulse in hollow- core- fiber. Moreover, we introduced a polarizer to improve the beam's linear polarization. However, the polarizer ( \(10\mathrm{mm}\) thick) introduced temporal broadening of the pulse to approximately 200 femtoseconds, rendering the results inconclusive. This is because the observed signal disappearance could either result from the extended pulse duration or the change in polarization. As an alternative, we employed a thin quarter- wave plate to generate a circularly polarized beam. In this case, the current oscillation \(\mathrm{I}_{\mathrm{E}}\) signal disappeared then reappeared again when switching back to a linearly polarized beam.
+
+<--- Page Split --->
+
+5- Also, the photon energy is above the bandgap of silicon, so I am not sure how the photo- excited carriers are not dominating. If the photon- energy of the light is smaller than the indirect bandgap of silicon, then it would be easier to understand.  
+
+The reviewer is correct, the photo- excited carriers dynamics in graphene (interband current \(\mathrm{I}_{\mathrm{p}}\) ) has the mean contribution in the measured light- induced current signal. In addition, we would like to emphasise that the measured \(\mathrm{I}_{\mathrm{E}}\) current in Fig. 2 is done after cancelling out the \(\mathrm{I}_{\mathrm{P}}\) major contribution by apply external voltage with opposite sign as explained in the revised version of our manuscript in page 5.  
+
+\(\Rightarrow\) "Accordingly, we modified our setup by splitting the input laser beam into two beams using a beamsplitter (Fig. 2a); each beam power has been set to have a similar estimated low field strength of \(\sim 0.85 \mathrm{V / nm}\) . Then, we recorded the current as a function of the time delay between the two pulses. The recorded current, when the two pulses are not overlapping in time, is \(75 \mathrm{nA}\) . Our setup's capability enabled the compensation (cancelling out) of the \(\mathrm{I}_{\mathrm{P}}\) current signal (generated by the interband dynamics)—which has the main contribution in \(\mathrm{I}_{\mathrm{L}}\) —by applying an external voltage ( \(V_{\mathrm{ext}}\) ) value until the output measured current is zero amperes."  
+
+Also we identified that the \(\mathrm{I}_{\mathrm{p}}\) has the main contribution in different position in the text of the revised manuscript  
+
+i.e. page 5  
+
+\(\Rightarrow\) "Our setup's capability enabled the compensation of the \(\mathrm{I}_{\mathrm{P}}\) current (generated by the interband dynamics)—which has the main contribution in \(\mathrm{I}_{\mathrm{L}}\) —by applying an external voltage ( \(V_{\mathrm{ext}}\) ) value until the output measured current is zero amperes."  
+
+page 6  
+
+\(\Rightarrow\) "The photo- induced \(\mathrm{I}_{\mathrm{P}}\) current signal has the main contribution to the \(\mathrm{I}_{\mathrm{L}}\) current."  
+
+page 7  
+
+\(\Rightarrow\) "Next, we studied the effect of the light- induced current and carrier excitation on the resistivity and photoconductivity, mainly driven by \(\mathrm{I}_{\mathrm{p}}\) of our phototransistor \(^{9}\) ."  
+
+In overall, in the present manuscript, while I find interesting experimental observations, I feel that the manuscript lack a clear interpretation of the data, and I do not find any similarities with IV curves of a typical transistor.  
+
+We hope the reviewer finds the revised interpretation and updated figures in the manuscript satisfactory. We sincerely appreciate the reviewer's insightful comments, which have greatly contributed to improving the clarity and overall presentation of both the text and figures.
+
+<--- Page Split --->
+
+## Reviewer #3 (Remarks to the Author):  
+
+Key results  
+
+A) The photoresponse of the planar-graphene tunneling phototransistor designed in this study is a result from the tunneling of photoexcited carriers between the graphene layers of a Gr-Si-Gr transistor. 
+B) Two components contributing to the total photoresponse are the direct tunneling of photoexcited charge carriers and the field-induced tunneling current which is carrier motion driven directly by the light field itself. The latter is attosecond scale response. 
+C) There is a notable photoconductivity enhancement at higher light field strength 
+D) This phototransistor is at the technology readiness level for developing attosecond and lightwave quantum optoelectronics.  
+
+## Comments on A)  
+
+Based on the current profile in Fig 1C and also the geometry of the device being graphene separated by a gap, it is reasonable to believe the current is due to tunneling photoexcited carriers. This device response is also consistent with other non-photoexcited lateral graphene tunneling devices (J. H. Yang et al., "Geometrically Enhanced Graphene Tunneling Diode with Lateral Nano-Scale Gap," IEEE Electron Device Letters, vol. 40, no. 11, pp. 1840- 1843, Nov. 2019, doi: 10.1109/LED.2019.2940818.).  
+
+We thank the reviewer for their valuable comment and for bringing the mentioned reference to our attention, which supports our findings. This important work (Ref. 57) has been cited in the revised version of the manuscript.  
+
+However, clarification on device configuration would be helpful. It would be helpful to the reader if the schematic in Fig S1a reflects the actual geometry of the device with the tunneling junction in the middle along with it the dimensions of this junction. The classification of this being a graphene- silicon- graphene photoresistor is also unclear, since the methods described in the text for creating the devices involve removing the middle section of the graphene and \(\mathrm{SiO_2}\) which means the Si is not in contact with the graphene in the lateral direction. Base on this description it would make the device a Graphene- vacuum/air- Graphene device. This may or may not affect the simulations outcomes. It would also be helpful to the reader to have the focused laser spot size and location described along with the tunneling junction dimensions. The size of the laser spot and also its location is important in understanding the origin of the tunneling current. For example, if the laser is focused such that it is only on one side of the graphene junction, then one might expect to only have field- induced current in one direction, so either it is always positive or always negative. This would help with visualizing the behavior of the field induced current.  
+
+We sincerely thank the reviewer for their valuable comments and suggestions.  
+
+In response, Fig. S1a and its caption have been revised as recommended. The updated illustration now clearly indicates the junction size, as well as the location and diameter of the laser beam. The silicon junction is formed using a tightly focused high- power laser beam (via a short focal length focusing mirror), which depletes the \(\mathrm{SiO_2}\) layer. Consequently, the graphene is likely in direct contact with the silicon junction at the edges as shown by Radman spectroscopy measurements.  
+
+Additionally, the laser beam illumination on both sides of the graphene is asymmetric, as evidenced by the results in Fig. 1c (where the IV curve is not symmetric around zero voltage)
+
+<--- Page Split --->
+
+and Fig. 2b (where the oscillations in the positive direction are larger than those in the negative direction). We strongly believe this asymmetry is the key factor enabling the tunneling current to flow, allowing us to measure it.  
+
+This now clarified in page 4 in the revised manuscript  
+
+\(\Rightarrow\) Moreover, the asymmetric of this obtain IV curve (around Vext=0 in Fig. 1c) suggests that the illumination of the laser beam of the two graphene sides is uneven (see Fig. S1a), which cause the symmetry breaking and explains the flow of the light-induced current in our setup.  
+
+![](images/Extended_Data_Figure_1.jpg)
+  
+
+Moreover, the field- induced current flows in  
+
+both directions of the graphene, as demonstrated by the results in Fig. 2a (now Fig. 2b in the revised version), where the current modulation is observed in both directions (see also the illustration below).  
+
+## Comments on B)  
+
+The tunneling configuration describe in this manuscript is expected to capture the field- induced current generated in graphene, since the junction would capture any charges moving into the junction and not away from it thus breaking the symmetry of the field- induced motion of the carriers. Moreover, this field- induced current has been observed previously in graphene under similar field intensities as such one would expect a field- induced current to contribute to the total photoresponse in the configuration as well. A petahertz responsive device with a simple easily reproducible geometry is indeed instrumental in studying sub- femtosecond dynamics.  
+
+On the other hand, more clarification maybe needed when describing the origins of the oscillating behavior of the field- induced current. The method in which the field- induced current was extracted was by removing the photo- induced current. This was done so by applying an external voltage across the contact of the device such that the total current is zero when the laser pulses are not overlapping. However, the magnitude of the photo- induce current could potentially vary when the laser pulses overlap due to the interference of the pulses which mean the photo- induced current may not be completely removed by applying a fixed voltage across the device. This would lead to a photo- induced current which also varies in time- delay of the pulses as the average number of excited carriers changes when the pulse interfere constructively or destructively. Thus, this current would also modulate similarly to the cross- correlation profile of the pulses. An explanation of the potential changes in the photo- induce current, or lack thereof, would be helpful to further solidify the conclusions of the field induced current.  
+
+We are grateful for the reviewer's very valuable comments and suggestions. The reviewer expectation is correct there are minimal contribution from the optical interference and the change of the IP value on the current signal oscillation shown in Fig. 2. To quantitively evaluate this interference effect, we measured the change power of the two beams as a function of the delay
+
+<--- Page Split --->
+
+between both of them. The results shows that a change of \(\pm 10\%\) power oscillation. Since our graphene dynamics remains in the linear regime (single photon excitation) as shown in Fig. 3, the current modulation due to optical interference should be in the same level \((\pm 10\%)\) . The minimal effect from the optical interference is because the two beams are non- collinearly propagated and has small incident angle \((< 5^{\circ})\) on the graphene transistor.  
+
+In response, we have considered the reviewer's suggestions and revised our explanation of the induced field current \((\mathrm{I}_{\mathrm{E}})\) results in the revised version on Pages 4 and 5. These revisions provide a clearer explanation of the \(\mathrm{I}_{\mathrm{E}}\) and discuss the effect of the \(\mathrm{I}_{\mathrm{P}}\) change on the modulation of \(\mathrm{I}_{\mathrm{E}}\) and the switching behavior, as follows:  
+
+\(\Rightarrow\) "The current tunnelling flow mechanism gates the generated current signal in time and allows us to measure and distinguish the instantaneous field- induced current \((\mathrm{I}_{\mathrm{E}})\) , which is generated due to the intraband dynamics in graphene. This current evolves during the laser pulse's existence time window. Hence, to measure \(\mathrm{I}_{\mathrm{E}}\) in real- time, we opted to perform a cross- correlation current measurement between two current signals generated by two pump laser pulses. Accordingly, we modified our setup by splitting the input laser beam into two beams using a beamsplitter (Fig. 2a); each beam power has been set to have a similar estimated low field strength of \(\sim 0.85 \mathrm{V / nm}\) . Then, we recorded the current as a function of the time delay between the two pulses. The recorded current, when the two pulses are not overlapping in time, is \(75 \mathrm{nA}\) . Our setup's capability enabled the compensation of the \(\mathrm{I}_{\mathrm{P}}\) current (generated by the interband dynamics)—which has the main contribution in \(\mathrm{I}_{\mathrm{L}}\) —by applying an external voltage \((\mathrm{V}_{\mathrm{ext}})\) value until the output measured current is zero amperes. The average of three first- order cross- correlation current measurements is shown in Fig. 2b (black dots connected with red lines). The absolute measured \(\mathrm{I}_{\mathrm{E}}\) current amplitude signal in real- time (plotted in Fig. 2c) switches from \(29 \mathrm{nA}\) (ON status) to \(< 1 \mathrm{nA}\) (OFF status) in 630 attoseconds (see the inset of Fig. 2c), demonstrating the attosecond current switching in our phototransistor. Please note, the modulation of the \(\mathrm{I}_{\mathrm{E}}\) oscillates in both negative and positive sides indicating that the \(\mathrm{I}_{\mathrm{E}}\) flows from the two graphene sides to the silicon junction every half- cycle of the driver field (depending on the driver field direction) causing the switching in subfemtosecond time window.  
+
+On another note, a minor contribution of the current amplitude oscillation (Fig. 2b) potentially originated from the optical interference, since we observed only a \(10\%\) oscillation in the power between the two pulses at the temporal overlap. Please note the two beams aren't collinearly propagating, and they incident on the sample with small angles \((< 5^{\circ})\) (Fig. 2a), which minimize the optical interference effect. Hence, we can estimate that \(10\%\) of the current modulation (shown in Fig. 2b) may come from the optical interface and the \(\mathrm{I}_{\mathrm{P}}\) change. Although, the contrast of the petahertz switching would remain in the range of 1 to \(25 \mathrm{nA}\) ."  
+
+## Comments on C)  
+
+Increase of the photoconductivity with intensity is expected due to excitation of carriers into the conduction band, this is also confirmed with the model calculations which seems reasonable. This gives a good metric for determining the efficiency of the logic circuit since it is expected to operate at high field intensities.  
+
+We would like to thank the reviewer for illustrating our results and related applications.
+
+<--- Page Split --->
+
+## Comments on D)  
+
+The technological readiness of this device would be clearer when the comments for the extraction of the field- induced current is addressed. The fabrication of the device also appears to be able to be scaled up given that a high- power scanning laser could produce the tunneling junction required. It would be helpful to give examples of methods to distinguish the ultrafast- intra- band current response from the slower overall inter- band current. As in how would one suppress the \(I_{\mathrm{E}}\) signal which changes sign to achieve an ultrafast logic gate states as described in table 1 and 2.  
+
+We would like to thank the reviewer for the helpful suggestion, which we have incorporated into our revised manuscript on page 7. The revised text now reads:  
+
+\(\Rightarrow\) "Furthermore, the petahertz logic gate can be demonstrated by using \(\mathbf{I}_{\mathrm{E}}\) . In this case, the delay between two laser pulses \(\tau\) is the input signal and the total \(\mathbf{I}_{\mathrm{E}}\) is the output signal (see Fig. 2b). When the delay between the two pulses \(\tau = 360\) as, the output \(\mathbf{I}_{\mathrm{E}}\) signal \(= 0\) . Alternatively, when the delay between the two pulses \(\tau = 0\) as, the output \(\mathbf{I}_{\mathrm{E}}\) has 29 nA (see Fig. 2 a) which present the status 1, demonstrating the NOT gate as summarized in table 4, open the door for establishing ultrafast optical computers."  
+
+and we provided and the new suggested example in table 4  
+
+Table 4   
+
+<table><tr><td rowspan="2">Signal A:<br>Delay τ</td><td colspan="2">Signal B:</td></tr><tr><td>IE</td><td></td></tr><tr><td>OFF</td><td>0</td><td>ON</td></tr><tr><td>ON</td><td>1</td><td>OFF</td></tr></table>  
+
+Table 4. Demonstration of the NOT logic gate. \(\mathbf{I}_{\mathrm{E}}\) is ON when the two pulse induced laser pulses are perfectly overlapped ( \(\tau = 0\) ). If the pulses are delayed by 630 as the output \(\mathbf{I}_{\mathrm{E}}\) is zero.  
+
+We sincerely thank the reviewer for their thoughtful comments and valuable suggestions, which have been instrumental in improving our manuscript. We believe the revisions have greatly enhanced the clarity and presentation of our work. We hope the revised version meets the reviewer's expectations, and we kindly request their recommendation for publication.
+
+<--- Page Split --->
+
+## Response to reviewers' comments - NCOMMS-24-50952A  
+
+The referees' comments are shown in bold and Italics, followed by the corresponding response in blue and the change in the revised version.  
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors adequately replied to many of my comments. I now understand physics better. The directional current in Fig. 1b is produced due to the unequal laser intensities on the two graphene- silicon interfaces. The offset of \(75\mathrm{nA}\) in Fig. 2b is canceled by an external DC voltage, and therefore the direction of the measured current ( \(29\mathrm{nA}\) amplitude) could be reversed. On the other hand, there are two remaining points which should be made clearer before publication.  
+
+1. The structure of the sample is still not clear to me. The authors removed graphene and SiO2 and exposed the silicon substrate. How can it make a graphene-silicon interface? From a three-dimensional perspective, isn't there at least \(90\mathrm{nm}\) distance ( \(= \mathrm{SiO}_2\) thickness) between graphene and silicon?  
+
+The formation of the Si junction is achieved by focusing an intense laser beam on the center of the chip to deplete the graphene and \(\mathrm{SiO}_2\) layers. The chip, mounted on a translation stage, is then moved along the x- axis. The laser beam follows a Gaussian intensity distribution, with the highest intensity at its peak, gradually decreasing toward the edges. As a result, depletion occurs in an arc shape, corresponding to the most intense region of the beam. Additionally, we anticipate that the Si melts and fills this arc due to the high laser power. Consequently, the graphene and \(\mathrm{SiO}_2\) layers may be connected, as confirmed by our Raman spectroscopy. The illustration below depicts the formation of the junction.  
+
+![](images/Extended_Data_Figure_1.jpg)
+  
+
+We appreciate the reviewer's comment! In response, we have clarified this point in the revised version of the SI and included this illustration in Fig. S1.
+
+<--- Page Split --->
+
+2. Regarding the 'first-order' cross-correlation, I mentioned 'cross-correlation' at a different place, which is in the sentence 'In our calculation, ... cross-correlation of the laser fields'. The question was asked because it was a bit of a surprise to me that the observed result in Fig. 2b can well be approximated by the first-order cross-correlation of the laser fields, meaning that the switching speed of the device is solely determined by the cycle period time of the laser field.  
+
+The authors excluded the contribution of optical interference (reply to reviewer #3) but the revised manuscript states that \(7.5 \mathrm{nA}\) ( \(= 10\%\) of the \(75 \mathrm{nA}\) ) of the \(29 \mathrm{nA}\) can come from the optical interference. On this point, the authors need to prove more carefully that the dominant effect is still the tunning, not the optical interference.  
+
+## My specific comments are  
+
+(2- a) An explanation should be added as to how and where the \(10\%\) power modulation was measured. Since two non- collinear beams can produce delay- dependent interference fringes in space, the local intensities on the graphene- silicon interfaces can be modulated by more than \(10\%\) . Since the current is driven by unequal local intensities, the interference fringes may cause a strong influence. Did the authors exclude this possibility, too?  
+
+We would like to thank the reviewer for their constructive comment!  
+
+Regarding the power modulation measurement, it was conducted in the same plane as the graphene transistor, where the measurements presented in Fig. 2a were taken. In response to the reviewer's comment, we have included this information in the revised version of the manuscript on page 5.  
+
+On another note, we replaced the phototransistor by a power meter, and we observed only a \(10\%\) oscillation in the power between the two pulses at the temporal overlap, indicating a minor contribution of the current amplitude oscillation (Fig. 2b) potentially originated from the optical interference.  
+
+Regarding spatial interference, its effect is minimal due to the very small angle ( \(< 5^{\circ}\) ) between the two beams. To verify this, we measured the beam profile using a beam profiler at the temporal overlap (at the focus plane, where the experiment takes place). The snapshots show no significant interference fringes in space (red circle).  
+
+Please note that the beam structure highlighted by the white square results from using a parabolic mirror to focus the beam. The astigmatism effect  
+
+![](images/Extended_Data_Figure_1.jpg)
+  
+
+occurs if the beam is not perfectly perpendicular to the mirror. However, this does not affect the current measurement since the focused beam in the red circle almost entirely covers the area of the graphene transistor.  
+
+Additionally, to demonstrate the minor contribution from optical interference, we conducted the experiment using a chirped laser pulse. The results are now presented in Extended Data Fig. 1 of the revised manuscript. Here, the current modulation is asymmetric at positive and negative delay sides, indicating that optical interference has only a minor contribution. If optical interference were the dominant effect, one would expect perfectly symmetric modulations on both sides (negative and positive).
+
+<--- Page Split --->
+![](images/Figure_unknown_1.jpg)
+
+<center>Extended Data Fig. 1. The measured cross-correlation \(\mathrm{I_E}\) current modulation driven by a field of chirped few-cycle pulse. </center>  
+
+(2- b) Is the noncollinear geometry considered in the theory with the dipole approximation? If not, the theory gives a strong optical interference effect and cannot support the interpretation. Additional explanation is needed on this point.  
+
+In our theory simulations, we used a simple theoretical model that can only take the amplitude and the polarization of the applied laser fields into account. Therefore, we cannot simulate scenarios like the noncollinear geometry requested by the reviewer. Furthermore, we believe these additional layers of complexity will not provide anything new for understating the nature of the observed phenomena. Taking more effects into consideration will only complicate the Hamiltonian of the system. However, all the observed physics comes from the analysis of the wavefunction/density matrix which still will contain information about both the optical interference and the generated tunneling current, as well as their possible dependence on each other.  
+
+Yet, there is a simple way we can distinguish the electric current generated due to the optical interference from that arising from the tunneling effect. We thank the reviewer for their stimulating comment which inspired us to perform additional calculations discussed in the revised main text page 6 and the last paragraph of the Methods section which we copy below.  
+
+"The persistent electric current obtained via solution of Eqs. (1) and (7) with the field synthetized from the two delayed Gaussian waveforms can, in principle, be attributed to two concurrent effects: optical interference of the applied fields and electron tunnelling through Gr- Si junction. To decipher these two mechanisms from each other, we performed additional simulations selectively switching off appropriate terms in Eq. (1) and analysing the obtained current. Therefore, in addition to two terminal cases described above, (a): full system with possibility of tunnelling and optical interference (Extended Data Fig. 3a), and (d): pure graphene with no junction (Extended Data Fig. 3d), we created two intermediate scenarios where we (b) prevent photon absorption by neglecting \(\mathbf{E}(\mathbf{t})\cdot \{\mathbf{D}(\mathbf{k}_t),\pmb {\rho}(\mathbf{k},\mathbf{t})\}_{m,n}\) term. This way we exclude any optical interference effects since the obtained density matrix will not, by definition, contain any contributions coming from the interaction with an optical field (Extended Data Fig. 3b).
+
+<--- Page Split --->
+
+Alternatively, we (c) prevent field acceleration of electrons by prohibiting changes of the crystal momentum frame in time: \(\mathbf{k}_{\mathrm{f}} = \mathbf{k}_{0}\) . This way we exclude tunnelling effects since the electrons will not be able to move through the Gr- Si junction (Extended Data Fig. 3c). As one can see, exclusion of the optical excitations and thus the optical interference from the consideration (Extended Data Fig. 3b) has a minor effect on the possibility to generate the electric current. This is because the number of electrons driven to the conduction band by the photon absorption is negligible in comparison to those already present there due to the thermal effects (see Eq. (3)) or originating from the Landau- Zener transitions. At the same time, excluding acceleration of electrons by the field and thus dramatically reducing the possibility of tunnelling, we see nearly complete suppression of the generated current (Extended Data Fig. 3c)."  
+
+We would like to thank the reviewer once again for their time and valuable comments, which have helped us improve our presentation and simulation. We sincerely appreciate the reviewer's recommendation to publish our manuscript in Nature Communications
+
+<--- Page Split --->
+
+Reviewer #3 (Remarks to the Author):  
+
+The follow up comments below should be addressed before publication can be recommended.  
+
+The main claims of the manuscript is restated below for reference, follow up comments follows.  
+
+A) The photoresponse of the planar-graphene tunneling phototransistor designed in this study is a result from the tunneling of photoexcited carriers between the graphene layers of a Gr-Si-Gr transistor.  
+
+B) Two components contributing to the total photoresponse are the direct tunneling of photoexcited charge carriers and the field-induced tunneling current which is carrier motion driven directly by the light field itself. The latter is attosecond scale response.  
+
+C) There is a notable photoconductivity enhancement at higher light field strength  
+
+D) This phototransistor is at the technology readiness level for developing attosecond and lightwave quantum optoelectronics.  
+
+## Follow up Comments on A)  
+
+The added scale for device in the revised manuscript is helpful. However, the overall geometry of the device is still unclear. I am unsure as to why the Graphene would be in contact with the silicon since the graphene should be on top of the SiO2 from the fabrication process. A cross-section schematic of the structure would be helpful here (see attached example). As for the Raman spectroscopy, while the silicon and graphene peaks are shown, that could potentially mean both the graphene and silicon are within the beam spot of the laser, but they could be vertically displaced.  
+
+The formation of the \(\mathrm{SiO_2}\) junction is achieved by focusing an intense laser beam on the center of the chip to deplete the graphene and \(\mathrm{SiO_2}\) layers. The chip, mounted on a translation stage, is then moved along the x- axis. The laser beam follows a Gaussian intensity distribution, with maximum intensity at its peak, gradually decreasing toward the edges. As a result, depletion occurs in an arc shape, corresponding to the most intense region of the beam.  
+
+Additionally, we anticipate that the Si melts and fills this arc due to the high laser power. Consequently, the graphene and \(\mathrm{SiO_2}\) layers may be connected, as confirmed by our Raman  
+
+![PLACEHOLDER_23_0]
+  
+
+spectroscopy scan. The illustration below depicts the formation of the junction.  
+
+We appreciate the reviewer's comment! In response, we have clarified this point in the revised version of the SI and included the cross- sectional illustration in Fig. S1.  
+
+"The \(\mathrm{SiO_2}\) junction is formed by a laser- induced depletion process. A focused laser beam with a Gaussian intensity profile is incident on the center of the chip, targeting the multilayer structure of graphene and \(\mathrm{SiO_2}\) . As the chip is translated along the x- axis, the central, high- intensity portion of the laser beam interacts with the material, causing localized depletion. This depletion results in a arc- shaped region where the graphene and \(\mathrm{SiO_2}\) layers are effectively removed (see cross- section illustration in Fig. S1). The formation of this junction is further supported by Raman spectroscopy scans, as we explain in the next section".
+
+<--- Page Split --->
+
+## Follow up Comments on B)  
+
+Since the light is exciting both graphene and silicon, is there a difference between the carriers tunneling from silicon to graphene vs graphene tunneling to silicon?  
+
+The tunneling direction depends on the direction of the excited field. As a result, electrons first tunnel from one side of the graphene to the silicon and then from the silicon to the opposite side of the graphene. This explains the observed current modulation in both positive and negative directions, as shown in Fig. 2b—otherwise, the current would be unidirectional. The measurements presented in this work suggest no significant difference between Si/Gr and Gr/Si tunneling.  
+
+It is also worth noting that the valence and conduction bands in graphene meet at the Dirac point, whereas silicon has a band gap of 1.12 eV. Consequently, the number of electrons in graphene's conduction band—both thermal electrons and those generated by Landau- Zener transitions—is significantly higher than the number of electrons that could be excited in silicon.  
+
+Also, in reference to the polarization testing mentioned by reviewer #1. It is not immediately clear why the circular polarization does not produce the same variation with since you can consider it as two orthogonal linear polarizations (one parallel to the stripe, and another perpendicular to the stripe).  
+
+This is because the test was conducted by rotating the quarter- wave plate while keeping the input beam power constant. As a result, the power is distributed between the two linear polarization components, leading to a lower induced- field current signal.  
+
+Lastly the linear polarization dependence is strong experimental evidence for this effect being field driven and should be included in the main text.  
+
+We appreciate the reviewer's suggestion! We have revised the manuscript and added the following statement on page 5:  
+
+"In addition, the induced effect (IE) is polarization- dependent, as the signal diminishes when the input linearly polarized beam is converted to circularly polarized light."  
+
+Additionally, we have included Extended Data Fig. 1, which shows asymmetric current modulation induced by a chirped pulse, as further evidence for the field- driven current measurement.  
+
+![PLACEHOLDER_24_0]
+
+<center>Extended Data Fig. 1. The measured cross-correlation \(\mathrm{I_E}\) current modulation driven by a field of chirped few-cycle pulse. </center>  
+
+Finally, we would like to thank the reviewer once again for their valuable comments during both the first and second rounds. We hope we have adequately addressed all the raised questions and comments. We appreciate the reviewer's recommendation to accept our manuscript for publication in Nature Communications.
+
+<--- Page Split --->
+
+Response to reviewers' comments – NCOMMS- 24- 50952B  
+
+The referees' comments are shown in bold and Italics, followed by the corresponding response in blue and the change in the revised version.  
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors have adequately responded to my last comments. But the new data, Extended Data 1, is difficult to understand. If my understanding is correct, the two pulses are nearly identical. If so, the same signals must be observed at plus and minus delay times because the two pulses are indistinguishable regardless of the chirp and the nonlinearity involved. If the two pulses are significantly unequal, the new result would not strongly support the claim. Additional explanation(s) should be added.  
+
+The measurement shown in Extended data Fig. 1 was performed using a 6.5 fs pulse (pulse 1) in one beam and a chirped pulse (pulse 2) in the other beam. We deconvoluted the measured trace in Extended Fig. 1 and plotted the results in the following figure: in (a) the waveform of the 6.5 fs pulse (pulse 1) and in (b) the chirped pulse (pulse 2). The calculated cross- correlation of pulse 1 & 2 (red line in (c)) is in acceptable agreement with the measured current modulation in Extended Data Fig. 1.  
+
+![PLACEHOLDER_25_0]
+
+<center>Fig. R1: The retrieved waveforms of the two pulses used to obtain the cross-correlation current modulation in Extended data Fig. 1: (a) pulse 1 (6.5 fs) (b) pulse 2 (chirped pulse). (c) The calculated cross-correlation of the two pulse is plotted in red in comparison with the measured current modulation presented in the Extended Data Fig. 1. </center>  
+
+We revised the manuscript accordingly to clarify this point in page 3  
+
+“Furthermore, we measured the \(I_{E}\) cross- correlation using a chirped input pulse in one beam and the 6.5 fs pulse in the other beam”
+
+<--- Page Split --->
+
+## Reviewer #3 (Remarks to the Author):  
+
+The comments on two of the main claims in the manuscripts was addressed in the previous rounds of comments. However more clarification is needed before publication to support the following two claims listed below (restated from the first round of comments for reference).  
+
+A) The photoresponse of the planar-graphene tunneling phototransistor designed in this study is a result from the tunneling of photoexcited carriers between the graphene layers of a Gr-Si-Gr transistor.  
+
+B) Two components contributing to the total photoresponse are the direct tunneling of photoexcited charge carriers and the field-induced tunneling current which is carrier motion driven directly by the light field itself. The latter is attosecond scale response.  
+
+(Follow up comments on A)  
+
+Both reviewer #1 and myself asked for clarification of the structure, which the authors has provided. While the provided explanation could explain the formation of the structure it is only a speculation on the melting of the Si. Perhaps an AFM measurement of the surface can show this effect.  
+
+We tried to perform the AFM measurement earlier however the structure of the commercial transistor (shown in the following photo) is covered by a black protection layer. The height of this layer prevented the AFM tip to reach the surface of graphene. We tried to remove the black protection layer without damaging the graphene structure, but we failed to remove it. Worth mentioning, our explanation of the structure is supported with the performed and provided Raman measurements and results.  
+
+## (Follow up comments on B)  
+
+A clarification on the asymmetric time delayed data using a chirped pulse, and a demonstration of the time- delay photocurrent data with different polarization is needed to support this effect being field driven.  
+
+A more clear explanation on why an asymmetric current modulation implies a field driven effect is needed. This asymmetry suggests the charge carriers cares about which of the two identical pulses arrives first, which intuitively even with a field driven affect should not occur.  
+
+We refer the reviewer to our answer to the reviewer 1 comment on the same point, in page 1.  
+
+Data showing the diminished tunneling current with respect to excitation by a pulse with linear- polarization parallel or semi parallel to the channel as mentioned in a previous rounds of comments should be included in the manuscript for this would be one of the hallmarks of this effect being field driven.  
+
+We included the requested data in the revised version in Fig. S3.  
+
+![PLACEHOLDER_26_0]
+
+<center>Fig. S3: The measured modulated current using circular polarized pulse. </center>
+
+<--- Page Split --->
+
+## Response to reviewers' comments - NCOMMS-24-50952C  
+
+The referees' comments are shown in bold and Italics, followed by the corresponding response in blue and the change in the revised version.  
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author).  
+
+Regarding the new information that a combination of chirped and un- chirped pulses was used for Extended Data 1, I cannot follow why and how the asymmetry in the data is connected to their conclusion of "minor effect from the optical interference". Optical interference can give asymmetric data because one of the pulses is asymmetric in time. A clear explanation should be added before publication.  
+
+\(\Rightarrow\) We agree with the reviewer that the asymmetry observed in the measurement shown in Extended Data Fig. 1 of the previous version (now Supplementary Fig. 4) may not fully rule out the possibility of an interference effect, which we acknowledged in our discussion in the main text. However, the results suggest that the measured current is field- sensitive. In response to the reviewers' suggestions, we have added the requested explanation to the revised manuscript (page 5).  
+
+"Furthermore, we measured the \(I_{\mathrm{E}}\) cross- correlation using a chirped input pulse in one beam and the 6.5 fs pulse in the other beam and the results are shown in Supplementary Figure 4 indicating that the \(I_{\mathrm{E}}\) is field sensitive."  
+
+In addition, the authors should describe in the method section how the pulse was chirped.  
+
+\(\Rightarrow\) We have considered the reviewer's suggestion, and the following sentence has been added to the Methods section page 9:  
+
+"For the measurements shown in Supplementary Figure 4, one of the pulses is chirped by propagating it through a thick piece of dispersive fused silica".  
+
+If the asymmetry in data cannot exclude the possibility of optical interference, the measurement result with a power meter ( \(10\%\) modulation) is the only direct evidence. To strengthen the authors' claim, I suggest adding a description of how this number ( \(10\%\) ) is translated into the amount of modulation of I E. If the modulation of I E is proportional to the laser intensity, I E should then be modulated by \(10\%\) due to optical interference. However, if I E varies nonlinearly with laser intensity, optical interference causes stronger modulation.  
+
+\(\Rightarrow\) As shown in Fig. 3, the measurements were conducted in the linear regime, where power (P) is directly proportional to intensity (I), according to the relation \(\mathrm{I} = \mathrm{P} / \mathrm{A}\) , where A is the area of the beam. Consequently, a \(10\%\) change in power results in a corresponding \(10\%\) change in intensity.
+
+<--- Page Split --->
+
+In response to the reviewer request we added the following in the revised version of the manuscript (page 5):  
+
+"Moreover, the measurements were conducted at low power in the linear regime, where power is directly proportional to intensity. Hence, we can estimate that \(10\%\) of the current modulation (shown in Fig. 2b) may come from the optical interface and the Ip change."  
+
+\(\spadesuit\) Lastly, although the authors replied to referee #3 and me previously that the connection between graphene and Si was confirmed by Raman measurements, I cannot find any proof that the SiO2 layer was removed. There is no Raman peaks assigned to SiO2 in Fig. S2. The authors should add an explanation before publication. If the removal of the SiO2 layer is a hypothetical level, it should be expressed as such.  
+
+\(\Rightarrow\) We considered the reviewer's suggestion and revised the corresponding section (Note 2) in the supplementary information to clarify the Raman measurement results. The updated version can now be found on page 8 of the SI:  
+
+"The graphene peaks are entirely absent in the spectrum of the junction area in the modified device, indicating a clear displacement of the graphene layer and likely suggesting the removal of the \(\mathrm{SiO_2}\) layer, thereby exposing the silicon substrate."
+
+<--- Page Split --->
+![PLACEHOLDER_29_0]
+
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__ad32fa1cc3a82ad82eec24ebd7c298574764a6e20564f3fb52074f431b1398f1/images_list.json

@@ -0,0 +1,205 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Fig. R1: Phylogenetic trees of the Oryza genus. allotetraploid wild rice genomes were split into sub-genomes to construction and the single copy genes were concatenated to generate the phylogenetic tree. Numbers at nodes indicate the median value for the divergence time (Mya) estimates for each clade.",
+    "footnote": [],
+    "bbox": [
+      [
+        200,
+        84,
+        850,
+        495
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Fig R2: Phylogenetic tree based on the chloroplast genome sequences. a. Phylogenetic tree based on 17 wild rice species chloroplast genome sequences (Z. may as outgroup). b. Phylogenetic tree constructed based on 18 wild rice (including a BB genome type O. punctata, KF359908) species chloroplast genome sequences.",
+    "footnote": [],
+    "bbox": [
+      [
+        155,
+        87,
+        842,
+        737
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Fig. R3: Characteristics of gene CNVs related to important agronomic traits in the Oryza genus super pangenome. Circle size shows the number of gene copies, color from grey to yellow represents the gene expression.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        490,
+        848,
+        755
+      ]
+    ],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Fig. R4: Collinearity of NLRs in various Oryza genomes within a specific region.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        560,
+        850,
+        735
+      ]
+    ],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "Fig. R5: A NLR synteny region and RNA evidence of the annotated NLR genes. The pink color in O. glumaepatula, O. minuta|Bt, O. latifolia NLR gene were validated by the RNA-seq in IGVtools.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        327,
+        850,
+        583
+      ]
+    ],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_5.jpg",
+    "caption": "Fig. R6: The Oryza genus super-pangenome constructed by using default parameters.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        313,
+        855,
+        670
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_6.jpg",
+    "caption": "Fig. R7: Landscape of the centromere in wild rice O. rhizomatis.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        280,
+        805,
+        600
+      ]
+    ],
+    "page_idx": 24
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_7.jpg",
+    "caption": "Fig. R1: Phylogenetic trees of the Oryza genus. allotetraploid wild rice genomes were split into sub-genomes to construction and the single copy genes were concatenated to generate the phylogenetic tree. Numbers at nodes indicate the median value for the divergence time (Mya) estimates for each clade.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 31
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_8.jpg",
+    "caption": "Fig R2: Phylogenetic tree based on the chloroplast genome sequences. a. Phylogenetic tree based on 17 wild rice species chloroplast genome sequences (Z. may as outgroup). b. Phylogenetic tree constructed based on 18 wild rice (including a BB genome type O. punctata, KF359908) species chloroplast genome sequences.",
+    "footnote": [],
+    "bbox": [
+      [
+        155,
+        87,
+        844,
+        737
+      ]
+    ],
+    "page_idx": 55
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_9.jpg",
+    "caption": "Fig. R3: Characteristics of gene CNVs related to important agronomic traits in the Oryza genus super pangenome. Circle size shows the number of gene copies, color from grey to yellow represents the gene expression.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        490,
+        848,
+        755
+      ]
+    ],
+    "page_idx": 58
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_10.jpg",
+    "caption": "Fig. R4: Collinearity of NLRs in various Oryza genomes within a specific region.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        562,
+        850,
+        733
+      ]
+    ],
+    "page_idx": 60
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_11.jpg",
+    "caption": "Fig. R5: A NLR synteny region and RNA evidence of the annotated NLR genes. The pink color in O. glumaepatula, O. minuta|Bt, O. latifolia NLR gene were validated by the RNA-seq in IGVtools.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        327,
+        850,
+        583
+      ]
+    ],
+    "page_idx": 61
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_12.jpg",
+    "caption": "Fig. R6: The Oryza genus super-pangenome constructed by using default parameters.",
+    "footnote": [],
+    "bbox": [
+      [
+        149,
+        313,
+        855,
+        670
+      ]
+    ],
+    "page_idx": 63
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_13.jpg",
+    "caption": "Fig. R7: Landscape of the centromere in wild rice O. rhizomatis.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        280,
+        805,
+        599
+      ]
+    ],
+    "page_idx": 70
+  }
+]

peer_reviews/supplementary_0_Peer Review file__b66bb4455ed4ce1821ad8154c79ec275ffb744ac8ee26be17226c6274257028b/images_list.json

@@ -0,0 +1,473 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Fig. 2 Spatiotemporal gene expression during six stages of cotton fiber development.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_8.jpg",
+    "caption": "Supplementary Fig. 8 Spatial transcriptomic analysis of 1-DPA cotton boll. a Left, t-SNE projection plot shows the clusters in a 1-DPA cotton boll. Right, the corresponding spatial map shows different clusters in the sample. The cell type",
+    "footnote": [],
+    "bbox": [
+      [
+        155,
+        92,
+        901,
+        840
+      ]
+    ],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4 Pseudotime analysis of all cotton ovules samples from \\(-1.5\\) -DPA to 1-DPA. a A pseudotime trajectory shows the distribution of cell clusters. The dashed arrow indicates the developmental direction of the pseudotime trajectory. b Proportion of pseudotime states in the four cotton ovule samples. c Pseudotime trajectories colored according to tissue type. d Heatmap clustering of key genes for developmental trajectories in all samples and",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1 A molecular approach towards fiber development atlas in cotton.",
+    "footnote": [],
+    "bbox": [
+      [
+        155,
+        370,
+        850,
+        707
+      ]
+    ],
+    "page_idx": 18
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_12.jpg",
+    "caption": "Supplementary Fig. 12 Cluster annotation of scRNA-seq data from cotton ovules at different developmental stages. (This figure is related to Fig. 3)",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4 Pseudotime analysis of all cotton ovules samples from \\(-1.5\\) -DPA to 1-DPA. a A pseudotime trajectory shows the distribution of cell clusters. The dashed arrow indicates the developmental direction of the pseudotime trajectory. b Proportion of pseudotime states in the four cotton ovule samples. c Pseudotime trajectories colored according to tissue type. d Heatmap clustering of key genes for developmental trajectories in all samples and",
+    "footnote": [],
+    "bbox": [
+      [
+        155,
+        90,
+        820,
+        789
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_7.jpg",
+    "caption": "Fig. 7 Spatiotemporal expression pattern of integument cell marker genes revealed by RNA in situ hybridization (ISH).",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        319,
+        850,
+        794
+      ]
+    ],
+    "page_idx": 25
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_27.jpg",
+    "caption": "Supplementary Fig. 27 RNA in situ hybridization (ISH) of SWEET15 and DOX2 during the early stages of cotton fiber development. (This figure is related to Fig. 7)",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        199,
+        883,
+        715
+      ]
+    ],
+    "page_idx": 26
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_28.jpg",
+    "caption": "Supplementary Fig. 28 Characterization of RNAi and overexpression lines of GhBEE3. (This figure is related to Fig. 8)",
+    "footnote": [],
+    "bbox": [
+      [
+        188,
+        110,
+        860,
+        700
+      ]
+    ],
+    "page_idx": 29
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_4.jpg",
+    "caption": "Supplementary Fig. 4 Verification of robustness of ST dataset. (This figure is related to Fig. 2)",
+    "footnote": [],
+    "bbox": [
+      [
+        156,
+        290,
+        909,
+        630
+      ]
+    ],
+    "page_idx": 31
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_13.jpg",
+    "caption": "Supplementary Fig. 13 Verification of robustness of scRNA-seq dataset. (This figure is related to Fig. 3)",
+    "footnote": [],
+    "bbox": [
+      [
+        155,
+        103,
+        860,
+        855
+      ]
+    ],
+    "page_idx": 34
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Fig. Global Spatiotemporal Analysis of Three Cardiac Developmental Stages (D) Dimensionality reduction and clustering of 3,115 spots from 19 tissue sections of 4.5–5, 6.5, and 9 PCW human embryonic hearts. Each cluster’s annotated anatomical region is indicated below. (F) Mapping of data points from all ten clusters to their spatial positions show that clusters are localized to distinct histological regions. Magnified images of the histological structures are shown in F1–F3. Scale bars, 100 mm. Asp M, Giacomello S, Larsson L, Wu C, Fürth D, Qian X, Wärdell E, Custodio J, Reimegård J, Salmén F, Österholm C, Ståhl PL, Sundström E, Åkesson E, Bergmann",
+    "footnote": [],
+    "bbox": [
+      [
+        156,
+        536,
+        836,
+        725
+      ]
+    ],
+    "page_idx": 35
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Fig. Single-cell mRNA (scRNA-seq) sequencing overview",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 37
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_7.jpg",
+    "caption": "Fig. 7 Spatiotemporal expression pattern of integument cell marker genes revealed by RNA in situ hybridization (ISH).",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 48
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_27.jpg",
+    "caption": "Supplementary Fig. 27 RNA in situ hybridization (ISH) of SWEET15 and DOX2 during the early stages of cotton fiber development. (This figure is related to Fig. 7)",
+    "footnote": [],
+    "bbox": [
+      [
+        160,
+        100,
+        884,
+        610
+      ]
+    ],
+    "page_idx": 49
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_28.jpg",
+    "caption": "Supplementary Fig. 28 Characterization of RNAi and overexpression lines of GhBEE3. (This figure is related to Fig. 8)",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 50
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_4.jpg",
+    "caption": "Supplementary Fig. 4 Verification of robustness of ST dataset. (This figure is related to Fig. 2)",
+    "footnote": [],
+    "bbox": [
+      [
+        156,
+        110,
+        909,
+        430
+      ]
+    ],
+    "page_idx": 51
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_13.jpg",
+    "caption": "Supplementary Fig. 13 Verification of robustness of scRNA-seq dataset. (This figure is related to Fig. 3)",
+    "footnote": [],
+    "bbox": [
+      [
+        152,
+        103,
+        864,
+        855
+      ]
+    ],
+    "page_idx": 53
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1 A molecular approach towards fiber development atlas in cotton.",
+    "footnote": [],
+    "bbox": [
+      [
+        153,
+        333,
+        850,
+        670
+      ]
+    ],
+    "page_idx": 55
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_7.jpg",
+    "caption": "Fig. 7 Spatiotemporal expression pattern of integument cell marker genes revealed by RNA in situ hybridization (ISH).",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 57
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_27.jpg",
+    "caption": "Supplementary Fig. 27 RNA in situ hybridization (ISH) of SWEET15 and DOX2 during the early stages of cotton fiber development. (This figure is related to Fig. 7)",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        255,
+        884,
+        770
+      ]
+    ],
+    "page_idx": 58
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_2.jpg",
+    "caption": "Supplementary Fig. 2 Detection map of penetration of cotton boll samples in ST.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        95,
+        840,
+        565
+      ]
+    ],
+    "page_idx": 60
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Pearson correlation heatmap of the clusters obtained by Seurat v4 and Seurat v3 (row column non clustering).",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 61
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Pearson correlation heatmap of the clusters obtained by Seurat v4 and Seurat v3 (row column clustering).",
+    "footnote": [],
+    "bbox": [
+      [
+        255,
+        488,
+        747,
+        830
+      ]
+    ],
+    "page_idx": 68
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1 A molecular approach towards fiber development atlas in cotton.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 69
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_72.jpg",
+    "caption": "Supplementary Fig. 72: Comparison of the analysis results of the new Seurat4/5 and the original Seurat3 in ST.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 72
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "b",
+    "footnote": [],
+    "bbox": [
+      [
+        198,
+        145,
+        890,
+        330
+      ]
+    ],
+    "page_idx": 73
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_5.jpg",
+    "caption": "Seurat4/ Seurat5",
+    "footnote": [],
+    "bbox": [
+      [
+        179,
+        365,
+        530,
+        601
+      ]
+    ],
+    "page_idx": 74
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_6.jpg",
+    "caption": "c",
+    "footnote": [],
+    "bbox": [
+      [
+        540,
+        365,
+        890,
+        601
+      ]
+    ],
+    "page_idx": 75
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_25.jpg",
+    "caption": "Supplementary Fig. 25: Identification of the representative metabolites by targeted metabolite analysis.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 83
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_18.jpg",
+    "caption": "Supplementary Fig. 18: Statistical analysis of the signal density of representative metabolite in embryo samples.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 84
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_7.jpg",
+    "caption": "Fig. R1: Identification of the representative metabolites by targeted metabolite analysis.",
+    "footnote": [],
+    "bbox": [
+      [
+        156,
+        90,
+        910,
+        415
+      ]
+    ],
+    "page_idx": 84
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_16.jpg",
+    "caption": "Supplementary Fig. 16: Spatial metabolomic analysis of cotton bolls from \\(-1.5\\) -",
+    "footnote": [],
+    "bbox": [
+      [
+        156,
+        99,
+        833,
+        767
+      ]
+    ],
+    "page_idx": 85
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_33.jpg",
+    "caption": "Supplementary Fig. 33: Differentially produced metabolites (DPMs) in 0-DPA cotton ovules compared with -1.5-DPA ovules.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        280,
+        840,
+        844
+      ]
+    ],
+    "page_idx": 85
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_34.jpg",
+    "caption": "Supplementary Fig. 34: Differentially produced metabolites (DPMs) in 3-DPA",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        316,
+        833,
+        870
+      ]
+    ],
+    "page_idx": 86
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_35.jpg",
+    "caption": "Supplementary Fig. 35: Identification of the representative metabolites by targeted metabolite analysis.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        92,
+        844,
+        759
+      ]
+    ],
+    "page_idx": 87
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_32.jpg",
+    "caption": "Supplementary Fig. 32: PCA analysis of metabolites in ovule at different developmental stages under positive and negative ion model. (This figure is related to Fig. 5)",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        100,
+        833,
+        715
+      ]
+    ],
+    "page_idx": 87
+  }
+]

peer_reviews/supplementary_0_Peer Review file__c2a72f84b368a919ac8ce5e3d5f8d13c56c21232e44ec50854b466f782040683/images_list.json

@@ -0,0 +1,608 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_57.jpg",
+    "caption": "Supplementary Fig. 57 \\(^1\\mathrm{H}\\) NMR spectral recorded for \\(\\mathrm{PEO}_{45 - b}\\mathrm{-P}(\\mathrm{Cy}_{73 - \\mathrm{CO}}\\mathrm{-TPE}_{34})\\) in tetrahydrofuran- \\(d_8 / \\mathrm{D}_2\\mathrm{O}\\) (1/1, v/v) in the absence (black line) and presence (red line) of \\(\\mathrm{SO}_2\\) , respectively.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_58.jpg",
+    "caption": "Supplementary Fig. 58 UV-Vis spectral monitoring the variation of \\(\\mathrm{PEO}_{45 - b}\\mathrm{-P}(\\mathrm{Cy}_{73 - \\mathrm{co}}\\) - \\(\\mathrm{TPE}_{34}\\) ) in the presence of \\(\\mathrm{SO}_2\\) gas stimulus at various concentrations (0 to \\(2\\mathrm{mM}\\) ) in THF/H2O (1/1, v/v).",
+    "footnote": [],
+    "bbox": [
+      [
+        272,
+        84,
+        725,
+        343
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Figure R1. (a) UV-Vis spectral variation of \\(\\mathrm{PEO_{45 - b - P(Cy_{73} - co - TPE_{34})}}\\) under increasing \\(\\mathrm{SO}_2\\) concentrations (0 to \\(8\\mathrm{mM}\\) , free blue to red curve) in \\(\\mathrm{CH_2Cl_2}\\) at a concentration of \\(1.0\\mathrm{mg / mL}\\) \\((2.6\\times 10 - 5\\mathrm{M}\\) , containing \\(\\sim 2\\mathrm{mM}\\) of \\(\\mathrm{SO}_2\\) -responsive Cy units). (b) UV-Vis spectral showing the reversible Ino-to-Cy conversion in the copolymer with \\(\\mathrm{H}_2\\mathrm{O}_2\\) peroxide treatment (0 to 20 mM). (c) UV-Vis spectral changes of \\(\\mathrm{PEO_{45 - b - P(Cy_{73} - co - TPE_{34})}}\\) copolymer in the presence of a group of endogenous analytes with similar nucleophilic ability including Cl, Br, I, \\(\\mathrm{S}_2\\mathrm{O}_3^{2 - }\\) \\(\\mathrm{NO}_3^-\\) , \\(\\mathrm{HPO}_4^{2 - }\\) , \\(\\mathrm{HCO}_3^-\\) , NO, \\(\\mathrm{N}_2\\mathrm{O}\\) , \\(\\mathrm{NH}_3\\) , CO, and \\(\\mathrm{H}_2\\mathrm{S}\\) . All these analytes are at the concentration of \\(160\\mathrm{mM}\\) (20 equiv. with respect to \\(\\mathrm{SO}_2\\) control). (d) Specific differentiation of \\(\\mathrm{PEO_{45 - b - P(Cy_{73} - co - TPE_{34})}}\\) copolymer to \\(\\mathrm{SO}_2\\) biosignal from other endogenous analytes.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Figure R2. TEM images of some bowl-shaped vesicles that can contain particles of sizes smaller than those of the cavities. Scale bar: \\(200 \\mathrm{nm}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        184,
+        408,
+        816,
+        560
+      ]
+    ],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_29.jpg",
+    "caption": "Supplementary Fig. 29 \\(^{1}\\mathrm{H}\\) NMR spectrum of \\(\\mathrm{PEO}_{45} - b\\) -P(Cy \\(_{54}\\) -co-TPE \\(_{5}\\) ) in \\(\\mathrm{CD}_{2}\\mathrm{Cl}_{2}\\) .",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_30.jpg",
+    "caption": "Supplementary Fig. 30 \\(^{1}\\mathrm{H}\\) NMR spectrum of \\(\\mathrm{PEO}_{45} - b\\) -P(Cy \\(_{26}\\) -co-TPE \\(_{56}\\) ) in \\(\\mathrm{CD}_{2}\\mathrm{Cl}_{2}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        160,
+        465,
+        837,
+        793
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_44.jpg",
+    "caption": "Supplementary Fig. 44 The morphology of polymer counterpart \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{78 - c0} - \\mathrm{TPE}_9)\\) . a. TEM image (shallow open mouth on membrane). b. Cross-sectional height analysis to show the overall diameter \\((D)\\) and the depth \\((H)\\) of membrane concaving. c. AFM and d. SEM image. In order to investigate the effect of the copolymer ratio on the assembly morphology, we have also prepared and analyzed the polymer counterpart \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{78 - c0} - \\mathrm{TPE}_9)\\) with a lower copolymer ratio than \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{73 - c0} - \\mathrm{TPE}_{34})\\) . It can be found that when the copolymer ratio was reduced from 0.47 to 0.12, \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{78 - c0} - \\mathrm{TPE}_9)\\) counterpart can form bowl-shaped vesicles, but the membrane concaving is relatively shallow. Their \\(D\\) is of \\(\\sim 497 \\mathrm{nm}\\) , close to that of \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{73 - c0} - \\mathrm{TPE}_{34})\\) ( \\(\\sim 520 \\mathrm{nm}\\) ); however, their \\(H\\) is only determined to be \\(\\sim 170 \\mathrm{nm}\\) , far lower than that of \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{73 - c0} - \\mathrm{TPE}_{34})\\) with deep invagination ( \\(\\sim 398 \\mathrm{nm}\\) ). This geometry corresponds to a shallow invaginated geometry.",
+    "footnote": [],
+    "bbox": [
+      [
+        211,
+        85,
+        783,
+        469
+      ]
+    ],
+    "page_idx": 18
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_45.jpg",
+    "caption": "Supplementary Fig. 45 The morphology of polymer counterpart \\(\\mathrm{PEO}_{45 - b} - \\mathrm{P}(\\mathrm{Cy}_{54} - \\mathrm{co} - \\mathrm{TPE}_5)\\)",
+    "footnote": [],
+    "bbox": [
+      [
+        177,
+        92,
+        835,
+        324
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_46.jpg",
+    "caption": "Supplementary Fig. 46 The morphology of polymer counterpart \\(\\mathrm{PEO}_{45 - b}\\mathrm{-P}(\\mathrm{Cy}_{28}\\mathrm{-co - TPE}_{30})\\) . a. TEM image (deep open mouth on membrane indicated by arrows). b. Cross-sectional height analysis to show the overall diameter \\((D)\\) and the depth \\((H)\\) of membrane concaving. c. AFM and d. SEM image.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        88,
+        845,
+        560
+      ]
+    ],
+    "page_idx": 23
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_47.jpg",
+    "caption": "Supplementary Fig. 47 The morphology of polymer counterpart \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{26 - c0 - }\\mathrm{TPE}_{56})\\) . a. TEM image (deep open mouth on membrane indicated by arrows). b and c. Cross-sectional height analysis to show the overall diameter (D) and the depth (H) of membrane concaving.",
+    "footnote": [],
+    "bbox": [
+      [
+        175,
+        121,
+        833,
+        350
+      ]
+    ],
+    "page_idx": 32
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Figure R3. Synthetic route of the block copolymer analogues PEO-\\(b\\) -P(CyOMe-co-TPE).",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 32
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Figure R4. \\(^1\\mathrm{H}\\) NMR spectrum of PEO-\\(b\\) -P(CyOMe-co-TPE) in \\(\\mathrm{CD_2Cl_2}\\)",
+    "footnote": [],
+    "bbox": [
+      [
+        200,
+        222,
+        792,
+        565
+      ]
+    ],
+    "page_idx": 34
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "Figure R5. (a-b) TEM images showing PEO-\\(b\\) -P(CyOMe-co-TPE) copolymer self-assembly vesicular structure in weak alkaline buffer \\(\\mathrm{(pH = 8.3}\\) , a) and weak acidic solution \\(\\mathrm{(pH = 5.5}\\) , b). c. UV-Vis spectral changes monitoring the supramolecular interactions during the self-assembly. Scale bar: \\(500\\mathrm{nm}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        592,
+        848,
+        750
+      ]
+    ],
+    "page_idx": 35
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_50.jpg",
+    "caption": "Supplementary Fig. 50 UV/Vis spectral changes monitoring the supramolecular interactions inside the polymer assemblies of \\(\\mathrm{PEO}_{45} - b - \\mathrm{PCy}_{70}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        232,
+        174,
+        764,
+        458
+      ]
+    ],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_51.jpg",
+    "caption": "Supplementary Fig. 51 (a-d) TEM images of the self-assembled morphology of \\(\\mathrm{PEO}_{45} - b - \\mathrm{PCy}_{70}\\) under various pH conditions: (a) \\(\\mathrm{pH} = 8.3\\) , (b) \\(\\mathrm{pH} = 7.6\\) , (c) \\(\\mathrm{pH} = 6.8\\) , and (d) \\(\\mathrm{pH} = 5.5\\) and",
+    "footnote": [],
+    "bbox": [
+      [
+        157,
+        551,
+        845,
+        840
+      ]
+    ],
+    "page_idx": 37
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_5.jpg",
+    "caption": "Figure R6. The changes in fluorescence luminescence spectra after varying the PBS buffer concentration of the assembled solutions under various pH conditions: (a) \\(\\mathrm{pH} = 8.3\\) . (b) \\(\\mathrm{pH} = 7.6\\) . (c) \\(\\mathrm{pH} = 6.8\\) and (d) \\(\\mathrm{pH} = 5.5\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        85,
+        850,
+        472
+      ]
+    ],
+    "page_idx": 38
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_6.jpg",
+    "caption": "Figure R7. (a-c) The polymer self-assembly systems at \\(\\mathrm{pH} = 7.6\\) , 6.8 and 5.5, respectively, and the assembly results obtained at PBS concentrations of \\(20 \\mathrm{mM}\\) , \\(50 \\mathrm{mM}\\) and \\(100 \\mathrm{mM}\\) , respectively are shown from left to right.",
+    "footnote": [],
+    "bbox": [
+      [
+        189,
+        98,
+        820,
+        504
+      ]
+    ],
+    "page_idx": 40
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_7.jpg",
+    "caption": "Figure R8. The Observed particles with multiple indentation sites. The images of b, d and f represent the results of processing the images a/b/c with the Interactive 3D Surface Plot function in the Image J software, which can assist in demonstrating the depressions on the particles. The",
+    "footnote": [],
+    "bbox": [
+      [
+        170,
+        455,
+        835,
+        808
+      ]
+    ],
+    "page_idx": 40
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_40.jpg",
+    "caption": "Supplementary Fig. 40 (a-d). TEM images and (e-h) SEM images of the self-assembly morphologies of \\(\\mathrm{PEO}_{45 - b}\\mathrm{-P}(\\mathrm{Cy}_{73 - c}\\mathrm{o}\\mathrm{-TPE}_{34})\\) at 0 minutes (a and e), 1.5 minutes (b and f), 6 minutes (c and g) and 21 minutes (d and h) following the addition of GL. Scale bar: 500 nm.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 40
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_49.jpg",
+    "caption": "Supplementary Fig. 49 pH titrations of \\(\\mathrm{PEO}_{45 - b - \\mathrm{P}(\\mathrm{Cy}_{73 - \\mathrm{CO} - \\mathrm{TPE}_{34})}}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        149,
+        334,
+        850,
+        540
+      ]
+    ],
+    "page_idx": 43
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_8.jpg",
+    "caption": "Fig. R9. The DSC curve of \\(\\mathrm{PEO}_{45 - b} - \\mathrm{P}(\\mathrm{Cy}_{73 - \\mathrm{CO}} - \\mathrm{TPE}_{34})\\) copolymer.",
+    "footnote": [],
+    "bbox": [
+      [
+        209,
+        430,
+        770,
+        720
+      ]
+    ],
+    "page_idx": 43
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_37.jpg",
+    "caption": "Supplementary Fig. 37 a. \\(^1\\mathrm{H}\\) NMR spectrum in \\(\\mathrm{CD}_2\\mathrm{Cl}_2\\) . b. The GPC traces and c. the DSC of \\(\\mathrm{PCy}_{67}\\) homopolymer.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        118,
+        855,
+        666
+      ]
+    ],
+    "page_idx": 45
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_38.jpg",
+    "caption": "Supplementary Fig. 38 a. \\(^1\\mathrm{H}\\) NMR spectrum in \\(\\mathrm{CD}_2\\mathrm{Cl}_2\\) . b. The GPC traces and c. the DSC of \\(\\mathrm{PTPE}_{26}\\) homopolymer.",
+    "footnote": [],
+    "bbox": [
+      [
+        168,
+        87,
+        836,
+        602
+      ]
+    ],
+    "page_idx": 46
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_9.jpg",
+    "caption": "Figure R8. The Observed particles with multiple indentation sites. The images of b, d and f represent the results of processing the images a/b/c with the Interactive 3D Surface Plot function",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 49
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_29.jpg",
+    "caption": "Supplementary Fig. 29 \\(^{1}\\mathrm{H}\\) NMR spectrum of \\(\\mathrm{PEO}_{45} - b - \\mathrm{P}(\\mathrm{Cy}_{54} - \\mathrm{co} - \\mathrm{TPE}_{5})\\) in \\(\\mathrm{CD}_{2}\\mathrm{Cl}_{2}\\) .",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 52
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_30.jpg",
+    "caption": "Supplementary Fig. 30 \\(^{1}\\mathrm{H}\\) NMR spectrum of \\(\\mathrm{PEO}_{45} - b - \\mathrm{P}(\\mathrm{Cy}_{26} - \\mathrm{co} - \\mathrm{TPE}_{5})\\) in \\(\\mathrm{CD}_{2}\\mathrm{Cl}_{2}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        161,
+        490,
+        837,
+        823
+      ]
+    ],
+    "page_idx": 53
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_44.jpg",
+    "caption": "Supplementary Fig. 44 The morphology of polymer counterpart \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{78 - c0} - \\mathrm{TPE}_9)\\) . a. TEM image (shallow open mouth on membrane). b. Cross-sectional height analysis to show the overall diameter \\((D)\\) and the depth \\((H)\\) of membrane concaving. c. AFM and d. SEM image. In order to investigate the effect of the copolymer ratio on the assembly morphology, we have also prepared and analyzed the polymer counterpart \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{78 - c0} - \\mathrm{TPE}_9)\\) with a lower copolymer ratio than \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{73 - c0} - \\mathrm{TPE}_{34})\\) . It can be found that when the copolymer ratio was reduced from 0.47 to 0.12, \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{78 - c0} - \\mathrm{TPE}_9)\\) counterpart can form bowl-shaped vesicles, but the membrane concaving is relatively shallow. Their \\(D\\) is of \\(\\sim 497 \\mathrm{nm}\\) , close to that of \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{73 - c0} - \\mathrm{TPE}_{34})\\) ( \\(\\sim 520 \\mathrm{nm}\\) ); however, their \\(H\\) is only determined to be \\(\\sim 170 \\mathrm{nm}\\) , far lower than that of \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{73 - c0} - \\mathrm{TPE}_{34})\\) with deep invagination ( \\(\\sim 398 \\mathrm{nm}\\) ). This geometry corresponds to a shallow invaginated geometry.",
+    "footnote": [],
+    "bbox": [
+      [
+        211,
+        85,
+        783,
+        469
+      ]
+    ],
+    "page_idx": 55
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_45.jpg",
+    "caption": "Supplementary Fig. 45 The morphology of polymer counterpart \\(\\mathrm{PEO}_{45 - b} - \\mathrm{P}(\\mathrm{Cy}_{54} - \\mathrm{co} - \\mathrm{TPE}_5)\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        177,
+        92,
+        835,
+        323
+      ]
+    ],
+    "page_idx": 57
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_46.jpg",
+    "caption": "Supplementary Fig. 46 The morphology of polymer counterpart \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{28 - c}\\mathrm{CO} - \\mathrm{TPE}_{30})\\) . a. TEM image (deep open mouth on membrane indicated by arrows). b. Cross-sectional height analysis to show the overall diameter \\((D)\\) and the depth \\((H)\\) of membrane concaving. c. AFM and d. SEM image.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        88,
+        844,
+        560
+      ]
+    ],
+    "page_idx": 58
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_47.jpg",
+    "caption": "Supplementary Fig. 47 The morphology of polymer counterpart \\(\\mathrm{PEO}_{45 - b - }\\mathrm{P}(\\mathrm{Cy}_{26 - c0 - }\\mathrm{TPE}_{56})\\) . a. TEM image (deep open mouth on membrane indicated by arrows). b and c. Cross-sectional height analysis to show the overall diameter (D) and the depth (H) of membrane concaving.",
+    "footnote": [],
+    "bbox": [
+      [
+        175,
+        120,
+        833,
+        350
+      ]
+    ],
+    "page_idx": 60
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_48.jpg",
+    "caption": "Supplementary Fig. 48 The variation of the plot of the H/D value of the bowl-shaped particles obtained by polymer self-assembly as a function of the ratio of the polymerization degree of TPE to Cy in the polymer chain. The two colors, blue and green, represent the formation of assemblies that are shallowly depressed and deeply depressed, respectively.",
+    "footnote": [],
+    "bbox": [
+      [
+        216,
+        90,
+        780,
+        412
+      ]
+    ],
+    "page_idx": 61
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_59.jpg",
+    "caption": "Supplementary Fig. 59 The TEM images of four vesicular morphologies assemblies (LCS, s-BV, m-BV and \\(d\\) -BV) of before (a-d) and after (e-h) treating with \\(2\\mathrm{mM}\\) \\(\\mathrm{SO}_2\\) . Scale bar: 200 nm.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 62
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_10.jpg",
+    "caption": "Figure R1. Additional Cryo-TEM observation of the self-assembly morphologies of \\(\\mathrm{PEO}_{45}\\text{-} b\\) \\(\\mathrm{P(Cy}_{73}\\text{-} co\\text{-} TPE_{34})\\) under varying pH values (from left to right, \\(\\mathrm{pH} = 8.3\\) , 7.6, 6.3 and 5.5) in parallel experiments.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 67
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_11.jpg",
+    "caption": "Figure R2. Additional DLS and SLS measurements of \\(\\mathrm{PEO}_{45 - b}\\mathrm{-P}(\\mathrm{Cy}_{73 - \\mathrm{co}}\\mathrm{-TPE}_{34})\\) assemblies under varying pH values for showing the variation of \\(R_{\\mathrm{g}}\\) , \\(R_{\\mathrm{h}}\\) and shape factor \\(\\rho\\) values.",
+    "footnote": [],
+    "bbox": [
+      [
+        260,
+        306,
+        736,
+        547
+      ]
+    ],
+    "page_idx": 68
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1 Illustration of traditional signal-to-response modes and the design concept of signal nonlinear amplification by regulation of vesicle membrane concavity. a. '1-to-1' response mode by side-group-type responsive polymers. b. '1-to-N' response mode by main-chain-type self-immolative polymers. c. This work presents '1-to-N' response mode by chain aggregation state mutation. d. Designed \\(\\mathrm{SO}_2\\) -biosignal-responsive block copolymer, PEO-\\(b\\) -P(Cy-co-TPE), and its self-assembly shape transformation with pH-tunable membrane concavity that confers to nonlinear amplifying effect of response ability towards \\(\\mathrm{SO}_2\\) biosignal.",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        92,
+        848,
+        412
+      ]
+    ],
+    "page_idx": 71
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1 Illustration of traditional signal-to-response modes and the design concept of signal nonlinear amplification by regulation of surface concavity of nanobowls. a. '1-to-1' response mode by side-group-type responsive polymers. b. '1-to-N' response mode by main-chain-type self-immolative polymers. c. This work presents '1-to-N' response mode by chain aggregation state mutation. d. Designed SO₂-biosignal-responsive block copolymer, PEO-b-P(Cy-co-TPE), and its self-assembly shape transformation with pH-tunable surface concavity that confers to nonlinear amplifying effect of response ability towards SO₂ biosignal.",
+    "footnote": [],
+    "bbox": [
+      [
+        171,
+        103,
+        850,
+        418
+      ]
+    ],
+    "page_idx": 73
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 Morphological evolution of bowl-shaped vesicles with variable membrane concavity.",
+    "footnote": [],
+    "bbox": [
+      [
+        146,
+        392,
+        846,
+        723
+      ]
+    ],
+    "page_idx": 77
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 Morphological evolution of nanobowls with variable surface concavity. Using different analysis and characterization methods, including (i) geometric illustration of surface invaginated structures, (ii) cryo-TEM, (iii) high-contrast TEM, (iv) electron-beam intensity profile analysis to individual assemblies, (v) tapping-mode AFM height images, and (vi) cross-sectional height analysis, to show the shape transformation of \\(\\mathrm{PEO}_{45 - b} - \\mathrm{P}(\\mathrm{Cy}_{73} - \\mathrm{co} - \\mathrm{TPE}_{34})\\) assemblies under various pH conditions: a. LCSs ( \\(\\mathrm{pH} = 8.3\\) ). b. s-NBs ( \\(\\mathrm{pH} = 7.6\\) ). c. m-NBs ( \\(\\mathrm{pH} = 6.8\\) ). d. d-NBs ( \\(\\mathrm{pH} = 5.5\\) ). All the polymer concentrations were at \\(1.0 \\mathrm{mg / mL}\\) . Scale bar in series (ii) and (iii) is \\(200 \\mathrm{nm}\\) , in series (v) is \\(500 \\mathrm{nm}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        170,
+        836,
+        499
+      ]
+    ],
+    "page_idx": 77
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4 Geometric parameter change of bowl-shaped vesicles and spectrometric monitoring of diverse nanobowl shapes. a. The width \\((W)\\) and height \\((H)\\) of open mouth in assemblies and their radius \\((R)\\) plotted against solution pH. b. The variation of membrane concavity \\((\\sigma)\\) of the assemblies as a function of solution pH. The border between two colors representing the pH point \\((-8.0)\\) at which the LCSs transform to the BVs. c. Infrared and d. UV-Vis spectral changes monitoring the supramolecular interactions inside the polymer assemblies with various shapes.",
+    "footnote": [],
+    "bbox": [
+      [
+        152,
+        180,
+        840,
+        519
+      ]
+    ],
+    "page_idx": 79
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4 Geometric parameter change and spectrometric monitoring of diverse nanobowl shapes. a. The width \\((W)\\) and height \\((H)\\) of open mouth in assemblies and their radius \\((R)\\) plotted against solution pH. b. The variation of surface concavity \\((\\sigma)\\) of the assemblies as a function of solution pH. The border between two colors representing the pH point \\((\\sim 8.0)\\) at which the LCSs transform to the NBs. c. Infrared and d. UV-Vis spectral changes monitoring the supramolecular interactions inside the polymer assemblies with various shapes.",
+    "footnote": [],
+    "bbox": [
+      [
+        152,
+        85,
+        848,
+        430
+      ]
+    ],
+    "page_idx": 80
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Fig. 5 Nonlinear amplification of nanobowl to \\(\\mathbf{SO}_2\\) biosignal. The fluorescent spectral change of a. LCS and b. d-NB assembled from \\(\\mathrm{PEO}_{45} - b\\text{-P(Cy}_{73} - c\\text{-O - TPE}_{34})\\) with \\(\\mathrm{SO}_2\\) concentration. c. Fluorescence intensity transition of four assemblies of different morphologies",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        288,
+        852,
+        808
+      ]
+    ],
+    "page_idx": 81
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_50.jpg",
+    "caption": "Supplementary Fig. 50 Schematic diagram of the variation of the degree of aggregation of the polymer chains inside the nanobowls during deformation. Grey sphere, deprotonated Cy unit; green sphere, protonated Cy unit; purple sphere, TPE unit; blue chain, hydrophobic block of polymer chain.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 82
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_12.jpg",
+    "caption": "Figure R1. The fluorescence emission spectra of polymer self-assembly solutions \\((E_{x} = 510\\) nm).",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 83
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_13.jpg",
+    "caption": "Figure R2. Confocal laser scanning microscopy (CLSM) images of polymer self-assembly solution at different \\(\\mathrm{pH}\\) : (a-c) \\(\\mathrm{pH} = 8.3\\) ; (d-f) \\(\\mathrm{pH} = 5.5\\) , where the first row is the bright field images, the second row is the images observed in the blue channel, the third row is the images observed in the red channel, and the fourth column is the merged result. Sale bar: \\(5\\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        456,
+        845,
+        720
+      ]
+    ],
+    "page_idx": 85
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5h.jpg",
+    "caption": "Fig 5h. Molecular simulation of chain aggregation state before and after \\(\\mathrm{SO}_2\\) trigger.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 95
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5h.jpg",
+    "caption": "Fig 5h. Molecular simulation of chain aggregation states in LCS and d-NB assemblies under",
+    "footnote": [],
+    "bbox": [
+      [
+        156,
+        541,
+        835,
+        863
+      ]
+    ],
+    "page_idx": 98
+  }
+]

peer_reviews/supplementary_0_Peer Review file__c2c345558b96db063604d2a55b798464f0913fd48543918fbef69a413db242a7/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review file__c2c345558b96db063604d2a55b798464f0913fd48543918fbef69a413db242a7/supplementary_0_Peer Review file__c2c345558b96db063604d2a55b798464f0913fd48543918fbef69a413db242a7.mmd

@@ -0,0 +1,568 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+# HSP90 as an Evolutionary Capacitor Drives Adaptive Eye Size Reduction via Atonal  
+
+Corresponding Author: Professor Joachim Kurtz  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author) Overview:  
+
+I was excited to read this manuscript because the abstract suggested a significant advance and proof- of- principle for heritable, adaptive uncovering of cryptic genetic variation via Hsp90 inhibition. By and large, I think this promise from the abstract was met by the study, I did very much enjoy reading the manuscript, and I think it has a lot to add to the field. However, there are several points that need to be clarified, potentially with additional data, for the manuscript to have an even more compelling story.  
+
+Major Comment:  
+
+The major shortcoming I see (that potentially extends to other studies of Hsp90 and CGV) is that we are not given evidence that Hsp90 expression or availability is affected by environmental change/stress. The authors functionally manipulate Hsp90 with RNAi and drugs, then assay fitness in different environments. We don't know however, if the different environment- continuous light- - affects Hsp90. I think the most compelling case that corresponds to that laid out in this study would require: 1) Establishing a link between environment and Hsp90 expression/availability 2) Establishing a link between environment and phenotypic variation 3) Establishing a link between Hsp90 expression and phenotypic variation 4) Establishing link between phenotypic variation and fitness, especially in the inducing environment 5) Establishing that phenotypic variation is inherited across generation  
+
+The authors address points 3- 4, but 1 & 2 remain untested, perhaps in any system. On L72- 74 for instance, it is unclear if the references refer to the first clause "However, HSP90 availability may become limited under stressful environmental conditions," or the second clause, "as it is required by numerous proteins that become damaged during stress" or both. If there are no references to support the first clause, then it seems like a major gap that seeks filling.  
+
+The authors should consider the effects of constant light itself of phenotype production and HSP90 expression. They ought to consider discussing how, in nature for example, a reduction of HSP90 expression (akin to their manipulations) could occur and have fitness differences be borne out in a novel environment (e.g. constant light) if that environment does or does not affect (we don't know from the current data) the expression itself.  
+
+Additional comments: L43: note what kind of variation is meant here  
+
+L102- 106: The reference to the previous study on social cues makes the reader think that this will be an important element of the current study even though it is not. The authors should consider ending with a brief overview of their study, its phenotypes, environments, and aims.  
+
+L108: "HSP90 reduction, mimicking what may occur during environmental stress, in ..." Again, evidence of HSP90 with environmental stress seems like a key point that is not well- established from what I see in the manuscript.  
+
+L119- 122 seems to disagree with the figure 1 legend. The figure says F2 had \(27.8\%\)
+
+<--- Page Split --->
+
+L138 Why not Hsp83? It seems odd that a different gene is being assayed here when everything else focused on Hsp83.  
+
+L158: How were reduced- and normal- eye beetles operationally delimited? Knowing the cut- off metric is important because much of the phenotypic and fitness data between the groups overlap (Fig. 1).  
+
+L210 Please note that the deviation was lower than expected.  
+
+L273- 274 I think omission of some commas in the sentence make it a bit awkwardly worded. Do you mean "Notably, our mapping analysis shows that the atonal gene, which consists of a single exon, overlaps with four synonymous polymorphic SNPs across the four pooled samples..."?  
+
+L270 & 273 Consider removing "Notably" from one of these sentences.  
+
+L279 I'm sure the authors wish they could pinpoint more closely candidate regulatory elements. On possibility is to bioinformatically identify differences in TF motifs between lines in the candidate region and how the presence/absence of SNPs affect motif occurrences. The online resource MEME Suite might be useful to explore and get a bit closer to the actual causal SNP(s). https://meme- suite.org/meme. Of course, CHIP- seq or ATAC- seq could also identify some candidate positions.  
+
+L373 This is the closest we get to understanding if HSP90 expression is affected by stress. It is unfortunate that it is not in the environment used in this study.  
+
+L601: I am confused about the abnormal leg phenotype and its being reported. Did this always co- occur with the reduced eyes? If not, how often was reduced- eyes by itself and why weren't these lines chosen instead of mixed phenotype lines? Why were leg phenotypes used to establish eye phenotype lines?  
+
+L641: Again, unclear why Hsp83 wasn't measured.  
+
+L680- 681: This is a clever design.  
+
+Fig. 1  
+
+I don't understand the diagram in 1a, what are the small arrows and red legs highlighting? It doesn't say in the legend. In 1b, % frequency for the y- axis is odd. I'm not sure how to interpret it. The legend says that the % eye ptype is above the gold bar, but those numbers are more likely sample sizes. I think something must be missing. In 1d, how were groups categorized, especially since there is overlap with some normal- eye having areas smaller than reduced- eye? Should it be treatment or line rather than ptype for the colors in panels in d- f?  
+
+## Reviewer #2  
+
+(Remarks to the Author)  
+
+The study by Sayed et al. explores the underappreciated evolutionary role of Hsp90 buffering in animals, identifying a beetle eye size trait whose variation is masked by Hsp90. Their findings suggest that Hsp90 enables the accumulation of cryptic genetic variation in natural populations. Transient Hsp90 inhibition (via chemical or siRNA) reveals a reduced- eye phenotype that persists in a subset of F1 and F2 progeny, can be driven to near fixation through crosses, and shows fitness advantages under constant light. A dimorphism is noted, and atonal is proposed as a phenocopy for Hsp90 inhibition. These findings are important and intriguing.  
+
+While this is a compelling example of Hsp90 buffering in animals, the manuscript suffers from overstated claims, limited novelty, and interpretational issues. The conclusions need to be tempered or supported with additional data. The introduction and discussion should better contextualize the work within existing evidence supporting the adaptive value of Hsp90- buffered variation, avoiding claims of controversy or primacy. The experimental rigor and statistical analyses also require improvement. That said, the manuscript is clearly written and presents an intriguing story. I would support publication following major revisions to address the concerns outlined below.  
+
+Major issues:  
+
+1. Conceptual Framing and Literature Context: The manuscript overstates the novelty and conceptual primacy of Hsp90 as an evolutionary capacitor. While this study advances our understanding of Hsp90-buffering in animals, the concept is well-established, particularly in yeast. The authors misrepresent the state of the field by framing the adaptive value of Hsp90-buffered variation as controversial. Multiple prior studies, including those by Lindquist and others (e.g., PMID: 16195452, 21205668, 39052788), have clearly demonstrated adaptive, genetically defined Hsp90-buffered traits. The manuscript also lacks appropriate citations and introduces misleading statements about the novelty and controversy of this mechanism. All novelty claims should be carefully rephrased to reflect that the contribution refers to Hsp90's role in animal evolution not evolution in general. I suggest revising lines 27-31, 37-38, 78-79, 86-87, 98, 169-170, 283-284, and 357-358 (these may not be all the sentences in need of revision to address this issue).
+
+<--- Page Split --->
+
+2. The ecological and adaptive relevance of the reduced-eye trait is inadequately supported. The authors test it under artificial, non-ecological conditions (e.g., continuous light), without exploring whether it is beneficial or deleterious under more natural environments. As a loss-of-function phenotype, reduced-eye is unlikely to confer broad fitness advantages, and its frequency in wild populations remains unexamined. Strong claims about its evolutionary significance require broader validation and testing of alternative hypotheses, including potential fitness costs. Overall, the trait's ecological relevance is unclear, and the study lacks the rigor seen in comparable work on Hsp90-buffered traits in yeast.  
+
+3. The claim that atonal mediates Hsp90-buffered eye-size variation is not sufficiently demonstrated. While atonal RNAi phenocopies the trait, there is no genetic or expression evidence linking atonal to Hsp90 buffering in the beetle. Allele-specific expression, response to Hsp90 inhibition, and functional SNP data are lacking. The role of HDACs is weakly supported and lacks a positive control. Without stronger evidence, the proposed mechanism remains speculative and should be presented as such.  
+
+4. Several conclusions overreach the presented data. Key findings (e.g., persistence of the trait, its selection, and its mechanistic basis) have low penetrance, limited replicates, and insufficient statistical support. Alternative hypotheses are not tested, and the environmental relevance of findings is unclear. Overall, stronger controls, clearer definitions, and more cautious interpretation are needed throughout. Statistical analyses are not robust. Effects on fitness are weak and may not be significant after multiple hypothesis testing.  
+
+Minor comments:  
+
+1. In place of "Hsp90-released trait" I suggest "Hsp90-buffered trait"; lines 38 and 74.  
+
+2. The keyword "plasticity" was not mentioned. Remove keyword or add explanation of the term in the text.  
+
+3. Statistical significance, \\* should be replaced with increments \\*\\*, \\*\\*\\*\\*\\*\\*\\*\\*\\*\\*\\* A single asterisk (\\*) indicating a p-value less than 0.05 (significant), two asterisks (\\*) a p-value less than 0.01 (highly significant), and three asterisks (\\*\\*) indicate a p-value less than 0.001 (very highly significant).  
+
+4. Hsp90 is an epistatic regulator but no epistasis experiments were performed.  
+
+5. Abstract lines 33-34: "stably inherited without continued HSP90 disruption" implies all revealed variation was stable but thew data show closer to \(20\%\) of the revealed variation is inherited and further declines in each generation so it is not that stable.  
+
+6. Figures should be revised to improve clarity and intuitiveness. In particular, the presentation of penetrance—such as in the HDAC inhibition example—is often unclear and could benefit from more straightforward visualization. A more quantitative trait would be the number of omatids per animal, which although labor intensive seems to present with a great dynamic range and could foster a more robust statistical analysis approach. I understand that counting the number of omatids in each animal may not be feasible, so this is only a minor comment/suggestion.  
+
+Reviewer #3  
+
+(Remarks to the Author)  
+
+I co-reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co-review manuscripts.  
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author) NCOMMS- 25- 2699A  
+
+My biggest concerns in the previous version of the manuscript were the links between Hsp90 levels, phenotypic variation, and environmental conditions. In their response letter, the authors do a convincing job of explaining that the conditions that affected Hsp90 availability need not be the same conditions that select upon phenotypes and genotypes revealed by Hsp90 depletion. Further, they explained the link between measuring expression level of Hsp68 instead of Hsp83. By and large, I am satisfied with their responses to my initial comments and suggestions, and I am pleased by their inclusion of additional data and consideration of my comments. I have only a few suggestions for further clarifying these points in the text.  
+
+L78: Consider mentioning here that selection could be acting in the Hsp90 affecting environment or a different environment. " ...differences on which selection can act either in the disrupting environmental conditions or other conditions altogether."  
+
+L186 consider adding a reminder here that continuous light COULD be encountered by these beetles in their human- commensal environment
+
+<--- Page Split --->
+
+L208 Consider changing "certain" to "some, albeit unknown, frequency"  
+
+L321 "...unmasking cryptic genetic variation and producing new phenotypes." To be explicit about the consequence of that CGV being uncovered.  
+
+(Remarks on code availability)  
+
+Reviewer #2  
+
+(Remarks to the Author) All my comments have been fully addressed. I congratulate the authors on their beautiful work and approve of its publication.  
+
+(Remarks on code availability)  
+
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+## RESPONSE TO THE REVIEWER COMMENTS  
+
+## Dear Reviewers,  
+
+We appreciate the opportunity to revise our manuscript entitled "HSP90 as an Evolutionary Capacitor Drives Adaptive Eye Size Reduction via atonal." We are grateful for the reviewers' insightful comments and suggestions, which have helped us to improve the clarity and quality of our work.  
+
+Reviewer #1 (Remarks to the Author):  
+
+## Overview:  
+
+I was excited to read this manuscript because the abstract suggested a significant advance and proof- of- principle for heritable, adaptive uncovering of cryptic genetic variation via Hsp90 inhibition. By and large, I think this promise from the abstract was met by the study, I did very much enjoy reading the manuscript, and I think it has a lot to add to the field. However, there are several points that need to be clarified, potentially with additional data, for the manuscript to have an even more compelling story.  
+
+## response:  
+
+Thank you very much for your positive feedback and for recognizing the strengths and significance of our study. We are pleased that you found the manuscript both promising and enjoyable, and we appreciate your suggestions for clarifications and additional data to further strengthen the work. As detailed in the following, we added additional data as suggested.  
+
+## Major Comment:  
+
+The major shortcoming I see (that potentially extends to other studies of Hsp90 and CGV) is that we are not given evidence that Hsp90 expression or availability is affected by environmental change/stress. The authors functionally manipulate Hsp90 with RNAi and drugs, then assay fitness in different environments. We don't know however, if the different environment- - - continuous light- - - affects Hsp90. I think the most compelling case that corresponds to that laid out in this study would require: 1) Establishing a link between environment and Hsp90 expression/availability 2) Establishing a link between environment and phenotypic variation 3) Establishing a link between Hsp90 expression and phenotypic variation 4) Establishing link between phenotypic variation and fitness, especially in the inducing environment 5) Establishing that phenotypic variation is inherited across generation  
+
+The authors address points 3- 4, but 1 & 2 remain untested, perhaps in any system. On L72- 74 for instance, it is unclear if the references refer to the first clause "However, HSP90 availability may become limited under stressful environmental conditions," or the second clause, "as it is required by numerous proteins that become damaged during stress" or both. If there are no references to support the first clause, then it seems like a major gap that seeks filling.  
+
+## response:
+
+<--- Page Split --->
+
+Thank you for this important and insightful comment. We have conducted additional experiments to further substantiate the link between HSP90 and the reduced- eye phenotype, now showing that RNAi knockdown of Hsp83 reduces atonal expression, supporting a regulatory link (Fig.3g) and that atonal is consistently downregulated in reduced- eye monomorphic lines (Fig.3f). More details on these data are provided in our response to Reviewer 2. While these additional data provide further support for point 3), we also carefully considered points 1) and 2) and conducted an additional experiment to address these points (see below). While agreeing that directly linking environmental changes to HSP90 expression or availability could strengthen both our manuscript and the broader understanding of HSP90's role in adaptation, we would like to provide some conceptual clarifications.  
+
+Our study is conceptually grounded in the mechanism whereby HSP90 becomes limiting primarily under physiological stresses, in particular those that cause protein damage or denaturation - such as thermal, toxic, pathogenic or osmotic stress or other proteotoxic conditions - where demand for HSP90 is elevated and its availability for other cellular processes becomes restricted. Under these conditions, HSP90 is required for the stabilization and refolding of numerous damaged proteins, which may temporarily reduce its availability and thereby unmask cryptic genetic variation. We would like to point out that according to this concept, the environment leading to reduced HSP90 availability is not necessarily the same as the environment in which released phenotypes may show fitness advantages. Specifically, the environments in points 1) and 2) are not expected to be the same as the environments where fitness benefits manifest (point 4).  
+
+We also acknowledge your point regarding the need for a clear reference supporting this mechanism. In response, we identified and now cite Alford and Brandman (2018), which directly supports the statement that HSP90 availability can become limited under stressful conditions that challenge proteostasis. We have revised the relevant sentence in the manuscript for clarity and included this reference at line 83- 86. The revised sentence now reads:  
+
+"However, under stressful environmental conditions that disrupt proteostasis, HSP90 availability may become limited due to its involvement in stabilizing numerous proteins that are damaged or denatured during stress (Alford and Brandman 2018; Borkovich et al. 1989; Chen and Wagner 2012; Peuss et al. 2015)".  
+
+The authors should consider the effects of constant light itself of phenotype production and HSP90 expression. They ought to consider discussing how, in nature for example, a reduction of HSP90 expression (akin to their manipulations) could occur and have fitness differences be borne out in a novel environment (e.g. constant light) if that environment does or does not affect (we don't know from the current data) the expression itself.  
+
+## response:  
+
+Thank you for raising this point regarding the potential effects of constant light on both phenotype production and HSP90 expression, as well as the ecological context for HSP90 reduction and fitness in novel environments. However, as noted above, please consider that it is conceptually not required that it is the same environmental condition that leads to reduced HSP90 availability and the phenotypes' fitness advantage.
+
+<--- Page Split --->
+
+- To nevertheless address your comment, we conducted an experiment to directly test whether constant light affects HSP90 expression. Our results showed no significant change in HSP90 expression under constant light, supporting our expectation that this condition alone does not challenge proteostasis or induce a heat shock response. We did not include these additional data into the manuscript but are happy to provide them if requested.  
+
+- Regarding the production of the reduced-eye phenotype, our data indicate that the phenotype is only revealed when HSP90 function is experimentally reduced (via RNAi or chemical inhibition), not by constant light alone. This suggests that constant light acts as a selective environment, but not directly as an inducer of the phenotype.  
+
+- In nature, HSP90 limitation is most likely to occur under environmental stresses that cause protein damage or denaturation, where HSP90 becomes limited due to increased demand. In such scenarios, hidden genetic variation could be revealed, and, as our study demonstrates, may confer a fitness advantage in specific novel environments like continuous light.  
+
+- We have expanded the discussion in the revised manuscript to clarify these points and to address how natural environmental stresses could lead to HSP90 limitation and the expression of adaptive phenotypes, even if the novel environment itself (e.g., constant light) does not directly affect HSP90 expression. We have added the following sentences at line 368 in the Discussion:  
+
+"It is well established that certain environmental stresses can limit HSP90 availability by increasing the demand for its chaperone function (Alford and Brandman 2018), thereby unmasking cryptic genetic variation (Rutherford and Lindquist 1998). Our findings suggest that once such variation is exposed—regardless of the initial stressor—it can become subject to selection in a novel environment, even if that environment does not directly affect HSP90 expression."  
+
+## Additional comments:  
+
+L43: note what kind of variation is meant here  
+
+## response:  
+
+We clarified the type of variation in the revised manuscript. The sentence now reads (line 51):  
+
+"Canalization buffers development against genetic and environmental disturbances, producing stable phenotypes despite underlying genetic and environmental variation (Waddington 1942)".  
+
+L102- 106: The reference to the previous study on social cues makes the reader think that this will be an important element of the current study even though it is not. The authors should consider ending with a brief overview of their study, its phenotypes, environments, and aims.  
+
+## response:  
+
+Thank you for pointing this out. To improve clarity and better guide the reader into our study, we have revised the final paragraph of the Introduction as suggested and now include a
+
+<--- Page Split --->
+
+concise overview of our study's main phenotypes, environments, and aims. This ensures a clear transition into the Results section without implying undue emphasis on social cues. Now this section reads _line 118:  
+
+"Here we used the red flour beetle, Tribolium castaneum as an alternative insect model (Brown et al. 2009; Pointer, Gage, and Spurgin 2021) to examine the genetic basis and fitness relevance of an HSP90- released phenotype. A previous study in T. castaneum found consistent downregulation of Hsp83, the primary HSP90- coding gene, in response to social cues mimicking a stressful environment. This suggests that HSP90 levels may be adaptively regulated, potentially facilitating the release of cryptic genetic variation when advantageous (Peuss et al. 2015). In the present study, we directly manipulated HSP90 function in T. castaneum and characterized the resulting reduced- eye phenotype. We identified the gene causing this phenotype and assessed its fitness consequences to evaluate the adaptive potential of HSP90- released variation."  
+
+L108: "HSP90 reduction, mimicking what may occur during environmental stress, in ..." Again, evidence of HSP90 with environmental stress seems like a key point that is not well- established from what I see in the manuscript.  
+
+## response:  
+
+Following your suggestion, we have revised the sentence at line 136 for greater clarity. As mentioned above, we have also addressed the evidence for HSP90 limitation under environmental stress and included the relevant reference (Alford and Brandman 2018) in the manuscript. The sentence now reads:  
+
+"To investigate the effects of reduced HSP90 function—mimicking conditions that may arise during environmental stress—we used RNA interference (RNAi) to target Hsp83 in T. castaneum, with the aim of releasing HSP90- buffered, selectable phenotypic variants in Cro1, a genetically heterogenous wildtype population of this beetle"  
+
+L119- 122 seems to disagree with the figure 1 legend. The figure says F2 had 27.8%  
+
+## response:  
+
+We appreciate your careful reading. The percentage shown in the original figure 1a (27.8%) represented the combined incidence of the reduced- eye phenotype in F2 offspring from the two families in which the trait was observed (12/47 and 20/68, totaling 32/115 beetles). To avoid confusion, we have now removed this percentage from Figure 1a and present the detailed incidence rates only in the text and Extended Data Table 1. The text now clearly states the incidence rates for each family (25.5% and 29.4%, respectively), as well as the overall frequency (4.2%) among all F2 offspring screened within this treatment as shown in Fig. 1b. This should resolve any inconsistency between the figure and the text.  
+
+L138 Why not Hsp83? It seems odd that a different gene is being assayed here when everything else focused on Hsp83.  
+
+## response:  
+
+We appreciate this observation. We focused on Hsp68a (HSP70 family) as a commonly used readout for HSP90 limitation, because 17- DMAG inhibits HSP90 (encoded by Hsp83) at the protein level rather than by reducing its mRNA expression. 17- DMAG is a well- characterized HSP90 inhibitor that binds to the ATP- binding pocket of the HSP90 protein, thereby blocking
+
+<--- Page Split --->
+
+its chaperone activity and leading to functional inhibition (Jez et al. 2003; Trepel et al. 2010). This inhibition does not typically reduce Hsp83 mRNA levels. Instead, inhibition of HSP90 activity is known to induce a compensatory heat shock response, including upregulation of HSP70 family genes such as Hsp68a (Kudryavtsev et al. 2017; Zhou et al. 2013). Thus, increased Hsp68a expression serves as a reliable molecular marker of successful HSP90 inhibition in our system. To clarify this, we have added the following sentence to the manuscript_line 181:  
+
+"17- DMAG inhibits HSP90 at the protein level by binding to its ATP- binding pocket, thereby blocking its chaperone activity without affecting Hsp83 mRNA levels (Jez et al. 2003; Trepel et al. 2010). As a result, upregulation of HSP70 family genes, such as Hsp68a, serves as a molecular marker of successful HSP90 inhibition (Jez et al. 2003)."  
+
+L158: How were reduced- and normal- eye beetles operationally delimited? Knowing the cut- off metric is important because much of the phenotypic and fitness data between the groups overlap (Fig. 1).  
+
+## response:  
+
+Thank you for pointing this out, it is an important point that needed clarification. During development, reduced- and normal- eye beetles could not be distinguished at the larval or pupal stages based on morphology alone. However, because individuals were tracked throughout development from larvae to adults, their adult eye phenotype could be retrospectively assigned to earlier stages. In the adult stage, classification into "reduced- eye" and "normal- eye" was based on visual inspection under a stereomicroscope, supported by quantitative image analysis of eye area. Reduced- eye beetles consistently showed visibly smaller eyes. For quantification, we normalized eye area within each developmental stage and used the adult phenotype (clearly visible) to assign individuals into the two groups. We have clarified this in the figure legend Fig1d as follows: line 1120  
+
+"Fig.1.....d, Normalized eye area in normal- eye and reduced- eye individuals across the three developmental stages, showing significant differences in eye size within stage. Normalization was done by stage's mean. Lines connect individual beetles, which were tracked from larval to adult stage. Phenotype classification (normal- vs. reduced- eye) was based on adult morphology, which allowed retrospective assignment of phenotype to earlier developmental stages."  
+
+And this sentence was added to the result part: line 205 "Because individuals were tracked throughout development, their adult phenotype could be retrospectively assigned to their larval and pupal stages."  
+
+L210 Please note that the deviation was lower than expected.  
+
+## response:  
+
+Indeed, the deviation was in the direction of a lower- than- expected frequency of the reduced- eye phenotype. We have clarified the sentence in the manuscript to reflect this more precisely. The updated sentence now reads: line 261  
+
+"However, in the crosses RNAi \(\mathcal{P}\) - Cro1 \(\sigma^r\) and RNAi \(\sigma^r\) - Cro1 \(\mathcal{P}\) , we observed a significantly lower proportion of the reduced- eye phenotype than the expected 25 %..."
+
+<--- Page Split --->
+
+L273- 274 I think omission of some commas in the sentence make it a bit awkwardly worded. Do you mean "Notably, our mapping analysis shows that the atonal gene, which consists of a single exon, overlaps with four synonymous polymorphic SNPs across the four pooled samples.."?  
+
+## response:  
+
+We have revised the sentence in L336 for improved clarity. The revised sentence now reads:  
+
+"Our mapping analysis shows that the atonal gene- which consists of a single exon- overlaps with four synonymous polymorphic SNPs across the four pooled samples"  
+
+L270 & 273 Consider removing "Notably" from one of these sentences.  
+
+## response:  
+
+Done! Line 337 the word "Notably" is removed.  
+
+L279 I'm sure the authors wish they could pinpoint more closely candidate regulatory elements. On possibility is to bioinformaticaly identify differences in TF motifs between lines in the candidate region and how the presence/absence of SNPs affect motif occurrences. The online resource MEME Suite might be useful to explore and get a bit closer to the actual causal SNP(s). https://meme- suite.org/meme. Of course, CHIP- seq or ATAC- seq could also identify some candidate positions.  
+
+## response:  
+
+We thank you for this suggestion. Following the recommendation, we performed motif discovery using the MEME Suite to identify potential transcription factor binding motifs in the candidate region. We also explored publicly available ATAC- seq and FAIRE- seq datasets from the iBeetleBase genome browser (https://ibeetle- base.uni- goettingen.de/genomebrowser/) to examine chromatin accessibility.  
+
+While the MEME analysis identified transcription factor motifs overlapping several genes in our candidate region, including atonal, the ATAC- seq and FAIRE- seq data revealed strong peaks of chromatin accessibility within the region lacking annotated genes but enriched for highly differentiated SNPs. These results support the hypothesis that a cis- regulatory element in this intergenic region may be influencing atonal expression. We have added these findings to the manuscript and included a new figure illustrating both the motif locations and peaks of chromatin accessibility. We now write in L341:  
+
+"This suggests that the phenotype may instead be caused by an as- yet- undiscovered regulatory sequence regulating atonal expression. We used two publicly available ATAC- seq (Mau et al. 2023) and FAIRE- seq (Lai et al. 2018) datasets to explore potential cis- regulatory regions. Particularly an area lacking annotated genes, yet enriched in highly differentiated SNPs, showed peaks of chromatin accessibility in both ATAC- seq and FAIRE- seq datasets (Extended Data Fig. 9)."  
+
+Moreover, we have added the following description to the Methods section detailing how MEME was ran_line: 954
+
+<--- Page Split --->
+
+"To identify potential regulatory motifs within the candidate region, we used the MEME Suite(Bailey and Elkan 1994) (https://meme- suite.org) to perform de novo motif discovery. We used the full 100 kb candidate region as input and ran the analysis with default parameters."  
+
+L373 This is the closest we get to understanding if HSP90 expression is affected by stress. It is unfortunate that it is not in the environment used in this study.  
+
+## response:  
+
+We recognize the importance of directly linking HSP90 regulation to specific environmental conditions. However, as mentioned before, it is conceptually not required that it is the same environmental condition that leads to reduced HSP90 availability and the phenotypes' fitness advantage. Our earlier work in T. castaneum demonstrated downregulation of HSP90 transcripts (Hsp83) in response to stressful social cues, specifically the presence of wounded conspecifics(Peuss et al. 2015). Other studies, in both T. castaneum and various insects, show that HSP90 expression is modulated by multiple forms of environmental stress, such as temperature extremes and chemical exposure (Brom et al. 2015; Ding et al. 2021; Liang et al. 2023). However, as you correctly note, our present study did not measure HSP90 levels under the exact selection regimes implemented here. We have now addressed this point in the revised discussion and have more clearly stated the need for future work that directly investigates HSP90 regulation in the context of these environments.  
+
+Line 469  
+
+"In a previous study in T. castaneum, HSP90 downregulation was observed in a risky environment, i.e., in cohabitation with wounded conspecifics(Peuss et al. 2015). Although this regulatory response was not directly addressed in the present study, the two studies together suggest a scenario that links evolutionary capacitance—and thus evolvability of a population—to environmental conditions...."  
+
+L601: I am confused about the abnormal leg phenotype and its being reported. Did this always co- occur with the reduced eyes? If not, how often was reduced- eyes by itself and why weren't these lines chosen instead of mixed phenotype lines? Why were leg phenotypes used to establish eye phenotype lines?  
+
+## response:  
+
+We appreciate the opportunity to clarify this point. Initially, F1 beetles exhibiting visible morphological abnormalities – most commonly leg malformations – were selected for generating F2 lines. This approach was primarily practical: such abnormalities served as visible markers of developmental disruption, suggesting that cryptic genetic variation might have been unmasked. The goal was to increase the chance of observing heritable phenotypes such as the reduced- eye trait in the F2 generation.  
+
+When the reduced- eye phenotype was first observed in F2 offspring from abnormal- leg F1 beetles, we tested whether this trait was linked to the leg malformation by also crossing their normal- legged full siblings from the same RNAi parental families. These crosses also yielded reduced- eye offspring in the F2, indicating that the reduced- eye and leg malformation phenotypes were not genetically linked.  
+
+It is also important to clarify that our reduced- eye lines were established only after observing this specific phenotype in the F2 generation. While these were initially derived
+
+<--- Page Split --->
+
+from families in which the F1 showed leg abnormalities, the reduced- eye lines themselves did not later display leg malformations, and the eye phenotype became monomorphic in subsequent generations.  
+
+To reflect this in the manuscript, we have added the following clarification in the Materials and Methods section: line 725  
+
+"... Notably, the reduced- eye phenotype was not genetically linked to leg malformations, as it also appeared in F2 offspring from normal- leg F1 siblings of the same RNAi- treated families (unpublished data)."  
+
+L641: Again, unclear why Hsp83 wasn't measured.  
+
+## response:  
+
+This point is now clarified in the Results section, where we explain that 17- DMAG inhibits HSP90 (encoded by Hsp83) at the protein level, not at the level of mRNA expression. Therefore, Hsp83 mRNA is not expected to respond directly to inhibition.  
+
+L680- 681: This is a clever design.  
+
+## response:  
+
+Thank you – we appreciate the positive feedback on our experimental design.  
+
+Fig. 1  
+
+I don't understand the diagram in 1a, what are the small arrows and red legs highlighting? It doesn't say in the legend.  
+
+In 1b, % frequency for the y- axis is odd. I'm not sure how to interpret it. The legend says that the % eye ptype is above the gold bar, but those numbers are more likely sample sizes. I think something must be missing.  
+
+In 1d, how were groups categorized, especially since there is overlap with some normal- eye having areas smaller than reduced- eye? Should it be treatment or line rather than ptype for the colors in panels in d- f?  
+
+## response:  
+
+Thank you for your careful evaluation and helpful observations regarding Figure 1. We address each point below:  
+
+1. Figure 1a – red arrows and legs: You are absolutely right – the small arrows were intended to indicate that the F2 reduced-eye phenotypes originated from F1 beetles with abnormal leg phenotypes (highlighted in red). These represented the families from which reduced-eye lines were established. However, as clarified in the text, we later confirmed that the reduced-eye phenotype is not genetically linked to leg malformations, since it also appeared in F2 offspring from normal-legged siblings of the same F1 families. To avoid confusion, we have removed these small arrows and red leg highlights from the updated version of the figure.  
+
+2. Figure 1b – % frequency and legend labeling: It is correct that the y-axis represents the frequency (%) of the reduced-eye phenotype within the screened F2 beetles in each treatment. The values above the bars correspond to sample sizes, not phenotypic percentages. The note in the legend stating that “% eye ptype is above
+
+<--- Page Split --->
+
+the gold bar" was left over from an earlier version of the figure and was mistakenly not removed when the plot format was updated. We have now corrected this in the revised legend.  
+
+3. Figure 1d - group categorization and classification: As noted earlier in this response letter and the manuscript, normalization of eye size was performed within each developmental stage. Individual beetles were tracked throughout development, and adult eye phenotype (normal or reduced) was assigned based on adult morphology using clear visual criteria, allowing retrospective classification across earlier stages. While there is some overlap in eye area between phenotypic classes, classification was based on adult appearance, which was visually distinct and confirmed by additional morphological analyses (see Fig. 1e for ommatidia counts). We've clarified this further in the legend to avoid ambiguity.  
+
+4. Color coding in panels d-f: We used eye phenotype (normal vs. reduced) as the common variable to categorize data in panels d-f to ensure consistency across the subfigures. Treatments or source lines varied across panels and were not consistently applicable, while eye phenotype provided a meaningful and common dimension for comparison. We note that:  
+
+- Panel d compares normalized eye area by phenotype across developmental stages- Panel e compares ommatidia counts by adult eye phenotypes- Panel f compares reproductive output by phenotype under two environmental conditions  
+
+## Reviewer #2 (Remarks to the Author):  
+
+The study by Sayed et al. explores the underappreciated evolutionary role of Hsp90 buffering in animals, identifying a beetle eye size trait whose variation is masked by Hsp90. Their findings suggest that Hsp90 enables the accumulation of cryptic genetic variation in natural populations. Transient Hsp90 inhibition (via chemical or siRNA) reveals a reduced-eye phenotype that persists in a subset of F1 and F2 progeny, can be driven to near fixation through crosses, and shows fitness advantages under constant light. A dimorphism is noted, and atonal is proposed as a phenocopy for Hsp90 inhibition. These findings are important and intriguing.  
+
+## response:  
+
+Thank you very much for your positive and thoughtful feedback on our work. We're glad that you found the conceptual approach and the main findings important and intriguing. Your encouraging comments were very motivating during the revision process and helped us further refine the manuscript.  
+
+While this is a compelling example of Hsp90 buffering in animals, the manuscript suffers from overstated claims, limited novelty, and interpretational issues. The conclusions need to be tempered or supported with additional data. The introduction and discussion should better contextualize the work within existing evidence supporting the adaptive value of Hsp90- buffered variation, avoiding claims of controversy or primacy. The experimental rigor and statistical analyses also require improvement. That said, the manuscript is clearly written and presents an intriguing story. I would support publication following major revisions to address the concerns outlined below.
+
+<--- Page Split --->
+
+## response:  
+
+Thank you for the thoughtful and constructive review. We have made substantial revisions to address your concerns:  
+
+- Tempered claims and improved context: We revised the Abstract, Introduction, and Discussion to avoid overstating novelty or controversy. The study is now clearly framed as an extension of previous work, highlighting HSP90's buffering in an animal under a certain environmental condition. We now more clearly acknowledge prior evidence for Hsp90 buffering and clarify where the novelty of our findings lies.  
+
+## - New supporting data:  
+
+- RNAi knockdown of Hsp83 reduces atonal expression, supporting a regulatory link (Fig.3g).- Atonal is consistently downregulated ( \(\sim 20\%\) ) in reduced-eye monomorphic lines (Fig.3f).- Computational analyses (Extended Data Fig. 9) reveal conserved cis-regulatory motifs within accessible chromatin near candidate loci.  
+
+These additions strengthen the mechanistic basis of our conclusions and address interpretational concerns. We appreciate your feedback and believe the manuscript is now significantly improved  
+
+## Major issues:  
+
+1. Conceptual Framing and Literature Context: The manuscript overstates the novelty and conceptual primacy of Hsp90 as an evolutionary capacitor. While this study advances our understanding of Hsp90-buffering in animals, the concept is well-established, particularly in yeast. The authors misrepresent the state of the field by framing the adaptive value of Hsp90-buffered variation as controversial. Multiple prior studies, including those by Lindquist and others (e.g., PMID: 16195452, 21205668, 39052788), have clearly demonstrated adaptive, genetically defined Hsp90-buffered traits. The manuscript also lacks appropriate citations and introduces misleading statements about the novelty and controversy of this mechanism. All novelty claims should be carefully rephrased to reflect that the contribution refers to Hsp90's role in animal evolution not evolution in general. I suggest revising lines 27-31, 37-38, 78-79, 86-87, 98, 169-170, 283-284, and 357-358 (these may not be all the sentences in need of revision to address this issue).  
+
+## response:  
+
+Tempering of claims and improved contextualization: We revised the Abstract, Introduction, and Discussion to avoid implying novelty or controversy where not appropriate. Statements suggesting primacy or debate (e.g., about the role of HSP90 buffering in animals) were rephrased to more clearly position our study within the broader literature. We agree that studies in yeast and other fungi have demonstrated adaptive, genetically defined HSP90- buffered traits, and make it now clearer that the novelty of our study lies in the fact that to our knowledge no study in any animal has so far connected a newly released, HSP90- buffered trait to its genetic basis and fitness consequences.  
+
+2. The ecological and adaptive relevance of the reduced-eye trait is inadequately supported. The authors test it under artificial, non-ecological conditions (e.g., continuous light), without exploring whether it is beneficial or deleterious under more natural
+
+<--- Page Split --->
+
+environments. As a loss- of- function phenotype, reduced- eye is unlikely to confer broad fitness advantages, and its frequency in wild populations remains unexamined. Strong claims about its evolutionary significance require broader validation and testing of alternative hypotheses, including potential fitness costs. Overall, the trait's ecological relevance is unclear, and the study lacks the rigor seen in comparable work on Hsp90- buffered traits in yeast.  
+
+## response:  
+
+We thank you for highlighting these important points, which we believe raise valid considerations regarding the ecological context and evolutionary relevance. Below, we provide our response to each point individually.  
+
+Regarding the relevance of our experimental conditions: While we acknowledge the concern regarding "artificial" or "non- ecological" laboratory conditions (e.g., continuous light), we respectfully note that \(T.\) castaneum is now a globally distributed species and a major pest of post- harvest grains and flour stored in human- made environments – including flour mills, warehouses, and storage bags – where beetles naturally encounter a range of light conditions (continuous, intermittent, or absent) and temperature fluctuations. Our laboratory setup, using flour as substrate and manipulating light and temperature, was specifically chosen to reflect these ecologically relevant, real- world habitats.  
+
+To further ensure relevance, we also evaluated fitness under constant darkness and mild heat stress \((+5^{\circ}C)\) – conditions also common in grain storage environments – but found no apparent benefit of the reduced- eye phenotype under these treatments (data not shown). Therefore, the observed fitness advantage under continuous light appears context- dependent, but still grounded in environmental conditions \(T.\) castaneum naturally experiences in its current human- associated niche.  
+
+We believe this experimental design effectively mimics the abiotic factors and substrate conditions typical of modern Tribolium habitats, providing a meaningful context to assess the adaptive value of the reduced- eye trait. Pointer et al. (2021) stated that "Owing to Tribolium's long human- commensal history ( \(\sim 70,000\) generations), the laboratory medium also has the advantage of very closely approximating its semi- wild habitat in food- storage facilities, allowing a lab environment that is less abstracted than that in other insect models".  
+
+Regarding the loss- of- function phenotype: While we agree that the reduced- eye phenotype – likely representing a loss- of- function – may not provide broad or universal fitness advantages, accumulating evidence suggests that loss or reduction of visual structures can be conditionally adaptive, especially in low- light or energetically constrained environments. Several studies have proposed that eye reduction or loss may be favored evolutionarily due to significant energy savings (Jeffery 2009; Moran, Softley, and Warrant 2015; Niven 2015; Ri as- Sánchez et al. 2025). Visual processing, including maintenance of photoreceptors and neural tissue, is metabolically costly: for example, cave fish devote 5–17% of total metabolic energy to vision (Moran et al. 2015), and in Drosophila melanogaster, photoreceptor activity alone accounts for around 8% of the organism's total energy usage (Laughlin, de Ruyter Van Steveninck, and Anderson 1998). Such costs rise in bright environments—energy demand may increase up to fourfold compared to darkness (Okawa et al. 2008).
+
+<--- Page Split --->
+
+Even minor reductions in eye size have been associated with decreased energetic expense through the positive scaling of visual neuropile and eye size (Niven 2015). Insects inhabiting dim or concealed microhabitats, such as the larval stages of \(D\) . melanogaster and \(T\) . castaneum, tend to exhibit behaviors and morphologies reflecting reduced reliance on vision (Busto, Iyengar, and Campos 1999; Park 1934). Since \(T\) . castaneum adults are also strongly photo- negative, reduced eye size could plausibly provide a selective advantage in the dark, resource- limited environments typical of grain storage habitats by minimizing the metabolic burden of maintaining unnecessary visual tissue.  
+
+- Regarding the frequency of the phenotype in wild populations: There is currently no published data on the frequency of the reduced-eye phenotype in wild or synanthropic populations of \(T\) . castaneum. While the trait has not been reported in field surveys, we believe it likely exists at very low frequency, given its spontaneous appearance in lab populations. Detecting it would require screening tens of thousands of individuals, which was beyond the scope of this study. Nonetheless, we agree this is an important and relevant direction for future research.  
+
+Finally, we recognize that the ecological relevance and evolutionary stability of this trait require further validation in field populations and across multiple environments, and we have revised the text to clarify that our conclusions concern context-dependent fitness effects, not broad adaptive significance. We also agree that testing for potential costs and evaluating trait frequency in natural populations remain important next steps.  
+
+2. The claim that atonal mediates Hsp90-buffered eye-size variation is not sufficiently demonstrated. While atonal RNAi phenocopies the trait, there is no genetic or expression evidence linking atonal to Hsp90 buffering in the beetle. Allele-specific expression, response to Hsp90 inhibition, and functional SNP data are lacking. The role of HDACs is weakly supported and lacks a positive control. Without stronger evidence, the proposed mechanism remains speculative and should be presented as such.  
+
+## response:  
+
+Thanks for these constructive comments. We have performed additional experiments to address them:  
+
+- Direct genetic and regulatory link: To strengthen our interpretation that HSP90 affects expression of atonal, we conducted new experiments and found that RNAi knockdown of Hsp83 significantly reduces atonal expression (new Fig. 3g), suggesting that HSP90 positively regulates atonal transcript levels.  
+
+- HDAC experiment and controls: The HDAC inhibition experiment is preliminary and was designed to explore whether epigenetic mechanisms may underlie the buffering of atonal expression. Beetles were fed flour discs containing either HDAC inhibitors or solvent-only (ethanol) as a control; additionally, we included loose flour (our standard maintenance medium) as a negative control.  
+
+Importantly, our new computational analyses (Extended Data Fig. 9) reveal conserved cis- regulatory motifs within accessible chromatin near candidate loci. We further tested atonal expression in our monomorphic reduced- eye lines and found it reduced compared to the ancestral Cro1 line (Fig. 3f). These results provide new evidence consistent with a regulatory link between chromatin state, Hsp83 function, and atonal expression.  
+
+- Caveats: We acknowledge that allele-specific expression data, natural-variant analysis, and functional SNP validation are currently lacking. These are all valuable directions for future work, which we plan to take into account in the next phase of the project. While these
+
+<--- Page Split --->
+
+data would further strengthen the proposed HSP90- atonal pathway, we believe our new results provide a solid foundation for this model in T. castaneum.  
+
+4. Several conclusions overreach the presented data. Key findings (e.g., persistence of the trait, its selection, and its mechanistic basis) have low penetrance, limited replicates, and insufficient statistical support. Alternative hypotheses are not tested, and the environmental relevance of findings is unclear. Overall, stronger controls, clearer definitions, and more cautious interpretation are needed throughout. Statistical analyses are not robust. Effects on fitness are weak and may not be significant after multiple hypothesis testing.  
+
+## response:  
+
+We believe that we could address these aspects with our revisions. Persistence of the reduced- eye trait was followed for many generations in our monomorphic lines, and stable phenotype frequencies were observed over many generations in our polymorphic lines, allowing us to use the F50 for Bulk Segregation Analysis. Even though effect sizes may appear to be weak, they are in an evolutionary relevant range, e.g. regarding fitness effects. Statistical analyses are not limited by insufficient replicate numbers, and multiple hypothesis testing was performed when needed.  
+
+Minor comments:  
+
+1. In place of "Hsp90-released trait" I suggest "Hsp90-buffered trait"; lines 38 and 74.  
+
+## response:  
+
+Done - "HSP90- released trait" has been replaced with "HSP90- buffered trait" at the indicated lines. Now lines 40, 66 and 104  
+
+2. The keyword "plasticity" was not mentioned. Remove keyword or add explanation of the term in the text.  
+
+## response:  
+
+Done - the keyword "plasticity" has been removed as requested.  
+
+3. Statistical significance, \* should be replaced with increments \*\*, \*\*\*, \*\*\*\*. A single asterisk (*) indicating a p-value less than 0.05 (significant), two asterisks (**) a p-value less than 0.01 (highly significant), and three asterisks (***) indicate a p-value less than 0.001 (very highly significant).  
+
+## response:  
+
+Done - we have updated all significance indicators in the figures and legends to follow the standard format: \(p < 0.05\) (*), \(p < 0.01\) (**) and \(p < 0.001\) (***), as requested.  
+
+4. Hsp90 is an epistatic regulator but no epistasis experiments were performed.  
+
+## response:  
+
+Thank you for mentioning this aspect. While we did not perform classical genetic interaction (epistasis) assays in this study, our results are consistent with HSP90 acting as an epistatic
+
+<--- Page Split --->
+
+regulator. Specifically, knockdown of Hsp83 released the reduced- eye phenotype; whole- genome sequencing and functional analysis identified the transcription factor atonal as the underlying gene; and direct knockdown of atonal robustly reproduced the same phenotype. Additionally, we observed a positive correlation between HSP90 and atonal expression, with atonal being downregulated after Hsp83 knockdown. Atonal likely interacts epistatically with further, potentially also HSP90- affected proteins, but it is complex to address this in T. castaneum and beyond the scope of our study.  
+
+5. Abstract lines 33-34: "stably inherited without continued HSP90 disruption" implies all revealed variation was stable but thew data show closer to \(20\%\) of the revealed variation is inherited and further declines in each generation so it is not that stable.  
+
+## response:  
+
+We appreciate this comment and understand the potential confusion, which likely stems from the polymorphic lines that harbour the reduced- eye allele at a certain gene frequency (Fig. 2a, lower two panels). It is important to note the lines are polymorphic populations and not clonal lines. In these lines, the allele is stably inherited, but the phenotype frequency depends on the allele frequency in these populations. Importantly, it was possible to select for the reduced- eye phenotype and produce monomorphic lines that are likely fixed for the reduced- eye allele and thus show close to \(100\%\) reduced- eye phenotypes (Fig. 2a upper panel), which was maintained over all successive generations up to now. The original phrasing "stably inherited without continued HSP90 disruption" was used to convey that once established, the reduced- eye phenotype continued to persist in subsequent generations, even though HSP90 function was no longer perturbed. We have revised the abstract language to clarify the aspect of persistence in the population.  
+
+The revised abstract line now reads: line 35  
+
+"...we consistently revealed a reduced- eye phenotype that persisted in descendant lines across generations without continued HSP90 disruption."  
+
+6. Figures should be revised to improve clarity and intuitiveness. In particular, the presentation of penetrance—such as in the HDAC inhibition example—is often unclear and could benefit from more straightforward visualization.  
+
+A more quantitative trait would be the number of omatids per animal, which although labor intensive seems to present with a great dynamic range and could foster a more robust statistical analysis approach. I understand that counting the number of omatids in each animal may not be feasible, so this is only a minor comment/suggestion.  
+
+## response:  
+
+Thank you for the thoughtful suggestions regarding data visualization and quantification. In response to the comment on HDAC inhibition (Extended Data Fig. 6): We have revised the figure to improve its clarity and informativeness. Specifically, we now provide the mean and standard error for each group in addition to individual datapoints. Furthermore, we adjusted the y- axis scale, setting the lower bound to \(50\%\) instead of \(0\%\) , to better focus on the region where all data points reside (60- 100% penetrance). This zoomed- in view makes the small yet statistically significant effects more visually interpretable. The figure caption has also been updated to explicitly state this.  
+
+Regarding the suggestion to use omnatidia number as a quantitative trait: We appreciate this idea and agree that omnatidial counts can, in many systems, provide a
+
+<--- Page Split --->
+
+detailed, quantitative metric. In our study, we performed a limited quantification of ommatidia number to support our conclusions (shown in Fig. 1e). Specifically, we manually counted the number of ommatidia in both compound eyes of adult beetles with clearly distinguishable eye morphology ( \(n = 8\) ) and used the average per beetle for analysis. The counts were compared between normal- and reduced- eye individuals using an unpaired two- sample t- test, which is regarded as a statistically robust method even with modest sample sizes.  
+
+However, we found that in \(T.\) castaneum, the number of ommatidia is relatively small and consistent between individuals of the same phenotype—approximately 40 ommatidia in normal eyes and \(\sim 20\) in reduced- eye phenotypes from ventral side—resulting in two discrete, stable categories, rather than a broad dynamic range. In addition, reliable ommatidial quantification was only possible in adults with well- structured compound eyes. Many reduced- eye animals showed fused or misshapen ommatidia, making accurate counting unreliable in a substantial proportion of samples.  
+
+Due to these biological and technical constraints, we used phenotypic penetrance as our main metric, which was more scalable and practical for quantifying trait expression across all treatments and lines. Nonetheless, ommatidial counts in Fig. 1e provide important validation of the morphological distinctions between phenotypes.  
+
+We hope these revisions have addressed your concerns, and we are grateful for your comments that helped us strengthen the manuscript considerably.  
+
+Reviewer #3 (Remarks to the Author):  
+
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+## response:  
+
+We thank Reviewer #3 for his/her contribution to the peer review process and for participating in the Nature Communications initiative supporting Early Career Researchers. We appreciate the time and effort dedicated to co- reviewing our manuscript.
+
+<--- Page Split --->
+
+## References  
+
+Alford, Brian D., and Onn Brandman. 2018. 'Quantification of Hsp90 Availability Reveals Differential Coupling to the Heat Shock Response'. Journal of Cell Biology 217(11). doi:10.1083/jcb.201803127. Bailey, Timothy L., and Charles Elkan. 1994. 'Fitting a Mixture Model by Expectation Maximization to Discover Motifs in Biopolymers'. in Proceedings of the 2nd International Conference on Intelligent Systems for Molecular Biology, ISMB 1994. Borkovich, K. A., F. W. Farrelly, D. B. Finkelstein, J. Taulien, and S. Lindquist. 1989. 'Hsp82 Is an Essential Protein That Is Required in Higher Concentrations for Growth of Cells at Higher Temperatures.' Molecular and Cellular Biology 9(9):3919- 30. doi:10.1128/MCB.9.9.3919. Updated. Brom, Krzysztof Roman, Bogdan Dolezych, Monika Tarnawska, Katarzyna Brzozowska, and Miroslaw Nakonieczny. 2015. 'Expression of the Hsp40, Hsp70 and Hsp90 Proteins in Colorado Potato Beetle (Leptinotarsa Decemlineata Say) after the Dimethoate Treatment'. Journal of the Entomological Research Society 17(2). Brown, Susan J., Teresa D. Shippy, Sherry Miller, Renata Bolognesi, Richard W. Beeman, Marcé D. Lorenzen, Gregor Bucher, Ernst a. Wimmer, and Martin Klingler. 2009. 'The Red Flour Beetle, Tribolium Castaneum (Coleoptera): A Model for Studies of Development and Pest Biology'. Cold Spring Harbor Protocols 4(8):1- 12. doi:10.1101/pdb.emo126. Busto, M., B. Iyengar, and a R. Campos. 1999. 'Genetic Dissection of Behavior: Modulation of Locomotion by Light in the Drosophila Melanogaster Larva Requires Genetically Distinct Visual System Functions'. The Journal of Neuroscience: The Official Journal of the Society for Neuroscience 19(9):3337- 44. doi:10.1523/JNEUROSCI.19- 09- 03337.1999. Chen, Bing, and Andreas Wagner. 2012. 'Hsp90 Is Important for Fecundity, Longevity, and Buffering of Cryptic Deleterious Variation in Wild Fly Populations'. BMC Evolutionary Biology 12(1):25. doi:10.1186/PREACCEPT- 2105524459276497. Ding, Jian Hao, Lu Xin Zheng, Jie Chu, Xin Hao Liang, Jun Wang, Xiao Wen Gao, Fu An Wu, and Sheng Sheng. 2021. 'Characterization, and Functional Analysis of Hsp70 and Hsp90 Gene Families in Glyphodes Pylolais Walker (Lepidoptera: Pyralidae)'. Frontiers in Physiology 12. doi:10.3389/fphys.2021.753914. Jeffery, William R. 2009. 'Regressive Evolution in Astyanax Cavefish'. Annual Review of Genetics 141(4):520- 29. doi:10.1016/j.surg.2006.10.010. Use. Jez, J. M., J. C. Chen, G. Rastelli, R. M. Stroud, and D. V Santi. 2003. 'Crystal Structure and Molecular Modeling of 17- DMAG in Complex with Human Hsp90'. Chem Biol 10(4):361- 68. http://www.ncbi.nlm.nih.gov/pubmed/12725864. Kudryavtsev, Vladimir A., Anna V. Khokhlova, Vera A. Mosina, Elena I. Selivanova, and Alexander E. Kabakov. 2017. 'Induction of Hsp70 in Tumor Cells Treated with Inhibitors of the Hsp90 Activity: A Predictive Marker and Promising Target for Radiosensitization'. PLoS ONE 12(3):1- 25. doi:10.1371/journal.pone.0173640. Lai, Yi Ting, Kevin D. Deem, Ferran Borras- Castells, Nagraj Sambrani, Heike Rudolf, Kushal Suryamohan, Ezzat El- Sherif, Marc S. Halfon, Daniel J. McKay, and Yoshinori Tomoyasu. 2018. 'Enhancer Identification and Activity Evaluation in the Red Flour Beetle, Tribolium Castaneum'. Development (Cambridge) 145(7). doi:10.1242/dev.160663. Laughlin, S. B., R. R. de Ruyter Van Steveninck, and J. C. Anderson. 1998. 'The Metabolic Cost of Neural Information'. Nature Neuroscience 1(1):36- 41. doi:10.1038/236. Liang, Chen, Lifang Li, Hang Zhao, Mingxian Lan, Yongyu Tang, Man Zhang, Deqiang Qin, Guoxing Wu, and Xi Gao. 2023. 'Identification and Expression Analysis of Heat Shock Protein Family Genes of Gall Fly (Procecidochares Utilis) under Temperature Stress'. Cell Stress and Chaperones 28(3). doi:10.1007/s12192- 023- 01338- 9.
+
+<--- Page Split --->
+
+Mau, Christine, Heike Rudolf, Frederic Strobl, Benjamin Schmid, Timo Regensburger, Ralf Palmisano, Ernst H. K. Stelzer, Leila Taher, and Ezzat El-Sherif. 2023. 'How Enhancers Regulate Wavelike Gene Expression Patterns'. ELife 12. doi:10.7554/eLife.84969. Moran, Damian, Rowan Softley, and Eric J. Warrant. 2015. 'The Energetic Cost of Vision and the Evolution of Eye-less Mexican Cavefish'. Science Advances 1(8):e1500363- e1500363. doi:10.1126/sciadv.1500363. Niven, Jeremy E. 2015. 'Neural Evolution: Costing the Benefits of Eye Loss'. Current Biology 25(19):R840- 41. doi:10.1016/j.cub.2015.08.050. Okawa, Haruhisa, Alapakkam P. Sampath, Simon B. Laughlin, and Gordon L. Fain. 2008. 'ATP Consumption by Mammalian Rod Photoreceptors in Darkness and in Light'. Current Biology 18(24):1917- 21. doi:10.1016/j.cub.2008.10.029. Park, T. 1934. 'Observations on the General Biology of the Flour Beetle, Tribolium Confusum'. Quarterly Review of Biology 9:36- 54. Peuss, R., Hendrik Eggert, Sophie A. O. Armitage, and Joachim Kurtz. 2015. 'Downregulation of the Evolutionary Capacitor Hsp90 Is Mediated by Social Cues'. Proceedings of the Royal Society B: Biological Sciences 282(1819):20152041- 20152041. doi:10.1098/rspb.2015.2041. Pointer, Michael D., Matthew J. G. Gage, and Lewis G. Spurgin. 2021. 'Tribolium Beetles as a Model System in Evolution and Ecology'. Heredity 126(6):869- 83. doi:10.1038/s41437- 021- 00420- 1. Ri as- Sánchez, David F., J. ake Morris, Camilo Salazar, Carolina P. ar do- Díaz, Richar M Merrill, Stephen H. Montgomery, and Richard M Merrill. 2025. 'Repea Ted e Volution of Reduced Visual in Vestment a t the Onset of Ecological Speciation in High- Altitude Heliconius Butterflies'. Evolution Letters 1- 9. doi:10.1093/el. Rutherford, S. L., and S. Lindquist. 1998. 'Hsp90 as a Capacitor for Morphological Evolution'. Nature 396(6709):336- 42. doi:10.1038/24550. Trepel, Jane, Mehdi Mollapour, Giuseppe Giaccone, and Len Neckers. 2010. 'Targeting the Dynamic HSP90 Complex in Cancer'. Nature Reviews Cancer 10(8):537- 49. doi:10.1038/nrc2887. Waddington, C. H. 1942. 'Canalization of Development and the Inheritance of Acquired Characters'. Nature 150(3811):563- 65. doi:10.1038/150563a0. Zhou, Dan, Yuan Liu, Josephine Ye, Weiwen Ying, Luisa Shin Ogawa, Takayo Inoue, Noriaki Tatsuta, Yumiko Wada, Keizo Koya, Qin Huang, Richard C. Bates, and Andrew J. Sonderfan. 2013. 'A Rat Retinal Damage Model Predicts for Potential Clinical Visual Disturbances Induced by Hsp90 Inhibitors'. Toxicology and Applied Pharmacology 273(2):401- 9. doi:10.1016/j.taap.2013.09.018.
+
+<--- Page Split --->
+
+## RESPONSE TO THE REVIEWER COMMENTS _2nd Round  
+
+## Dear Reviewers,  
+
+We sincerely thank you for your continued evaluation of our manuscript entitled "HSP90 as an Evolutionary Capacitor Drives Adaptive Eye Size Reduction via Atonal." We appreciate the additional feedback provided and have addressed the remaining points to further improve the clarity and robustness of our work.  
+
+Reviewer #1 (Remarks to the Author):  
+
+My biggest concerns in the previous version of the manuscript were the links between Hsp90 levels, phenotypic variation, and environmental conditions. In their response letter, the authors do a convincing job of explaining that the conditions that affected Hsp90 availability need not be the same conditions that select upon phenotypes and genotypes revealed by Hsp90 depletion. Further, they explained the link between measuring expression level of Hsp68 instead of Hsp83. By and large, I am satisfied with their responses to my initial comments and suggestions, and I am pleased by their inclusion of additional data and consideration of my comments. I have only a few suggestions for further clarifying these points in the text.  
+
+## response:  
+
+We are grateful for the reviewer's positive evaluation and are pleased that our previous revisions addressed the major concerns. As suggested, we have incorporated additional clarifications in the relevant sections to further strengthen the connections between HSP90 availability, phenotypic variation, and environmental context.  
+
+L87: Consider mentioning here that selection could be acting in the Hsp90 affecting environment or a different environment. " ...differences on which selection can act either in the disrupting environmental conditions or other conditions altogether."  
+
+## response:  
+
+The sentence has been revised as follows (L51 with "No Markup" as the adjusted option): "...differences on which selection can act either in the disrupting environmental conditions or other conditions altogether."  
+
+L186 consider adding a reminder here that continuous light COULD be encountered by these beetles in their human- commensal environment.  
+
+## response:  
+
+We added the following (L160 with "No Markup" option): "...continuous light stress, a situation that could be encountered by these beetles in their human- commensal environment."  
+
+L208 Consider changing "certain" to "some, albeit unknown, frequency"  
+
+## response:
+
+<--- Page Split --->
+
+This change has been made (L180 with "No Markup" option).  
+
+L321 "...unmasking cryptic genetic variation and producing new phenotypes." To be explicit about the consequence of that CGV being uncovered.  
+
+## response:  
+
+We added "and producing new phenotypes" (L293 with "No Markup" option).  
+
+## Reviewer #2 (Remarks to the Author):  
+
+All my comments have been fully addressed. I congratulate the authors on their beautiful work and approve of its publication.  
+
+## response:  
+
+We sincerely thank the reviewer for their very helpful previous comments that substantially improved our manuscript and are grateful for this positive feedback.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__c2c345558b96db063604d2a55b798464f0913fd48543918fbef69a413db242a7/supplementary_0_Peer Review file__c2c345558b96db063604d2a55b798464f0913fd48543918fbef69a413db242a7_det.mmd

@@ -0,0 +1,806 @@
+<|ref|>title<|/ref|><|det|>[[72, 53, 295, 80]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[74, 96, 296, 118]]<|/det|>
+Peer Review File  
+
+<|ref|>title<|/ref|><|det|>[[73, 161, 864, 210]]<|/det|>
+# HSP90 as an Evolutionary Capacitor Drives Adaptive Eye Size Reduction via Atonal  
+
+<|ref|>text<|/ref|><|det|>[[73, 224, 487, 240]]<|/det|>
+Corresponding Author: Professor Joachim Kurtz  
+
+<|ref|>text<|/ref|><|det|>[[72, 274, 864, 289]]<|/det|>
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+<|ref|>text<|/ref|><|det|>[[73, 327, 144, 340]]<|/det|>
+Version 0:  
+
+<|ref|>text<|/ref|><|det|>[[73, 354, 219, 367]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 379, 160, 393]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 404, 238, 431]]<|/det|>
+(Remarks to the Author) Overview:  
+
+<|ref|>text<|/ref|><|det|>[[73, 432, 896, 499]]<|/det|>
+I was excited to read this manuscript because the abstract suggested a significant advance and proof- of- principle for heritable, adaptive uncovering of cryptic genetic variation via Hsp90 inhibition. By and large, I think this promise from the abstract was met by the study, I did very much enjoy reading the manuscript, and I think it has a lot to add to the field. However, there are several points that need to be clarified, potentially with additional data, for the manuscript to have an even more compelling story.  
+
+<|ref|>text<|/ref|><|det|>[[73, 511, 187, 524]]<|/det|>
+Major Comment:  
+
+<|ref|>text<|/ref|><|det|>[[72, 535, 923, 655]]<|/det|>
+The major shortcoming I see (that potentially extends to other studies of Hsp90 and CGV) is that we are not given evidence that Hsp90 expression or availability is affected by environmental change/stress. The authors functionally manipulate Hsp90 with RNAi and drugs, then assay fitness in different environments. We don't know however, if the different environment- continuous light- - affects Hsp90. I think the most compelling case that corresponds to that laid out in this study would require: 1) Establishing a link between environment and Hsp90 expression/availability 2) Establishing a link between environment and phenotypic variation 3) Establishing a link between Hsp90 expression and phenotypic variation 4) Establishing link between phenotypic variation and fitness, especially in the inducing environment 5) Establishing that phenotypic variation is inherited across generation  
+
+<|ref|>text<|/ref|><|det|>[[72, 666, 920, 720]]<|/det|>
+The authors address points 3- 4, but 1 & 2 remain untested, perhaps in any system. On L72- 74 for instance, it is unclear if the references refer to the first clause "However, HSP90 availability may become limited under stressful environmental conditions," or the second clause, "as it is required by numerous proteins that become damaged during stress" or both. If there are no references to support the first clause, then it seems like a major gap that seeks filling.  
+
+<|ref|>text<|/ref|><|det|>[[72, 730, 916, 784]]<|/det|>
+The authors should consider the effects of constant light itself of phenotype production and HSP90 expression. They ought to consider discussing how, in nature for example, a reduction of HSP90 expression (akin to their manipulations) could occur and have fitness differences be borne out in a novel environment (e.g. constant light) if that environment does or does not affect (we don't know from the current data) the expression itself.  
+
+<|ref|>text<|/ref|><|det|>[[72, 796, 384, 822]]<|/det|>
+Additional comments: L43: note what kind of variation is meant here  
+
+<|ref|>text<|/ref|><|det|>[[72, 834, 923, 875]]<|/det|>
+L102- 106: The reference to the previous study on social cues makes the reader think that this will be an important element of the current study even though it is not. The authors should consider ending with a brief overview of their study, its phenotypes, environments, and aims.  
+
+<|ref|>text<|/ref|><|det|>[[70, 886, 884, 914]]<|/det|>
+L108: "HSP90 reduction, mimicking what may occur during environmental stress, in ..." Again, evidence of HSP90 with environmental stress seems like a key point that is not well- established from what I see in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[72, 925, 647, 940]]<|/det|>
+L119- 122 seems to disagree with the figure 1 legend. The figure says F2 had \(27.8\%\)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 60, 897, 74]]<|/det|>
+L138 Why not Hsp83? It seems odd that a different gene is being assayed here when everything else focused on Hsp83.  
+
+<|ref|>text<|/ref|><|det|>[[70, 86, 912, 114]]<|/det|>
+L158: How were reduced- and normal- eye beetles operationally delimited? Knowing the cut- off metric is important because much of the phenotypic and fitness data between the groups overlap (Fig. 1).  
+
+<|ref|>text<|/ref|><|det|>[[72, 125, 494, 139]]<|/det|>
+L210 Please note that the deviation was lower than expected.  
+
+<|ref|>text<|/ref|><|det|>[[72, 152, 914, 192]]<|/det|>
+L273- 274 I think omission of some commas in the sentence make it a bit awkwardly worded. Do you mean "Notably, our mapping analysis shows that the atonal gene, which consists of a single exon, overlaps with four synonymous polymorphic SNPs across the four pooled samples..."?  
+
+<|ref|>text<|/ref|><|det|>[[72, 204, 555, 218]]<|/det|>
+L270 & 273 Consider removing "Notably" from one of these sentences.  
+
+<|ref|>text<|/ref|><|det|>[[72, 230, 918, 296]]<|/det|>
+L279 I'm sure the authors wish they could pinpoint more closely candidate regulatory elements. On possibility is to bioinformatically identify differences in TF motifs between lines in the candidate region and how the presence/absence of SNPs affect motif occurrences. The online resource MEME Suite might be useful to explore and get a bit closer to the actual causal SNP(s). https://meme- suite.org/meme. Of course, CHIP- seq or ATAC- seq could also identify some candidate positions.  
+
+<|ref|>text<|/ref|><|det|>[[70, 308, 900, 335]]<|/det|>
+L373 This is the closest we get to understanding if HSP90 expression is affected by stress. It is unfortunate that it is not in the environment used in this study.  
+
+<|ref|>text<|/ref|><|det|>[[72, 347, 901, 387]]<|/det|>
+L601: I am confused about the abnormal leg phenotype and its being reported. Did this always co- occur with the reduced eyes? If not, how often was reduced- eyes by itself and why weren't these lines chosen instead of mixed phenotype lines? Why were leg phenotypes used to establish eye phenotype lines?  
+
+<|ref|>text<|/ref|><|det|>[[72, 398, 422, 412]]<|/det|>
+L641: Again, unclear why Hsp83 wasn't measured.  
+
+<|ref|>text<|/ref|><|det|>[[72, 425, 305, 438]]<|/det|>
+L680- 681: This is a clever design.  
+
+<|ref|>text<|/ref|><|det|>[[72, 451, 111, 464]]<|/det|>
+Fig. 1  
+
+<|ref|>text<|/ref|><|det|>[[72, 464, 899, 530]]<|/det|>
+I don't understand the diagram in 1a, what are the small arrows and red legs highlighting? It doesn't say in the legend. In 1b, % frequency for the y- axis is odd. I'm not sure how to interpret it. The legend says that the % eye ptype is above the gold bar, but those numbers are more likely sample sizes. I think something must be missing. In 1d, how were groups categorized, especially since there is overlap with some normal- eye having areas smaller than reduced- eye? Should it be treatment or line rather than ptype for the colors in panels in d- f?  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 580, 161, 593]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[72, 606, 238, 619]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 619, 921, 698]]<|/det|>
+The study by Sayed et al. explores the underappreciated evolutionary role of Hsp90 buffering in animals, identifying a beetle eye size trait whose variation is masked by Hsp90. Their findings suggest that Hsp90 enables the accumulation of cryptic genetic variation in natural populations. Transient Hsp90 inhibition (via chemical or siRNA) reveals a reduced- eye phenotype that persists in a subset of F1 and F2 progeny, can be driven to near fixation through crosses, and shows fitness advantages under constant light. A dimorphism is noted, and atonal is proposed as a phenocopy for Hsp90 inhibition. These findings are important and intriguing.  
+
+<|ref|>text<|/ref|><|det|>[[72, 709, 920, 789]]<|/det|>
+While this is a compelling example of Hsp90 buffering in animals, the manuscript suffers from overstated claims, limited novelty, and interpretational issues. The conclusions need to be tempered or supported with additional data. The introduction and discussion should better contextualize the work within existing evidence supporting the adaptive value of Hsp90- buffered variation, avoiding claims of controversy or primacy. The experimental rigor and statistical analyses also require improvement. That said, the manuscript is clearly written and presents an intriguing story. I would support publication following major revisions to address the concerns outlined below.  
+
+<|ref|>text<|/ref|><|det|>[[72, 802, 165, 815]]<|/det|>
+Major issues:  
+
+<|ref|>text<|/ref|><|det|>[[72, 827, 917, 945]]<|/det|>
+1. Conceptual Framing and Literature Context: The manuscript overstates the novelty and conceptual primacy of Hsp90 as an evolutionary capacitor. While this study advances our understanding of Hsp90-buffering in animals, the concept is well-established, particularly in yeast. The authors misrepresent the state of the field by framing the adaptive value of Hsp90-buffered variation as controversial. Multiple prior studies, including those by Lindquist and others (e.g., PMID: 16195452, 21205668, 39052788), have clearly demonstrated adaptive, genetically defined Hsp90-buffered traits. The manuscript also lacks appropriate citations and introduces misleading statements about the novelty and controversy of this mechanism. All novelty claims should be carefully rephrased to reflect that the contribution refers to Hsp90's role in animal evolution not evolution in general. I suggest revising lines 27-31, 37-38, 78-79, 86-87, 98, 169-170, 283-284, and 357-358 (these may not be all the sentences in need of revision to address this issue).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 60, 923, 140]]<|/det|>
+2. The ecological and adaptive relevance of the reduced-eye trait is inadequately supported. The authors test it under artificial, non-ecological conditions (e.g., continuous light), without exploring whether it is beneficial or deleterious under more natural environments. As a loss-of-function phenotype, reduced-eye is unlikely to confer broad fitness advantages, and its frequency in wild populations remains unexamined. Strong claims about its evolutionary significance require broader validation and testing of alternative hypotheses, including potential fitness costs. Overall, the trait's ecological relevance is unclear, and the study lacks the rigor seen in comparable work on Hsp90-buffered traits in yeast.  
+
+<|ref|>text<|/ref|><|det|>[[72, 152, 923, 218]]<|/det|>
+3. The claim that atonal mediates Hsp90-buffered eye-size variation is not sufficiently demonstrated. While atonal RNAi phenocopies the trait, there is no genetic or expression evidence linking atonal to Hsp90 buffering in the beetle. Allele-specific expression, response to Hsp90 inhibition, and functional SNP data are lacking. The role of HDACs is weakly supported and lacks a positive control. Without stronger evidence, the proposed mechanism remains speculative and should be presented as such.  
+
+<|ref|>text<|/ref|><|det|>[[72, 230, 907, 296]]<|/det|>
+4. Several conclusions overreach the presented data. Key findings (e.g., persistence of the trait, its selection, and its mechanistic basis) have low penetrance, limited replicates, and insufficient statistical support. Alternative hypotheses are not tested, and the environmental relevance of findings is unclear. Overall, stronger controls, clearer definitions, and more cautious interpretation are needed throughout. Statistical analyses are not robust. Effects on fitness are weak and may not be significant after multiple hypothesis testing.  
+
+<|ref|>text<|/ref|><|det|>[[72, 308, 192, 321]]<|/det|>
+Minor comments:  
+
+<|ref|>text<|/ref|><|det|>[[72, 333, 645, 348]]<|/det|>
+1. In place of "Hsp90-released trait" I suggest "Hsp90-buffered trait"; lines 38 and 74.  
+
+<|ref|>text<|/ref|><|det|>[[72, 359, 795, 374]]<|/det|>
+2. The keyword "plasticity" was not mentioned. Remove keyword or add explanation of the term in the text.  
+
+<|ref|>text<|/ref|><|det|>[[72, 385, 920, 426]]<|/det|>
+3. Statistical significance, \\* should be replaced with increments \\*\\*, \\*\\*\\*\\*\\*\\*\\*\\*\\*\\*\\* A single asterisk (\\*) indicating a p-value less than 0.05 (significant), two asterisks (\\*) a p-value less than 0.01 (highly significant), and three asterisks (\\*\\*) indicate a p-value less than 0.001 (very highly significant).  
+
+<|ref|>text<|/ref|><|det|>[[72, 437, 613, 452]]<|/det|>
+4. Hsp90 is an epistatic regulator but no epistasis experiments were performed.  
+
+<|ref|>text<|/ref|><|det|>[[72, 463, 905, 503]]<|/det|>
+5. Abstract lines 33-34: "stably inherited without continued HSP90 disruption" implies all revealed variation was stable but thew data show closer to \(20\%\) of the revealed variation is inherited and further declines in each generation so it is not that stable.  
+
+<|ref|>text<|/ref|><|det|>[[72, 515, 905, 580]]<|/det|>
+6. Figures should be revised to improve clarity and intuitiveness. In particular, the presentation of penetrance—such as in the HDAC inhibition example—is often unclear and could benefit from more straightforward visualization. A more quantitative trait would be the number of omatids per animal, which although labor intensive seems to present with a great dynamic range and could foster a more robust statistical analysis approach. I understand that counting the number of omatids in each animal may not be feasible, so this is only a minor comment/suggestion.  
+
+<|ref|>text<|/ref|><|det|>[[72, 592, 162, 605]]<|/det|>
+Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[72, 618, 237, 631]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 632, 864, 672]]<|/det|>
+I co-reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co-review manuscripts.  
+
+<|ref|>text<|/ref|><|det|>[[72, 684, 144, 697]]<|/det|>
+Version 1:  
+
+<|ref|>text<|/ref|><|det|>[[72, 709, 219, 722]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[72, 735, 160, 748]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[72, 761, 237, 787]]<|/det|>
+(Remarks to the Author) NCOMMS- 25- 2699A  
+
+<|ref|>text<|/ref|><|det|>[[72, 789, 920, 867]]<|/det|>
+My biggest concerns in the previous version of the manuscript were the links between Hsp90 levels, phenotypic variation, and environmental conditions. In their response letter, the authors do a convincing job of explaining that the conditions that affected Hsp90 availability need not be the same conditions that select upon phenotypes and genotypes revealed by Hsp90 depletion. Further, they explained the link between measuring expression level of Hsp68 instead of Hsp83. By and large, I am satisfied with their responses to my initial comments and suggestions, and I am pleased by their inclusion of additional data and consideration of my comments. I have only a few suggestions for further clarifying these points in the text.  
+
+<|ref|>text<|/ref|><|det|>[[70, 878, 923, 906]]<|/det|>
+L78: Consider mentioning here that selection could be acting in the Hsp90 affecting environment or a different environment. " ...differences on which selection can act either in the disrupting environmental conditions or other conditions altogether."  
+
+<|ref|>text<|/ref|><|det|>[[70, 918, 870, 945]]<|/det|>
+L186 consider adding a reminder here that continuous light COULD be encountered by these beetles in their human- commensal environment
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 60, 561, 74]]<|/det|>
+L208 Consider changing "certain" to "some, albeit unknown, frequency"  
+
+<|ref|>text<|/ref|><|det|>[[70, 86, 907, 114]]<|/det|>
+L321 "...unmasking cryptic genetic variation and producing new phenotypes." To be explicit about the consequence of that CGV being uncovered.  
+
+<|ref|>text<|/ref|><|det|>[[73, 165, 283, 179]]<|/det|>
+(Remarks on code availability)  
+
+<|ref|>text<|/ref|><|det|>[[73, 203, 162, 216]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[72, 230, 922, 258]]<|/det|>
+(Remarks to the Author) All my comments have been fully addressed. I congratulate the authors on their beautiful work and approve of its publication.  
+
+<|ref|>text<|/ref|><|det|>[[73, 269, 283, 283]]<|/det|>
+(Remarks on code availability)  
+
+<|ref|>text<|/ref|><|det|>[[72, 650, 916, 702]]<|/det|>
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+<|ref|>text<|/ref|><|det|>[[72, 702, 796, 715]]<|/det|>
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+<|ref|>text<|/ref|><|det|>[[72, 715, 911, 767]]<|/det|>
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+<|ref|>text<|/ref|><|det|>[[72, 767, 618, 780]]<|/det|>
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[107, 50, 496, 67]]<|/det|>
+## RESPONSE TO THE REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[107, 103, 257, 119]]<|/det|>
+## Dear Reviewers,  
+
+<|ref|>text<|/ref|><|det|>[[107, 137, 892, 206]]<|/det|>
+We appreciate the opportunity to revise our manuscript entitled "HSP90 as an Evolutionary Capacitor Drives Adaptive Eye Size Reduction via atonal." We are grateful for the reviewers' insightful comments and suggestions, which have helped us to improve the clarity and quality of our work.  
+
+<|ref|>text<|/ref|><|det|>[[108, 242, 428, 258]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>sub_title<|/ref|><|det|>[[107, 277, 195, 292]]<|/det|>
+## Overview:  
+
+<|ref|>text<|/ref|><|det|>[[107, 294, 885, 399]]<|/det|>
+I was excited to read this manuscript because the abstract suggested a significant advance and proof- of- principle for heritable, adaptive uncovering of cryptic genetic variation via Hsp90 inhibition. By and large, I think this promise from the abstract was met by the study, I did very much enjoy reading the manuscript, and I think it has a lot to add to the field. However, there are several points that need to be clarified, potentially with additional data, for the manuscript to have an even more compelling story.  
+
+<|ref|>text<|/ref|><|det|>[[107, 414, 220, 430]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[107, 445, 889, 532]]<|/det|>
+Thank you very much for your positive feedback and for recognizing the strengths and significance of our study. We are pleased that you found the manuscript both promising and enjoyable, and we appreciate your suggestions for clarifications and additional data to further strengthen the work. As detailed in the following, we added additional data as suggested.  
+
+<|ref|>sub_title<|/ref|><|det|>[[107, 546, 255, 563]]<|/det|>
+## Major Comment:  
+
+<|ref|>text<|/ref|><|det|>[[105, 576, 890, 788]]<|/det|>
+The major shortcoming I see (that potentially extends to other studies of Hsp90 and CGV) is that we are not given evidence that Hsp90 expression or availability is affected by environmental change/stress. The authors functionally manipulate Hsp90 with RNAi and drugs, then assay fitness in different environments. We don't know however, if the different environment- - - continuous light- - - affects Hsp90. I think the most compelling case that corresponds to that laid out in this study would require: 1) Establishing a link between environment and Hsp90 expression/availability 2) Establishing a link between environment and phenotypic variation 3) Establishing a link between Hsp90 expression and phenotypic variation 4) Establishing link between phenotypic variation and fitness, especially in the inducing environment 5) Establishing that phenotypic variation is inherited across generation  
+
+<|ref|>text<|/ref|><|det|>[[105, 803, 888, 909]]<|/det|>
+The authors address points 3- 4, but 1 & 2 remain untested, perhaps in any system. On L72- 74 for instance, it is unclear if the references refer to the first clause "However, HSP90 availability may become limited under stressful environmental conditions," or the second clause, "as it is required by numerous proteins that become damaged during stress" or both. If there are no references to support the first clause, then it seems like a major gap that seeks filling.  
+
+<|ref|>text<|/ref|><|det|>[[107, 924, 220, 939]]<|/det|>
+## response:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[106, 50, 892, 241]]<|/det|>
+Thank you for this important and insightful comment. We have conducted additional experiments to further substantiate the link between HSP90 and the reduced- eye phenotype, now showing that RNAi knockdown of Hsp83 reduces atonal expression, supporting a regulatory link (Fig.3g) and that atonal is consistently downregulated in reduced- eye monomorphic lines (Fig.3f). More details on these data are provided in our response to Reviewer 2. While these additional data provide further support for point 3), we also carefully considered points 1) and 2) and conducted an additional experiment to address these points (see below). While agreeing that directly linking environmental changes to HSP90 expression or availability could strengthen both our manuscript and the broader understanding of HSP90's role in adaptation, we would like to provide some conceptual clarifications.  
+
+<|ref|>text<|/ref|><|det|>[[106, 256, 892, 448]]<|/det|>
+Our study is conceptually grounded in the mechanism whereby HSP90 becomes limiting primarily under physiological stresses, in particular those that cause protein damage or denaturation - such as thermal, toxic, pathogenic or osmotic stress or other proteotoxic conditions - where demand for HSP90 is elevated and its availability for other cellular processes becomes restricted. Under these conditions, HSP90 is required for the stabilization and refolding of numerous damaged proteins, which may temporarily reduce its availability and thereby unmask cryptic genetic variation. We would like to point out that according to this concept, the environment leading to reduced HSP90 availability is not necessarily the same as the environment in which released phenotypes may show fitness advantages. Specifically, the environments in points 1) and 2) are not expected to be the same as the environments where fitness benefits manifest (point 4).  
+
+<|ref|>text<|/ref|><|det|>[[106, 462, 892, 566]]<|/det|>
+We also acknowledge your point regarding the need for a clear reference supporting this mechanism. In response, we identified and now cite Alford and Brandman (2018), which directly supports the statement that HSP90 availability can become limited under stressful conditions that challenge proteostasis. We have revised the relevant sentence in the manuscript for clarity and included this reference at line 83- 86. The revised sentence now reads:  
+
+<|ref|>text<|/ref|><|det|>[[107, 580, 875, 650]]<|/det|>
+"However, under stressful environmental conditions that disrupt proteostasis, HSP90 availability may become limited due to its involvement in stabilizing numerous proteins that are damaged or denatured during stress (Alford and Brandman 2018; Borkovich et al. 1989; Chen and Wagner 2012; Peuss et al. 2015)".  
+
+<|ref|>text<|/ref|><|det|>[[107, 681, 877, 768]]<|/det|>
+The authors should consider the effects of constant light itself of phenotype production and HSP90 expression. They ought to consider discussing how, in nature for example, a reduction of HSP90 expression (akin to their manipulations) could occur and have fitness differences be borne out in a novel environment (e.g. constant light) if that environment does or does not affect (we don't know from the current data) the expression itself.  
+
+<|ref|>text<|/ref|><|det|>[[107, 784, 220, 799]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[107, 814, 892, 901]]<|/det|>
+Thank you for raising this point regarding the potential effects of constant light on both phenotype production and HSP90 expression, as well as the ecological context for HSP90 reduction and fitness in novel environments. However, as noted above, please consider that it is conceptually not required that it is the same environmental condition that leads to reduced HSP90 availability and the phenotypes' fitness advantage.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[135, 50, 893, 155]]<|/det|>
+- To nevertheless address your comment, we conducted an experiment to directly test whether constant light affects HSP90 expression. Our results showed no significant change in HSP90 expression under constant light, supporting our expectation that this condition alone does not challenge proteostasis or induce a heat shock response. We did not include these additional data into the manuscript but are happy to provide them if requested.  
+
+<|ref|>text<|/ref|><|det|>[[135, 171, 893, 243]]<|/det|>
+- Regarding the production of the reduced-eye phenotype, our data indicate that the phenotype is only revealed when HSP90 function is experimentally reduced (via RNAi or chemical inhibition), not by constant light alone. This suggests that constant light acts as a selective environment, but not directly as an inducer of the phenotype.  
+
+<|ref|>text<|/ref|><|det|>[[135, 259, 893, 346]]<|/det|>
+- In nature, HSP90 limitation is most likely to occur under environmental stresses that cause protein damage or denaturation, where HSP90 becomes limited due to increased demand. In such scenarios, hidden genetic variation could be revealed, and, as our study demonstrates, may confer a fitness advantage in specific novel environments like continuous light.  
+
+<|ref|>text<|/ref|><|det|>[[135, 363, 893, 450]]<|/det|>
+- We have expanded the discussion in the revised manuscript to clarify these points and to address how natural environmental stresses could lead to HSP90 limitation and the expression of adaptive phenotypes, even if the novel environment itself (e.g., constant light) does not directly affect HSP90 expression. We have added the following sentences at line 368 in the Discussion:  
+
+<|ref|>text<|/ref|><|det|>[[135, 468, 893, 572]]<|/det|>
+"It is well established that certain environmental stresses can limit HSP90 availability by increasing the demand for its chaperone function (Alford and Brandman 2018), thereby unmasking cryptic genetic variation (Rutherford and Lindquist 1998). Our findings suggest that once such variation is exposed—regardless of the initial stressor—it can become subject to selection in a novel environment, even if that environment does not directly affect HSP90 expression."  
+
+<|ref|>sub_title<|/ref|><|det|>[[108, 587, 300, 602]]<|/det|>
+## Additional comments:  
+
+<|ref|>text<|/ref|><|det|>[[108, 604, 500, 620]]<|/det|>
+L43: note what kind of variation is meant here  
+
+<|ref|>text<|/ref|><|det|>[[107, 637, 220, 652]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 668, 872, 702]]<|/det|>
+We clarified the type of variation in the revised manuscript. The sentence now reads (line 51):  
+
+<|ref|>text<|/ref|><|det|>[[106, 717, 891, 769]]<|/det|>
+"Canalization buffers development against genetic and environmental disturbances, producing stable phenotypes despite underlying genetic and environmental variation (Waddington 1942)".  
+
+<|ref|>text<|/ref|><|det|>[[106, 783, 880, 851]]<|/det|>
+L102- 106: The reference to the previous study on social cues makes the reader think that this will be an important element of the current study even though it is not. The authors should consider ending with a brief overview of their study, its phenotypes, environments, and aims.  
+
+<|ref|>text<|/ref|><|det|>[[106, 867, 220, 883]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 899, 891, 933]]<|/det|>
+Thank you for pointing this out. To improve clarity and better guide the reader into our study, we have revised the final paragraph of the Introduction as suggested and now include a
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[107, 50, 891, 102]]<|/det|>
+concise overview of our study's main phenotypes, environments, and aims. This ensures a clear transition into the Results section without implying undue emphasis on social cues. Now this section reads _line 118:  
+
+<|ref|>text<|/ref|><|det|>[[106, 116, 892, 291]]<|/det|>
+"Here we used the red flour beetle, Tribolium castaneum as an alternative insect model (Brown et al. 2009; Pointer, Gage, and Spurgin 2021) to examine the genetic basis and fitness relevance of an HSP90- released phenotype. A previous study in T. castaneum found consistent downregulation of Hsp83, the primary HSP90- coding gene, in response to social cues mimicking a stressful environment. This suggests that HSP90 levels may be adaptively regulated, potentially facilitating the release of cryptic genetic variation when advantageous (Peuss et al. 2015). In the present study, we directly manipulated HSP90 function in T. castaneum and characterized the resulting reduced- eye phenotype. We identified the gene causing this phenotype and assessed its fitness consequences to evaluate the adaptive potential of HSP90- released variation."  
+
+<|ref|>text<|/ref|><|det|>[[106, 305, 888, 356]]<|/det|>
+L108: "HSP90 reduction, mimicking what may occur during environmental stress, in ..." Again, evidence of HSP90 with environmental stress seems like a key point that is not well- established from what I see in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[106, 372, 220, 388]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 404, 891, 473]]<|/det|>
+Following your suggestion, we have revised the sentence at line 136 for greater clarity. As mentioned above, we have also addressed the evidence for HSP90 limitation under environmental stress and included the relevant reference (Alford and Brandman 2018) in the manuscript. The sentence now reads:  
+
+<|ref|>text<|/ref|><|det|>[[106, 488, 891, 557]]<|/det|>
+"To investigate the effects of reduced HSP90 function—mimicking conditions that may arise during environmental stress—we used RNA interference (RNAi) to target Hsp83 in T. castaneum, with the aim of releasing HSP90- buffered, selectable phenotypic variants in Cro1, a genetically heterogenous wildtype population of this beetle"  
+
+<|ref|>text<|/ref|><|det|>[[105, 571, 812, 588]]<|/det|>
+L119- 122 seems to disagree with the figure 1 legend. The figure says F2 had 27.8%  
+
+<|ref|>text<|/ref|><|det|>[[106, 604, 220, 619]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[105, 634, 892, 774]]<|/det|>
+We appreciate your careful reading. The percentage shown in the original figure 1a (27.8%) represented the combined incidence of the reduced- eye phenotype in F2 offspring from the two families in which the trait was observed (12/47 and 20/68, totaling 32/115 beetles). To avoid confusion, we have now removed this percentage from Figure 1a and present the detailed incidence rates only in the text and Extended Data Table 1. The text now clearly states the incidence rates for each family (25.5% and 29.4%, respectively), as well as the overall frequency (4.2%) among all F2 offspring screened within this treatment as shown in Fig. 1b. This should resolve any inconsistency between the figure and the text.  
+
+<|ref|>text<|/ref|><|det|>[[105, 788, 832, 821]]<|/det|>
+L138 Why not Hsp83? It seems odd that a different gene is being assayed here when everything else focused on Hsp83.  
+
+<|ref|>text<|/ref|><|det|>[[106, 838, 220, 853]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 869, 891, 937]]<|/det|>
+We appreciate this observation. We focused on Hsp68a (HSP70 family) as a commonly used readout for HSP90 limitation, because 17- DMAG inhibits HSP90 (encoded by Hsp83) at the protein level rather than by reducing its mRNA expression. 17- DMAG is a well- characterized HSP90 inhibitor that binds to the ATP- binding pocket of the HSP90 protein, thereby blocking
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[106, 50, 892, 172]]<|/det|>
+its chaperone activity and leading to functional inhibition (Jez et al. 2003; Trepel et al. 2010). This inhibition does not typically reduce Hsp83 mRNA levels. Instead, inhibition of HSP90 activity is known to induce a compensatory heat shock response, including upregulation of HSP70 family genes such as Hsp68a (Kudryavtsev et al. 2017; Zhou et al. 2013). Thus, increased Hsp68a expression serves as a reliable molecular marker of successful HSP90 inhibition in our system. To clarify this, we have added the following sentence to the manuscript_line 181:  
+
+<|ref|>text<|/ref|><|det|>[[106, 186, 892, 255]]<|/det|>
+"17- DMAG inhibits HSP90 at the protein level by binding to its ATP- binding pocket, thereby blocking its chaperone activity without affecting Hsp83 mRNA levels (Jez et al. 2003; Trepel et al. 2010). As a result, upregulation of HSP70 family genes, such as Hsp68a, serves as a molecular marker of successful HSP90 inhibition (Jez et al. 2003)."  
+
+<|ref|>text<|/ref|><|det|>[[106, 271, 884, 322]]<|/det|>
+L158: How were reduced- and normal- eye beetles operationally delimited? Knowing the cut- off metric is important because much of the phenotypic and fitness data between the groups overlap (Fig. 1).  
+
+<|ref|>text<|/ref|><|det|>[[106, 337, 220, 353]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 368, 892, 543]]<|/det|>
+Thank you for pointing this out, it is an important point that needed clarification. During development, reduced- and normal- eye beetles could not be distinguished at the larval or pupal stages based on morphology alone. However, because individuals were tracked throughout development from larvae to adults, their adult eye phenotype could be retrospectively assigned to earlier stages. In the adult stage, classification into "reduced- eye" and "normal- eye" was based on visual inspection under a stereomicroscope, supported by quantitative image analysis of eye area. Reduced- eye beetles consistently showed visibly smaller eyes. For quantification, we normalized eye area within each developmental stage and used the adult phenotype (clearly visible) to assign individuals into the two groups. We have clarified this in the figure legend Fig1d as follows: line 1120  
+
+<|ref|>text<|/ref|><|det|>[[106, 557, 892, 660]]<|/det|>
+"Fig.1.....d, Normalized eye area in normal- eye and reduced- eye individuals across the three developmental stages, showing significant differences in eye size within stage. Normalization was done by stage's mean. Lines connect individual beetles, which were tracked from larval to adult stage. Phenotype classification (normal- vs. reduced- eye) was based on adult morphology, which allowed retrospective assignment of phenotype to earlier developmental stages."  
+
+<|ref|>text<|/ref|><|det|>[[106, 676, 892, 727]]<|/det|>
+And this sentence was added to the result part: line 205 "Because individuals were tracked throughout development, their adult phenotype could be retrospectively assigned to their larval and pupal stages."  
+
+<|ref|>text<|/ref|><|det|>[[106, 742, 632, 757]]<|/det|>
+L210 Please note that the deviation was lower than expected.  
+
+<|ref|>text<|/ref|><|det|>[[106, 774, 220, 789]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 805, 892, 856]]<|/det|>
+Indeed, the deviation was in the direction of a lower- than- expected frequency of the reduced- eye phenotype. We have clarified the sentence in the manuscript to reflect this more precisely. The updated sentence now reads: line 261  
+
+<|ref|>text<|/ref|><|det|>[[106, 872, 858, 907]]<|/det|>
+"However, in the crosses RNAi \(\mathcal{P}\) - Cro1 \(\sigma^r\) and RNAi \(\sigma^r\) - Cro1 \(\mathcal{P}\) , we observed a significantly lower proportion of the reduced- eye phenotype than the expected 25 %..."
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[106, 50, 872, 120]]<|/det|>
+L273- 274 I think omission of some commas in the sentence make it a bit awkwardly worded. Do you mean "Notably, our mapping analysis shows that the atonal gene, which consists of a single exon, overlaps with four synonymous polymorphic SNPs across the four pooled samples.."?  
+
+<|ref|>text<|/ref|><|det|>[[106, 153, 220, 169]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 184, 838, 217]]<|/det|>
+We have revised the sentence in L336 for improved clarity. The revised sentence now reads:  
+
+<|ref|>text<|/ref|><|det|>[[106, 233, 831, 266]]<|/det|>
+"Our mapping analysis shows that the atonal gene- which consists of a single exon- overlaps with four synonymous polymorphic SNPs across the four pooled samples"  
+
+<|ref|>text<|/ref|><|det|>[[106, 281, 715, 297]]<|/det|>
+L270 & 273 Consider removing "Notably" from one of these sentences.  
+
+<|ref|>text<|/ref|><|det|>[[106, 314, 220, 330]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 345, 511, 362]]<|/det|>
+Done! Line 337 the word "Notably" is removed.  
+
+<|ref|>text<|/ref|><|det|>[[106, 376, 880, 480]]<|/det|>
+L279 I'm sure the authors wish they could pinpoint more closely candidate regulatory elements. On possibility is to bioinformaticaly identify differences in TF motifs between lines in the candidate region and how the presence/absence of SNPs affect motif occurrences. The online resource MEME Suite might be useful to explore and get a bit closer to the actual causal SNP(s). https://meme- suite.org/meme. Of course, CHIP- seq or ATAC- seq could also identify some candidate positions.  
+
+<|ref|>text<|/ref|><|det|>[[106, 496, 220, 511]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 527, 892, 614]]<|/det|>
+We thank you for this suggestion. Following the recommendation, we performed motif discovery using the MEME Suite to identify potential transcription factor binding motifs in the candidate region. We also explored publicly available ATAC- seq and FAIRE- seq datasets from the iBeetleBase genome browser (https://ibeetle- base.uni- goettingen.de/genomebrowser/) to examine chromatin accessibility.  
+
+<|ref|>text<|/ref|><|det|>[[106, 628, 892, 749]]<|/det|>
+While the MEME analysis identified transcription factor motifs overlapping several genes in our candidate region, including atonal, the ATAC- seq and FAIRE- seq data revealed strong peaks of chromatin accessibility within the region lacking annotated genes but enriched for highly differentiated SNPs. These results support the hypothesis that a cis- regulatory element in this intergenic region may be influencing atonal expression. We have added these findings to the manuscript and included a new figure illustrating both the motif locations and peaks of chromatin accessibility. We now write in L341:  
+
+<|ref|>text<|/ref|><|det|>[[106, 764, 892, 867]]<|/det|>
+"This suggests that the phenotype may instead be caused by an as- yet- undiscovered regulatory sequence regulating atonal expression. We used two publicly available ATAC- seq (Mau et al. 2023) and FAIRE- seq (Lai et al. 2018) datasets to explore potential cis- regulatory regions. Particularly an area lacking annotated genes, yet enriched in highly differentiated SNPs, showed peaks of chromatin accessibility in both ATAC- seq and FAIRE- seq datasets (Extended Data Fig. 9)."  
+
+<|ref|>text<|/ref|><|det|>[[106, 882, 890, 916]]<|/det|>
+Moreover, we have added the following description to the Methods section detailing how MEME was ran_line: 954
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[107, 50, 892, 119]]<|/det|>
+"To identify potential regulatory motifs within the candidate region, we used the MEME Suite(Bailey and Elkan 1994) (https://meme- suite.org) to perform de novo motif discovery. We used the full 100 kb candidate region as input and ran the analysis with default parameters."  
+
+<|ref|>text<|/ref|><|det|>[[105, 134, 890, 170]]<|/det|>
+L373 This is the closest we get to understanding if HSP90 expression is affected by stress. It is unfortunate that it is not in the environment used in this study.  
+
+<|ref|>text<|/ref|><|det|>[[107, 185, 220, 201]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 215, 892, 425]]<|/det|>
+We recognize the importance of directly linking HSP90 regulation to specific environmental conditions. However, as mentioned before, it is conceptually not required that it is the same environmental condition that leads to reduced HSP90 availability and the phenotypes' fitness advantage. Our earlier work in T. castaneum demonstrated downregulation of HSP90 transcripts (Hsp83) in response to stressful social cues, specifically the presence of wounded conspecifics(Peuss et al. 2015). Other studies, in both T. castaneum and various insects, show that HSP90 expression is modulated by multiple forms of environmental stress, such as temperature extremes and chemical exposure (Brom et al. 2015; Ding et al. 2021; Liang et al. 2023). However, as you correctly note, our present study did not measure HSP90 levels under the exact selection regimes implemented here. We have now addressed this point in the revised discussion and have more clearly stated the need for future work that directly investigates HSP90 regulation in the context of these environments.  
+
+<|ref|>text<|/ref|><|det|>[[106, 439, 182, 455]]<|/det|>
+Line 469  
+
+<|ref|>text<|/ref|><|det|>[[106, 470, 892, 558]]<|/det|>
+"In a previous study in T. castaneum, HSP90 downregulation was observed in a risky environment, i.e., in cohabitation with wounded conspecifics(Peuss et al. 2015). Although this regulatory response was not directly addressed in the present study, the two studies together suggest a scenario that links evolutionary capacitance—and thus evolvability of a population—to environmental conditions...."  
+
+<|ref|>text<|/ref|><|det|>[[106, 572, 874, 640]]<|/det|>
+L601: I am confused about the abnormal leg phenotype and its being reported. Did this always co- occur with the reduced eyes? If not, how often was reduced- eyes by itself and why weren't these lines chosen instead of mixed phenotype lines? Why were leg phenotypes used to establish eye phenotype lines?  
+
+<|ref|>text<|/ref|><|det|>[[106, 655, 220, 671]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 686, 892, 791]]<|/det|>
+We appreciate the opportunity to clarify this point. Initially, F1 beetles exhibiting visible morphological abnormalities – most commonly leg malformations – were selected for generating F2 lines. This approach was primarily practical: such abnormalities served as visible markers of developmental disruption, suggesting that cryptic genetic variation might have been unmasked. The goal was to increase the chance of observing heritable phenotypes such as the reduced- eye trait in the F2 generation.  
+
+<|ref|>text<|/ref|><|det|>[[106, 805, 892, 892]]<|/det|>
+When the reduced- eye phenotype was first observed in F2 offspring from abnormal- leg F1 beetles, we tested whether this trait was linked to the leg malformation by also crossing their normal- legged full siblings from the same RNAi parental families. These crosses also yielded reduced- eye offspring in the F2, indicating that the reduced- eye and leg malformation phenotypes were not genetically linked.  
+
+<|ref|>text<|/ref|><|det|>[[106, 905, 892, 941]]<|/det|>
+It is also important to clarify that our reduced- eye lines were established only after observing this specific phenotype in the F2 generation. While these were initially derived
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[106, 50, 891, 102]]<|/det|>
+from families in which the F1 showed leg abnormalities, the reduced- eye lines themselves did not later display leg malformations, and the eye phenotype became monomorphic in subsequent generations.  
+
+<|ref|>text<|/ref|><|det|>[[106, 116, 891, 151]]<|/det|>
+To reflect this in the manuscript, we have added the following clarification in the Materials and Methods section: line 725  
+
+<|ref|>text<|/ref|><|det|>[[106, 165, 891, 217]]<|/det|>
+"... Notably, the reduced- eye phenotype was not genetically linked to leg malformations, as it also appeared in F2 offspring from normal- leg F1 siblings of the same RNAi- treated families (unpublished data)."  
+
+<|ref|>text<|/ref|><|det|>[[106, 232, 544, 249]]<|/det|>
+L641: Again, unclear why Hsp83 wasn't measured.  
+
+<|ref|>text<|/ref|><|det|>[[106, 265, 220, 281]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 295, 891, 348]]<|/det|>
+This point is now clarified in the Results section, where we explain that 17- DMAG inhibits HSP90 (encoded by Hsp83) at the protein level, not at the level of mRNA expression. Therefore, Hsp83 mRNA is not expected to respond directly to inhibition.  
+
+<|ref|>text<|/ref|><|det|>[[106, 362, 392, 378]]<|/det|>
+L680- 681: This is a clever design.  
+
+<|ref|>text<|/ref|><|det|>[[106, 394, 220, 410]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 425, 769, 442]]<|/det|>
+Thank you – we appreciate the positive feedback on our experimental design.  
+
+<|ref|>text<|/ref|><|det|>[[106, 458, 154, 473]]<|/det|>
+Fig. 1  
+
+<|ref|>text<|/ref|><|det|>[[106, 475, 890, 509]]<|/det|>
+I don't understand the diagram in 1a, what are the small arrows and red legs highlighting? It doesn't say in the legend.  
+
+<|ref|>text<|/ref|><|det|>[[106, 510, 890, 562]]<|/det|>
+In 1b, % frequency for the y- axis is odd. I'm not sure how to interpret it. The legend says that the % eye ptype is above the gold bar, but those numbers are more likely sample sizes. I think something must be missing.  
+
+<|ref|>text<|/ref|><|det|>[[106, 563, 870, 614]]<|/det|>
+In 1d, how were groups categorized, especially since there is overlap with some normal- eye having areas smaller than reduced- eye? Should it be treatment or line rather than ptype for the colors in panels in d- f?  
+
+<|ref|>text<|/ref|><|det|>[[106, 629, 220, 645]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 659, 890, 693]]<|/det|>
+Thank you for your careful evaluation and helpful observations regarding Figure 1. We address each point below:  
+
+<|ref|>text<|/ref|><|det|>[[137, 708, 892, 848]]<|/det|>
+1. Figure 1a – red arrows and legs: You are absolutely right – the small arrows were intended to indicate that the F2 reduced-eye phenotypes originated from F1 beetles with abnormal leg phenotypes (highlighted in red). These represented the families from which reduced-eye lines were established. However, as clarified in the text, we later confirmed that the reduced-eye phenotype is not genetically linked to leg malformations, since it also appeared in F2 offspring from normal-legged siblings of the same F1 families. To avoid confusion, we have removed these small arrows and red leg highlights from the updated version of the figure.  
+
+<|ref|>text<|/ref|><|det|>[[137, 862, 892, 932]]<|/det|>
+2. Figure 1b – % frequency and legend labeling: It is correct that the y-axis represents the frequency (%) of the reduced-eye phenotype within the screened F2 beetles in each treatment. The values above the bars correspond to sample sizes, not phenotypic percentages. The note in the legend stating that “% eye ptype is above
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[166, 50, 891, 102]]<|/det|>
+the gold bar" was left over from an earlier version of the figure and was mistakenly not removed when the plot format was updated. We have now corrected this in the revised legend.  
+
+<|ref|>text<|/ref|><|det|>[[137, 117, 892, 274]]<|/det|>
+3. Figure 1d - group categorization and classification: As noted earlier in this response letter and the manuscript, normalization of eye size was performed within each developmental stage. Individual beetles were tracked throughout development, and adult eye phenotype (normal or reduced) was assigned based on adult morphology using clear visual criteria, allowing retrospective classification across earlier stages. While there is some overlap in eye area between phenotypic classes, classification was based on adult appearance, which was visually distinct and confirmed by additional morphological analyses (see Fig. 1e for ommatidia counts). We've clarified this further in the legend to avoid ambiguity.  
+
+<|ref|>text<|/ref|><|det|>[[137, 288, 892, 375]]<|/det|>
+4. Color coding in panels d-f: We used eye phenotype (normal vs. reduced) as the common variable to categorize data in panels d-f to ensure consistency across the subfigures. Treatments or source lines varied across panels and were not consistently applicable, while eye phenotype provided a meaningful and common dimension for comparison. We note that:  
+
+<|ref|>text<|/ref|><|det|>[[195, 389, 891, 476]]<|/det|>
+- Panel d compares normalized eye area by phenotype across developmental stages- Panel e compares ommatidia counts by adult eye phenotypes- Panel f compares reproductive output by phenotype under two environmental conditions  
+
+<|ref|>sub_title<|/ref|><|det|>[[107, 490, 445, 507]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[106, 525, 888, 664]]<|/det|>
+The study by Sayed et al. explores the underappreciated evolutionary role of Hsp90 buffering in animals, identifying a beetle eye size trait whose variation is masked by Hsp90. Their findings suggest that Hsp90 enables the accumulation of cryptic genetic variation in natural populations. Transient Hsp90 inhibition (via chemical or siRNA) reveals a reduced-eye phenotype that persists in a subset of F1 and F2 progeny, can be driven to near fixation through crosses, and shows fitness advantages under constant light. A dimorphism is noted, and atonal is proposed as a phenocopy for Hsp90 inhibition. These findings are important and intriguing.  
+
+<|ref|>text<|/ref|><|det|>[[107, 680, 220, 695]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 710, 891, 779]]<|/det|>
+Thank you very much for your positive and thoughtful feedback on our work. We're glad that you found the conceptual approach and the main findings important and intriguing. Your encouraging comments were very motivating during the revision process and helped us further refine the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[106, 793, 891, 933]]<|/det|>
+While this is a compelling example of Hsp90 buffering in animals, the manuscript suffers from overstated claims, limited novelty, and interpretational issues. The conclusions need to be tempered or supported with additional data. The introduction and discussion should better contextualize the work within existing evidence supporting the adaptive value of Hsp90- buffered variation, avoiding claims of controversy or primacy. The experimental rigor and statistical analyses also require improvement. That said, the manuscript is clearly written and presents an intriguing story. I would support publication following major revisions to address the concerns outlined below.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[105, 50, 220, 66]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[105, 82, 880, 116]]<|/det|>
+Thank you for the thoughtful and constructive review. We have made substantial revisions to address your concerns:  
+
+<|ref|>text<|/ref|><|det|>[[135, 131, 893, 220]]<|/det|>
+- Tempered claims and improved context: We revised the Abstract, Introduction, and Discussion to avoid overstating novelty or controversy. The study is now clearly framed as an extension of previous work, highlighting HSP90's buffering in an animal under a certain environmental condition. We now more clearly acknowledge prior evidence for Hsp90 buffering and clarify where the novelty of our findings lies.  
+
+<|ref|>sub_title<|/ref|><|det|>[[137, 222, 364, 237]]<|/det|>
+## - New supporting data:  
+
+<|ref|>text<|/ref|><|det|>[[135, 238, 892, 344]]<|/det|>
+- RNAi knockdown of Hsp83 reduces atonal expression, supporting a regulatory link (Fig.3g).- Atonal is consistently downregulated ( \(\sim 20\%\) ) in reduced-eye monomorphic lines (Fig.3f).- Computational analyses (Extended Data Fig. 9) reveal conserved cis-regulatory motifs within accessible chromatin near candidate loci.  
+
+<|ref|>text<|/ref|><|det|>[[106, 357, 892, 409]]<|/det|>
+These additions strengthen the mechanistic basis of our conclusions and address interpretational concerns. We appreciate your feedback and believe the manuscript is now significantly improved  
+
+<|ref|>sub_title<|/ref|><|det|>[[106, 423, 223, 439]]<|/det|>
+## Major issues:  
+
+<|ref|>text<|/ref|><|det|>[[105, 456, 893, 666]]<|/det|>
+1. Conceptual Framing and Literature Context: The manuscript overstates the novelty and conceptual primacy of Hsp90 as an evolutionary capacitor. While this study advances our understanding of Hsp90-buffering in animals, the concept is well-established, particularly in yeast. The authors misrepresent the state of the field by framing the adaptive value of Hsp90-buffered variation as controversial. Multiple prior studies, including those by Lindquist and others (e.g., PMID: 16195452, 21205668, 39052788), have clearly demonstrated adaptive, genetically defined Hsp90-buffered traits. The manuscript also lacks appropriate citations and introduces misleading statements about the novelty and controversy of this mechanism. All novelty claims should be carefully rephrased to reflect that the contribution refers to Hsp90's role in animal evolution not evolution in general. I suggest revising lines 27-31, 37-38, 78-79, 86-87, 98, 169-170, 283-284, and 357-358 (these may not be all the sentences in need of revision to address this issue).  
+
+<|ref|>text<|/ref|><|det|>[[106, 681, 220, 696]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[135, 712, 892, 867]]<|/det|>
+Tempering of claims and improved contextualization: We revised the Abstract, Introduction, and Discussion to avoid implying novelty or controversy where not appropriate. Statements suggesting primacy or debate (e.g., about the role of HSP90 buffering in animals) were rephrased to more clearly position our study within the broader literature. We agree that studies in yeast and other fungi have demonstrated adaptive, genetically defined HSP90- buffered traits, and make it now clearer that the novelty of our study lies in the fact that to our knowledge no study in any animal has so far connected a newly released, HSP90- buffered trait to its genetic basis and fitness consequences.  
+
+<|ref|>text<|/ref|><|det|>[[105, 885, 881, 937]]<|/det|>
+2. The ecological and adaptive relevance of the reduced-eye trait is inadequately supported. The authors test it under artificial, non-ecological conditions (e.g., continuous light), without exploring whether it is beneficial or deleterious under more natural
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[105, 50, 860, 155]]<|/det|>
+environments. As a loss- of- function phenotype, reduced- eye is unlikely to confer broad fitness advantages, and its frequency in wild populations remains unexamined. Strong claims about its evolutionary significance require broader validation and testing of alternative hypotheses, including potential fitness costs. Overall, the trait's ecological relevance is unclear, and the study lacks the rigor seen in comparable work on Hsp90- buffered traits in yeast.  
+
+<|ref|>text<|/ref|><|det|>[[106, 170, 220, 186]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[105, 201, 892, 253]]<|/det|>
+We thank you for highlighting these important points, which we believe raise valid considerations regarding the ecological context and evolutionary relevance. Below, we provide our response to each point individually.  
+
+<|ref|>text<|/ref|><|det|>[[77, 268, 892, 407]]<|/det|>
+Regarding the relevance of our experimental conditions: While we acknowledge the concern regarding "artificial" or "non- ecological" laboratory conditions (e.g., continuous light), we respectfully note that \(T.\) castaneum is now a globally distributed species and a major pest of post- harvest grains and flour stored in human- made environments – including flour mills, warehouses, and storage bags – where beetles naturally encounter a range of light conditions (continuous, intermittent, or absent) and temperature fluctuations. Our laboratory setup, using flour as substrate and manipulating light and temperature, was specifically chosen to reflect these ecologically relevant, real- world habitats.  
+
+<|ref|>text<|/ref|><|det|>[[105, 421, 892, 525]]<|/det|>
+To further ensure relevance, we also evaluated fitness under constant darkness and mild heat stress \((+5^{\circ}C)\) – conditions also common in grain storage environments – but found no apparent benefit of the reduced- eye phenotype under these treatments (data not shown). Therefore, the observed fitness advantage under continuous light appears context- dependent, but still grounded in environmental conditions \(T.\) castaneum naturally experiences in its current human- associated niche.  
+
+<|ref|>text<|/ref|><|det|>[[105, 540, 892, 661]]<|/det|>
+We believe this experimental design effectively mimics the abiotic factors and substrate conditions typical of modern Tribolium habitats, providing a meaningful context to assess the adaptive value of the reduced- eye trait. Pointer et al. (2021) stated that "Owing to Tribolium's long human- commensal history ( \(\sim 70,000\) generations), the laboratory medium also has the advantage of very closely approximating its semi- wild habitat in food- storage facilities, allowing a lab environment that is less abstracted than that in other insect models".  
+
+<|ref|>text<|/ref|><|det|>[[77, 677, 892, 903]]<|/det|>
+Regarding the loss- of- function phenotype: While we agree that the reduced- eye phenotype – likely representing a loss- of- function – may not provide broad or universal fitness advantages, accumulating evidence suggests that loss or reduction of visual structures can be conditionally adaptive, especially in low- light or energetically constrained environments. Several studies have proposed that eye reduction or loss may be favored evolutionarily due to significant energy savings (Jeffery 2009; Moran, Softley, and Warrant 2015; Niven 2015; Ri as- Sánchez et al. 2025). Visual processing, including maintenance of photoreceptors and neural tissue, is metabolically costly: for example, cave fish devote 5–17% of total metabolic energy to vision (Moran et al. 2015), and in Drosophila melanogaster, photoreceptor activity alone accounts for around 8% of the organism's total energy usage (Laughlin, de Ruyter Van Steveninck, and Anderson 1998). Such costs rise in bright environments—energy demand may increase up to fourfold compared to darkness (Okawa et al. 2008).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[105, 50, 892, 189]]<|/det|>
+Even minor reductions in eye size have been associated with decreased energetic expense through the positive scaling of visual neuropile and eye size (Niven 2015). Insects inhabiting dim or concealed microhabitats, such as the larval stages of \(D\) . melanogaster and \(T\) . castaneum, tend to exhibit behaviors and morphologies reflecting reduced reliance on vision (Busto, Iyengar, and Campos 1999; Park 1934). Since \(T\) . castaneum adults are also strongly photo- negative, reduced eye size could plausibly provide a selective advantage in the dark, resource- limited environments typical of grain storage habitats by minimizing the metabolic burden of maintaining unnecessary visual tissue.  
+
+<|ref|>text<|/ref|><|det|>[[77, 205, 892, 326]]<|/det|>
+- Regarding the frequency of the phenotype in wild populations: There is currently no published data on the frequency of the reduced-eye phenotype in wild or synanthropic populations of \(T\) . castaneum. While the trait has not been reported in field surveys, we believe it likely exists at very low frequency, given its spontaneous appearance in lab populations. Detecting it would require screening tens of thousands of individuals, which was beyond the scope of this study. Nonetheless, we agree this is an important and relevant direction for future research.  
+
+<|ref|>text<|/ref|><|det|>[[105, 341, 870, 428]]<|/det|>
+Finally, we recognize that the ecological relevance and evolutionary stability of this trait require further validation in field populations and across multiple environments, and we have revised the text to clarify that our conclusions concern context-dependent fitness effects, not broad adaptive significance. We also agree that testing for potential costs and evaluating trait frequency in natural populations remain important next steps.  
+
+<|ref|>text<|/ref|><|det|>[[105, 442, 892, 546]]<|/det|>
+2. The claim that atonal mediates Hsp90-buffered eye-size variation is not sufficiently demonstrated. While atonal RNAi phenocopies the trait, there is no genetic or expression evidence linking atonal to Hsp90 buffering in the beetle. Allele-specific expression, response to Hsp90 inhibition, and functional SNP data are lacking. The role of HDACs is weakly supported and lacks a positive control. Without stronger evidence, the proposed mechanism remains speculative and should be presented as such.  
+
+<|ref|>text<|/ref|><|det|>[[105, 562, 220, 577]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[100, 593, 890, 627]]<|/det|>
+Thanks for these constructive comments. We have performed additional experiments to address them:  
+
+<|ref|>text<|/ref|><|det|>[[77, 629, 892, 699]]<|/det|>
+- Direct genetic and regulatory link: To strengthen our interpretation that HSP90 affects expression of atonal, we conducted new experiments and found that RNAi knockdown of Hsp83 significantly reduces atonal expression (new Fig. 3g), suggesting that HSP90 positively regulates atonal transcript levels.  
+
+<|ref|>text<|/ref|><|det|>[[77, 700, 892, 785]]<|/det|>
+- HDAC experiment and controls: The HDAC inhibition experiment is preliminary and was designed to explore whether epigenetic mechanisms may underlie the buffering of atonal expression. Beetles were fed flour discs containing either HDAC inhibitors or solvent-only (ethanol) as a control; additionally, we included loose flour (our standard maintenance medium) as a negative control.  
+
+<|ref|>text<|/ref|><|det|>[[105, 786, 892, 871]]<|/det|>
+Importantly, our new computational analyses (Extended Data Fig. 9) reveal conserved cis- regulatory motifs within accessible chromatin near candidate loci. We further tested atonal expression in our monomorphic reduced- eye lines and found it reduced compared to the ancestral Cro1 line (Fig. 3f). These results provide new evidence consistent with a regulatory link between chromatin state, Hsp83 function, and atonal expression.  
+
+<|ref|>text<|/ref|><|det|>[[77, 873, 892, 925]]<|/det|>
+- Caveats: We acknowledge that allele-specific expression data, natural-variant analysis, and functional SNP validation are currently lacking. These are all valuable directions for future work, which we plan to take into account in the next phase of the project. While these
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[105, 50, 891, 85]]<|/det|>
+data would further strengthen the proposed HSP90- atonal pathway, we believe our new results provide a solid foundation for this model in T. castaneum.  
+
+<|ref|>text<|/ref|><|det|>[[105, 103, 892, 225]]<|/det|>
+4. Several conclusions overreach the presented data. Key findings (e.g., persistence of the trait, its selection, and its mechanistic basis) have low penetrance, limited replicates, and insufficient statistical support. Alternative hypotheses are not tested, and the environmental relevance of findings is unclear. Overall, stronger controls, clearer definitions, and more cautious interpretation are needed throughout. Statistical analyses are not robust. Effects on fitness are weak and may not be significant after multiple hypothesis testing.  
+
+<|ref|>text<|/ref|><|det|>[[106, 240, 220, 255]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[105, 271, 892, 392]]<|/det|>
+We believe that we could address these aspects with our revisions. Persistence of the reduced- eye trait was followed for many generations in our monomorphic lines, and stable phenotype frequencies were observed over many generations in our polymorphic lines, allowing us to use the F50 for Bulk Segregation Analysis. Even though effect sizes may appear to be weak, they are in an evolutionary relevant range, e.g. regarding fitness effects. Statistical analyses are not limited by insufficient replicate numbers, and multiple hypothesis testing was performed when needed.  
+
+<|ref|>text<|/ref|><|det|>[[106, 428, 262, 443]]<|/det|>
+Minor comments:  
+
+<|ref|>text<|/ref|><|det|>[[106, 444, 840, 461]]<|/det|>
+1. In place of "Hsp90-released trait" I suggest "Hsp90-buffered trait"; lines 38 and 74.  
+
+<|ref|>text<|/ref|><|det|>[[106, 477, 220, 492]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 508, 830, 542]]<|/det|>
+Done - "HSP90- released trait" has been replaced with "HSP90- buffered trait" at the indicated lines. Now lines 40, 66 and 104  
+
+<|ref|>text<|/ref|><|det|>[[105, 560, 888, 593]]<|/det|>
+2. The keyword "plasticity" was not mentioned. Remove keyword or add explanation of the term in the text.  
+
+<|ref|>text<|/ref|><|det|>[[106, 610, 220, 625]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 641, 661, 658]]<|/det|>
+Done - the keyword "plasticity" has been removed as requested.  
+
+<|ref|>text<|/ref|><|det|>[[106, 675, 888, 744]]<|/det|>
+3. Statistical significance, \* should be replaced with increments \*\*, \*\*\*, \*\*\*\*. A single asterisk (*) indicating a p-value less than 0.05 (significant), two asterisks (**) a p-value less than 0.01 (highly significant), and three asterisks (***) indicate a p-value less than 0.001 (very highly significant).  
+
+<|ref|>text<|/ref|><|det|>[[106, 761, 220, 776]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 792, 884, 825]]<|/det|>
+Done - we have updated all significance indicators in the figures and legends to follow the standard format: \(p < 0.05\) (*), \(p < 0.01\) (**) and \(p < 0.001\) (***), as requested.  
+
+<|ref|>text<|/ref|><|det|>[[106, 841, 792, 857]]<|/det|>
+4. Hsp90 is an epistatic regulator but no epistasis experiments were performed.  
+
+<|ref|>text<|/ref|><|det|>[[106, 873, 220, 888]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 904, 891, 937]]<|/det|>
+Thank you for mentioning this aspect. While we did not perform classical genetic interaction (epistasis) assays in this study, our results are consistent with HSP90 acting as an epistatic
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[106, 50, 892, 172]]<|/det|>
+regulator. Specifically, knockdown of Hsp83 released the reduced- eye phenotype; whole- genome sequencing and functional analysis identified the transcription factor atonal as the underlying gene; and direct knockdown of atonal robustly reproduced the same phenotype. Additionally, we observed a positive correlation between HSP90 and atonal expression, with atonal being downregulated after Hsp83 knockdown. Atonal likely interacts epistatically with further, potentially also HSP90- affected proteins, but it is complex to address this in T. castaneum and beyond the scope of our study.  
+
+<|ref|>text<|/ref|><|det|>[[106, 190, 888, 240]]<|/det|>
+5. Abstract lines 33-34: "stably inherited without continued HSP90 disruption" implies all revealed variation was stable but thew data show closer to \(20\%\) of the revealed variation is inherited and further declines in each generation so it is not that stable.  
+
+<|ref|>text<|/ref|><|det|>[[106, 257, 220, 272]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[105, 288, 892, 496]]<|/det|>
+We appreciate this comment and understand the potential confusion, which likely stems from the polymorphic lines that harbour the reduced- eye allele at a certain gene frequency (Fig. 2a, lower two panels). It is important to note the lines are polymorphic populations and not clonal lines. In these lines, the allele is stably inherited, but the phenotype frequency depends on the allele frequency in these populations. Importantly, it was possible to select for the reduced- eye phenotype and produce monomorphic lines that are likely fixed for the reduced- eye allele and thus show close to \(100\%\) reduced- eye phenotypes (Fig. 2a upper panel), which was maintained over all successive generations up to now. The original phrasing "stably inherited without continued HSP90 disruption" was used to convey that once established, the reduced- eye phenotype continued to persist in subsequent generations, even though HSP90 function was no longer perturbed. We have revised the abstract language to clarify the aspect of persistence in the population.  
+
+<|ref|>text<|/ref|><|det|>[[106, 511, 480, 527]]<|/det|>
+The revised abstract line now reads: line 35  
+
+<|ref|>text<|/ref|><|det|>[[106, 543, 867, 576]]<|/det|>
+"...we consistently revealed a reduced- eye phenotype that persisted in descendant lines across generations without continued HSP90 disruption."  
+
+<|ref|>text<|/ref|><|det|>[[105, 592, 892, 660]]<|/det|>
+6. Figures should be revised to improve clarity and intuitiveness. In particular, the presentation of penetrance—such as in the HDAC inhibition example—is often unclear and could benefit from more straightforward visualization.  
+
+<|ref|>text<|/ref|><|det|>[[106, 642, 892, 712]]<|/det|>
+A more quantitative trait would be the number of omatids per animal, which although labor intensive seems to present with a great dynamic range and could foster a more robust statistical analysis approach. I understand that counting the number of omatids in each animal may not be feasible, so this is only a minor comment/suggestion.  
+
+<|ref|>text<|/ref|><|det|>[[106, 728, 220, 743]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[105, 759, 892, 897]]<|/det|>
+Thank you for the thoughtful suggestions regarding data visualization and quantification. In response to the comment on HDAC inhibition (Extended Data Fig. 6): We have revised the figure to improve its clarity and informativeness. Specifically, we now provide the mean and standard error for each group in addition to individual datapoints. Furthermore, we adjusted the y- axis scale, setting the lower bound to \(50\%\) instead of \(0\%\) , to better focus on the region where all data points reside (60- 100% penetrance). This zoomed- in view makes the small yet statistically significant effects more visually interpretable. The figure caption has also been updated to explicitly state this.  
+
+<|ref|>text<|/ref|><|det|>[[106, 913, 892, 946]]<|/det|>
+Regarding the suggestion to use omnatidia number as a quantitative trait: We appreciate this idea and agree that omnatidial counts can, in many systems, provide a
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[106, 50, 893, 171]]<|/det|>
+detailed, quantitative metric. In our study, we performed a limited quantification of ommatidia number to support our conclusions (shown in Fig. 1e). Specifically, we manually counted the number of ommatidia in both compound eyes of adult beetles with clearly distinguishable eye morphology ( \(n = 8\) ) and used the average per beetle for analysis. The counts were compared between normal- and reduced- eye individuals using an unpaired two- sample t- test, which is regarded as a statistically robust method even with modest sample sizes.  
+
+<|ref|>text<|/ref|><|det|>[[106, 186, 892, 307]]<|/det|>
+However, we found that in \(T.\) castaneum, the number of ommatidia is relatively small and consistent between individuals of the same phenotype—approximately 40 ommatidia in normal eyes and \(\sim 20\) in reduced- eye phenotypes from ventral side—resulting in two discrete, stable categories, rather than a broad dynamic range. In addition, reliable ommatidial quantification was only possible in adults with well- structured compound eyes. Many reduced- eye animals showed fused or misshapen ommatidia, making accurate counting unreliable in a substantial proportion of samples.  
+
+<|ref|>text<|/ref|><|det|>[[106, 322, 892, 392]]<|/det|>
+Due to these biological and technical constraints, we used phenotypic penetrance as our main metric, which was more scalable and practical for quantifying trait expression across all treatments and lines. Nonetheless, ommatidial counts in Fig. 1e provide important validation of the morphological distinctions between phenotypes.  
+
+<|ref|>text<|/ref|><|det|>[[106, 406, 833, 441]]<|/det|>
+We hope these revisions have addressed your concerns, and we are grateful for your comments that helped us strengthen the manuscript considerably.  
+
+<|ref|>text<|/ref|><|det|>[[108, 459, 428, 476]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[106, 494, 886, 563]]<|/det|>
+I co- reviewed this manuscript with one of the reviewers who provided the listed reports. This is part of the Nature Communications initiative to facilitate training in peer review and to provide appropriate recognition for Early Career Researchers who co- review manuscripts.  
+
+<|ref|>text<|/ref|><|det|>[[106, 579, 220, 594]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 609, 892, 661]]<|/det|>
+We thank Reviewer #3 for his/her contribution to the peer review process and for participating in the Nature Communications initiative supporting Early Career Researchers. We appreciate the time and effort dedicated to co- reviewing our manuscript.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[106, 51, 213, 66]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[100, 68, 895, 920]]<|/det|>
+Alford, Brian D., and Onn Brandman. 2018. 'Quantification of Hsp90 Availability Reveals Differential Coupling to the Heat Shock Response'. Journal of Cell Biology 217(11). doi:10.1083/jcb.201803127. Bailey, Timothy L., and Charles Elkan. 1994. 'Fitting a Mixture Model by Expectation Maximization to Discover Motifs in Biopolymers'. in Proceedings of the 2nd International Conference on Intelligent Systems for Molecular Biology, ISMB 1994. Borkovich, K. A., F. W. Farrelly, D. B. Finkelstein, J. Taulien, and S. Lindquist. 1989. 'Hsp82 Is an Essential Protein That Is Required in Higher Concentrations for Growth of Cells at Higher Temperatures.' Molecular and Cellular Biology 9(9):3919- 30. doi:10.1128/MCB.9.9.3919. Updated. Brom, Krzysztof Roman, Bogdan Dolezych, Monika Tarnawska, Katarzyna Brzozowska, and Miroslaw Nakonieczny. 2015. 'Expression of the Hsp40, Hsp70 and Hsp90 Proteins in Colorado Potato Beetle (Leptinotarsa Decemlineata Say) after the Dimethoate Treatment'. Journal of the Entomological Research Society 17(2). Brown, Susan J., Teresa D. Shippy, Sherry Miller, Renata Bolognesi, Richard W. Beeman, Marcé D. Lorenzen, Gregor Bucher, Ernst a. Wimmer, and Martin Klingler. 2009. 'The Red Flour Beetle, Tribolium Castaneum (Coleoptera): A Model for Studies of Development and Pest Biology'. Cold Spring Harbor Protocols 4(8):1- 12. doi:10.1101/pdb.emo126. Busto, M., B. Iyengar, and a R. Campos. 1999. 'Genetic Dissection of Behavior: Modulation of Locomotion by Light in the Drosophila Melanogaster Larva Requires Genetically Distinct Visual System Functions'. The Journal of Neuroscience: The Official Journal of the Society for Neuroscience 19(9):3337- 44. doi:10.1523/JNEUROSCI.19- 09- 03337.1999. Chen, Bing, and Andreas Wagner. 2012. 'Hsp90 Is Important for Fecundity, Longevity, and Buffering of Cryptic Deleterious Variation in Wild Fly Populations'. BMC Evolutionary Biology 12(1):25. doi:10.1186/PREACCEPT- 2105524459276497. Ding, Jian Hao, Lu Xin Zheng, Jie Chu, Xin Hao Liang, Jun Wang, Xiao Wen Gao, Fu An Wu, and Sheng Sheng. 2021. 'Characterization, and Functional Analysis of Hsp70 and Hsp90 Gene Families in Glyphodes Pylolais Walker (Lepidoptera: Pyralidae)'. Frontiers in Physiology 12. doi:10.3389/fphys.2021.753914. Jeffery, William R. 2009. 'Regressive Evolution in Astyanax Cavefish'. Annual Review of Genetics 141(4):520- 29. doi:10.1016/j.surg.2006.10.010. Use. Jez, J. M., J. C. Chen, G. Rastelli, R. M. Stroud, and D. V Santi. 2003. 'Crystal Structure and Molecular Modeling of 17- DMAG in Complex with Human Hsp90'. Chem Biol 10(4):361- 68. http://www.ncbi.nlm.nih.gov/pubmed/12725864. Kudryavtsev, Vladimir A., Anna V. Khokhlova, Vera A. Mosina, Elena I. Selivanova, and Alexander E. Kabakov. 2017. 'Induction of Hsp70 in Tumor Cells Treated with Inhibitors of the Hsp90 Activity: A Predictive Marker and Promising Target for Radiosensitization'. PLoS ONE 12(3):1- 25. doi:10.1371/journal.pone.0173640. Lai, Yi Ting, Kevin D. Deem, Ferran Borras- Castells, Nagraj Sambrani, Heike Rudolf, Kushal Suryamohan, Ezzat El- Sherif, Marc S. Halfon, Daniel J. McKay, and Yoshinori Tomoyasu. 2018. 'Enhancer Identification and Activity Evaluation in the Red Flour Beetle, Tribolium Castaneum'. Development (Cambridge) 145(7). doi:10.1242/dev.160663. Laughlin, S. B., R. R. de Ruyter Van Steveninck, and J. C. Anderson. 1998. 'The Metabolic Cost of Neural Information'. Nature Neuroscience 1(1):36- 41. doi:10.1038/236. Liang, Chen, Lifang Li, Hang Zhao, Mingxian Lan, Yongyu Tang, Man Zhang, Deqiang Qin, Guoxing Wu, and Xi Gao. 2023. 'Identification and Expression Analysis of Heat Shock Protein Family Genes of Gall Fly (Procecidochares Utilis) under Temperature Stress'. Cell Stress and Chaperones 28(3). doi:10.1007/s12192- 023- 01338- 9.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[101, 49, 895, 680]]<|/det|>
+Mau, Christine, Heike Rudolf, Frederic Strobl, Benjamin Schmid, Timo Regensburger, Ralf Palmisano, Ernst H. K. Stelzer, Leila Taher, and Ezzat El-Sherif. 2023. 'How Enhancers Regulate Wavelike Gene Expression Patterns'. ELife 12. doi:10.7554/eLife.84969. Moran, Damian, Rowan Softley, and Eric J. Warrant. 2015. 'The Energetic Cost of Vision and the Evolution of Eye-less Mexican Cavefish'. Science Advances 1(8):e1500363- e1500363. doi:10.1126/sciadv.1500363. Niven, Jeremy E. 2015. 'Neural Evolution: Costing the Benefits of Eye Loss'. Current Biology 25(19):R840- 41. doi:10.1016/j.cub.2015.08.050. Okawa, Haruhisa, Alapakkam P. Sampath, Simon B. Laughlin, and Gordon L. Fain. 2008. 'ATP Consumption by Mammalian Rod Photoreceptors in Darkness and in Light'. Current Biology 18(24):1917- 21. doi:10.1016/j.cub.2008.10.029. Park, T. 1934. 'Observations on the General Biology of the Flour Beetle, Tribolium Confusum'. Quarterly Review of Biology 9:36- 54. Peuss, R., Hendrik Eggert, Sophie A. O. Armitage, and Joachim Kurtz. 2015. 'Downregulation of the Evolutionary Capacitor Hsp90 Is Mediated by Social Cues'. Proceedings of the Royal Society B: Biological Sciences 282(1819):20152041- 20152041. doi:10.1098/rspb.2015.2041. Pointer, Michael D., Matthew J. G. Gage, and Lewis G. Spurgin. 2021. 'Tribolium Beetles as a Model System in Evolution and Ecology'. Heredity 126(6):869- 83. doi:10.1038/s41437- 021- 00420- 1. Ri as- Sánchez, David F., J. ake Morris, Camilo Salazar, Carolina P. ar do- Díaz, Richar M Merrill, Stephen H. Montgomery, and Richard M Merrill. 2025. 'Repea Ted e Volution of Reduced Visual in Vestment a t the Onset of Ecological Speciation in High- Altitude Heliconius Butterflies'. Evolution Letters 1- 9. doi:10.1093/el. Rutherford, S. L., and S. Lindquist. 1998. 'Hsp90 as a Capacitor for Morphological Evolution'. Nature 396(6709):336- 42. doi:10.1038/24550. Trepel, Jane, Mehdi Mollapour, Giuseppe Giaccone, and Len Neckers. 2010. 'Targeting the Dynamic HSP90 Complex in Cancer'. Nature Reviews Cancer 10(8):537- 49. doi:10.1038/nrc2887. Waddington, C. H. 1942. 'Canalization of Development and the Inheritance of Acquired Characters'. Nature 150(3811):563- 65. doi:10.1038/150563a0. Zhou, Dan, Yuan Liu, Josephine Ye, Weiwen Ying, Luisa Shin Ogawa, Takayo Inoue, Noriaki Tatsuta, Yumiko Wada, Keizo Koya, Qin Huang, Richard C. Bates, and Andrew J. Sonderfan. 2013. 'A Rat Retinal Damage Model Predicts for Potential Clinical Visual Disturbances Induced by Hsp90 Inhibitors'. Toxicology and Applied Pharmacology 273(2):401- 9. doi:10.1016/j.taap.2013.09.018.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[107, 50, 607, 68]]<|/det|>
+## RESPONSE TO THE REVIEWER COMMENTS _2nd Round  
+
+<|ref|>sub_title<|/ref|><|det|>[[107, 103, 257, 119]]<|/det|>
+## Dear Reviewers,  
+
+<|ref|>text<|/ref|><|det|>[[107, 137, 891, 206]]<|/det|>
+We sincerely thank you for your continued evaluation of our manuscript entitled "HSP90 as an Evolutionary Capacitor Drives Adaptive Eye Size Reduction via Atonal." We appreciate the additional feedback provided and have addressed the remaining points to further improve the clarity and robustness of our work.  
+
+<|ref|>text<|/ref|><|det|>[[107, 242, 428, 258]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[106, 273, 892, 428]]<|/det|>
+My biggest concerns in the previous version of the manuscript were the links between Hsp90 levels, phenotypic variation, and environmental conditions. In their response letter, the authors do a convincing job of explaining that the conditions that affected Hsp90 availability need not be the same conditions that select upon phenotypes and genotypes revealed by Hsp90 depletion. Further, they explained the link between measuring expression level of Hsp68 instead of Hsp83. By and large, I am satisfied with their responses to my initial comments and suggestions, and I am pleased by their inclusion of additional data and consideration of my comments. I have only a few suggestions for further clarifying these points in the text.  
+
+<|ref|>text<|/ref|><|det|>[[107, 445, 220, 460]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 476, 891, 545]]<|/det|>
+We are grateful for the reviewer's positive evaluation and are pleased that our previous revisions addressed the major concerns. As suggested, we have incorporated additional clarifications in the relevant sections to further strengthen the connections between HSP90 availability, phenotypic variation, and environmental context.  
+
+<|ref|>text<|/ref|><|det|>[[106, 560, 891, 611]]<|/det|>
+L87: Consider mentioning here that selection could be acting in the Hsp90 affecting environment or a different environment. " ...differences on which selection can act either in the disrupting environmental conditions or other conditions altogether."  
+
+<|ref|>text<|/ref|><|det|>[[106, 626, 220, 642]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 658, 875, 709]]<|/det|>
+The sentence has been revised as follows (L51 with "No Markup" as the adjusted option): "...differences on which selection can act either in the disrupting environmental conditions or other conditions altogether."  
+
+<|ref|>text<|/ref|><|det|>[[106, 741, 891, 775]]<|/det|>
+L186 consider adding a reminder here that continuous light COULD be encountered by these beetles in their human- commensal environment.  
+
+<|ref|>text<|/ref|><|det|>[[106, 792, 220, 807]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[106, 823, 875, 874]]<|/det|>
+We added the following (L160 with "No Markup" option): "...continuous light stress, a situation that could be encountered by these beetles in their human- commensal environment."  
+
+<|ref|>text<|/ref|><|det|>[[106, 889, 728, 906]]<|/det|>
+L208 Consider changing "certain" to "some, albeit unknown, frequency"  
+
+<|ref|>text<|/ref|><|det|>[[106, 921, 220, 936]]<|/det|>
+## response:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[105, 50, 630, 68]]<|/det|>
+This change has been made (L180 with "No Markup" option).  
+
+<|ref|>text<|/ref|><|det|>[[105, 113, 890, 149]]<|/det|>
+L321 "...unmasking cryptic genetic variation and producing new phenotypes." To be explicit about the consequence of that CGV being uncovered.  
+
+<|ref|>text<|/ref|><|det|>[[106, 164, 221, 180]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[105, 194, 768, 212]]<|/det|>
+We added "and producing new phenotypes" (L293 with "No Markup" option).  
+
+<|ref|>sub_title<|/ref|><|det|>[[106, 257, 445, 275]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[105, 293, 876, 327]]<|/det|>
+All my comments have been fully addressed. I congratulate the authors on their beautiful work and approve of its publication.  
+
+<|ref|>text<|/ref|><|det|>[[106, 343, 221, 359]]<|/det|>
+## response:  
+
+<|ref|>text<|/ref|><|det|>[[105, 373, 891, 408]]<|/det|>
+We sincerely thank the reviewer for their very helpful previous comments that substantially improved our manuscript and are grateful for this positive feedback.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__c3902e7c221d3e3f06a8e7be186d078007da23cdc319c5fd0c6ad9a338e3cb91/images_list.json

@@ -0,0 +1,17 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Impact of Confidence Thresholding on Sample Classification",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        405,
+        878,
+        638
+      ]
+    ],
+    "page_idx": 22
+  }
+]

peer_reviews/supplementary_0_Peer Review file__caa685a328daac33d8261fb19023ca5fa2ca99c25c1ffc44e94d273ac99b6bb0/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review file__caa685a328daac33d8261fb19023ca5fa2ca99c25c1ffc44e94d273ac99b6bb0/supplementary_0_Peer Review file__caa685a328daac33d8261fb19023ca5fa2ca99c25c1ffc44e94d273ac99b6bb0.mmd

@@ -0,0 +1,225 @@
+
+# nature portfolio  
+
+# Peer Review File  
+
+# Immunomics-guided biomarker discovery for human liver fluke infection and infection-associated cholangiocarcinoma  
+
+Corresponding Author: Professor Alex Loukas  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+Food- borne trematodiases pose a serious public health problem, affecting more than 50 million people worldwide. Liver flukes, including Opisthorchis viverrini, Opisthorchis felineus, and Clonorchis sinensis, are not only parasitic infections but also carcinogenic, contributing to bile duct cancer and other related malignancies.  
+
+The prevention and control of liver fluke infections in both endemic and non- endemic areas rely on parasite detection. However, traditional detection methods, while considered the gold standard for diagnosis, are labor- intensive, have low sensitivity, and require skilled technicians.  
+
+This study presents promising results, demonstrating the potential for a rapid ELISA- based screening method that could be effectively applied for field detection in South Asia.  
+
+The study is a valuable contribution to field detection for the parasitic and parasite- related diseases. Compared to existing literature, it introduces an innovative design and a novel biomarker for rapid schistosomiasis screening, making it an original and significant advancement in the field.  
+
+The manuscript presents results based on two biomarkers for O. viverrini infection and infection- associated CCA, evaluated in a cohort of patients. However, the study's sample size appears to be a limitation, which may affect the persuasiveness of the findings. I believe that increasing the sample size would strengthen the conclusions and enhance the reliability of the results.  
+
+In this study, patients with CCA caused by chronic liver fluke infection were enrolled. However, the interpretation of CCA needs to be described in greater detail in the methods section. Were all patients, especially CCAs, verified through other diagnostic examinations? Please clarify this in the manuscript to ensure the accuracy and reliability of the findings.  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+The manuscript relates to the identification of excretory- secretory proteins of O. viverrini and generating a protein array of selected targets from the excretory- secretory proteins. These protein arrays were screened with infected and un- infected sera to identify potential biomarkers. This is a well written paper and the flow is good. It would have been helpful if the manuscript included line numbers to help with the comments.  
+
+Comments:  
+
+1. In the introduction, the authors mention that there have been other tests (antigen-based and antibody-based) that were developed. However, I do not see any comparison or discussion of the current proposed biomarkers (P1 and P9) with either of the available assays in the manuscript. Were the same samples tested for antigen by any of the methods? Ideally, a new assay needs to be compared with existing assays. Although the assays cited are antigen-based, and the current manuscript is antibody based, from a public health perspective, it would finally come down to which assay has better sensitivities in detecting infections while maintaining specificity. 
+2. In the abbreviated methods, the detailed methods are supposedly in 'Appendix 1'. But the supplemental methods are not labeled as such. 
+3. The methods mention excretory/secretory (ES) and extracellular vesicles (EVs). But the supplemental methods mention
+
+<--- Page Split --->
+
+ES, EVs, ELVs and MVs. This needs to be clarified. Based on the supplementary methods, it is not clear as to which datasets are from this study and which ones are from prior studies.  
+
+4. The related supplemental figure (S2) shows the VENN diagram with ES, ELVs and MVs. Where do the P1 to P9 fall in the VENN diagram? Along same line, Table S3 does not really need the last column 'Selected for the array' if all of the proteins are being printed. Further, the title of Table S3 needs to be modified as it is not a list of all the 825 proteins detected by MS. 
+5. The second half of the study design repeats itself in the array probing methodology. This redundancy can be cleaned up. 
+6. On a minor note, what is the composition of the 'array blocking buffer'? There is no reference or mention of it in the supplemental methods.  
+
+7. Fig 1.  
+
+a. The ES part needs to be clarified in the text (all through) if it is going to be used to encompass all forms of excretory-secretory proteins.  
+
+b. While it says 245 of the 278 proteins were printed on the array, the last line of the results section on the characterization of the secretome reads "... a total of 278 printed proteins or protein fragments". In addition, previously published diagnostic targets of O. viverrini and Clonorchis spp resulted in a total of 249 proteins that were selected to be printed. This needs to be clarified.  
+
+8. In the 'immunomics' section of the results, the authors describe that P3 was recognized exclusively by sera of individuals with C. sinensis infection. How do you explain this? I do not see much of a discussion as to why 'thioredoxin' that is almost universally present is picked up by sera from C. sinensis infections. Also, if sera from individuals with CCA are also positive for P3, then is it really 'exclusive'?  
+
+9. Fig 2.  
+
+a. The results section mentions 36-antigens (excluding OvES and OvSo) were the targets of significantly elevated IgG1 or IgG4 of which 20 were targets with significant values for both IgG1 and IgG4. The legend says top 40 protein hits and the order of them is different. I do not see the data for P3, P6 and P7 in 2A and 2B. While this is totally up to the authors, I feel that having all the protein names in the middle with their corresponding IgG1 histograms on the top and IgG4 histograms on the bottom, would be easier to see which of the 20 targets were commonly found to have significant. Or it can be flipped vertically so that the naming is easier to read, with IgG1 and IgG4 data to the left and right.  
+
+b. How do you explain the relative lack of signal in IgG4 for the OvES? How much of the protein was printed on the array? Assuming equal amounts of proteins across, what is the relative abundance of the selected proteins in the ES?  
+
+c. Was there a correlation between the array signals and ICT data?  
+
+10. Pg 8. Why were the analyses moved from IgG1 to total IgG for the ICT? I do not see any discussion pertaining to this switch. Also going back, the abstract says IgG reactive antigens were taken forward for recombinant expression. This needs to be corrected.  
+
+11. Is there a TPP of some sort CCA/Ov infections? What would be an ideal situations for the two case scenarios? 
+12. Pg 11. Why is it that the CCA cases were 'suspected' to be caused by chronic O. viverrini infection? Were they not diagnosed as Ov?  
+
+13. In the discussion, I don't think it is appropriate to say a 'genome-scale proteome microarray' was constructed. The protein array was specific to the ES.  
+
+14. The authors do allude to it, but the discussion on EPG between C. sinensis and O. viverrini is at best is speculative as the correlations between egg burden and signal weak to modest.  
+
+15. Pg 14. The starting lines on the seropositivity endemic control populations. During the screening process, were the  
+
+individuals not asked if they have been tested before? Was this a totally new region that was previously never studied?  
+
+16. Maybe I missed something, but looked like all the samples were egg positive cases. And the discussion on CCA cases being in a 'non-infected' state was confusing for me. Were the blood and stool sampled at different times? What are the kinetics of antibody levels? Do they drop off quickly too?  
+
+## Reviewer #3  
+
+(Remarks to the Author)  
+
+The manuscript title, "Immunomics-guided biomarker discovery for human liver fluke infection and infection-associated cholangiocarcinoma" aims to develop a diagnostic tool, specifically an immunochromatography strip, for the detection of Opisthocis viverrini infection and O. viverrini-associated cholangiocarcinoma.  
+
+The manuscript is well-structured and informative, and the experimental design is clearly outlined. However, one important question arises regarding How to interpretation the results? Specifically, it is unclear how to interpret, when it is positive results indicating Opisthocis viverrini infection, Opisthocis viverrini infection in the past, or Opisthocis viverrini-related cholangiocarcinoma. This question should be further discussed in the discussion section.  
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+This manuscript presents valuable findings in the field of Opisthocris viverrini infection and associated cholangiocarcinoma (CCA). The study offers important insights into chronic liver fluke infection, highlighting a research area that deserves greater attention given its status as a neglected tropical disease. The work is well- conducted and makes a significant contribution to the field. I recommend its publication in Nature Communications after addressing minor revisions (if any are required by
+
+<--- Page Split --->
+
+other reviewers).  
+
+Reviewer #2  
+
+(Remarks to the Author) I think the authors have answered satisfactorily to all the questions/comments. A couple of minor queries.  
+
+When the authors say 'routinely finding' anti- IgG1 secondary antibodies as sub- optimal, is it across the board for all such studies? If so, why use it in the first place? Secondly, since the authors increased the sample numbers by utilizing the same samples used for array screening, how do the results correlate between the IgG1 (array) and the ICT (IgG)?  
+
+Reviewer #3  
+
+(Remarks to the Author) All of the raised question have been satisfactorily addressed by authors.  
+
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+Point by point response  
+
+## Reviewer 1.  
+
+1. The manuscript presents results based on two biomarkers for O. viverrini infection and infection-associated CCA, evaluated in a cohort of patients. However, the study's sample size appears to be a limitation, which may affect the persuasiveness of the findings. I believe that increasing the sample size would strengthen the conclusions and enhance the reliability of the results.  
+
+Response: We emphasised geographic diversity of serum samples from flukeinfected subjects used in the study (Thailand, Laos, China). The reviewer's point is well taken, so we used the two best ICTs to screen another 50 serum samples from O. viverrini-infected subjects where diagnosis was noted as egg-positive or eggnegative, and egg counts were not quantified as they were for the subjects we currently report in the manuscript. Nonetheless, we thought this would be a valuable addition to the paper and have included the information as Supplementary Figure 4. Please note, this cohort was one of those used to screen the proteome microarray (where we were not concerned with quantified EPG), but we had not previously used this cohort to screen the ICTs. With this new cohort of subjects, we observed similar sensitivity and specificity of the ICTs to that reported for subjects where egg counts had been quantified.  
+
+2. In this study, patients with CCA caused by chronic liver fluke infection were enrolled. However, the interpretation of CCA needs to be described in greater detail in the methods section. Were all patients, especially CCAs, verified through other diagnostic examinations? Please clarify this in the manuscript to ensure the accuracy and reliability of the findings.  
+
+Response: The requested information on CCA patient clinical diagnosis is now provided in Supplementary Data 3. In addition, we have added a new panel to Figure 5 (Fig. 5E) that shows the positivity rate for serodiagnosis of CCA at different clinical stages of malignancy, and we have discussed the findings at the end of the Results (page 5).  
+
+## Reviewer 2.  
+
+1. In the introduction, the authors mention that there have been other tests (antigen-based and antibody-based) that were developed. However, I do not see any comparison or discussion of the current proposed biomarkers (P1 and P9) with either of the available assays in the manuscript. Were the same samples tested for antigen by any of the methods? Ideally, a new assay needs to be compared with existing assays. Although the assays cited are antigen-based, and the current manuscript is antibody based, from a public health perspective, it would finally come down to which assay has better sensitivities in detecting infections while maintaining specificity. Response: The antigen detection RDT for opisthorchiasis relies on detection of antigen in urine of infected subjects. Antigen is detected using a monoclonal antibody raised against an undefined antigen, probably glycan in nature based on its Western blot profile. We did not collect urine from our subjects due to ethical constraints. Moreover, the urine antigen test is not widely used now. The gold standard for diagnosing O. viverrini infection is the formalin ether concentration technique (FECT). We therefore benchmarked our new antibody ICTs against FECT, as well as an antibody ICT using crude fluke excretory/secretory (ES) products. While we note percentage positivity comparisons in numerous places throughout the
+
+<--- Page Split --->
+
+ms, we had not conducted a kappa test to determine the extent of agreement between the two tests. We have now conducted a kappa test comparing the predictive value of our tests with FECT (new Supplementary Table 3) and we have modified the Results to compare positivity rates of our antibody tests with the urine antigen RDT (lines 234- 238, page 5).  
+
+2. In the abbreviated methods, the detailed methods are supposedly in 'Appendix 1'. But the supplemental methods are not labeled as such.  
+
+Response: We apologise for the oversight; we have moved the supplementary methods into the main body of the text such that all the methods are now in the main manuscript.  
+
+3. The methods mention excretory/secretory (ES) and extracellular vesicles (EVs). But the supplemental methods mention ES, EVs, ELVs and MVs. This needs to be clarified. Based on the supplementary methods, it is not clear as to which datasets are from this study and which ones are from prior studies.  
+
+Response: We have revised the Methods to ensure consistency and make it clear which datasets are new, and which are historical - see section starting at line 471. 4. The related supplemental figure (S2) shows the VENN diagram with ES, ELVs and MVs. Where do the P1 to P9 fall in the VENN diagram? Along same line, Table S3 does not really need the last column 'Selected for the array' if all of the proteins are being printed. Further, the title of Table S3 needs to be modified as it is not a list of all the 825 proteins detected by MS.  
+
+Response: As per the journal formatting instructions, we have revised the supplementary information such that tables longer than one A4 page are now referred to as Supplemental Data rather than Supplemental Tables. Supplementary Table 3 in the previous version of the ms is now Supplementary Data 1, and it shows the presence in the three different fluke secreted extracts (ES, ELVs and MVs) of the different proteins. Supplementary Data 2 shows the proteins selected for printing on the array and provides their sequences (listed by accession number). In the ms (page 11, lines 615- 617) we had noted the accession numbers for P1 and P9, and these accession numbers can be searched in Supplementary Data 1 and 2 to readily identify the proteins. Nonetheless, to make it easier for readers to access this information, we have revised the Results section to describe that P1 and P9 were detected in all three fluke secretome extracts - ES, ELVs and MVs (lines 215- 217). We have also removed the "selected for array" column in Supplemental Data 1 as requested and replaced it with the P1- 9 protein codes. We also highlighted P1- 9 in the Venn diagram (Supplementary Figure 2).  
+
+5. The second half of the study design repeats itself in the array probing methodology. This redundancy can be cleaned up.  
+
+Response: We acknowledge the reviewer's concern here but point out that the latter part of the Results section focuses on serodiagnosis of CCA as opposed to the first part of the Results focusing on serodiagnosis of fluke infection. These are two very different objectives that we chose to consider separately to avoid any confusion.  
+
+6. On a minor note, what is the composition of the 'array blocking buffer'? There is no reference or mention of it in the supplemental methods.  
+
+Response: The blocking buffer is a commercial product of undisclosed composition. We have modified the relevant section in the Methods on page 10 (lines 586- 587) to provide the details of the product and manufacturer.  
+
+7. Fig 1. a. The ES part needs to be clarified in the text (all through) if it is going to be used to encompass all forms of excretory-secretory proteins.
+
+<--- Page Split --->
+
+Response: Thanks for pointing this out. We have rectified the confusing nomenclature as follows: the entire secreted complement (which contains soluble and vesicular proteins) is now referred to as "secretome"; the soluble proteins are "Excretory/Secretory, or ES; exosome like vesicles are ELVs; microvesicles are MVs. b. While it says 245 of the 278 proteins were printed on the array, the last line of the results section on the characterization of the secretome reads "... a total of 278 printed proteins or protein fragments". In addition, previously published diagnostic targets of O. viverrini and Clonorchis spp resulted in a total of 249 proteins that were selected to be printed. This needs to be clarified.  
+
+Response: We selected 245 proteins for in vitro transcription/translation and printing on the array (as per Supplementary Data 1 and 2). We also included several control proteins that had been expressed in E. coli and purified, including Ov- MUC60- 1, OvMUC60- 2, Cs- glutathione transferase omega- 1 and Cs- glutathione transferase omega- 2. This totals 249 proteins. Some proteins were too large for in vitro transcription- translation and were divided into two ORFs. We have revised Figure 1 and its legend to explain this more carefully.  
+
+8. In the 'immunomics' section of the results, the authors describe that P3 was recognized exclusively by sera of individuals with C. sinensis infection. How do you explain this? I do not see much of a discussion as to why 'thioredoxin' that is almost universally present is picked up by sera from C. sinensis infections. Also, if sera from individuals with CCA are also positive for P3, then is it really 'exclusive'?  
+
+Response: this is a good point, and our wording was poorly chosen. We believe that P3 is the target of an antibody response in O. viverrini infected subjects, but the differences in mean IgG1 and IgG4 levels between infected and uninfected subjects did not reach significance. This is likely because most of our O. viverrini infected subjects had low or medium intensity infections as opposed to the majority of C. sinensis infected subjects having high intensity infection based on FECT. Our CCA patients were mostly FECT negative by the time they were diagnosed with advanced CCA but given that CCA is likely the result of years of chronic (heavy) infection, this is not unexpected. When transitioned to the ICT format, P3 did not perform well compared to P1 and P9. This could have been due to the refolded recombinant protein produced in E. coli presenting different epitopes to the in vitro transcription/translation product used to print the proteome arrays, or the different substrate and reagents used in the different assays. We have modified the text in the Results (page 4, lines 176- 189) to explain this finding.  
+
+9. Fig 2a. The results section mentions 36-antigens (excluding OvES and OvSo) were the targets of significantly elevated IgG1 or IgG4 of which 20 were targets with significant values for both IgG1 and IgG4. The legend says top 40 protein hits and the order of them is different. I do not see the data for P3, P6 and P7 in 2A and 2B. While this is totally up to the authors, I feel that having all the protein names in the middle with their corresponding IgG1 histograms on the top and IgG4 histograms on the bottom, would be easier to see which of the 20 targets were commonly found to have significant. Or it can be flipped vertically so that the naming is easier to read, with IgG1 and IgG4 data to the left and right.  
+
+Response: we like this suggestion by the reviewer and have redrawn Figure 2A/B to present the data as recommended. Panels A and B are now merged into a single panel A. Proteins were ranked based on IgG4 reactivity P values.  
+
+Fig 2b. How do you explain the relative lack of signal in IgG4 for the OvES? How much of the protein was printed on the array? Assuming equal amounts of proteins across, what is the relative abundance of the selected proteins in the ES?
+
+<--- Page Split --->
+
+Response: This is possibly due to a small amount of ES being printed on the array and under- representation of IgG4 reactive proteins in this complex mixture. It might also be explained by the relatively light- to- moderate infection intensity of our O. viverrini infected subjects, hence the greater reaction with IgG1. This weak IgG4 reactivity to ES products was also seen with PoC ICTs containing crude ES products (Figure 5).  
+
+Fig 2c. Was there a correlation between the array signals and ICT data?  
+
+Response: we did not detect significant correlations between array and ICT data. The most logical explanation for this is the different matrices used in protein microarray and ICTs, and the different buffer optimisation approaches employed. Many different buffering conditions were assessed in the optimisation process for the ICTs, and some proteins that were reactive on the arrays did not retain equally robust reactivity in ICT format.  
+
+10. Pg 8. Why were the analyses moved from IgG1 to total IgG for the ICT? I do not see any discussion pertaining to this switch. Also going back, the abstract says IgG reactive antigens were taken forward for recombinant expression. This needs to be corrected.  
+
+Response: When optimising the conditions for printing ICTs, we routinely experienced higher background levels with non- endemic control sera for IgG1 compared to IgG. We have added a statement to this effect in the Results section at the bottom of page 4 (lines 187- 189).  
+
+11. Is there a TPP of some sort CCA/Ov infections? What would be an ideal situations for the two case scenarios?  
+
+Response: We are not aware (from literature searches and discussion with KoLs) of a TPP for serodiagnosis of human liver fluke infections. There are clearly, however, key differences in TPP for the two case scenarios. An infection- focused test should be highly sensitive and field- deployable, while a CCA- specific test should be highly specific (e.g., Needs to differentiate from unrelated hepatobiliary conditions and malignancies) and integrated with clinical workflows for managing and treating cholangiocarcinoma. Such assays will significantly enhance surveillance and early diagnosis, ultimately reducing the burden of opisthorchiasis- related complications in endemic regions. We have expanded the Discussion (lines 411- 432) to address this concern.  
+
+12. Pg 11. Why is it that the CCA cases were 'suspected' to be caused by chronic O. viverrini infection? Were they not diagnosed as Ov?  
+
+Response: One can never be certain (with humans) of the cause of cancer. These subjects resided in O. viverrini endemic areas, ate raw fish, and some at least had a history of positive FECT diagnosis for O. viverrini infection. At the point of clinical diagnosis, many CCA patients no longer harbour flukes, probably due to the inhospitable environment in the cancerous liver. Moreover, CCA is relatively rare in areas where liver flukes are not endemic, and numbers are dwarfed by hepatocellular carcinoma. In unpublished work, we have shown that patients with HCC from fluke non-endemic sites do not have anti- P1 or - P9 antibodies.  
+
+13. In the discussion, I don't think it is appropriate to say a 'genome-scale proteome microarray' was constructed. The protein array was specific to the ES.  
+
+Response: We have revised the text from "genome" to "secretome" on page 6, line 346.  
+
+14. The authors do allude to it, but the discussion on EPG between C. sinensis and O. viverrini is at best is speculative as the correlations between egg burden and signal weak to modest.
+
+<--- Page Split --->
+
+Response: We agree with the reviewer but feel that the Discussion is a suitable place for at least some level of speculation. We noted in the original text that "we are hesitant to make quantitative comparisons because the average infection intensities for the C. sinensis infected subjects (average 5,320 EPG, median 1,224) was higher than that for O. viverrini infected subjects (average 748 EPG, median 294)." 15. Pg 14. The starting lines on the seropositivity endemic control populations. During the screening process, were the individuals not asked if they have been tested before? Was this a totally new region that was previously never studied? Response: Non- endemic control cases were collected from southern and central regions of Thailand, non- endemic areas for opisthorchiasis, and no parasite material was detected in stool examinations using the concentration method. These populations also had no history of consuming raw fish. The endemic control cases were collected from the northeastern part of Thailand from an opisthorchiasis endemic area, but these subjects were negative for parasite infection by stool examination (Elkins et al 1986 – new reference 38). These subjects were also interviewed and had no history of consuming raw fish. Blood was collected at the same time as feces. Subjects who were negative for parasite eggs by fecal examination (FECT) but were seropositive could be explained as subclinical cases and/or harbouring low intensity infections that were not detected by FECT, or they remained seropositive from an earlier infection. We have revised the Methods section to include this information at the top of page 10, lines 543- 551. 16. Maybe I missed something, but looked like all the samples were egg positive cases. And the discussion on CCA cases being in a 'non- infected' state was confusing for me. Were the blood and stool sampled at different times? What are the kinetics of antibody levels? Do they drop off quickly too? Response: For the serodiagnosis of infection component of the study, all "infected" subjects were diagnosed with the gold standard FECT. Endemic and non- endemic controls were FECT- negative. For the serodiagnosis of CCA cases, blood and stool samples were collected at the same time at the hospital. Studies on antibody kinetics were not performed. Diagnosis of CCA cases was based on gross and histopathological examinations. All cases were located in opisthorchiasis endemic areas and patients all confirmed eating raw cyprinoid fish (O. viverrini intermediate host). Two CCA cases were positive for O. viverrini eggs by stool concentration (Elkins et al 1986, ref 38). Normally, CCA associated opisthorchiasis is characterized by an absence of O. viverrini egg in stool samples, possibly due to chronic opisthorchiasis and associated chronic biliary tract inflammation caused by choledocholithiasis, cholelithiasis, or primary sclerosing cholangitis and bile duct obstruction. We have now included a new Supplementary Data 3 with the stage of CCA diagnosis.  
+
+## Reviewer 3.  
+
+1. ....one important question arises regarding How to interpretation the results? Specifically, it is unclear how to interpret, when it is positive results indicating Opisthocis viverrini infection, Opisthocis viverrini infection in the past, or Opisthocis viverrini-related cholangiocarcinoma. This question should be further discussed in the discussion section. Response: We have provided a new section in the Discussion (lines 411-432) focusing on Target Product Profile which addresses the reviewer's concern.
+
+<--- Page Split --->
+
+Point by point response  
+
+## Reviewers 1 and 3 did not have any remaining questions.  
+
+## Reviewer 2.  
+
+1. When the authors say 'routinely finding' anti-IgG1 secondary antibodies as suboptimal, is it across the board for all such studies? If so, why use it in the first place?Response: We were referring to this particular study and noted high levels of background with this particular secondary antiserum. Given the quality and informative nature of the other isotypes, we proceeded without specifically looking at IgG1.2. Since the authors increased the sample numbers by utilizing the same samples used for array screening, how do the results correlate between the IgG1 (array) and the ICT (IgG).Response: We did not detect a significant correlation between IgG1 array and IgG ICT data. We did however detect a significant correlation between IgG4 array and ICT results.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__caa685a328daac33d8261fb19023ca5fa2ca99c25c1ffc44e94d273ac99b6bb0/supplementary_0_Peer Review file__caa685a328daac33d8261fb19023ca5fa2ca99c25c1ffc44e94d273ac99b6bb0_det.mmd

@@ -0,0 +1,317 @@
+<|ref|>title<|/ref|><|det|>[[72, 50, 295, 80]]<|/det|>
+# nature portfolio  
+
+<|ref|>title<|/ref|><|det|>[[74, 96, 296, 120]]<|/det|>
+# Peer Review File  
+
+<|ref|>title<|/ref|><|det|>[[73, 161, 910, 211]]<|/det|>
+# Immunomics-guided biomarker discovery for human liver fluke infection and infection-associated cholangiocarcinoma  
+
+<|ref|>text<|/ref|><|det|>[[73, 224, 470, 241]]<|/det|>
+Corresponding Author: Professor Alex Loukas  
+
+<|ref|>text<|/ref|><|det|>[[72, 274, 864, 290]]<|/det|>
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+<|ref|>text<|/ref|><|det|>[[73, 325, 145, 339]]<|/det|>
+Version 0:  
+
+<|ref|>text<|/ref|><|det|>[[73, 352, 220, 366]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 378, 160, 392]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 404, 238, 417]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 417, 910, 457]]<|/det|>
+Food- borne trematodiases pose a serious public health problem, affecting more than 50 million people worldwide. Liver flukes, including Opisthorchis viverrini, Opisthorchis felineus, and Clonorchis sinensis, are not only parasitic infections but also carcinogenic, contributing to bile duct cancer and other related malignancies.  
+
+<|ref|>text<|/ref|><|det|>[[72, 457, 893, 496]]<|/det|>
+The prevention and control of liver fluke infections in both endemic and non- endemic areas rely on parasite detection. However, traditional detection methods, while considered the gold standard for diagnosis, are labor- intensive, have low sensitivity, and require skilled technicians.  
+
+<|ref|>text<|/ref|><|det|>[[70, 495, 911, 523]]<|/det|>
+This study presents promising results, demonstrating the potential for a rapid ELISA- based screening method that could be effectively applied for field detection in South Asia.  
+
+<|ref|>text<|/ref|><|det|>[[72, 534, 920, 575]]<|/det|>
+The study is a valuable contribution to field detection for the parasitic and parasite- related diseases. Compared to existing literature, it introduces an innovative design and a novel biomarker for rapid schistosomiasis screening, making it an original and significant advancement in the field.  
+
+<|ref|>text<|/ref|><|det|>[[72, 586, 917, 640]]<|/det|>
+The manuscript presents results based on two biomarkers for O. viverrini infection and infection- associated CCA, evaluated in a cohort of patients. However, the study's sample size appears to be a limitation, which may affect the persuasiveness of the findings. I believe that increasing the sample size would strengthen the conclusions and enhance the reliability of the results.  
+
+<|ref|>text<|/ref|><|det|>[[72, 651, 899, 692]]<|/det|>
+In this study, patients with CCA caused by chronic liver fluke infection were enrolled. However, the interpretation of CCA needs to be described in greater detail in the methods section. Were all patients, especially CCAs, verified through other diagnostic examinations? Please clarify this in the manuscript to ensure the accuracy and reliability of the findings.  
+
+<|ref|>text<|/ref|><|det|>[[72, 704, 161, 717]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 730, 238, 744]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 744, 899, 796]]<|/det|>
+The manuscript relates to the identification of excretory- secretory proteins of O. viverrini and generating a protein array of selected targets from the excretory- secretory proteins. These protein arrays were screened with infected and un- infected sera to identify potential biomarkers. This is a well written paper and the flow is good. It would have been helpful if the manuscript included line numbers to help with the comments.  
+
+<|ref|>text<|/ref|><|det|>[[72, 809, 152, 821]]<|/det|>
+Comments:  
+
+<|ref|>text<|/ref|><|det|>[[70, 821, 919, 936]]<|/det|>
+1. In the introduction, the authors mention that there have been other tests (antigen-based and antibody-based) that were developed. However, I do not see any comparison or discussion of the current proposed biomarkers (P1 and P9) with either of the available assays in the manuscript. Were the same samples tested for antigen by any of the methods? Ideally, a new assay needs to be compared with existing assays. Although the assays cited are antigen-based, and the current manuscript is antibody based, from a public health perspective, it would finally come down to which assay has better sensitivities in detecting infections while maintaining specificity. 
+2. In the abbreviated methods, the detailed methods are supposedly in 'Appendix 1'. But the supplemental methods are not labeled as such. 
+3. The methods mention excretory/secretory (ES) and extracellular vesicles (EVs). But the supplemental methods mention
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[70, 48, 868, 75]]<|/det|>
+ES, EVs, ELVs and MVs. This needs to be clarified. Based on the supplementary methods, it is not clear as to which datasets are from this study and which ones are from prior studies.  
+
+<|ref|>text<|/ref|><|det|>[[70, 74, 920, 155]]<|/det|>
+4. The related supplemental figure (S2) shows the VENN diagram with ES, ELVs and MVs. Where do the P1 to P9 fall in the VENN diagram? Along same line, Table S3 does not really need the last column 'Selected for the array' if all of the proteins are being printed. Further, the title of Table S3 needs to be modified as it is not a list of all the 825 proteins detected by MS. 
+5. The second half of the study design repeats itself in the array probing methodology. This redundancy can be cleaned up. 
+6. On a minor note, what is the composition of the 'array blocking buffer'? There is no reference or mention of it in the supplemental methods.  
+
+<|ref|>text<|/ref|><|det|>[[72, 155, 130, 167]]<|/det|>
+7. Fig 1.  
+
+<|ref|>text<|/ref|><|det|>[[70, 166, 884, 193]]<|/det|>
+a. The ES part needs to be clarified in the text (all through) if it is going to be used to encompass all forms of excretory-secretory proteins.  
+
+<|ref|>text<|/ref|><|det|>[[70, 193, 920, 245]]<|/det|>
+b. While it says 245 of the 278 proteins were printed on the array, the last line of the results section on the characterization of the secretome reads "... a total of 278 printed proteins or protein fragments". In addition, previously published diagnostic targets of O. viverrini and Clonorchis spp resulted in a total of 249 proteins that were selected to be printed. This needs to be clarified.  
+
+<|ref|>text<|/ref|><|det|>[[70, 244, 917, 296]]<|/det|>
+8. In the 'immunomics' section of the results, the authors describe that P3 was recognized exclusively by sera of individuals with C. sinensis infection. How do you explain this? I do not see much of a discussion as to why 'thioredoxin' that is almost universally present is picked up by sera from C. sinensis infections. Also, if sera from individuals with CCA are also positive for P3, then is it really 'exclusive'?  
+
+<|ref|>text<|/ref|><|det|>[[72, 296, 130, 309]]<|/det|>
+9. Fig 2.  
+
+<|ref|>text<|/ref|><|det|>[[70, 308, 916, 385]]<|/det|>
+a. The results section mentions 36-antigens (excluding OvES and OvSo) were the targets of significantly elevated IgG1 or IgG4 of which 20 were targets with significant values for both IgG1 and IgG4. The legend says top 40 protein hits and the order of them is different. I do not see the data for P3, P6 and P7 in 2A and 2B. While this is totally up to the authors, I feel that having all the protein names in the middle with their corresponding IgG1 histograms on the top and IgG4 histograms on the bottom, would be easier to see which of the 20 targets were commonly found to have significant. Or it can be flipped vertically so that the naming is easier to read, with IgG1 and IgG4 data to the left and right.  
+
+<|ref|>text<|/ref|><|det|>[[70, 384, 904, 411]]<|/det|>
+b. How do you explain the relative lack of signal in IgG4 for the OvES? How much of the protein was printed on the array? Assuming equal amounts of proteins across, what is the relative abundance of the selected proteins in the ES?  
+
+<|ref|>text<|/ref|><|det|>[[72, 411, 533, 424]]<|/det|>
+c. Was there a correlation between the array signals and ICT data?  
+
+<|ref|>text<|/ref|><|det|>[[70, 424, 918, 463]]<|/det|>
+10. Pg 8. Why were the analyses moved from IgG1 to total IgG for the ICT? I do not see any discussion pertaining to this switch. Also going back, the abstract says IgG reactive antigens were taken forward for recombinant expression. This needs to be corrected.  
+
+<|ref|>text<|/ref|><|det|>[[70, 463, 884, 503]]<|/det|>
+11. Is there a TPP of some sort CCA/Ov infections? What would be an ideal situations for the two case scenarios? 
+12. Pg 11. Why is it that the CCA cases were 'suspected' to be caused by chronic O. viverrini infection? Were they not diagnosed as Ov?  
+
+<|ref|>text<|/ref|><|det|>[[70, 502, 880, 529]]<|/det|>
+13. In the discussion, I don't think it is appropriate to say a 'genome-scale proteome microarray' was constructed. The protein array was specific to the ES.  
+
+<|ref|>text<|/ref|><|det|>[[70, 529, 904, 555]]<|/det|>
+14. The authors do allude to it, but the discussion on EPG between C. sinensis and O. viverrini is at best is speculative as the correlations between egg burden and signal weak to modest.  
+
+<|ref|>text<|/ref|><|det|>[[70, 555, 910, 571]]<|/det|>
+15. Pg 14. The starting lines on the seropositivity endemic control populations. During the screening process, were the  
+
+<|ref|>text<|/ref|><|det|>[[70, 570, 910, 585]]<|/det|>
+individuals not asked if they have been tested before? Was this a totally new region that was previously never studied?  
+
+<|ref|>text<|/ref|><|det|>[[70, 585, 910, 624]]<|/det|>
+16. Maybe I missed something, but looked like all the samples were egg positive cases. And the discussion on CCA cases being in a 'non-infected' state was confusing for me. Were the blood and stool sampled at different times? What are the kinetics of antibody levels? Do they drop off quickly too?  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 648, 163, 661]]<|/det|>
+## Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[72, 675, 238, 688]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 688, 894, 727]]<|/det|>
+The manuscript title, "Immunomics-guided biomarker discovery for human liver fluke infection and infection-associated cholangiocarcinoma" aims to develop a diagnostic tool, specifically an immunochromatography strip, for the detection of Opisthocis viverrini infection and O. viverrini-associated cholangiocarcinoma.  
+
+<|ref|>text<|/ref|><|det|>[[72, 737, 910, 791]]<|/det|>
+The manuscript is well-structured and informative, and the experimental design is clearly outlined. However, one important question arises regarding How to interpretation the results? Specifically, it is unclear how to interpret, when it is positive results indicating Opisthocis viverrini infection, Opisthocis viverrini infection in the past, or Opisthocis viverrini-related cholangiocarcinoma. This question should be further discussed in the discussion section.  
+
+<|ref|>text<|/ref|><|det|>[[72, 803, 144, 816]]<|/det|>
+Version 1:  
+
+<|ref|>text<|/ref|><|det|>[[72, 829, 219, 842]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[72, 855, 160, 868]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[72, 881, 238, 894]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 894, 923, 947]]<|/det|>
+This manuscript presents valuable findings in the field of Opisthocris viverrini infection and associated cholangiocarcinoma (CCA). The study offers important insights into chronic liver fluke infection, highlighting a research area that deserves greater attention given its status as a neglected tropical disease. The work is well- conducted and makes a significant contribution to the field. I recommend its publication in Nature Communications after addressing minor revisions (if any are required by
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 47, 188, 60]]<|/det|>
+other reviewers).  
+
+<|ref|>text<|/ref|><|det|>[[72, 72, 161, 86]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[72, 99, 788, 128]]<|/det|>
+(Remarks to the Author) I think the authors have answered satisfactorily to all the questions/comments. A couple of minor queries.  
+
+<|ref|>text<|/ref|><|det|>[[72, 139, 915, 180]]<|/det|>
+When the authors say 'routinely finding' anti- IgG1 secondary antibodies as sub- optimal, is it across the board for all such studies? If so, why use it in the first place? Secondly, since the authors increased the sample numbers by utilizing the same samples used for array screening, how do the results correlate between the IgG1 (array) and the ICT (IgG)?  
+
+<|ref|>text<|/ref|><|det|>[[72, 203, 161, 216]]<|/det|>
+Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[72, 230, 565, 257]]<|/det|>
+(Remarks to the Author) All of the raised question have been satisfactorily addressed by authors.  
+
+<|ref|>text<|/ref|><|det|>[[72, 609, 916, 662]]<|/det|>
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+<|ref|>text<|/ref|><|det|>[[72, 662, 796, 676]]<|/det|>
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+<|ref|>text<|/ref|><|det|>[[72, 675, 910, 728]]<|/det|>
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+<|ref|>text<|/ref|><|det|>[[72, 728, 618, 741]]<|/det|>
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[95, 85, 306, 100]]<|/det|>
+Point by point response  
+
+<|ref|>sub_title<|/ref|><|det|>[[95, 117, 205, 133]]<|/det|>
+## Reviewer 1.  
+
+<|ref|>text<|/ref|><|det|>[[95, 149, 868, 231]]<|/det|>
+1. The manuscript presents results based on two biomarkers for O. viverrini infection and infection-associated CCA, evaluated in a cohort of patients. However, the study's sample size appears to be a limitation, which may affect the persuasiveness of the findings. I believe that increasing the sample size would strengthen the conclusions and enhance the reliability of the results.  
+
+<|ref|>text<|/ref|><|det|>[[115, 231, 875, 427]]<|/det|>
+Response: We emphasised geographic diversity of serum samples from flukeinfected subjects used in the study (Thailand, Laos, China). The reviewer's point is well taken, so we used the two best ICTs to screen another 50 serum samples from O. viverrini-infected subjects where diagnosis was noted as egg-positive or eggnegative, and egg counts were not quantified as they were for the subjects we currently report in the manuscript. Nonetheless, we thought this would be a valuable addition to the paper and have included the information as Supplementary Figure 4. Please note, this cohort was one of those used to screen the proteome microarray (where we were not concerned with quantified EPG), but we had not previously used this cohort to screen the ICTs. With this new cohort of subjects, we observed similar sensitivity and specificity of the ICTs to that reported for subjects where egg counts had been quantified.  
+
+<|ref|>text<|/ref|><|det|>[[95, 428, 878, 510]]<|/det|>
+2. In this study, patients with CCA caused by chronic liver fluke infection were enrolled. However, the interpretation of CCA needs to be described in greater detail in the methods section. Were all patients, especially CCAs, verified through other diagnostic examinations? Please clarify this in the manuscript to ensure the accuracy and reliability of the findings.  
+
+<|ref|>text<|/ref|><|det|>[[117, 510, 878, 592]]<|/det|>
+Response: The requested information on CCA patient clinical diagnosis is now provided in Supplementary Data 3. In addition, we have added a new panel to Figure 5 (Fig. 5E) that shows the positivity rate for serodiagnosis of CCA at different clinical stages of malignancy, and we have discussed the findings at the end of the Results (page 5).  
+
+<|ref|>sub_title<|/ref|><|det|>[[95, 608, 205, 623]]<|/det|>
+## Reviewer 2.  
+
+<|ref|>text<|/ref|><|det|>[[95, 624, 880, 904]]<|/det|>
+1. In the introduction, the authors mention that there have been other tests (antigen-based and antibody-based) that were developed. However, I do not see any comparison or discussion of the current proposed biomarkers (P1 and P9) with either of the available assays in the manuscript. Were the same samples tested for antigen by any of the methods? Ideally, a new assay needs to be compared with existing assays. Although the assays cited are antigen-based, and the current manuscript is antibody based, from a public health perspective, it would finally come down to which assay has better sensitivities in detecting infections while maintaining specificity. Response: The antigen detection RDT for opisthorchiasis relies on detection of antigen in urine of infected subjects. Antigen is detected using a monoclonal antibody raised against an undefined antigen, probably glycan in nature based on its Western blot profile. We did not collect urine from our subjects due to ethical constraints. Moreover, the urine antigen test is not widely used now. The gold standard for diagnosing O. viverrini infection is the formalin ether concentration technique (FECT). We therefore benchmarked our new antibody ICTs against FECT, as well as an antibody ICT using crude fluke excretory/secretory (ES) products. While we note percentage positivity comparisons in numerous places throughout the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 860, 168]]<|/det|>
+ms, we had not conducted a kappa test to determine the extent of agreement between the two tests. We have now conducted a kappa test comparing the predictive value of our tests with FECT (new Supplementary Table 3) and we have modified the Results to compare positivity rates of our antibody tests with the urine antigen RDT (lines 234- 238, page 5).  
+
+<|ref|>text<|/ref|><|det|>[[117, 168, 870, 201]]<|/det|>
+2. In the abbreviated methods, the detailed methods are supposedly in 'Appendix 1'. But the supplemental methods are not labeled as such.  
+
+<|ref|>text<|/ref|><|det|>[[117, 201, 875, 249]]<|/det|>
+Response: We apologise for the oversight; we have moved the supplementary methods into the main body of the text such that all the methods are now in the main manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[117, 249, 860, 313]]<|/det|>
+3. The methods mention excretory/secretory (ES) and extracellular vesicles (EVs). But the supplemental methods mention ES, EVs, ELVs and MVs. This needs to be clarified. Based on the supplementary methods, it is not clear as to which datasets are from this study and which ones are from prior studies.  
+
+<|ref|>text<|/ref|><|det|>[[117, 314, 870, 425]]<|/det|>
+Response: We have revised the Methods to ensure consistency and make it clear which datasets are new, and which are historical - see section starting at line 471. 4. The related supplemental figure (S2) shows the VENN diagram with ES, ELVs and MVs. Where do the P1 to P9 fall in the VENN diagram? Along same line, Table S3 does not really need the last column 'Selected for the array' if all of the proteins are being printed. Further, the title of Table S3 needs to be modified as it is not a list of all the 825 proteins detected by MS.  
+
+<|ref|>text<|/ref|><|det|>[[117, 426, 875, 675]]<|/det|>
+Response: As per the journal formatting instructions, we have revised the supplementary information such that tables longer than one A4 page are now referred to as Supplemental Data rather than Supplemental Tables. Supplementary Table 3 in the previous version of the ms is now Supplementary Data 1, and it shows the presence in the three different fluke secreted extracts (ES, ELVs and MVs) of the different proteins. Supplementary Data 2 shows the proteins selected for printing on the array and provides their sequences (listed by accession number). In the ms (page 11, lines 615- 617) we had noted the accession numbers for P1 and P9, and these accession numbers can be searched in Supplementary Data 1 and 2 to readily identify the proteins. Nonetheless, to make it easier for readers to access this information, we have revised the Results section to describe that P1 and P9 were detected in all three fluke secretome extracts - ES, ELVs and MVs (lines 215- 217). We have also removed the "selected for array" column in Supplemental Data 1 as requested and replaced it with the P1- 9 protein codes. We also highlighted P1- 9 in the Venn diagram (Supplementary Figure 2).  
+
+<|ref|>text<|/ref|><|det|>[[118, 675, 761, 707]]<|/det|>
+5. The second half of the study design repeats itself in the array probing methodology. This redundancy can be cleaned up.  
+
+<|ref|>text<|/ref|><|det|>[[117, 708, 870, 771]]<|/det|>
+Response: We acknowledge the reviewer's concern here but point out that the latter part of the Results section focuses on serodiagnosis of CCA as opposed to the first part of the Results focusing on serodiagnosis of fluke infection. These are two very different objectives that we chose to consider separately to avoid any confusion.  
+
+<|ref|>text<|/ref|><|det|>[[117, 772, 855, 804]]<|/det|>
+6. On a minor note, what is the composition of the 'array blocking buffer'? There is no reference or mention of it in the supplemental methods.  
+
+<|ref|>text<|/ref|><|det|>[[117, 805, 870, 853]]<|/det|>
+Response: The blocking buffer is a commercial product of undisclosed composition. We have modified the relevant section in the Methods on page 10 (lines 586- 587) to provide the details of the product and manufacturer.  
+
+<|ref|>text<|/ref|><|det|>[[117, 854, 875, 887]]<|/det|>
+7. Fig 1. a. The ES part needs to be clarified in the text (all through) if it is going to be used to encompass all forms of excretory-secretory proteins.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 875, 232]]<|/det|>
+Response: Thanks for pointing this out. We have rectified the confusing nomenclature as follows: the entire secreted complement (which contains soluble and vesicular proteins) is now referred to as "secretome"; the soluble proteins are "Excretory/Secretory, or ES; exosome like vesicles are ELVs; microvesicles are MVs. b. While it says 245 of the 278 proteins were printed on the array, the last line of the results section on the characterization of the secretome reads "... a total of 278 printed proteins or protein fragments". In addition, previously published diagnostic targets of O. viverrini and Clonorchis spp resulted in a total of 249 proteins that were selected to be printed. This needs to be clarified.  
+
+<|ref|>text<|/ref|><|det|>[[117, 232, 873, 346]]<|/det|>
+Response: We selected 245 proteins for in vitro transcription/translation and printing on the array (as per Supplementary Data 1 and 2). We also included several control proteins that had been expressed in E. coli and purified, including Ov- MUC60- 1, OvMUC60- 2, Cs- glutathione transferase omega- 1 and Cs- glutathione transferase omega- 2. This totals 249 proteins. Some proteins were too large for in vitro transcription- translation and were divided into two ORFs. We have revised Figure 1 and its legend to explain this more carefully.  
+
+<|ref|>text<|/ref|><|det|>[[117, 346, 875, 430]]<|/det|>
+8. In the 'immunomics' section of the results, the authors describe that P3 was recognized exclusively by sera of individuals with C. sinensis infection. How do you explain this? I do not see much of a discussion as to why 'thioredoxin' that is almost universally present is picked up by sera from C. sinensis infections. Also, if sera from individuals with CCA are also positive for P3, then is it really 'exclusive'?  
+
+<|ref|>text<|/ref|><|det|>[[117, 430, 875, 657]]<|/det|>
+Response: this is a good point, and our wording was poorly chosen. We believe that P3 is the target of an antibody response in O. viverrini infected subjects, but the differences in mean IgG1 and IgG4 levels between infected and uninfected subjects did not reach significance. This is likely because most of our O. viverrini infected subjects had low or medium intensity infections as opposed to the majority of C. sinensis infected subjects having high intensity infection based on FECT. Our CCA patients were mostly FECT negative by the time they were diagnosed with advanced CCA but given that CCA is likely the result of years of chronic (heavy) infection, this is not unexpected. When transitioned to the ICT format, P3 did not perform well compared to P1 and P9. This could have been due to the refolded recombinant protein produced in E. coli presenting different epitopes to the in vitro transcription/translation product used to print the proteome arrays, or the different substrate and reagents used in the different assays. We have modified the text in the Results (page 4, lines 176- 189) to explain this finding.  
+
+<|ref|>text<|/ref|><|det|>[[117, 658, 870, 805]]<|/det|>
+9. Fig 2a. The results section mentions 36-antigens (excluding OvES and OvSo) were the targets of significantly elevated IgG1 or IgG4 of which 20 were targets with significant values for both IgG1 and IgG4. The legend says top 40 protein hits and the order of them is different. I do not see the data for P3, P6 and P7 in 2A and 2B. While this is totally up to the authors, I feel that having all the protein names in the middle with their corresponding IgG1 histograms on the top and IgG4 histograms on the bottom, would be easier to see which of the 20 targets were commonly found to have significant. Or it can be flipped vertically so that the naming is easier to read, with IgG1 and IgG4 data to the left and right.  
+
+<|ref|>text<|/ref|><|det|>[[117, 806, 870, 850]]<|/det|>
+Response: we like this suggestion by the reviewer and have redrawn Figure 2A/B to present the data as recommended. Panels A and B are now merged into a single panel A. Proteins were ranked based on IgG4 reactivity P values.  
+
+<|ref|>text<|/ref|><|det|>[[117, 851, 860, 902]]<|/det|>
+Fig 2b. How do you explain the relative lack of signal in IgG4 for the OvES? How much of the protein was printed on the array? Assuming equal amounts of proteins across, what is the relative abundance of the selected proteins in the ES?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 874, 184]]<|/det|>
+Response: This is possibly due to a small amount of ES being printed on the array and under- representation of IgG4 reactive proteins in this complex mixture. It might also be explained by the relatively light- to- moderate infection intensity of our O. viverrini infected subjects, hence the greater reaction with IgG1. This weak IgG4 reactivity to ES products was also seen with PoC ICTs containing crude ES products (Figure 5).  
+
+<|ref|>text<|/ref|><|det|>[[118, 184, 764, 201]]<|/det|>
+Fig 2c. Was there a correlation between the array signals and ICT data?  
+
+<|ref|>text<|/ref|><|det|>[[117, 201, 876, 297]]<|/det|>
+Response: we did not detect significant correlations between array and ICT data. The most logical explanation for this is the different matrices used in protein microarray and ICTs, and the different buffer optimisation approaches employed. Many different buffering conditions were assessed in the optimisation process for the ICTs, and some proteins that were reactive on the arrays did not retain equally robust reactivity in ICT format.  
+
+<|ref|>text<|/ref|><|det|>[[118, 297, 867, 362]]<|/det|>
+10. Pg 8. Why were the analyses moved from IgG1 to total IgG for the ICT? I do not see any discussion pertaining to this switch. Also going back, the abstract says IgG reactive antigens were taken forward for recombinant expression. This needs to be corrected.  
+
+<|ref|>text<|/ref|><|det|>[[118, 363, 869, 428]]<|/det|>
+Response: When optimising the conditions for printing ICTs, we routinely experienced higher background levels with non- endemic control sera for IgG1 compared to IgG. We have added a statement to this effect in the Results section at the bottom of page 4 (lines 187- 189).  
+
+<|ref|>text<|/ref|><|det|>[[118, 428, 795, 460]]<|/det|>
+11. Is there a TPP of some sort CCA/Ov infections? What would be an ideal situations for the two case scenarios?  
+
+<|ref|>text<|/ref|><|det|>[[117, 461, 869, 624]]<|/det|>
+Response: We are not aware (from literature searches and discussion with KoLs) of a TPP for serodiagnosis of human liver fluke infections. There are clearly, however, key differences in TPP for the two case scenarios. An infection- focused test should be highly sensitive and field- deployable, while a CCA- specific test should be highly specific (e.g., Needs to differentiate from unrelated hepatobiliary conditions and malignancies) and integrated with clinical workflows for managing and treating cholangiocarcinoma. Such assays will significantly enhance surveillance and early diagnosis, ultimately reducing the burden of opisthorchiasis- related complications in endemic regions. We have expanded the Discussion (lines 411- 432) to address this concern.  
+
+<|ref|>text<|/ref|><|det|>[[118, 625, 870, 658]]<|/det|>
+12. Pg 11. Why is it that the CCA cases were 'suspected' to be caused by chronic O. viverrini infection? Were they not diagnosed as Ov?  
+
+<|ref|>text<|/ref|><|det|>[[117, 658, 869, 789]]<|/det|>
+Response: One can never be certain (with humans) of the cause of cancer. These subjects resided in O. viverrini endemic areas, ate raw fish, and some at least had a history of positive FECT diagnosis for O. viverrini infection. At the point of clinical diagnosis, many CCA patients no longer harbour flukes, probably due to the inhospitable environment in the cancerous liver. Moreover, CCA is relatively rare in areas where liver flukes are not endemic, and numbers are dwarfed by hepatocellular carcinoma. In unpublished work, we have shown that patients with HCC from fluke non-endemic sites do not have anti- P1 or - P9 antibodies.  
+
+<|ref|>text<|/ref|><|det|>[[118, 790, 867, 822]]<|/det|>
+13. In the discussion, I don't think it is appropriate to say a 'genome-scale proteome microarray' was constructed. The protein array was specific to the ES.  
+
+<|ref|>text<|/ref|><|det|>[[118, 822, 860, 853]]<|/det|>
+Response: We have revised the text from "genome" to "secretome" on page 6, line 346.  
+
+<|ref|>text<|/ref|><|det|>[[118, 855, 864, 904]]<|/det|>
+14. The authors do allude to it, but the discussion on EPG between C. sinensis and O. viverrini is at best is speculative as the correlations between egg burden and signal weak to modest.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 875, 730]]<|/det|>
+Response: We agree with the reviewer but feel that the Discussion is a suitable place for at least some level of speculation. We noted in the original text that "we are hesitant to make quantitative comparisons because the average infection intensities for the C. sinensis infected subjects (average 5,320 EPG, median 1,224) was higher than that for O. viverrini infected subjects (average 748 EPG, median 294)." 15. Pg 14. The starting lines on the seropositivity endemic control populations. During the screening process, were the individuals not asked if they have been tested before? Was this a totally new region that was previously never studied? Response: Non- endemic control cases were collected from southern and central regions of Thailand, non- endemic areas for opisthorchiasis, and no parasite material was detected in stool examinations using the concentration method. These populations also had no history of consuming raw fish. The endemic control cases were collected from the northeastern part of Thailand from an opisthorchiasis endemic area, but these subjects were negative for parasite infection by stool examination (Elkins et al 1986 – new reference 38). These subjects were also interviewed and had no history of consuming raw fish. Blood was collected at the same time as feces. Subjects who were negative for parasite eggs by fecal examination (FECT) but were seropositive could be explained as subclinical cases and/or harbouring low intensity infections that were not detected by FECT, or they remained seropositive from an earlier infection. We have revised the Methods section to include this information at the top of page 10, lines 543- 551. 16. Maybe I missed something, but looked like all the samples were egg positive cases. And the discussion on CCA cases being in a 'non- infected' state was confusing for me. Were the blood and stool sampled at different times? What are the kinetics of antibody levels? Do they drop off quickly too? Response: For the serodiagnosis of infection component of the study, all "infected" subjects were diagnosed with the gold standard FECT. Endemic and non- endemic controls were FECT- negative. For the serodiagnosis of CCA cases, blood and stool samples were collected at the same time at the hospital. Studies on antibody kinetics were not performed. Diagnosis of CCA cases was based on gross and histopathological examinations. All cases were located in opisthorchiasis endemic areas and patients all confirmed eating raw cyprinoid fish (O. viverrini intermediate host). Two CCA cases were positive for O. viverrini eggs by stool concentration (Elkins et al 1986, ref 38). Normally, CCA associated opisthorchiasis is characterized by an absence of O. viverrini egg in stool samples, possibly due to chronic opisthorchiasis and associated chronic biliary tract inflammation caused by choledocholithiasis, cholelithiasis, or primary sclerosing cholangitis and bile duct obstruction. We have now included a new Supplementary Data 3 with the stage of CCA diagnosis.  
+
+<|ref|>sub_title<|/ref|><|det|>[[95, 757, 205, 772]]<|/det|>
+## Reviewer 3.  
+
+<|ref|>text<|/ref|><|det|>[[115, 773, 844, 888]]<|/det|>
+1. ....one important question arises regarding How to interpretation the results? Specifically, it is unclear how to interpret, when it is positive results indicating Opisthocis viverrini infection, Opisthocis viverrini infection in the past, or Opisthocis viverrini-related cholangiocarcinoma. This question should be further discussed in the discussion section. Response: We have provided a new section in the Discussion (lines 411-432) focusing on Target Product Profile which addresses the reviewer's concern.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[95, 84, 308, 100]]<|/det|>
+Point by point response  
+
+<|ref|>sub_title<|/ref|><|det|>[[95, 117, 643, 135]]<|/det|>
+## Reviewers 1 and 3 did not have any remaining questions.  
+
+<|ref|>sub_title<|/ref|><|det|>[[95, 150, 206, 166]]<|/det|>
+## Reviewer 2.  
+
+<|ref|>text<|/ref|><|det|>[[95, 167, 880, 380]]<|/det|>
+1. When the authors say 'routinely finding' anti-IgG1 secondary antibodies as suboptimal, is it across the board for all such studies? If so, why use it in the first place?Response: We were referring to this particular study and noted high levels of background with this particular secondary antiserum. Given the quality and informative nature of the other isotypes, we proceeded without specifically looking at IgG1.2. Since the authors increased the sample numbers by utilizing the same samples used for array screening, how do the results correlate between the IgG1 (array) and the ICT (IgG).Response: We did not detect a significant correlation between IgG1 array and IgG ICT data. We did however detect a significant correlation between IgG4 array and ICT results.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__ceb012ed561eda660b4d01889149b0af06ed7f147b3579ba0b2d245783ecf797/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review file__ceb012ed561eda660b4d01889149b0af06ed7f147b3579ba0b2d245783ecf797/supplementary_0_Peer Review file__ceb012ed561eda660b4d01889149b0af06ed7f147b3579ba0b2d245783ecf797.mmd

@@ -0,0 +1,58 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+# Engineering synthetic signaling receptors to enable erythropoietin-free erythropoiesis  
+
+Corresponding Author: Dr M. Cromer  
+
+This manuscript has been previously reviewed at another journal. This document only contains information relating to versions considered at Nature Communications.  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author) The authors did a good job addressing my comments. I don't have any other critiques.  
+
+Reviewer #4  
+
+(Remarks to the Author)  
+
+In this revised version the authors properly addressed the most relevant critiques raised by the reviewer, resulting in a significantly improved manuscript. My only minor concern relates to the lack of clarity of graph legends: in detail, while it is clear that \(+Epo\) and \(+BB\) indicate samples treated with each single agent, it is not immediate to understand what \(+ / +\) and \(- / -\) mean. I would suggest to include a description in the figure legend. Also, in Figure S14C the annotation of the \(x\) axis is missing.
+
+<--- Page Split --->
+
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+Dear Nature Communications Editorial Staff,  
+
+We appreciate the time and effort that you and the reviewers have dedicated to providing your valuable feedback on our manuscript. We have incorporated changes to reflect the suggestions provided by all reviewers and have itemized a point- by- point response to the additional comments and suggestions.  
+
+## Reviewer #1 (Remarks to the Author):  
+
+The authors did a good job addressing my comments. I don't have any other critiques.  
+
+We appreciate the positive feedback. We again thank the reviewers for their insightful comments which have helped further improve our manuscript.  
+
+## Reviewer #4 (Remarks to the Author):  
+
+In this revised version the authors properly addressed the most relevant critiques raised by the reviewer, resulting in a significantly improved manuscript.  
+
+My only minor concern relates to the lack of clarity of graph legends: in detail, while it is clear that +Epo and +BB indicate samples treated with each single agent, it is not immediate to understand what \(+ / +\) and - /- mean. I would suggest to include a description in the figure legend. Also, in Figure S14C the annotation of the x axis is missing.  
+
+This has been corrected.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__ceb012ed561eda660b4d01889149b0af06ed7f147b3579ba0b2d245783ecf797/supplementary_0_Peer Review file__ceb012ed561eda660b4d01889149b0af06ed7f147b3579ba0b2d245783ecf797_det.mmd

@@ -0,0 +1,81 @@
+<|ref|>title<|/ref|><|det|>[[72, 53, 295, 78]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[73, 95, 297, 118]]<|/det|>
+Peer Review File  
+
+<|ref|>title<|/ref|><|det|>[[73, 160, 787, 210]]<|/det|>
+# Engineering synthetic signaling receptors to enable erythropoietin-free erythropoiesis  
+
+<|ref|>text<|/ref|><|det|>[[73, 223, 394, 240]]<|/det|>
+Corresponding Author: Dr M. Cromer  
+
+<|ref|>text<|/ref|><|det|>[[72, 274, 875, 302]]<|/det|>
+This manuscript has been previously reviewed at another journal. This document only contains information relating to versions considered at Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[72, 313, 866, 329]]<|/det|>
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+<|ref|>text<|/ref|><|det|>[[73, 365, 145, 379]]<|/det|>
+Version 0:  
+
+<|ref|>text<|/ref|><|det|>[[73, 391, 220, 405]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 417, 161, 431]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 443, 660, 471]]<|/det|>
+(Remarks to the Author) The authors did a good job addressing my comments. I don't have any other critiques.  
+
+<|ref|>text<|/ref|><|det|>[[73, 482, 161, 496]]<|/det|>
+Reviewer #4  
+
+<|ref|>text<|/ref|><|det|>[[73, 508, 240, 522]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 520, 920, 590]]<|/det|>
+In this revised version the authors properly addressed the most relevant critiques raised by the reviewer, resulting in a significantly improved manuscript. My only minor concern relates to the lack of clarity of graph legends: in detail, while it is clear that \(+Epo\) and \(+BB\) indicate samples treated with each single agent, it is not immediate to understand what \(+ / +\) and \(- / -\) mean. I would suggest to include a description in the figure legend. Also, in Figure S14C the annotation of the \(x\) axis is missing.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 46, 916, 100]]<|/det|>
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+<|ref|>text<|/ref|><|det|>[[72, 100, 796, 113]]<|/det|>
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+<|ref|>text<|/ref|><|det|>[[72, 113, 910, 166]]<|/det|>
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+<|ref|>text<|/ref|><|det|>[[72, 166, 618, 180]]<|/det|>
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 137, 470, 153]]<|/det|>
+Dear Nature Communications Editorial Staff,  
+
+<|ref|>text<|/ref|><|det|>[[114, 172, 884, 257]]<|/det|>
+We appreciate the time and effort that you and the reviewers have dedicated to providing your valuable feedback on our manuscript. We have incorporated changes to reflect the suggestions provided by all reviewers and have itemized a point- by- point response to the additional comments and suggestions.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 310, 437, 327]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 346, 798, 364]]<|/det|>
+The authors did a good job addressing my comments. I don't have any other critiques.  
+
+<|ref|>text<|/ref|><|det|>[[115, 381, 884, 421]]<|/det|>
+We appreciate the positive feedback. We again thank the reviewers for their insightful comments which have helped further improve our manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 483, 437, 500]]<|/det|>
+## Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 519, 883, 559]]<|/det|>
+In this revised version the authors properly addressed the most relevant critiques raised by the reviewer, resulting in a significantly improved manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[114, 577, 884, 660]]<|/det|>
+My only minor concern relates to the lack of clarity of graph legends: in detail, while it is clear that +Epo and +BB indicate samples treated with each single agent, it is not immediate to understand what \(+ / +\) and - /- mean. I would suggest to include a description in the figure legend. Also, in Figure S14C the annotation of the x axis is missing.  
+
+<|ref|>text<|/ref|><|det|>[[115, 679, 315, 695]]<|/det|>
+This has been corrected.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__d65a0990f78c1f9e5e22fc4adc62b98b33a6430b24bdf15838130b71ceecd3aa/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review file__d65a0990f78c1f9e5e22fc4adc62b98b33a6430b24bdf15838130b71ceecd3aa/supplementary_0_Peer Review file__d65a0990f78c1f9e5e22fc4adc62b98b33a6430b24bdf15838130b71ceecd3aa.mmd

@@ -0,0 +1,405 @@
+
+# HuD impairs neuromuscular junctions and induces apoptosis in in vitro and in vivo ALS models  
+
+Corresponding Author: Professor Alessandro Rosa  
+
+This manuscript has been previously reviewed at another journal. This document only contains reviewer comments, rebuttal and decision letters for versions considered at Nature Communications.  
+
+Parts of this Peer Review File have been redacted as indicated to remove third- party material.  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+The manuscript by Silvestri et al. describes the effect of the gene HuD (ELAVL4) on ALS- related phenotypes in human iPSC- derived motor neurons and Drosophila models. It is shown that HuD overexpression causes a motor phenotype in flies, and neuromuscular (NMJ) defects and apoptosis in human motor neurons. Moreover, HuD knockdown alleviates FUS- induced phenotypes. Last but not least, HuD levels and function are suggested to be related to oxidative stress. Overall, this is an interesting study suggesting that HuD may underlie NMJ defects and apoptosis in FUS- ALS. However, some relevant questions remain unanswered and all the statistics should be done properly.  
+
+Major remarks:  
+
+- The title promises much more than actually presented in this manuscript. In line with previous publications from the same group, it is shown that HuD (ELAVL4) plays an important role in mutant FUS related ALS. Stating that this is the case for 'familial and sporadic ALS models' is a clear overstatement.  
+
+- The statistics used to determine significant differences in the iPSC experiments (Student's t-tests) is not correct at all as in most cases more than two groups are compared within one experiment. Extra attention should be devoted to those experiments where the measurements are not independent (and show no variation). This is particularly the case for Fig. 3B. As similar remark can be made for Fig. 4B. In addition, it should be indicated what type of error bars are always shown and the statistical numbers should be always replaced by stars. Not clear which statistical analysis was used in Fig. 7.  
+
+- The introduction doesn't help at all to understand the context of the current study. It also contains too many irrelevant details. It should place the current study in a broader perspective, rather than elaborating on previous findings of which it not always clear how these relate to the current study (and why it is relevant to know it). Overall, the introduction is overloaded with too many details, making it extremely difficult for the reader to understand the key point(s) the authors want to make.  
+
+- In supplementary figure 6, it is shown that HuD mRNA levels decrease after siRNA treatment. However, discrepancies between the abundance of mRNA and protein levels are sometimes observed. Therefore, a Western blot is needed to show the effect of the siRNAs on HuD at the protein level.  
+
+- Concerning the Drosophila experiments, it is not mentioned in the methods section whether the stocks had the same genetic background (e.g. stocks 28371, 38400...). If not, the flies should be crossed into the same control stock (e.g.w1118) for at least 6 generations, and then the experiments should be repeated to ensure that the effects observed by overexpression or RNAi are not due to differences in genetic backgrounds among the stocks. In addition, the number of flies and the number of experimental replicates should be increased to observe clearer effects (for example in Fig6B). Moreover, it is not clear whether similar levels of FUS expression were present in the WT and the mutant flies. Where these transgenes inserted into similar safe harbor loci and was it checked whether the different crosses didn't influence the expression level of WT or mutant FUS?  
+
+Minor remarks:  
+
+- The abstract should be rewritten. The major findings and conclusions should be summarized in a more concise way and also here the claims are much broader than what can actually be concluded from this study.
+
+<--- Page Split --->
+
+- The information on the genetics of ALS is incomplete and even misleading. It looks as if all genetic forms of ALS are related to mutations in RNA binding proteins, which is clearly not the case.  
+- The findings related to oxidative stress are mentioned in the abstract but not at all in the last paragraph of the introduction, summarizing the findings. Overall, these oxidative stress data (induced by arsenic) are interesting but not very well integrated into the story.  
+- The first paragraph of the result section doesn't contain relevant information. The fact that cytosine arabinoside was added can be mentioned in one sentence and should be described in the M&M.  
+- In Fig. 1D, it is not clear why the measurements of each condition are connected with lines.  
+- Visually, the amount of bungarotoxin positive dots looks very similar in the WT and in the mutant FUS condition (while it is clearly lower in the HuD condition).  
+- The length of the Y axes in Fig. 2C should be similar for day 14 and day 28.  
+- While overexpression of HuD has a negative effect on the P525L flies, this condition is missing in the first part of the manuscript. Which is the effect of HuD overexpression in the P525L iPSC line? Similar as in the flies?  
+- In the results section, when \(\alpha\) -amanitin is first mentioned, it is not explained what it is and how it is related to the studied proteins.  
+- The indications of the conditions in Fig. 6A are unclear. It is also not clear why the temperature is indicated as it is always \(25^{\circ}C\). Not so clear what the 15 s graph (panel D) adds to the story. Especially taking into account that some of the numbers are lower than after 10 s.  
+- The references should be carefully checked as in some cases important info is missing. Moreover, it is weird that 'De Santis et al.' is alphabetically classified at the S and 'de Winter et al.' at the W.  
+
+## Reviewer #2  
+
+(Remarks to the Author)  
+
+Silvestri and collaborators provide in this study a very comprehensive set of results converging to ascribe an important role of the RNA binding protein HuD in FUS ALS. This conclusion comes from convincing cell biology experiments, and are backed up in vivo in a drosophila model.  
+
+First the authors set up a co- culture system allowing long term study of MN/muscle co- cultures. They show that muscles are innervated by motor neurons and that this innervation leads to nicotine- dependent muscle contractions. They went on to confirm previous studies showing that FUS mutation in motor neurons is sufficient to lead to NMJ defects, and that this is phenocopied in wild type motor neurons overexpressing HuD. Importantly, either FUS mutation or HuD overexpression was sufficient to lead to motor neuron apoptosis. Conversely, knockdown of HuD or of RRN1 (a major HuD target) was sufficient to rescue mutant FUS mediated MN apoptosis. Furthermore, the dependency of FUS toxicity to HuD was confirmed in a Drosophila model using both gain and loss of function experiments. In an attempt to understand whether this pathway could be relevant to sporadic ALS, the authors then show that arsenic treatment was able to increase HuD levels, and that HuD and its targets might be upregulated in bulk RNA from various ALS patients group.  
+
+In all this is a very interesting manuscript, that shows convincing and converging evidence, and this might have consequences for the development of FUS- ALS therapies.  
+
+A major strength of the manuscript is their use of multiple orthogonal approaches (overexpression or knock- down, iPSC and fly models). There are some points that could be improved to solidify the conclusions of this already strong candidate for publication.  
+
+## Major points:  
+
+1) The characterization of HuD-overexpressing iPSC motor neurons is limited, and it cannot be excluded that at least part of the toxic effects could be due to a shift in motor neuron differentiation or identity as a consequence of HuD overexpression. Could the authors show that motor neurons of HuD overexpressing cells are similar in terms of differentiation or identity (eg via RNAseq)? It would also thicken their plot if HuD overexpressing wild type motor neurons would be transcriptionally more similar to mutant FUS motor neurons than to wild type motor neurons.  
+
+2) The analogy between iPSC models and fly models is unfortunately not followed up until NMJ defects. Is elav knockdown able to rescue NMJ defects of FUS overexpressing flies? Conversely is elav overexpression damaging NMJs in a similar manner than FUS overexpression?  
+
+3) The last figure on oxidative stress and HuD expression in sporadic ALS currently poorly articulated with the rest of the study, and. First, arsenic treatment is also leading to stress granule accumulation, which might confound the oxidative stress hypothesis of the authors. Other oxidative stressors could be tested. Second, there is no mechanistic link made here with FUS. Could the authors show if FUS knockdown alters arsenic-dependent HuD upregulation? That would strongly support their hypothesis. Also, could the authors relate HuD, RRN1 or GAP43 levels to FUS expression levels (or FUS introns 6/7 retention levels) in Figure 7G? It might be possible that FUS altered function is heterogeneous across patients and a stronger relation between HuD, its target and FUS could be highlighted through correlation analysis at individual patient level.  
+
+4) It is not clear what individual points represent in each of the figures. Are these individual experiments (ie biological replicates) or technical replicates? If this is biological replicates, the number of technical replicates per experiment should be indicated, and statistical analysis should be performed pairing per experiment. Furthermore, in Figure 6, there are 3 points (presumably 3 independent experiments), but the legend states \(n = 6 - 10\) (presumably the number of animals per experiment). This should be clarified.  
+
+5) In addition, all experiments in iPSCs appear to have been analyzed using Student's t test, while there are more than 2 groups (eg Figures 2C, 4D...). In some instances (eg Figure 4B), a two way ANOVA (with time and genotype) should be applied. In general, the statistics should be revised and use the appropriate tests for the experiments. As the results are robust, this would not likely change the statistical significance but is important for scientific robustness of the study.
+
+<--- Page Split --->
+
+## Reviewer #3  
+
+(Remarks to the Author)  
+
+Silvestri and colleagues use an iPSC- based NMJ model to assess contributions of an aggressive FUS mutation and an overexpression of ELAVL4. They argue that the in both of these experimental conditions there are NMJ defects and apoptosis, and they claim similarity of the two conditions. They evaluate the relationship between the genes further in flies, and find that elav knockdown rescues the FUS phenotype. Finally, they propose an oxidative stress phenotype.  
+
+Unfortunately, this reviewer finds several major and broad concerns that undermine the results and its conclusions.  
+
+First, there is lack of rigor in the description of techniques and details of analyses that spans all iPSC modeling studies. What are the replicates? How many fields were chosen for image analysis of contractions and IHC? How were these fields chosen? For contraction analysis, what was the timing of glutamate application and the effect. In terms of the "isogenic", did the control go through editing process to control for such effects? This seems hard then to add an additional true isogenic control for the prior isogenic pair. How were \(\%\) of BTX positive fibers calculated?  
+
+Second, there is insufficient biological evaluation, too many unwarranted assumptions, and too many unexplained phenomena. For example, one needs co- localized presynaptic markers to interpret BTX staining as synaptic. MN appearance is remarkable different in eg Fig 1, Fig 2, and Fig 4c. In Fig 4, MNs only contain lots of DAPI that is MAP- . Much of the CC- 3 staining is outside MAP2 staining. What are the dead cells? If there is indeed a neuronal toxicity brought out by muscle co- culture, how does one explain this mechanistically?  
+
+The fly experiment is the strongest result, although again what the 3 replicates means is not clarified.  
+
+The statistical analysis in the bioinformatic data does not appear to be adjusted for multiple comparisons.  
+
+Version 2:  
+
+Reviewer comments:  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+This is a manuscript that has been nicely reviewed, and the authors have adequately answered my previous comments. There are still few typos:  
+
+- Figure 1E: glutamate is spelled twice with 2 m (glutamate)- Figure 5A/B: "non targeting" is misspelled (non targeting)- fonts are non homogenous from one figure to another, as well as size of letterings- Figure 7G could be labelled with more self explanatory annotations.  
+
+Reviewer #3  
+
+(Remarks to the Author)  
+
+Unfortunately this reviewer does not find sufficient improvement in the fundamental stem cell experiments of the manuscript.  
+
+The staining of MyHC seems very weak compared to other examples of stem cell- derived muscle (Choi et al, 2016; Martins et al, 2020; Bar- Nur et al 2018). Furthermore, the bright MyHC spots are outside the muscle fibers, and do not seem consistent with patterns or intensity seen in other iPSC or primary muscle cell types. There is no demonstration of striations with sacromeric alpha- actininin (as in Osaki et al 2018; Pereira et al. 2021).  
+
+Images in figure 1D do not convincingly demonstrate co- colocalization of synaptophysin with bungarotoxin.  
+
+The description of the contractions is not sufficient and would require video or dynamic collection of images to be convincing. Furthermore, the time course of glutamate effects, which could be ionotropic or metabotropic, are not sufficiently resolved. How do we know that the duration of glutamate effect doesn't taper in 5- 10 minutes, as there is no vehicle control for the addition of curare?  
+
+Unfortunately, give the concerns about the fundamental representation of the bungarotoxin staining and its relation to presynaptic NMJ components, this reviewer cannot interpret the results in the subsequent iPSC figures. In particular, the overlap of bungarotoxin on the muscle fibers is not clear in Figure 2 and in Figure 3 it seems clear that BTX puncta are not on the muscle? Thus it's not clear that the bungarotoxin puncta are directly connected to the muscle or putative NMJ.  
+
+For the cell death experiments, one needs to understand more about the identity of the CC- 3 positive cells. For example, are these the muscle cells or not?
+
+<--- Page Split --->
+
+Version 3:  
+
+Reviewer comments:  
+
+Reviewer #3  
+
+(Remarks to the Author) I appreciate the efforts put into the manuscript but remain unpersuaded of the results.  
+
+I believe substantial issues remain in terms of the veracity of the model and analysis. The bugarotoxin staining and proximity to a pre- synaptic marker is not convincing. In particular, one needs to see the synapsin staining in larger areas of neurons and neurons alone (ie, not in co- culture). Similarly, the majority of the bugarotoxin is not sufficiently proximal to botulinum puncta. Indeed, bugarotoxin of the muscle alone (not in co- culture) should also be shown for comparison. Because the extent of the bugarotoxin staining outside the muscle is so extensive, it is hard to be confident that simply restricting analysis to the staining over the muscle is sufficient to eliminate artifacts. One would need to see specific (which is not evident in the synapsin image in Fig 2D) co- localization of synapsin and bugarotoxin - from the images provided, it may be that synapsin is present in that pattern throughout the neuron. One might also consider staining for bassoon, which may be more focal. As an aside, imaging is based on only 5- 6 fields per differentiation replicate - this seems quite low - given concerns about variation in such a model.  
+
+If these issues could be addressed, one could then assess the conclusions, the level of evidence supporting them, and their implications. These include studies mainly from a single pair of isogenic FUS mutations (WT, FUS P525L) and WT+Syn::HuD, without assessing the effect of HuD in the FUS mutation. Some similar results are achieved with introduction of FUS P525L and WT+syn::HuD from another line (WTSI), although the late introduction of this second set of cells makes it difficult to compare effects, particularly as quantification is not provided for many components of the WTSI analysis. Notably, there are no details given regarding these new (WTSI) clones, quality control, karyotype, etc.  
+
+The authors observe a toxicity to neurons elicited by co- culture with muscle (evident in Fig 4 D/E but quite notable in image from Fig S4). This toxicity is exacerbated by FUS mutation or HuD overexpression. Because only wild- type FUS muscle is used throughout the paper, the fundamental question of non- cell autonomous contribution raised in the Dupuis model - which would require mutation of FUS in the muscle, isn't addressed.  
+
+Furthermore, there is insufficient mechanistic insight, particularly as the apoptosis is occurring in both motor neurons and muscle. While the effects of arsenic treatment do include oxidative stress, the results from the last section and the inference from the human ALS subtypes seem premature, undeveloped, and insufficiently supported.  
+
+Other points:  
+
+When all control replicates are normalized to 1, the reader cannot appreciate how the within- group variance compares to the effect size.  
+
+Version 4:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author) NA
+
+<--- Page Split --->
+
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+We thank the reviewers for their suggestions and constructive criticism that have helped improving our paper. In response to their requests, we have performed additional experiments and analyses, which have been included in the new version of the manuscript, as detailed here.  
+
+Reviewer #1 (Remarks to the Author):  
+
+The manuscript by Silvestri et al. describes the effect of the gene HuD (ELAVL4) on ALS- related phenotypes in human iPSC- derived motor neurons and Drosophila models. It is shown that HuD overexpression causes a motor phenotype in flies, and neuromuscular (NMJ) defects and apoptosis in human motor neurons. Moreover, HuD knockdown alleviates FUS- induced phenotypes. Last but not least, HuD levels and function are suggested to be related to oxidative stress. Overall, this is an interesting study suggesting that HuD may underlie NMJ defects and apoptosis in FUS- ALS. However, some relevant questions remain unanswered and all the statistics should be done properly.  
+
+We thank this reviewer for his/her interest in our study and for suggesting important controls that have helped to strengthen the solidity of our conclusions, and changes in the text to ameliorate the readability of the paper.  
+
+## Major remarks:  
+
+1) The title promises much more than actually presented in this manuscript. In line with previous publications from the same group, it is shown that HuD (ELAVL4) plays an important role in mutant FUS related ALS. Stating that this is the case for 'familial and sporadic ALS models' is a clear overstatement.  
+
+We have changed the title as follows: "HuD (ELAVL4) gain-of-function impairs neuromuscular junctions and induces apoptosis in in vitro and in vivo models of amyotrophic lateral sclerosis"  
+
+2) The statistics used to determine significant differences in the iPSC experiments (Student's t-tests) is not correct at all as in most cases more than two groups are compared within one experiment. Extra attention should be devoted to those experiments where the measurements are not independent (and show no variation). This is particularly the case for Fig. 3B. As similar remark can be made for Fig. 4B. In addition, it should be indicated what type of error bars are always shown and the statistical numbers should be always replaced by stars. Not clear which statistical analysis was used in Fig. 7.  
+
+This point has been raised also by the other referees. As also suggested by reviewer 2 (point 5), we have applied an ordinary one- way ANOVA in cases when more than two groups are compared for one factor, and a two- way ANOVA in Figures 1E, 2C, 4B, 4D, 7C, 7F. For Figure 3B, we have used a Fisher's exact test with Bonferroni correction for multiple testing. For what concerns replacing statistical numbers with stars, we followed the guidelines of the journal, requesting to provide the actual values for \(p\) (even if it is not significant). For what concerns Figure 7, as also requested by reviewer 3, we have better clarified that the \(p\) values we have provided are "adjusted", meaning that they were corrected for multiple comparisons via the Benjamini- Hochberg procedure. We recognize that our prior explanation may have lacked clarity and we have added this detail in the "Statistics and reproducibility" section.
+
+<--- Page Split --->
+
+3) The introduction doesn't help at all to understand the context of the current study. It also contains too many irrelevant details. It should place the current study in a broader perspective, rather than elaborating on previous findings of which it not always clear how these relate to the current study (and why it is relevant to know it). Overall, the introduction is overloaded with too many details, making it extremely difficult for the reader to understand the key point(s) the authors want to make.  
+
+We have substantially simplified the introduction as suggested, by removing most of the details from previous findings.  
+
+4) In supplementary figure 6, it is shown that HuD mRNA levels decrease after siRNA treatment. However, discrepancies between the abundance of mRNA and protein levels are sometimes observed. Therefore, a Western blot is needed to show the effect of the siRNAs on HuD at the protein level.  
+
+The Western blot has been added to Supplementary Figure S6. Consistent with the quantification of mRNA levels, HuD protein levels in \(\mathrm{FUS}^{p525\mathrm{L}}\) cells MNs transfected with HuD siRNAs were decreased to levels comparable to untransfected \(\mathrm{FUS}^{WT}\) samples.  
+
+5) Concerning the Drosophila experiments, it is not mentioned in the methods section whether the stocks had the same genetic background (e.g., stocks 28371, 38400)? If not, the flies should be crossed into the same control stock (e.g.w1118) for at least 6 generations, and then the experiments should be repeated to ensure that the effects observed by overexpression or RNAi are not due to differences in genetic backgrounds among the stocks.  
+
+The reviewer made a valid point, and we fully agree. The elav RNAi (stock#28371) and the UAS- elav (Stock#38400) do not have the same genetic background. The elav RNAi is in the vermillion (v) background while the UAS elav is on the white (w1118) background. We obtained an RNAi line from the Vienna Drosophila Research Center and crossed it with the gal4 line in the same background. The results of the new experiments in this genetic background are shown in the revised Figure 6. The rescue of the motor phenotype of flies expressing either WT or mutant FUS by ELAV RNAi was confirmed (new Fig. 6B). Moreover, we added the analysis of climbing velocity, which was significantly ameliorated by elav RNAi in both FUS lines (new Fig. 6C). In addition to the abovementioned in vivo data, we found that overexpression of ELAV in the motor neurons produces a motor phenotype itself (Figure 6A). Since ELAV overexpression leads to a phenotype per se, this line is not useful for genetic interaction studies as the resulting phenotype(s) would be synergistic. Therefore, in the revised paper we are providing data showing elav RNAi KD in FUS lines and ELAV overexpression data itself in the revised figure 6.  
+
+In addition, the number of flies and the number of experimental replicates should be increased to observe clearer effects (for example in Fig6B).  
+
+We thank the reviewer for raising this question. Just to clarify, in both the control and experimental conditions, each fly of 6- 10 in numbers is considered one data point and we performed 3 trials or technical replicates which means our N=3 with 6- 10 flies each. We strongly believe that we used appropriate number of animals and sufficient replicates for drawing a conclusion.  
+
+Moreover, it is not clear whether similar levels of FUS expression were present in the WT and the mutant flies. Where these transgenes inserted into similar safe harbor loci and was it checked whether the different crosses didn't influence the expression level of WT or mutant FUS?
+
+<--- Page Split --->
+
+The fly lines used in the current manuscript were generated using site- specific integration approach to ensure they give a similar FUS expression. We already tested the FUS expression levels in these lines and published (Casci et al., 2019; Anderson et al., 2018). The UAS- FUS WT and the UAS- FUS P525L lines were generated through site specific insertion of the transgene at BestGene Inc. using the (attp2) integration vector and were previously described (Casci et al., 2019; Anderson et al., 2018). The combination lines FUS- WT; elavGS and FUS- P525L; elavGS used for the climbing assay are random insertion lines. In the revised paper we show that expression levels of FUS, checked by western blot, were not significantly different in WT and P525L flies (new Suppl. Fig. 8A).  
+
+## Minor remarks:  
+
+- The abstract should be rewritten. The major findings and conclusions should be summarized in a more concise way and also here the claims are much broader than what can actually be concluded from this study.  
+
+We have substantially changed the abstract following the reviewer's suggestions, in particular by summarizing findings and conclusions and dampening the claims.  
+
+- The information on the genetics of ALS is incomplete and even misleading. It looks as if all genetic forms of ALS are related to mutations in RNA binding proteins, which is clearly not the case.  
+
+We have changed the introduction to better explain this point.  
+
+- The findings related to oxidative stress are mentioned in the abstract but not at all in the last paragraph of the introduction, summarizing the findings. Overall, these oxidative stress data (induced by arsenic) are interesting but not very well integrated into the story.  
+
+We have implemented the last paragraph of the introduction with the description of the findings related to oxidative stress and sALS.  
+
+- The first paragraph of the result section doesn't contain relevant information. The fact that cytosine arabinoside was added can be mentioned in one sentence and should be described in the M&M.  
+
+We have moved this paragraph to M&M as suggested.  
+
+- In Fig. 1D, it is not clear why the measurements of each condition are connected with lines. For better clarity, we have changed this graph (Figure 1E in the revised paper) to represent that contractions were counted in 3 randomly selected fields (each represented by the points connected with a line) of 3 independent co-cultures. For each individual field, the number of contractions per minute was measured before treatment, 5' upon glutamate addition, and 5' upon further addition of tubocurarine. To take into account such multiple levels and the fact that each individual field has been analyzed in the 3 different conditions, we have performed a two-way repeated measures ANOVA.  
+
+- Visually, the amount of bungarotoxin positive dots looks very similar in the WT and in the mutant FUS condition (while it is clearly lower in the HuD condition). We have replaced this image with a more representative one.
+
+<--- Page Split --->
+
+- The length of the Y axes in Fig. 2C should be similar for day 14 and day 28. We have changed all the panels in which the same analysis had been done at multiple time points by grouping them in one graph (Fig. 2C, 4D, 4E).  
+
+- While overexpression of HuD has a negative effect on the P525L flies, this condition is missing in the first part of the manuscript. Which is the effect of HuD overexpression in the P525L iPSC line? Similar as in the flies?  
+
+As discussed above, overexpression of ELAV caused a motor dysfunction itself making this line not suitable for genetic interaction studies. Furthermore, we have removed the results of the combined elav & FUS overexpression in flies as it was in a different genetic background. The comment of the reviewer is anyway interesting. While we haven't explored here if HuD overexpression in a FUS mutant iPSC line could exacerbate the detrimental effects of the P525L mutation, in our previous work we have shown that co-expression of mutant FUS and HuD in Hela caused a marked increase of HuD levels in cytoplasmic speckles, which were stiff structures with peculiar viscoelastic characteristics (De Santis et al., 2019).  
+
+- In the results section, when \(\alpha\) -amanitin is first mentioned, it is not explained what it is and how it is related to the studied proteins.  
+
+We have explained in the text that \(\alpha\) -amanitin is a potent toxin that selectively inhibits RNA polymerase II and was used here to investigate the consequences of preventing new transcription during oxidative stress treatments. This experiment allowed us to assess that oxidative stress-induced upregulation of HuD could be dependent on new transcription.  
+
+- The indications of the conditions in Fig. 6A are unclear. It is also not clear why the temperature is indicated as it is always \(25^{\circ}C\). Not so clear what the 15 s graph (panel D) adds to the story. Especially taking into account that some of the numbers are lower than after 10 s.  
+
+The flies were allowed to mate and propagate under the experimental condition of \(25^{\circ}C\) in temperature. The motor function phenotypes of all control and diseased conditions were all collected at \(25^{\circ}C\). As suggested, we have kept in the revised manuscript only the analysis at 10 s.  
+
+- The references should be carefully checked as in some cases important info is missing. Moreover, it is weird that 'De Santis et al.' is alphabetically classified at the S and 'de Winter et al.' at the W. References have been fixed  
+
+Reviewer #2 (Remarks to the Author):  
+
+Silvestri and collaborators provide in this study a very comprehensive set of results converging to ascribe an important role of the RNA binding protein HuD in FUS ALS. This conclusion comes from convincing cell biology experiments, and are backed up in vivo in a drosophila model.  
+
+First the authors set up a co- culture system allowing long term study of MN/muscle co- cultures. They show that muscles are innervated by motor neurons and that this innervation leads to nicotine- dependent muscle contractions. They went on to confirm previous studies showing that FUS mutation in motor neurons is sufficient to lead to NMJ defects, and that this is phenocopied in wild type motor neurons overexpressing HuD. Importantly, either FUS mutation or HuD overexpression was sufficient to lead to motor neuron apoptosis. Conversely, knockdown of HuD
+
+<--- Page Split --->
+
+or of NRN1 (a major HuD target) was sufficient to rescue mutant FUS mediated MN apoptosis. Furthermore, the dependency of FUS toxicity to HuD was confirmed in a Drosophila model using both gain and loss of function experiments. In an attempt to understand whether this pathway could be relevant to sporadic ALS, the authors then show that arsenite treatment was able to increase HuD levels, and that HuD and its targets might be upregulated in bulk RNA from various ALS patients group.  
+
+In all this is a very interesting manuscript, that shows convincing and converging evidence, and this might have consequences for the development of FUS- ALS therapies.  
+
+A major strength of the manuscript is their use of multiple orthogonal approaches (overexpression or knock- down, iPSC and fly models). There are some points that could be improved to solidify the conclusions of this already strong candidate for publication.  
+
+We are grateful to this reviewer for considering our paper as a strong candidate for publication and for his/her suggestions of important experiments and analyses, whose results have been included in the revised version.  
+
+## Major points:  
+
+1) The characterization of HuD-overexpressing iPSC motor neurons is limited, and it cannot be excluded that at least part of the toxic effects could be due to a shift in motor neuron differentiation or identity as a consequence of HuD overexpression. Could the authors show that motor neurons of HuD overexpressing cells are similar in terms of differentiation or identity (eg via RNAseq)? It would also thicken their plot if HuD overexpressing wild type motor neurons would be transcriptionally more similar to mutant FUS motor neurons than to wild type motor neurons.  
+
+Our recent publication (Garone et al., 2022) demonstrated that increased HuD levels in MNs indeed mimic the effects of ALS mutant FUS. We investigated the gene expression of FUS<sup>WT</sup>, FUSP<sup>525L</sup>, and SYN1::HuD MNs by leveraging RNA profiling using a digital colour-coded molecular barcoding (Nanostring). The gene expression analysis revealed that SYN1::HuD MNs show a similar expression profile to FUSP<sup>525L</sup> MNs in both soma and neurite compartments. We focused on targets found altered in FUS<sup>525L</sup> MNs (De Santis et al., 2019) and involved in neurodevelopment, cytoskeleton and synapses via a mechanism previously demonstrated in Garone et al., 2021. As a result, such previous work showed that SYN1::HuD MNs were similar in differentiation and identity (gene expression analysis on neurodevelopment targets) but also suggested that HuD plays a critical role in ALS neuromuscular junctions (gene expression analysis on cytoskeleton and synapse targets). This hypothesis has been experimentally addressed in the present work.  
+
+2) The analogy between iPSC models and fly models is unfortunately not followed up until NMJ defects. Is elav knockdown able to rescue NMJ defects of FUS overexpressing flies? Conversely is elav overexpression damaging NMJs in a similar manner than FUS overexpression? We thank the reviewer for this suggestion. In the revised paper we have addressed this point by dissecting third instar larvae and staining for NMJs markers. The results of this analysis are shown in the new Fig. 6D,E and Suppl. Fig. S8B,C panels. Upon elav knockdown, we found a significant shift towards mature NMJs (reduction of the % of satellite boutons and increase of the % of mature boutons) in flies overexpressing FUS-WT. However, we did not observe a similar effect when elav levels were reduced in FUS-P525L overexpressing flies. Moreover, we found a trend towards an increase of the % of satellite boutons and decrease of the % of mature boutons in elav overexpressing flies, even though in this case statistical significance was not
+
+<--- Page Split --->
+
+reached. Taken together, these results suggest that in the in vivo model the mechanisms underlying the motor phenotype upon elav perturbation in neurons cannot be solely explained by the NMJ defect. A comment on this has been added to the DISCUSSION.  
+
+3) The last figure on oxidative stress and HuD expression in sporadic ALS currently poorly articulated with the rest of the study, and. First, arsenite treatment is also leading to stress granule accumulation, which might confound the oxidative stress hypothesis of the authors. Other oxidative stressors could be tested. Second, there is no mechanistic link made here with FUS. Could the authors show if FUS knockdown alters arsenite-dependent HuD upregulation? That would strongly support their hypothesis. Also, could the authors relate HuD, NRN1 or GAP43 levels to FUS expression levels (or FUS introns 6/7 retention levels) in Figure 7G? It might be possible that FUS altered function is heterogeneous across patients and a stronger relation between HuD, its target and FUS could be highlighted through correlation analysis at individual patient level. As also mentioned in the text, others have already shown HuD upregulation upon stress with hydrogen peroxide, \(\mathsf{H}_2\mathsf{O}_2\) (Dell'Orco et al., 2021). As correctly pointed out by the referee, this evidence is relevant because, as opposed to arsenite, \(\mathsf{H}_2\mathsf{O}_2\) - induced oxidative stress occurs by elF2alpha-independent mechanisms (Emara et al., 2012). In depth characterization of the molecular mechanisms leading to HuD increase during stress is beyond the scope of this paper and will be pursued in future work. To this regard, the suggestion of the referee on a mechanistic link with FUS is intriguing, but unlikely. We have previously shown that the regulation of HuD levels by FUS occurs at the post-transcriptional level via miR-375 and 3'UTR binding (De Santis et al., 2017; De Santis et al., 2019; Garone et al., 2021). Instead, here we show that the effect of arsenite stress was abolished upon treatment with alpha-amanitin (inhibition of new transcription). However, as suggested by the reviewer, we have analyzed the correlation between levels of HuD (and its targets) and FUS in sporadic ALS patients. This analysis has been performed as suggested at individual patient level for all groups, including the controls. The correlation matrix reporting the Spearman's correlation coefficients calculated between the expression levels (FPKM) of FUS and those of HuD, NRN1 and GAP43 in the samples from the NYGC consortium is shown in the new Figure S9B. The correlation values were calculated across various groups: OND, CTR, ALS_Glia, ALS_Ox, ALS_TE, and CTR+ALS (including CTR and all ALS groups). The results suggest that there is no correlation between FUS and HuD expression levels in OND and ALS_Ox groups, low positive correlation in CTR and ALS_TE groups, and low negative correlation in the ALS_Glia group. For what concerns NRN1 and GAP43, we observed positive correlation with FUS in all groups except ALS_Glia. Finally, when all groups are merged (CTR+ALS), there is no correlation for FUS and HuD and low positive correlation between FUS and NRN1 and FUS and GAP43. These results suggest that altered levels of FUS are unlikely the reason why HuD is upregulated in sporadic ALS patients with an oxidative stress signature. More in general, while altered FUS activities in FUS mutant models lead to HuD upregulation (De Santis et al., 2017; De Santis et al., 2019; Garone et al., 2021) there is no obvious correlation between FUS and HuD levels in non-FUS ALS sporadic patients and controls.  
+
+We have better explained in the DISCUSSION that our previous and present results point to two independent mechanisms underlying HuD upregulation in ALS: one is FUS- dependent and can be observed in FUS ALS models and patients; the second could be oxidative- stress dependent and can be reproduced by inducing oxidative stress in in vitro models.  
+
+4) It is not clear what individual points represent in each of the figures. Are these individual experiments (ie biological replicates) or technical replicates? If this is biological replicates, the
+
+<--- Page Split --->
+
+number of technical replicates per experiment should be indicated, and statistical analysis should be performed pairing per experiment.  
+
+A similar point has been raised by Reviewer #3. Unfortunately, in studies based on human iPSCs, the definition of what can be considered a bona fide biological replicate is still debated (see for example: Chan & Teo, Stem Cells 2020; DOI: 10.1002/stem.3237; section 5.4.2 of the ISSCR "Standards for Human Stem Cell Use in Research" document, https://www.isscr.org/standards- document/reporting). Prompted by the comments of Reviewers 2 and 3, in the revised version of the paper we have decided to use the term "biological" to indicate replicates in the strictest way, i.e. to define the two sets of human iPSC lines derived from two different individuals: the WT- I- based set (Lenzi et al., 2015; Garone et al., 2021) and the set based on the EBiSC line WTSii004- A (this paper). Within each line, each experiment has been performed on individual batches of differentiated iPSCs. For example, for image analyses of Fig. 2- 5 and Suppl. Fig. S3- S4, the experiment was performed on 3 batches of differentiated cells and a number of randomly selected fields per replicate indicated in the legends. Each dot in the graphs represents the average of the fields for each replicate.  
+
+The definition of what is considered here as biological replicates is provided in the MATERIALS AND METHODS section, and the information on the number of differentiated batches and fields within each individual batch used for the analyses is provided in each figure legend.  
+
+Furthermore, in Figure 6, there are 3 points (presumably 3 independent experiments), but the legend states \(n = 6 - 10\) (presumably the number of animals per experiment). This should be clarified. In each of the control and experimental conditions each fly of 6- 10 in numbers are considered one data point and we performed the experiments in three trials which make our \(N = 3\) , with 6- 10 flies. This makes our total flies to \(\sim 30\) flies. Also kindly check the response to the first reviewer.  
+
+5) In addition, all experiments in iPSCs appear to have been analyzed using Student's t test, while there are more than 2 groups (eg Figures 2C, 4D...). In some instances (eg Figure 4B), a two way ANOVA (with time and genotype) should be applied. In general, the statistics should be revised and use the appropriate tests for the experiments. As the results are robust, this would not likely change the statistical significance but is important for scientific robustness of the study.  
+
+This point has been raised also by reviewer #1. As suggested, we have now applied an ordinary one- way ANOVA in cases when more than two groups are compared for one factor, and a two- way ANOVA (with time and genotype considered two factors) in Figures 2C, 4B, and 4D. In figure 1E we have performed a two- way repeated measures ANOVA to take into account multiple variables (treatment and replicate) and the fact that, within each replicate, each individual field has been analyzed in the 3 different treatment conditions.  
+
+As predicted by this reviewer, overall the statistical significance has not substantially changed. However, this suggestion allowed us to drive interesting conclusions from the results of Figure 4, i.e. that, longitudinally, a significant increase of dead cells from the initial time point (day 7) was observed for the \(\mathrm{FUS}^{WT}\) co- cultures only at the latest time point (day 28; possibly due to intrinsic fragility of iPSC- derived MNs and muscle cells in culture), while it was evident at earlier time points for \(\mathrm{FUS}^{P525L}\) (from day 14 onward) and \(\mathrm{FUS}^{WT} + \mathrm{HuD}\) (from day 21 onward) (Fig. 4B). Moreover, from day 14 to day 28, the number of apoptotic cells significantly increased in the co- cultures made by \(\mathrm{FUS}^{P525L}\) and \(\mathrm{FUS}^{WT} + \mathrm{HuD}\) , but not \(\mathrm{FUS}^{WT}\) , MNs (Fig. 4E). These observations, in line with our conclusions, have been included in the RESULTS.
+
+<--- Page Split --->
+
+Reviewer #3 (Remarks to the Author):  
+
+Silvestri and colleagues use an iPSC- based NMJ model to assess contributions of an aggressive FUS mutation and an overexpression of ELAVL4. They argue that the in both of these experimental conditions there are NMJ defects and apoptosis, and they claim similarity of the two conditions. They evaluate the relationship between the genes further in flies, and find that elav knockdown rescues the FUS phenotype. Finally, they propose an oxidative stress phenotype.  
+
+Unfortunately, this reviewer finds several major and broad concerns that undermine the results and its conclusions.  
+
+We thank the reviewer for his/her constructive criticisms. We apologize for lack of clarity in important aspects, e.g. the definition of replicates, which has been better explained in the revised text. Moreover, we have improved the statistical analysis, as also requested by the other reviewers.  
+
+1) First, there is lack of rigor in the description of techniques and details of analyses that spans all iPSC modeling studies. What are the replicates? How many fields were chosen for image analysis of contractions and IHC? How were these fields chosen? For contraction analysis, what was the timing of glutamate application and the effect. In terms of the "isogenic", did the control go through editing process to control for such effects? This seems hard then to add an additional true isogenic control for the prior isogenic pair.  
+
+A similar point has been raised by Reviewer #2. Unfortunately, in studies based on human iPSCs, the definition of what can be considered a bona fide biological replicate is still debated (see for example: Chan & Teo, Stem Cells 2020; DOI: 10.1002/stem.3237; section 5.4.2 of the ISSCR "Standards for Human Stem Cell Use in Research" document, https://www.isscr.org/standards- document/reporting). Prompted by the comments of Reviewers 2 and 3, in the revised version of the paper we have decided to use the term "biological" to indicate replicates in the strictest way, i.e. to define the two sets of human iPSC lines derived from two different individuals: the WT- I- based set (Lenzi et al., 2015; Garone et al., 2021) and the set based on the EBiSC line WTSii004- A (this paper). In particular:  
+
+a) the WT-I line, from a healthy individual, described in Lenzi et al., 2015 (FUSwT) 
+b) a FUS mutant line generated from a) by gene editing as described in Lenzi et al., 2015 (FUSPS25L) 
+c) a line generated from a) by stably inserting a transgene overexpressing HuD (described in Garone et al. 2021) 
+d) the WTSii004-A line, from a healthy individual, obtained by EBiSC (here named WTSI-FUSwT) 
+e) a FUS mutant line generated from d) by gene editing (WTSI-FUSPS25L) 
+f) a line generated from d) by stably inserting a transgene overexpressing HuD (WTSI-FUSwT+HuD)  
+
+iPSC lines a) and b) are isogenic and went through the same editing process; the same for lines d) and e). Lines c) and f) have not been obtained by gene editing, but by piggyBac- mediated integration of a HuD transgene under the control of the human SYN1 promoter. We have better clarified this in the text (RESULTS and MATERIALS AND METHODS).  
+
+Within each line, each experiment has been performed on individual batches of differentiated iPSCs. For example, for image analyses of Fig. 2- 5 and Suppl. Fig. S3- S4, the experiment was performed on 3 batches of differentiated cells, and a number of randomly selected fields per
+
+<--- Page Split --->
+
+replicate was analyzed, as indicated in the legends. Each dot in the graphs represents the average of the fields for each replicate. In the contraction analysis of Figure 1E, we performed the experiment 3 times using different batches of NIL iPSCs and MB iPSCs. During the experiment, for each batch, 3 fields were randomly selected. The glutamate was added to the cells and the effect was measured after 5 minutes, then we added tubocurarine and measured contractions after 5 minutes, as specified in the revised text (MATERIALS AND METHODS and FIGURE LEGEND).  
+
+The definition of what is considered here as biological replicates is provided in the MATERIALS AND METHODS section, and the information on the number of differentiated batches and fields within each individual batch used for the analyses is provided in each figure legend.  
+
+How were \(\%\) of BTX positive fibers calculated?  
+
+As specified in the revised METHODS section, MyHC+ muscle fibers have been identified and counted from multiple randomly selected fields for each biological replicate. In the same fields, the total number of muscle fibers showing BTX signal has been assessed to calculate the \(\%\) .  
+
+2) Second, there is insufficient biological evaluation, too many unwarranted assumptions, and too many unexplained phenomena. For example, one needs co-localized presynaptic markers to interpret BTX staining as synaptic. MN appearance is remarkable different in eg Fig 1, Fig 2, and Fig 4c. In Fig 4, MNs only contain lots of DAPI that is MAP-. Much of the CC-3 staining is outside MAP2 staining. What are the dead cells? If there is indeed a neuronal toxicity brought out by muscle co- culture, how does one explain this mechanistically?  
+
+In order to validate BTX staining as synaptic, we have performed a triple staining that included a presynaptic marker (SYNAPTOPHYSIN), together with BTX and MyHC. Representative images are included in the new Figure 1D and show close proximity of the BTX and SYNAPTOPHYSIN signals. For what concerns the different appearance of MNs, the appearance of the MNs in Fig1 and 2 might seem different because of the different magnification. Fig.4C shows a staining with MAP2, which was used instead of TUJ1 in the co-staining with CC-3 due to antibody compatibility. We realized that the MAP2 staining was weaker in the original images of Figure 4C (single panels in supplementary Fig. S3), so we have replaced the "MNs only" panels of this figure with more representative images. As correctly noticed by the reviewer, much of the CC-3 signal is outside MAP2 staining in the co- cultures panels ("MNs & FUS<sup>WT</sup> SkMCs"). This is due to the fact that not only MNs, but also muscle cells die by apoptosis in these conditions. When cultured alone, a basal level of muscle cell death is always observed over time in our experiments, possibly due to intrinsic fragility of iPSC-derived skeletal muscle in vitro. MNs are known to release trophic factors that support the survival and growth of nearby cells. Therefore, the presence of MNs in our co- cultures can enhance the survival of skeletal muscle cells and contribute to their overall health. In Figure 4C,D we show that the number of apoptotic cells (both MNs and muscle) is significantly increased in presence of FUS<sup>P525L</sup> and FUS<sup>WT</sup>+HuD MNs in the co- cultures (compared to FUS<sup>WT</sup> MNs). We hypothesize that this could be due to the fact that upon MNs death and/or failure to form mature NMJs, their support to muscle survival in vitro fails. We have better explained this in the revised DISCUSSION. For what concerns the CC-3 signal outside of MAP2 staining in MNs monocultures, despite our differentiation method is quite efficient in terms of the fraction of post-mitotic MNs over the total number of differentiated cells (usually >80% of Isl1+ and CHAT+ cells, see De santis et al., 2018; Garone et al., 2019), in long term cultures (beyond 2 weeks), a subpopulation of proliferating non neuronal cells - conceivably MAP2- negative undifferentiated progenitors - becomes increasingly evident.
+
+<--- Page Split --->
+
+4) The fly experiment is the strongest result, although again what the 3 replicates means is not clarified.  
+
+In each of the control and experimental conditions each fly of 6- 10 in numbers are considered one data point and we performed the experiments in three trials which make our \(N = 3\) , with 6- 10 flies. This makes our total flies to \(\sim 30\) flies. Note that in response to a request by the first reviewer we have repeated and confirmed the rescue experiments in a different genetic context and added the measure of climbing velocity (new Fig. 6B,C).  
+
+5) The statistical analysis in the bioinformatic data does not appear to be adjusted for multiple comparisons.  
+
+As indicated in the captions for Figure 7G and Supplementary Figure S8A, the p-values we have provided are "adjusted", meaning that they were corrected for multiple comparisons via the Benjamini-Hochberg procedure. We recognize that our prior explanation may have lacked clarity and we have added this detail in the "Statistics and reproducibility" section.
+
+<--- Page Split --->
+
+Reviewers' comments:  
+
+Reviewer #2 (Remarks to the Author):  
+
+This is a manuscript that has been nicely reviewed, and the authors have adequately answered my previous comments.  
+
+There are still few typos:  
+
+- Figure 1E: glutamate is spelled twice with 2 m (glutammate)- Figure 5A/B: "non targeting" is misspelled (non targeting)- fonts are non homogenous from one figure to another, as well as size of letterings- Figure 7G could be labelled with more self explanatory annotations.  
+
+Typos have been corrected and fonts have been homogenized. We have added the self- explanatory annotation of the patients' groups in Figure 7G.  
+
+Reviewer #3 (Remarks to the Author):  
+
+Unfortunately this reviewer does not find sufficient improvement in the fundamental stem cell experiments of the manuscript.  
+
+The staining of MyHC seems very weak compared to other examples of stem cell- derived muscle (Choi et al, 2016; Martins et al, 2020; Bar- Nur et al 2018). Furthermore, the bright MyHC spots are outside the muscle fibers, and do not seem consistent with patterns or intensity seen in other iPSC or primary muscle cell types.  
+
+If the reviewer is referring to the intensity of the signal in the images, this cannot be directly compared in different publications in which different microscopes, parameters for the acquisition, kind of primary/secondary antibodies, etc. are used. If the comment is instead on the percentage of cells that are positive for MyHC, we had assessed in our previous publication that the original protocol yielded around \(77\%\) of MyHC- positive cells (Lenzi et al., 2016; DOI 10.1016/j.scr.2016.06.003). Prompted by the comment of the referee we have quantified the percentage of MyHC- positive cells with the improved protocol described here (Fig. S1D) and assessed that it is \(86.5 \pm 4.6\%\) at day 9 and \(90.0 \pm 2.5\%\) at day 13, which is in line with the literature (e.g. \(63.6\% \pm 9.68\%\) in Choi et al, 2016). This result has been included in the revised version of the paper (Supplementary Fig. S1 legend). For what concerns the MyHC signal outside the muscle fibers, this is due to antibody binding to cell debris which are present in long term co- cultures.  
+
+There is no demonstration of striations with sarcomeric alpha- actininin (as in Osaki et al 2018; Pereira et. al. 2021).  
+
+Demonstration of striations with sarcomeric \(\alpha\) - actininin in skeletal muscle monocultures and in co- cultures is shown in the new Supplementary Fig. S2 (panels A and B). The pattern is very similar to what reported in the literature, and in particular by Osaki et al. (2018) - Figures redacted due to lack of 3rd Party Rights
+
+<--- Page Split --->
+
+Images in figure 1D do not convincingly demonstrate co- colocalization of synaptophysin with bungarotoxin.  
+
+The co- staining with \(\alpha\) - bungarotoxin ( \(\alpha\) - BTX; marking the postsynaptic side of the myoblast) and an antibody for the presynaptic protein synaptophysin (SYP; marking synaptic vesicles) is often used for characterization of NMJ formed by in vitro human iPSC- derived cultures. At a confocal microscope, these signals are not expected to be co- localized, being respectively post- and pre- synaptic, but located in proximity. The images in Figure 1D and new representative images shown in the new Supplementary Fig. S2C show indeed this situation, which is very similar to previous work by others,  
+
+Figures by others redacted due to lack of 3rd Party Rights
+
+<--- Page Split --->
+
+The description of the contractions is not sufficient and would require video or dynamic collection of images to be convincing. Furthermore, the time course of glutamate effects, which could be ionotropic or metabotropic, are not sufficiently resolved. How do we know that the duration of glutamate effect doesn't taper in 5- 10 minutes, as there is no vehicle control for the addition of curare?  
+
+Video recording of contractions is now included as a new supplementary material (Supplementary Video S1). Such spontaneous contractions were absent in presence of curare (Supplementary Video S2). It should be noted that in this case we did not measure contractions induced by glutamate, but spontaneous ones, to avoid any possible issue due to attenuation of glutamate effects over time.  
+
+Unfortunately, give the concerns about the fundamental representation of the bungarotoxin staining and its relation to presynaptic NMJ components, this reviewer cannot interpret the results in the subsequent iPSC figures. In particular, the overlap of bungarotoxin on the muscle fibers is not clear in Figure 2 and in Figure 3 it seems clear that BTX puncta are not on the muscle? Thus it's not clear that the bungarotoxin puncta are directly connected to the muscle or putative NMJ.  
+
+We agree with the reviewer that in some figures part of the BTX signal is outside the muscle fibers, sometimes staining cell debris. However, we would like to clarify that for the quantification of the \(\alpha\) BTX- positive fibers of Figure 2 (and elsewhere in the paper for such kind of analysis) we have exclusively taken into consideration the specific BTX signal localized in bona fide MyHC- positive muscle fibers. This is shown in the images below, representing magnification of Figure 2 panels.
+
+<--- Page Split --->
+![PLACEHOLDER_18_0]
+  
+
+For what concerns Figure 3, while it is clear that the \(\alpha\) BTX signal is overlapped with the MyHC signal in the panels relative to "Diffuse puncta", "Dense" and "Cluster", the concern of the reviewer about the "Aligned puncta" panels is probably due to the fact that in these images the MyHC staining is weaker at the periphery of the fiber. Here below we show a magnification of the same merged image in which we have increased the brightness of the MyHC staining. In this image the green \(\alpha\) BTX spots are clearly inside the fiber. We prefer to keep the original image in the paper since the MyHC signal in the image below is oversaturated at the center of the fiber.  
+
+![PLACEHOLDER_18_1]
+
+
+<--- Page Split --->
+
+For the cell death experiments, one needs to understand more about the identity of the CC- 3 positive cells. For example, are these the muscle cells or not?  
+
+This point is addressed in the discussion as follows:  
+
+"In the co- cultures, we observed a significant increase on cell death in both neural and muscle populations upon MN- restricted FUS mutation or HuD overexpression. Both in vivo and in vitro, MNs are recognized for their ability to release trophic factors that support the survival and growth of nearby cells. The presence of MNs in the co- culture can thus enhance the survival of skeletal muscle cells and contribute to their overall health. We hypothesize that upon increased FUS<sup>P525L</sup> and FUS<sup>WT</sup>+HuD MNs death and/or failure to form mature NMJs, their support to muscle survival in vitro fails, thus explaining the increased apoptosis of muscle fibers."
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__d65a0990f78c1f9e5e22fc4adc62b98b33a6430b24bdf15838130b71ceecd3aa/supplementary_0_Peer Review file__d65a0990f78c1f9e5e22fc4adc62b98b33a6430b24bdf15838130b71ceecd3aa_det.mmd

@@ -0,0 +1,557 @@
+<|ref|>title<|/ref|><|det|>[[72, 161, 920, 209]]<|/det|>
+# HuD impairs neuromuscular junctions and induces apoptosis in in vitro and in vivo ALS models  
+
+<|ref|>text<|/ref|><|det|>[[73, 224, 512, 240]]<|/det|>
+Corresponding Author: Professor Alessandro Rosa  
+
+<|ref|>text<|/ref|><|det|>[[72, 274, 918, 302]]<|/det|>
+This manuscript has been previously reviewed at another journal. This document only contains reviewer comments, rebuttal and decision letters for versions considered at Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[73, 314, 712, 328]]<|/det|>
+Parts of this Peer Review File have been redacted as indicated to remove third- party material.  
+
+<|ref|>text<|/ref|><|det|>[[72, 340, 866, 354]]<|/det|>
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+<|ref|>text<|/ref|><|det|>[[73, 392, 144, 405]]<|/det|>
+Version 1:  
+
+<|ref|>text<|/ref|><|det|>[[73, 418, 219, 431]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 444, 160, 457]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 470, 238, 482]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 483, 920, 562]]<|/det|>
+The manuscript by Silvestri et al. describes the effect of the gene HuD (ELAVL4) on ALS- related phenotypes in human iPSC- derived motor neurons and Drosophila models. It is shown that HuD overexpression causes a motor phenotype in flies, and neuromuscular (NMJ) defects and apoptosis in human motor neurons. Moreover, HuD knockdown alleviates FUS- induced phenotypes. Last but not least, HuD levels and function are suggested to be related to oxidative stress. Overall, this is an interesting study suggesting that HuD may underlie NMJ defects and apoptosis in FUS- ALS. However, some relevant questions remain unanswered and all the statistics should be done properly.  
+
+<|ref|>text<|/ref|><|det|>[[72, 576, 177, 588]]<|/det|>
+Major remarks:  
+
+<|ref|>text<|/ref|><|det|>[[72, 589, 914, 627]]<|/det|>
+- The title promises much more than actually presented in this manuscript. In line with previous publications from the same group, it is shown that HuD (ELAVL4) plays an important role in mutant FUS related ALS. Stating that this is the case for 'familial and sporadic ALS models' is a clear overstatement.  
+
+<|ref|>text<|/ref|><|det|>[[72, 627, 916, 691]]<|/det|>
+- The statistics used to determine significant differences in the iPSC experiments (Student's t-tests) is not correct at all as in most cases more than two groups are compared within one experiment. Extra attention should be devoted to those experiments where the measurements are not independent (and show no variation). This is particularly the case for Fig. 3B. As similar remark can be made for Fig. 4B. In addition, it should be indicated what type of error bars are always shown and the statistical numbers should be always replaced by stars. Not clear which statistical analysis was used in Fig. 7.  
+
+<|ref|>text<|/ref|><|det|>[[72, 691, 920, 744]]<|/det|>
+- The introduction doesn't help at all to understand the context of the current study. It also contains too many irrelevant details. It should place the current study in a broader perspective, rather than elaborating on previous findings of which it not always clear how these relate to the current study (and why it is relevant to know it). Overall, the introduction is overloaded with too many details, making it extremely difficult for the reader to understand the key point(s) the authors want to make.  
+
+<|ref|>text<|/ref|><|det|>[[72, 744, 916, 783]]<|/det|>
+- In supplementary figure 6, it is shown that HuD mRNA levels decrease after siRNA treatment. However, discrepancies between the abundance of mRNA and protein levels are sometimes observed. Therefore, a Western blot is needed to show the effect of the siRNAs on HuD at the protein level.  
+
+<|ref|>text<|/ref|><|det|>[[72, 783, 920, 886]]<|/det|>
+- Concerning the Drosophila experiments, it is not mentioned in the methods section whether the stocks had the same genetic background (e.g. stocks 28371, 38400...). If not, the flies should be crossed into the same control stock (e.g.w1118) for at least 6 generations, and then the experiments should be repeated to ensure that the effects observed by overexpression or RNAi are not due to differences in genetic backgrounds among the stocks. In addition, the number of flies and the number of experimental replicates should be increased to observe clearer effects (for example in Fig6B). Moreover, it is not clear whether similar levels of FUS expression were present in the WT and the mutant flies. Where these transgenes inserted into similar safe harbor loci and was it checked whether the different crosses didn't influence the expression level of WT or mutant FUS?  
+
+<|ref|>text<|/ref|><|det|>[[72, 900, 176, 911]]<|/det|>
+Minor remarks:  
+
+<|ref|>text<|/ref|><|det|>[[72, 912, 904, 938]]<|/det|>
+- The abstract should be rewritten. The major findings and conclusions should be summarized in a more concise way and also here the claims are much broader than what can actually be concluded from this study.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[70, 46, 920, 310]]<|/det|>
+- The information on the genetics of ALS is incomplete and even misleading. It looks as if all genetic forms of ALS are related to mutations in RNA binding proteins, which is clearly not the case.  
+- The findings related to oxidative stress are mentioned in the abstract but not at all in the last paragraph of the introduction, summarizing the findings. Overall, these oxidative stress data (induced by arsenic) are interesting but not very well integrated into the story.  
+- The first paragraph of the result section doesn't contain relevant information. The fact that cytosine arabinoside was added can be mentioned in one sentence and should be described in the M&M.  
+- In Fig. 1D, it is not clear why the measurements of each condition are connected with lines.  
+- Visually, the amount of bungarotoxin positive dots looks very similar in the WT and in the mutant FUS condition (while it is clearly lower in the HuD condition).  
+- The length of the Y axes in Fig. 2C should be similar for day 14 and day 28.  
+- While overexpression of HuD has a negative effect on the P525L flies, this condition is missing in the first part of the manuscript. Which is the effect of HuD overexpression in the P525L iPSC line? Similar as in the flies?  
+- In the results section, when \(\alpha\) -amanitin is first mentioned, it is not explained what it is and how it is related to the studied proteins.  
+- The indications of the conditions in Fig. 6A are unclear. It is also not clear why the temperature is indicated as it is always \(25^{\circ}C\). Not so clear what the 15 s graph (panel D) adds to the story. Especially taking into account that some of the numbers are lower than after 10 s.  
+- The references should be carefully checked as in some cases important info is missing. Moreover, it is weird that 'De Santis et al.' is alphabetically classified at the S and 'de Winter et al.' at the W.  
+
+<|ref|>sub_title<|/ref|><|det|>[[73, 333, 162, 346]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 358, 238, 371]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 371, 915, 411]]<|/det|>
+Silvestri and collaborators provide in this study a very comprehensive set of results converging to ascribe an important role of the RNA binding protein HuD in FUS ALS. This conclusion comes from convincing cell biology experiments, and are backed up in vivo in a drosophila model.  
+
+<|ref|>text<|/ref|><|det|>[[72, 410, 920, 540]]<|/det|>
+First the authors set up a co- culture system allowing long term study of MN/muscle co- cultures. They show that muscles are innervated by motor neurons and that this innervation leads to nicotine- dependent muscle contractions. They went on to confirm previous studies showing that FUS mutation in motor neurons is sufficient to lead to NMJ defects, and that this is phenocopied in wild type motor neurons overexpressing HuD. Importantly, either FUS mutation or HuD overexpression was sufficient to lead to motor neuron apoptosis. Conversely, knockdown of HuD or of RRN1 (a major HuD target) was sufficient to rescue mutant FUS mediated MN apoptosis. Furthermore, the dependency of FUS toxicity to HuD was confirmed in a Drosophila model using both gain and loss of function experiments. In an attempt to understand whether this pathway could be relevant to sporadic ALS, the authors then show that arsenic treatment was able to increase HuD levels, and that HuD and its targets might be upregulated in bulk RNA from various ALS patients group.  
+
+<|ref|>text<|/ref|><|det|>[[73, 538, 833, 565]]<|/det|>
+In all this is a very interesting manuscript, that shows convincing and converging evidence, and this might have consequences for the development of FUS- ALS therapies.  
+
+<|ref|>text<|/ref|><|det|>[[72, 554, 920, 595]]<|/det|>
+A major strength of the manuscript is their use of multiple orthogonal approaches (overexpression or knock- down, iPSC and fly models). There are some points that could be improved to solidify the conclusions of this already strong candidate for publication.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 607, 162, 620]]<|/det|>
+## Major points:  
+
+<|ref|>text<|/ref|><|det|>[[72, 620, 920, 686]]<|/det|>
+1) The characterization of HuD-overexpressing iPSC motor neurons is limited, and it cannot be excluded that at least part of the toxic effects could be due to a shift in motor neuron differentiation or identity as a consequence of HuD overexpression. Could the authors show that motor neurons of HuD overexpressing cells are similar in terms of differentiation or identity (eg via RNAseq)? It would also thicken their plot if HuD overexpressing wild type motor neurons would be transcriptionally more similar to mutant FUS motor neurons than to wild type motor neurons.  
+
+<|ref|>text<|/ref|><|det|>[[72, 685, 916, 725]]<|/det|>
+2) The analogy between iPSC models and fly models is unfortunately not followed up until NMJ defects. Is elav knockdown able to rescue NMJ defects of FUS overexpressing flies? Conversely is elav overexpression damaging NMJs in a similar manner than FUS overexpression?  
+
+<|ref|>text<|/ref|><|det|>[[72, 724, 914, 826]]<|/det|>
+3) The last figure on oxidative stress and HuD expression in sporadic ALS currently poorly articulated with the rest of the study, and. First, arsenic treatment is also leading to stress granule accumulation, which might confound the oxidative stress hypothesis of the authors. Other oxidative stressors could be tested. Second, there is no mechanistic link made here with FUS. Could the authors show if FUS knockdown alters arsenic-dependent HuD upregulation? That would strongly support their hypothesis. Also, could the authors relate HuD, RRN1 or GAP43 levels to FUS expression levels (or FUS introns 6/7 retention levels) in Figure 7G? It might be possible that FUS altered function is heterogeneous across patients and a stronger relation between HuD, its target and FUS could be highlighted through correlation analysis at individual patient level.  
+
+<|ref|>text<|/ref|><|det|>[[72, 826, 923, 892]]<|/det|>
+4) It is not clear what individual points represent in each of the figures. Are these individual experiments (ie biological replicates) or technical replicates? If this is biological replicates, the number of technical replicates per experiment should be indicated, and statistical analysis should be performed pairing per experiment. Furthermore, in Figure 6, there are 3 points (presumably 3 independent experiments), but the legend states \(n = 6 - 10\) (presumably the number of animals per experiment). This should be clarified.  
+
+<|ref|>text<|/ref|><|det|>[[72, 892, 899, 945]]<|/det|>
+5) In addition, all experiments in iPSCs appear to have been analyzed using Student's t test, while there are more than 2 groups (eg Figures 2C, 4D...). In some instances (eg Figure 4B), a two way ANOVA (with time and genotype) should be applied. In general, the statistics should be revised and use the appropriate tests for the experiments. As the results are robust, this would not likely change the statistical significance but is important for scientific robustness of the study.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[73, 74, 162, 87]]<|/det|>
+## Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[73, 100, 238, 112]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 113, 911, 166]]<|/det|>
+Silvestri and colleagues use an iPSC- based NMJ model to assess contributions of an aggressive FUS mutation and an overexpression of ELAVL4. They argue that the in both of these experimental conditions there are NMJ defects and apoptosis, and they claim similarity of the two conditions. They evaluate the relationship between the genes further in flies, and find that elav knockdown rescues the FUS phenotype. Finally, they propose an oxidative stress phenotype.  
+
+<|ref|>text<|/ref|><|det|>[[72, 177, 857, 191]]<|/det|>
+Unfortunately, this reviewer finds several major and broad concerns that undermine the results and its conclusions.  
+
+<|ref|>text<|/ref|><|det|>[[73, 203, 914, 269]]<|/det|>
+First, there is lack of rigor in the description of techniques and details of analyses that spans all iPSC modeling studies. What are the replicates? How many fields were chosen for image analysis of contractions and IHC? How were these fields chosen? For contraction analysis, what was the timing of glutamate application and the effect. In terms of the "isogenic", did the control go through editing process to control for such effects? This seems hard then to add an additional true isogenic control for the prior isogenic pair. How were \(\%\) of BTX positive fibers calculated?  
+
+<|ref|>text<|/ref|><|det|>[[73, 281, 916, 346]]<|/det|>
+Second, there is insufficient biological evaluation, too many unwarranted assumptions, and too many unexplained phenomena. For example, one needs co- localized presynaptic markers to interpret BTX staining as synaptic. MN appearance is remarkable different in eg Fig 1, Fig 2, and Fig 4c. In Fig 4, MNs only contain lots of DAPI that is MAP- . Much of the CC- 3 staining is outside MAP2 staining. What are the dead cells? If there is indeed a neuronal toxicity brought out by muscle co- culture, how does one explain this mechanistically?  
+
+<|ref|>text<|/ref|><|det|>[[73, 358, 760, 373]]<|/det|>
+The fly experiment is the strongest result, although again what the 3 replicates means is not clarified.  
+
+<|ref|>text<|/ref|><|det|>[[73, 384, 789, 399]]<|/det|>
+The statistical analysis in the bioinformatic data does not appear to be adjusted for multiple comparisons.  
+
+<|ref|>text<|/ref|><|det|>[[73, 437, 145, 450]]<|/det|>
+Version 2:  
+
+<|ref|>text<|/ref|><|det|>[[73, 463, 219, 476]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 489, 162, 501]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 515, 237, 528]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 528, 890, 555]]<|/det|>
+This is a manuscript that has been nicely reviewed, and the authors have adequately answered my previous comments. There are still few typos:  
+
+<|ref|>text<|/ref|><|det|>[[73, 555, 639, 606]]<|/det|>
+- Figure 1E: glutamate is spelled twice with 2 m (glutamate)- Figure 5A/B: "non targeting" is misspelled (non targeting)- fonts are non homogenous from one figure to another, as well as size of letterings- Figure 7G could be labelled with more self explanatory annotations.  
+
+<|ref|>text<|/ref|><|det|>[[73, 633, 162, 646]]<|/det|>
+Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[73, 659, 238, 672]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 672, 916, 685]]<|/det|>
+Unfortunately this reviewer does not find sufficient improvement in the fundamental stem cell experiments of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[73, 697, 914, 750]]<|/det|>
+The staining of MyHC seems very weak compared to other examples of stem cell- derived muscle (Choi et al, 2016; Martins et al, 2020; Bar- Nur et al 2018). Furthermore, the bright MyHC spots are outside the muscle fibers, and do not seem consistent with patterns or intensity seen in other iPSC or primary muscle cell types. There is no demonstration of striations with sacromeric alpha- actininin (as in Osaki et al 2018; Pereira et al. 2021).  
+
+<|ref|>text<|/ref|><|det|>[[73, 761, 803, 775]]<|/det|>
+Images in figure 1D do not convincingly demonstrate co- colocalization of synaptophysin with bungarotoxin.  
+
+<|ref|>text<|/ref|><|det|>[[73, 788, 916, 839]]<|/det|>
+The description of the contractions is not sufficient and would require video or dynamic collection of images to be convincing. Furthermore, the time course of glutamate effects, which could be ionotropic or metabotropic, are not sufficiently resolved. How do we know that the duration of glutamate effect doesn't taper in 5- 10 minutes, as there is no vehicle control for the addition of curare?  
+
+<|ref|>text<|/ref|><|det|>[[73, 852, 911, 905]]<|/det|>
+Unfortunately, give the concerns about the fundamental representation of the bungarotoxin staining and its relation to presynaptic NMJ components, this reviewer cannot interpret the results in the subsequent iPSC figures. In particular, the overlap of bungarotoxin on the muscle fibers is not clear in Figure 2 and in Figure 3 it seems clear that BTX puncta are not on the muscle? Thus it's not clear that the bungarotoxin puncta are directly connected to the muscle or putative NMJ.  
+
+<|ref|>text<|/ref|><|det|>[[70, 917, 920, 944]]<|/det|>
+For the cell death experiments, one needs to understand more about the identity of the CC- 3 positive cells. For example, are these the muscle cells or not?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[73, 112, 145, 125]]<|/det|>
+Version 3:  
+
+<|ref|>text<|/ref|><|det|>[[73, 138, 220, 152]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 164, 163, 178]]<|/det|>
+Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[73, 190, 650, 218]]<|/det|>
+(Remarks to the Author) I appreciate the efforts put into the manuscript but remain unpersuaded of the results.  
+
+<|ref|>text<|/ref|><|det|>[[72, 229, 923, 360]]<|/det|>
+I believe substantial issues remain in terms of the veracity of the model and analysis. The bugarotoxin staining and proximity to a pre- synaptic marker is not convincing. In particular, one needs to see the synapsin staining in larger areas of neurons and neurons alone (ie, not in co- culture). Similarly, the majority of the bugarotoxin is not sufficiently proximal to botulinum puncta. Indeed, bugarotoxin of the muscle alone (not in co- culture) should also be shown for comparison. Because the extent of the bugarotoxin staining outside the muscle is so extensive, it is hard to be confident that simply restricting analysis to the staining over the muscle is sufficient to eliminate artifacts. One would need to see specific (which is not evident in the synapsin image in Fig 2D) co- localization of synapsin and bugarotoxin - from the images provided, it may be that synapsin is present in that pattern throughout the neuron. One might also consider staining for bassoon, which may be more focal. As an aside, imaging is based on only 5- 6 fields per differentiation replicate - this seems quite low - given concerns about variation in such a model.  
+
+<|ref|>text<|/ref|><|det|>[[72, 371, 920, 452]]<|/det|>
+If these issues could be addressed, one could then assess the conclusions, the level of evidence supporting them, and their implications. These include studies mainly from a single pair of isogenic FUS mutations (WT, FUS P525L) and WT+Syn::HuD, without assessing the effect of HuD in the FUS mutation. Some similar results are achieved with introduction of FUS P525L and WT+syn::HuD from another line (WTSI), although the late introduction of this second set of cells makes it difficult to compare effects, particularly as quantification is not provided for many components of the WTSI analysis. Notably, there are no details given regarding these new (WTSI) clones, quality control, karyotype, etc.  
+
+<|ref|>text<|/ref|><|det|>[[72, 463, 915, 518]]<|/det|>
+The authors observe a toxicity to neurons elicited by co- culture with muscle (evident in Fig 4 D/E but quite notable in image from Fig S4). This toxicity is exacerbated by FUS mutation or HuD overexpression. Because only wild- type FUS muscle is used throughout the paper, the fundamental question of non- cell autonomous contribution raised in the Dupuis model - which would require mutation of FUS in the muscle, isn't addressed.  
+
+<|ref|>text<|/ref|><|det|>[[72, 529, 914, 570]]<|/det|>
+Furthermore, there is insufficient mechanistic insight, particularly as the apoptosis is occurring in both motor neurons and muscle. While the effects of arsenic treatment do include oxidative stress, the results from the last section and the inference from the human ALS subtypes seem premature, undeveloped, and insufficiently supported.  
+
+<|ref|>text<|/ref|><|det|>[[72, 582, 161, 595]]<|/det|>
+Other points:  
+
+<|ref|>text<|/ref|><|det|>[[70, 595, 920, 622]]<|/det|>
+When all control replicates are normalized to 1, the reader cannot appreciate how the within- group variance compares to the effect size.  
+
+<|ref|>text<|/ref|><|det|>[[72, 648, 144, 661]]<|/det|>
+Version 4:  
+
+<|ref|>text<|/ref|><|det|>[[72, 674, 220, 686]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[72, 699, 160, 712]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[72, 725, 238, 750]]<|/det|>
+(Remarks to the Author) NA
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 45, 916, 99]]<|/det|>
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.  
+
+<|ref|>text<|/ref|><|det|>[[72, 100, 797, 113]]<|/det|>
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.  
+
+<|ref|>text<|/ref|><|det|>[[72, 113, 911, 166]]<|/det|>
+The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+<|ref|>text<|/ref|><|det|>[[72, 166, 618, 180]]<|/det|>
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 100, 905, 152]]<|/det|>
+We thank the reviewers for their suggestions and constructive criticism that have helped improving our paper. In response to their requests, we have performed additional experiments and analyses, which have been included in the new version of the manuscript, as detailed here.  
+
+<|ref|>text<|/ref|><|det|>[[94, 187, 403, 203]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[92, 221, 896, 362]]<|/det|>
+The manuscript by Silvestri et al. describes the effect of the gene HuD (ELAVL4) on ALS- related phenotypes in human iPSC- derived motor neurons and Drosophila models. It is shown that HuD overexpression causes a motor phenotype in flies, and neuromuscular (NMJ) defects and apoptosis in human motor neurons. Moreover, HuD knockdown alleviates FUS- induced phenotypes. Last but not least, HuD levels and function are suggested to be related to oxidative stress. Overall, this is an interesting study suggesting that HuD may underlie NMJ defects and apoptosis in FUS- ALS. However, some relevant questions remain unanswered and all the statistics should be done properly.  
+
+<|ref|>text<|/ref|><|det|>[[92, 379, 881, 431]]<|/det|>
+We thank this reviewer for his/her interest in our study and for suggesting important controls that have helped to strengthen the solidity of our conclusions, and changes in the text to ameliorate the readability of the paper.  
+
+<|ref|>sub_title<|/ref|><|det|>[[93, 449, 220, 465]]<|/det|>
+## Major remarks:  
+
+<|ref|>text<|/ref|><|det|>[[92, 483, 888, 551]]<|/det|>
+1) The title promises much more than actually presented in this manuscript. In line with previous publications from the same group, it is shown that HuD (ELAVL4) plays an important role in mutant FUS related ALS. Stating that this is the case for 'familial and sporadic ALS models' is a clear overstatement.  
+
+<|ref|>text<|/ref|><|det|>[[92, 552, 884, 587]]<|/det|>
+We have changed the title as follows: "HuD (ELAVL4) gain-of-function impairs neuromuscular junctions and induces apoptosis in in vitro and in vivo models of amyotrophic lateral sclerosis"  
+
+<|ref|>text<|/ref|><|det|>[[92, 605, 880, 726]]<|/det|>
+2) The statistics used to determine significant differences in the iPSC experiments (Student's t-tests) is not correct at all as in most cases more than two groups are compared within one experiment. Extra attention should be devoted to those experiments where the measurements are not independent (and show no variation). This is particularly the case for Fig. 3B. As similar remark can be made for Fig. 4B. In addition, it should be indicated what type of error bars are always shown and the statistical numbers should be always replaced by stars. Not clear which statistical analysis was used in Fig. 7.  
+
+<|ref|>text<|/ref|><|det|>[[91, 727, 901, 901]]<|/det|>
+This point has been raised also by the other referees. As also suggested by reviewer 2 (point 5), we have applied an ordinary one- way ANOVA in cases when more than two groups are compared for one factor, and a two- way ANOVA in Figures 1E, 2C, 4B, 4D, 7C, 7F. For Figure 3B, we have used a Fisher's exact test with Bonferroni correction for multiple testing. For what concerns replacing statistical numbers with stars, we followed the guidelines of the journal, requesting to provide the actual values for \(p\) (even if it is not significant). For what concerns Figure 7, as also requested by reviewer 3, we have better clarified that the \(p\) values we have provided are "adjusted", meaning that they were corrected for multiple comparisons via the Benjamini- Hochberg procedure. We recognize that our prior explanation may have lacked clarity and we have added this detail in the "Statistics and reproducibility" section.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 100, 888, 204]]<|/det|>
+3) The introduction doesn't help at all to understand the context of the current study. It also contains too many irrelevant details. It should place the current study in a broader perspective, rather than elaborating on previous findings of which it not always clear how these relate to the current study (and why it is relevant to know it). Overall, the introduction is overloaded with too many details, making it extremely difficult for the reader to understand the key point(s) the authors want to make.  
+
+<|ref|>text<|/ref|><|det|>[[92, 205, 895, 240]]<|/det|>
+We have substantially simplified the introduction as suggested, by removing most of the details from previous findings.  
+
+<|ref|>text<|/ref|><|det|>[[92, 257, 880, 326]]<|/det|>
+4) In supplementary figure 6, it is shown that HuD mRNA levels decrease after siRNA treatment. However, discrepancies between the abundance of mRNA and protein levels are sometimes observed. Therefore, a Western blot is needed to show the effect of the siRNAs on HuD at the protein level.  
+
+<|ref|>text<|/ref|><|det|>[[92, 327, 870, 379]]<|/det|>
+The Western blot has been added to Supplementary Figure S6. Consistent with the quantification of mRNA levels, HuD protein levels in \(\mathrm{FUS}^{p525\mathrm{L}}\) cells MNs transfected with HuD siRNAs were decreased to levels comparable to untransfected \(\mathrm{FUS}^{WT}\) samples.  
+
+<|ref|>text<|/ref|><|det|>[[92, 396, 900, 484]]<|/det|>
+5) Concerning the Drosophila experiments, it is not mentioned in the methods section whether the stocks had the same genetic background (e.g., stocks 28371, 38400)? If not, the flies should be crossed into the same control stock (e.g.w1118) for at least 6 generations, and then the experiments should be repeated to ensure that the effects observed by overexpression or RNAi are not due to differences in genetic backgrounds among the stocks.  
+
+<|ref|>text<|/ref|><|det|>[[91, 484, 901, 710]]<|/det|>
+The reviewer made a valid point, and we fully agree. The elav RNAi (stock#28371) and the UAS- elav (Stock#38400) do not have the same genetic background. The elav RNAi is in the vermillion (v) background while the UAS elav is on the white (w1118) background. We obtained an RNAi line from the Vienna Drosophila Research Center and crossed it with the gal4 line in the same background. The results of the new experiments in this genetic background are shown in the revised Figure 6. The rescue of the motor phenotype of flies expressing either WT or mutant FUS by ELAV RNAi was confirmed (new Fig. 6B). Moreover, we added the analysis of climbing velocity, which was significantly ameliorated by elav RNAi in both FUS lines (new Fig. 6C). In addition to the abovementioned in vivo data, we found that overexpression of ELAV in the motor neurons produces a motor phenotype itself (Figure 6A). Since ELAV overexpression leads to a phenotype per se, this line is not useful for genetic interaction studies as the resulting phenotype(s) would be synergistic. Therefore, in the revised paper we are providing data showing elav RNAi KD in FUS lines and ELAV overexpression data itself in the revised figure 6.  
+
+<|ref|>text<|/ref|><|det|>[[92, 727, 897, 761]]<|/det|>
+In addition, the number of flies and the number of experimental replicates should be increased to observe clearer effects (for example in Fig6B).  
+
+<|ref|>text<|/ref|><|det|>[[92, 761, 900, 849]]<|/det|>
+We thank the reviewer for raising this question. Just to clarify, in both the control and experimental conditions, each fly of 6- 10 in numbers is considered one data point and we performed 3 trials or technical replicates which means our N=3 with 6- 10 flies each. We strongly believe that we used appropriate number of animals and sufficient replicates for drawing a conclusion.  
+
+<|ref|>text<|/ref|><|det|>[[92, 866, 891, 918]]<|/det|>
+Moreover, it is not clear whether similar levels of FUS expression were present in the WT and the mutant flies. Where these transgenes inserted into similar safe harbor loci and was it checked whether the different crosses didn't influence the expression level of WT or mutant FUS?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 82, 904, 240]]<|/det|>
+The fly lines used in the current manuscript were generated using site- specific integration approach to ensure they give a similar FUS expression. We already tested the FUS expression levels in these lines and published (Casci et al., 2019; Anderson et al., 2018). The UAS- FUS WT and the UAS- FUS P525L lines were generated through site specific insertion of the transgene at BestGene Inc. using the (attp2) integration vector and were previously described (Casci et al., 2019; Anderson et al., 2018). The combination lines FUS- WT; elavGS and FUS- P525L; elavGS used for the climbing assay are random insertion lines. In the revised paper we show that expression levels of FUS, checked by western blot, were not significantly different in WT and P525L flies (new Suppl. Fig. 8A).  
+
+<|ref|>sub_title<|/ref|><|det|>[[93, 276, 220, 292]]<|/det|>
+## Minor remarks:  
+
+<|ref|>text<|/ref|><|det|>[[92, 309, 901, 360]]<|/det|>
+- The abstract should be rewritten. The major findings and conclusions should be summarized in a more concise way and also here the claims are much broader than what can actually be concluded from this study.  
+
+<|ref|>text<|/ref|><|det|>[[92, 361, 884, 396]]<|/det|>
+We have substantially changed the abstract following the reviewer's suggestions, in particular by summarizing findings and conclusions and dampening the claims.  
+
+<|ref|>text<|/ref|><|det|>[[92, 412, 866, 464]]<|/det|>
+- The information on the genetics of ALS is incomplete and even misleading. It looks as if all genetic forms of ALS are related to mutations in RNA binding proteins, which is clearly not the case.  
+
+<|ref|>text<|/ref|><|det|>[[92, 465, 618, 483]]<|/det|>
+We have changed the introduction to better explain this point.  
+
+<|ref|>text<|/ref|><|det|>[[92, 500, 866, 551]]<|/det|>
+- The findings related to oxidative stress are mentioned in the abstract but not at all in the last paragraph of the introduction, summarizing the findings. Overall, these oxidative stress data (induced by arsenic) are interesting but not very well integrated into the story.  
+
+<|ref|>text<|/ref|><|det|>[[92, 552, 836, 586]]<|/det|>
+We have implemented the last paragraph of the introduction with the description of the findings related to oxidative stress and sALS.  
+
+<|ref|>text<|/ref|><|det|>[[92, 604, 900, 656]]<|/det|>
+- The first paragraph of the result section doesn't contain relevant information. The fact that cytosine arabinoside was added can be mentioned in one sentence and should be described in the M&M.  
+
+<|ref|>text<|/ref|><|det|>[[92, 657, 549, 674]]<|/det|>
+We have moved this paragraph to M&M as suggested.  
+
+<|ref|>text<|/ref|><|det|>[[92, 691, 895, 831]]<|/det|>
+- In Fig. 1D, it is not clear why the measurements of each condition are connected with lines. For better clarity, we have changed this graph (Figure 1E in the revised paper) to represent that contractions were counted in 3 randomly selected fields (each represented by the points connected with a line) of 3 independent co-cultures. For each individual field, the number of contractions per minute was measured before treatment, 5' upon glutamate addition, and 5' upon further addition of tubocurarine. To take into account such multiple levels and the fact that each individual field has been analyzed in the 3 different conditions, we have performed a two-way repeated measures ANOVA.  
+
+<|ref|>text<|/ref|><|det|>[[92, 848, 901, 901]]<|/det|>
+- Visually, the amount of bungarotoxin positive dots looks very similar in the WT and in the mutant FUS condition (while it is clearly lower in the HuD condition). We have replaced this image with a more representative one.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 100, 860, 152]]<|/det|>
+- The length of the Y axes in Fig. 2C should be similar for day 14 and day 28. We have changed all the panels in which the same analysis had been done at multiple time points by grouping them in one graph (Fig. 2C, 4D, 4E).  
+
+<|ref|>text<|/ref|><|det|>[[92, 170, 895, 221]]<|/det|>
+- While overexpression of HuD has a negative effect on the P525L flies, this condition is missing in the first part of the manuscript. Which is the effect of HuD overexpression in the P525L iPSC line? Similar as in the flies?  
+
+<|ref|>text<|/ref|><|det|>[[92, 223, 900, 360]]<|/det|>
+As discussed above, overexpression of ELAV caused a motor dysfunction itself making this line not suitable for genetic interaction studies. Furthermore, we have removed the results of the combined elav & FUS overexpression in flies as it was in a different genetic background. The comment of the reviewer is anyway interesting. While we haven't explored here if HuD overexpression in a FUS mutant iPSC line could exacerbate the detrimental effects of the P525L mutation, in our previous work we have shown that co-expression of mutant FUS and HuD in Hela caused a marked increase of HuD levels in cytoplasmic speckles, which were stiff structures with peculiar viscoelastic characteristics (De Santis et al., 2019).  
+
+<|ref|>text<|/ref|><|det|>[[92, 379, 900, 412]]<|/det|>
+- In the results section, when \(\alpha\) -amanitin is first mentioned, it is not explained what it is and how it is related to the studied proteins.  
+
+<|ref|>text<|/ref|><|det|>[[92, 414, 864, 482]]<|/det|>
+We have explained in the text that \(\alpha\) -amanitin is a potent toxin that selectively inhibits RNA polymerase II and was used here to investigate the consequences of preventing new transcription during oxidative stress treatments. This experiment allowed us to assess that oxidative stress-induced upregulation of HuD could be dependent on new transcription.  
+
+<|ref|>text<|/ref|><|det|>[[92, 500, 901, 550]]<|/det|>
+- The indications of the conditions in Fig. 6A are unclear. It is also not clear why the temperature is indicated as it is always \(25^{\circ}C\). Not so clear what the 15 s graph (panel D) adds to the story. Especially taking into account that some of the numbers are lower than after 10 s.  
+
+<|ref|>text<|/ref|><|det|>[[92, 551, 901, 603]]<|/det|>
+The flies were allowed to mate and propagate under the experimental condition of \(25^{\circ}C\) in temperature. The motor function phenotypes of all control and diseased conditions were all collected at \(25^{\circ}C\). As suggested, we have kept in the revised manuscript only the analysis at 10 s.  
+
+<|ref|>text<|/ref|><|det|>[[92, 622, 903, 673]]<|/det|>
+- The references should be carefully checked as in some cases important info is missing. Moreover, it is weird that 'De Santis et al.' is alphabetically classified at the S and 'de Winter et al.' at the W. References have been fixed  
+
+<|ref|>text<|/ref|><|det|>[[93, 727, 404, 742]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[92, 761, 900, 820]]<|/det|>
+Silvestri and collaborators provide in this study a very comprehensive set of results converging to ascribe an important role of the RNA binding protein HuD in FUS ALS. This conclusion comes from convincing cell biology experiments, and are backed up in vivo in a drosophila model.  
+
+<|ref|>text<|/ref|><|det|>[[92, 820, 902, 918]]<|/det|>
+First the authors set up a co- culture system allowing long term study of MN/muscle co- cultures. They show that muscles are innervated by motor neurons and that this innervation leads to nicotine- dependent muscle contractions. They went on to confirm previous studies showing that FUS mutation in motor neurons is sufficient to lead to NMJ defects, and that this is phenocopied in wild type motor neurons overexpressing HuD. Importantly, either FUS mutation or HuD overexpression was sufficient to lead to motor neuron apoptosis. Conversely, knockdown of HuD
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 83, 888, 187]]<|/det|>
+or of NRN1 (a major HuD target) was sufficient to rescue mutant FUS mediated MN apoptosis. Furthermore, the dependency of FUS toxicity to HuD was confirmed in a Drosophila model using both gain and loss of function experiments. In an attempt to understand whether this pathway could be relevant to sporadic ALS, the authors then show that arsenite treatment was able to increase HuD levels, and that HuD and its targets might be upregulated in bulk RNA from various ALS patients group.  
+
+<|ref|>text<|/ref|><|det|>[[92, 188, 904, 222]]<|/det|>
+In all this is a very interesting manuscript, that shows convincing and converging evidence, and this might have consequences for the development of FUS- ALS therapies.  
+
+<|ref|>text<|/ref|><|det|>[[92, 223, 901, 274]]<|/det|>
+A major strength of the manuscript is their use of multiple orthogonal approaches (overexpression or knock- down, iPSC and fly models). There are some points that could be improved to solidify the conclusions of this already strong candidate for publication.  
+
+<|ref|>text<|/ref|><|det|>[[92, 291, 891, 343]]<|/det|>
+We are grateful to this reviewer for considering our paper as a strong candidate for publication and for his/her suggestions of important experiments and analyses, whose results have been included in the revised version.  
+
+<|ref|>sub_title<|/ref|><|det|>[[93, 362, 203, 378]]<|/det|>
+## Major points:  
+
+<|ref|>text<|/ref|><|det|>[[92, 396, 904, 500]]<|/det|>
+1) The characterization of HuD-overexpressing iPSC motor neurons is limited, and it cannot be excluded that at least part of the toxic effects could be due to a shift in motor neuron differentiation or identity as a consequence of HuD overexpression. Could the authors show that motor neurons of HuD overexpressing cells are similar in terms of differentiation or identity (eg via RNAseq)? It would also thicken their plot if HuD overexpressing wild type motor neurons would be transcriptionally more similar to mutant FUS motor neurons than to wild type motor neurons.  
+
+<|ref|>text<|/ref|><|det|>[[92, 501, 905, 708]]<|/det|>
+Our recent publication (Garone et al., 2022) demonstrated that increased HuD levels in MNs indeed mimic the effects of ALS mutant FUS. We investigated the gene expression of FUS<sup>WT</sup>, FUSP<sup>525L</sup>, and SYN1::HuD MNs by leveraging RNA profiling using a digital colour-coded molecular barcoding (Nanostring). The gene expression analysis revealed that SYN1::HuD MNs show a similar expression profile to FUSP<sup>525L</sup> MNs in both soma and neurite compartments. We focused on targets found altered in FUS<sup>525L</sup> MNs (De Santis et al., 2019) and involved in neurodevelopment, cytoskeleton and synapses via a mechanism previously demonstrated in Garone et al., 2021. As a result, such previous work showed that SYN1::HuD MNs were similar in differentiation and identity (gene expression analysis on neurodevelopment targets) but also suggested that HuD plays a critical role in ALS neuromuscular junctions (gene expression analysis on cytoskeleton and synapse targets). This hypothesis has been experimentally addressed in the present work.  
+
+<|ref|>text<|/ref|><|det|>[[92, 726, 896, 917]]<|/det|>
+2) The analogy between iPSC models and fly models is unfortunately not followed up until NMJ defects. Is elav knockdown able to rescue NMJ defects of FUS overexpressing flies? Conversely is elav overexpression damaging NMJs in a similar manner than FUS overexpression? We thank the reviewer for this suggestion. In the revised paper we have addressed this point by dissecting third instar larvae and staining for NMJs markers. The results of this analysis are shown in the new Fig. 6D,E and Suppl. Fig. S8B,C panels. Upon elav knockdown, we found a significant shift towards mature NMJs (reduction of the % of satellite boutons and increase of the % of mature boutons) in flies overexpressing FUS-WT. However, we did not observe a similar effect when elav levels were reduced in FUS-P525L overexpressing flies. Moreover, we found a trend towards an increase of the % of satellite boutons and decrease of the % of mature boutons in elav overexpressing flies, even though in this case statistical significance was not
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 83, 890, 135]]<|/det|>
+reached. Taken together, these results suggest that in the in vivo model the mechanisms underlying the motor phenotype upon elav perturbation in neurons cannot be solely explained by the NMJ defect. A comment on this has been added to the DISCUSSION.  
+
+<|ref|>text<|/ref|><|det|>[[90, 150, 904, 799]]<|/det|>
+3) The last figure on oxidative stress and HuD expression in sporadic ALS currently poorly articulated with the rest of the study, and. First, arsenite treatment is also leading to stress granule accumulation, which might confound the oxidative stress hypothesis of the authors. Other oxidative stressors could be tested. Second, there is no mechanistic link made here with FUS. Could the authors show if FUS knockdown alters arsenite-dependent HuD upregulation? That would strongly support their hypothesis. Also, could the authors relate HuD, NRN1 or GAP43 levels to FUS expression levels (or FUS introns 6/7 retention levels) in Figure 7G? It might be possible that FUS altered function is heterogeneous across patients and a stronger relation between HuD, its target and FUS could be highlighted through correlation analysis at individual patient level. As also mentioned in the text, others have already shown HuD upregulation upon stress with hydrogen peroxide, \(\mathsf{H}_2\mathsf{O}_2\) (Dell'Orco et al., 2021). As correctly pointed out by the referee, this evidence is relevant because, as opposed to arsenite, \(\mathsf{H}_2\mathsf{O}_2\) - induced oxidative stress occurs by elF2alpha-independent mechanisms (Emara et al., 2012). In depth characterization of the molecular mechanisms leading to HuD increase during stress is beyond the scope of this paper and will be pursued in future work. To this regard, the suggestion of the referee on a mechanistic link with FUS is intriguing, but unlikely. We have previously shown that the regulation of HuD levels by FUS occurs at the post-transcriptional level via miR-375 and 3'UTR binding (De Santis et al., 2017; De Santis et al., 2019; Garone et al., 2021). Instead, here we show that the effect of arsenite stress was abolished upon treatment with alpha-amanitin (inhibition of new transcription). However, as suggested by the reviewer, we have analyzed the correlation between levels of HuD (and its targets) and FUS in sporadic ALS patients. This analysis has been performed as suggested at individual patient level for all groups, including the controls. The correlation matrix reporting the Spearman's correlation coefficients calculated between the expression levels (FPKM) of FUS and those of HuD, NRN1 and GAP43 in the samples from the NYGC consortium is shown in the new Figure S9B. The correlation values were calculated across various groups: OND, CTR, ALS_Glia, ALS_Ox, ALS_TE, and CTR+ALS (including CTR and all ALS groups). The results suggest that there is no correlation between FUS and HuD expression levels in OND and ALS_Ox groups, low positive correlation in CTR and ALS_TE groups, and low negative correlation in the ALS_Glia group. For what concerns NRN1 and GAP43, we observed positive correlation with FUS in all groups except ALS_Glia. Finally, when all groups are merged (CTR+ALS), there is no correlation for FUS and HuD and low positive correlation between FUS and NRN1 and FUS and GAP43. These results suggest that altered levels of FUS are unlikely the reason why HuD is upregulated in sporadic ALS patients with an oxidative stress signature. More in general, while altered FUS activities in FUS mutant models lead to HuD upregulation (De Santis et al., 2017; De Santis et al., 2019; Garone et al., 2021) there is no obvious correlation between FUS and HuD levels in non-FUS ALS sporadic patients and controls.  
+
+<|ref|>text<|/ref|><|det|>[[92, 777, 902, 846]]<|/det|>
+We have better explained in the DISCUSSION that our previous and present results point to two independent mechanisms underlying HuD upregulation in ALS: one is FUS- dependent and can be observed in FUS ALS models and patients; the second could be oxidative- stress dependent and can be reproduced by inducing oxidative stress in in vitro models.  
+
+<|ref|>text<|/ref|><|det|>[[92, 865, 863, 899]]<|/det|>
+4) It is not clear what individual points represent in each of the figures. Are these individual experiments (ie biological replicates) or technical replicates? If this is biological replicates, the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 83, 895, 118]]<|/det|>
+number of technical replicates per experiment should be indicated, and statistical analysis should be performed pairing per experiment.  
+
+<|ref|>text<|/ref|><|det|>[[92, 118, 901, 345]]<|/det|>
+A similar point has been raised by Reviewer #3. Unfortunately, in studies based on human iPSCs, the definition of what can be considered a bona fide biological replicate is still debated (see for example: Chan & Teo, Stem Cells 2020; DOI: 10.1002/stem.3237; section 5.4.2 of the ISSCR "Standards for Human Stem Cell Use in Research" document, https://www.isscr.org/standards- document/reporting). Prompted by the comments of Reviewers 2 and 3, in the revised version of the paper we have decided to use the term "biological" to indicate replicates in the strictest way, i.e. to define the two sets of human iPSC lines derived from two different individuals: the WT- I- based set (Lenzi et al., 2015; Garone et al., 2021) and the set based on the EBiSC line WTSii004- A (this paper). Within each line, each experiment has been performed on individual batches of differentiated iPSCs. For example, for image analyses of Fig. 2- 5 and Suppl. Fig. S3- S4, the experiment was performed on 3 batches of differentiated cells and a number of randomly selected fields per replicate indicated in the legends. Each dot in the graphs represents the average of the fields for each replicate.  
+
+<|ref|>text<|/ref|><|det|>[[92, 345, 904, 398]]<|/det|>
+The definition of what is considered here as biological replicates is provided in the MATERIALS AND METHODS section, and the information on the number of differentiated batches and fields within each individual batch used for the analyses is provided in each figure legend.  
+
+<|ref|>text<|/ref|><|det|>[[92, 414, 904, 501]]<|/det|>
+Furthermore, in Figure 6, there are 3 points (presumably 3 independent experiments), but the legend states \(n = 6 - 10\) (presumably the number of animals per experiment). This should be clarified. In each of the control and experimental conditions each fly of 6- 10 in numbers are considered one data point and we performed the experiments in three trials which make our \(N = 3\) , with 6- 10 flies. This makes our total flies to \(\sim 30\) flies. Also kindly check the response to the first reviewer.  
+
+<|ref|>text<|/ref|><|det|>[[92, 518, 895, 604]]<|/det|>
+5) In addition, all experiments in iPSCs appear to have been analyzed using Student's t test, while there are more than 2 groups (eg Figures 2C, 4D...). In some instances (eg Figure 4B), a two way ANOVA (with time and genotype) should be applied. In general, the statistics should be revised and use the appropriate tests for the experiments. As the results are robust, this would not likely change the statistical significance but is important for scientific robustness of the study.  
+
+<|ref|>text<|/ref|><|det|>[[92, 605, 901, 710]]<|/det|>
+This point has been raised also by reviewer #1. As suggested, we have now applied an ordinary one- way ANOVA in cases when more than two groups are compared for one factor, and a two- way ANOVA (with time and genotype considered two factors) in Figures 2C, 4B, and 4D. In figure 1E we have performed a two- way repeated measures ANOVA to take into account multiple variables (treatment and replicate) and the fact that, within each replicate, each individual field has been analyzed in the 3 different treatment conditions.  
+
+<|ref|>text<|/ref|><|det|>[[92, 710, 904, 866]]<|/det|>
+As predicted by this reviewer, overall the statistical significance has not substantially changed. However, this suggestion allowed us to drive interesting conclusions from the results of Figure 4, i.e. that, longitudinally, a significant increase of dead cells from the initial time point (day 7) was observed for the \(\mathrm{FUS}^{WT}\) co- cultures only at the latest time point (day 28; possibly due to intrinsic fragility of iPSC- derived MNs and muscle cells in culture), while it was evident at earlier time points for \(\mathrm{FUS}^{P525L}\) (from day 14 onward) and \(\mathrm{FUS}^{WT} + \mathrm{HuD}\) (from day 21 onward) (Fig. 4B). Moreover, from day 14 to day 28, the number of apoptotic cells significantly increased in the co- cultures made by \(\mathrm{FUS}^{P525L}\) and \(\mathrm{FUS}^{WT} + \mathrm{HuD}\) , but not \(\mathrm{FUS}^{WT}\) , MNs (Fig. 4E). These observations, in line with our conclusions, have been included in the RESULTS.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[93, 83, 404, 99]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[92, 117, 900, 205]]<|/det|>
+Silvestri and colleagues use an iPSC- based NMJ model to assess contributions of an aggressive FUS mutation and an overexpression of ELAVL4. They argue that the in both of these experimental conditions there are NMJ defects and apoptosis, and they claim similarity of the two conditions. They evaluate the relationship between the genes further in flies, and find that elav knockdown rescues the FUS phenotype. Finally, they propose an oxidative stress phenotype.  
+
+<|ref|>text<|/ref|><|det|>[[92, 223, 880, 256]]<|/det|>
+Unfortunately, this reviewer finds several major and broad concerns that undermine the results and its conclusions.  
+
+<|ref|>text<|/ref|><|det|>[[92, 274, 864, 344]]<|/det|>
+We thank the reviewer for his/her constructive criticisms. We apologize for lack of clarity in important aspects, e.g. the definition of replicates, which has been better explained in the revised text. Moreover, we have improved the statistical analysis, as also requested by the other reviewers.  
+
+<|ref|>text<|/ref|><|det|>[[92, 361, 901, 466]]<|/det|>
+1) First, there is lack of rigor in the description of techniques and details of analyses that spans all iPSC modeling studies. What are the replicates? How many fields were chosen for image analysis of contractions and IHC? How were these fields chosen? For contraction analysis, what was the timing of glutamate application and the effect. In terms of the "isogenic", did the control go through editing process to control for such effects? This seems hard then to add an additional true isogenic control for the prior isogenic pair.  
+
+<|ref|>text<|/ref|><|det|>[[91, 466, 900, 620]]<|/det|>
+A similar point has been raised by Reviewer #2. Unfortunately, in studies based on human iPSCs, the definition of what can be considered a bona fide biological replicate is still debated (see for example: Chan & Teo, Stem Cells 2020; DOI: 10.1002/stem.3237; section 5.4.2 of the ISSCR "Standards for Human Stem Cell Use in Research" document, https://www.isscr.org/standards- document/reporting). Prompted by the comments of Reviewers 2 and 3, in the revised version of the paper we have decided to use the term "biological" to indicate replicates in the strictest way, i.e. to define the two sets of human iPSC lines derived from two different individuals: the WT- I- based set (Lenzi et al., 2015; Garone et al., 2021) and the set based on the EBiSC line WTSii004- A (this paper). In particular:  
+
+<|ref|>text<|/ref|><|det|>[[120, 621, 900, 792]]<|/det|>
+a) the WT-I line, from a healthy individual, described in Lenzi et al., 2015 (FUSwT) 
+b) a FUS mutant line generated from a) by gene editing as described in Lenzi et al., 2015 (FUSPS25L) 
+c) a line generated from a) by stably inserting a transgene overexpressing HuD (described in Garone et al. 2021) 
+d) the WTSii004-A line, from a healthy individual, obtained by EBiSC (here named WTSI-FUSwT) 
+e) a FUS mutant line generated from d) by gene editing (WTSI-FUSPS25L) 
+f) a line generated from d) by stably inserting a transgene overexpressing HuD (WTSI-FUSwT+HuD)  
+
+<|ref|>text<|/ref|><|det|>[[91, 793, 904, 863]]<|/det|>
+iPSC lines a) and b) are isogenic and went through the same editing process; the same for lines d) and e). Lines c) and f) have not been obtained by gene editing, but by piggyBac- mediated integration of a HuD transgene under the control of the human SYN1 promoter. We have better clarified this in the text (RESULTS and MATERIALS AND METHODS).  
+
+<|ref|>text<|/ref|><|det|>[[92, 864, 884, 916]]<|/det|>
+Within each line, each experiment has been performed on individual batches of differentiated iPSCs. For example, for image analyses of Fig. 2- 5 and Suppl. Fig. S3- S4, the experiment was performed on 3 batches of differentiated cells, and a number of randomly selected fields per
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 82, 888, 204]]<|/det|>
+replicate was analyzed, as indicated in the legends. Each dot in the graphs represents the average of the fields for each replicate. In the contraction analysis of Figure 1E, we performed the experiment 3 times using different batches of NIL iPSCs and MB iPSCs. During the experiment, for each batch, 3 fields were randomly selected. The glutamate was added to the cells and the effect was measured after 5 minutes, then we added tubocurarine and measured contractions after 5 minutes, as specified in the revised text (MATERIALS AND METHODS and FIGURE LEGEND).  
+
+<|ref|>text<|/ref|><|det|>[[92, 205, 904, 258]]<|/det|>
+The definition of what is considered here as biological replicates is provided in the MATERIALS AND METHODS section, and the information on the number of differentiated batches and fields within each individual batch used for the analyses is provided in each figure legend.  
+
+<|ref|>text<|/ref|><|det|>[[92, 274, 473, 290]]<|/det|>
+How were \(\%\) of BTX positive fibers calculated?  
+
+<|ref|>text<|/ref|><|det|>[[92, 290, 895, 342]]<|/det|>
+As specified in the revised METHODS section, MyHC+ muscle fibers have been identified and counted from multiple randomly selected fields for each biological replicate. In the same fields, the total number of muscle fibers showing BTX signal has been assessed to calculate the \(\%\) .  
+
+<|ref|>text<|/ref|><|det|>[[92, 377, 903, 482]]<|/det|>
+2) Second, there is insufficient biological evaluation, too many unwarranted assumptions, and too many unexplained phenomena. For example, one needs co-localized presynaptic markers to interpret BTX staining as synaptic. MN appearance is remarkable different in eg Fig 1, Fig 2, and Fig 4c. In Fig 4, MNs only contain lots of DAPI that is MAP-. Much of the CC-3 staining is outside MAP2 staining. What are the dead cells? If there is indeed a neuronal toxicity brought out by muscle co- culture, how does one explain this mechanistically?  
+
+<|ref|>text<|/ref|><|det|>[[91, 483, 904, 917]]<|/det|>
+In order to validate BTX staining as synaptic, we have performed a triple staining that included a presynaptic marker (SYNAPTOPHYSIN), together with BTX and MyHC. Representative images are included in the new Figure 1D and show close proximity of the BTX and SYNAPTOPHYSIN signals. For what concerns the different appearance of MNs, the appearance of the MNs in Fig1 and 2 might seem different because of the different magnification. Fig.4C shows a staining with MAP2, which was used instead of TUJ1 in the co-staining with CC-3 due to antibody compatibility. We realized that the MAP2 staining was weaker in the original images of Figure 4C (single panels in supplementary Fig. S3), so we have replaced the "MNs only" panels of this figure with more representative images. As correctly noticed by the reviewer, much of the CC-3 signal is outside MAP2 staining in the co- cultures panels ("MNs & FUS<sup>WT</sup> SkMCs"). This is due to the fact that not only MNs, but also muscle cells die by apoptosis in these conditions. When cultured alone, a basal level of muscle cell death is always observed over time in our experiments, possibly due to intrinsic fragility of iPSC-derived skeletal muscle in vitro. MNs are known to release trophic factors that support the survival and growth of nearby cells. Therefore, the presence of MNs in our co- cultures can enhance the survival of skeletal muscle cells and contribute to their overall health. In Figure 4C,D we show that the number of apoptotic cells (both MNs and muscle) is significantly increased in presence of FUS<sup>P525L</sup> and FUS<sup>WT</sup>+HuD MNs in the co- cultures (compared to FUS<sup>WT</sup> MNs). We hypothesize that this could be due to the fact that upon MNs death and/or failure to form mature NMJs, their support to muscle survival in vitro fails. We have better explained this in the revised DISCUSSION. For what concerns the CC-3 signal outside of MAP2 staining in MNs monocultures, despite our differentiation method is quite efficient in terms of the fraction of post-mitotic MNs over the total number of differentiated cells (usually >80% of Isl1+ and CHAT+ cells, see De santis et al., 2018; Garone et al., 2019), in long term cultures (beyond 2 weeks), a subpopulation of proliferating non neuronal cells - conceivably MAP2- negative undifferentiated progenitors - becomes increasingly evident.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 100, 873, 134]]<|/det|>
+4) The fly experiment is the strongest result, although again what the 3 replicates means is not clarified.  
+
+<|ref|>text<|/ref|><|det|>[[91, 135, 904, 223]]<|/det|>
+In each of the control and experimental conditions each fly of 6- 10 in numbers are considered one data point and we performed the experiments in three trials which make our \(N = 3\) , with 6- 10 flies. This makes our total flies to \(\sim 30\) flies. Note that in response to a request by the first reviewer we have repeated and confirmed the rescue experiments in a different genetic context and added the measure of climbing velocity (new Fig. 6B,C).  
+
+<|ref|>text<|/ref|><|det|>[[90, 240, 870, 274]]<|/det|>
+5) The statistical analysis in the bioinformatic data does not appear to be adjusted for multiple comparisons.  
+
+<|ref|>text<|/ref|><|det|>[[91, 275, 904, 345]]<|/det|>
+As indicated in the captions for Figure 7G and Supplementary Figure S8A, the p-values we have provided are "adjusted", meaning that they were corrected for multiple comparisons via the Benjamini-Hochberg procedure. We recognize that our prior explanation may have lacked clarity and we have added this detail in the "Statistics and reproducibility" section.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[94, 85, 290, 100]]<|/det|>
+Reviewers' comments:  
+
+<|ref|>text<|/ref|><|det|>[[94, 118, 415, 135]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[93, 154, 822, 188]]<|/det|>
+This is a manuscript that has been nicely reviewed, and the authors have adequately answered my previous comments.  
+
+<|ref|>text<|/ref|><|det|>[[94, 206, 303, 222]]<|/det|>
+There are still few typos:  
+
+<|ref|>text<|/ref|><|det|>[[93, 223, 805, 292]]<|/det|>
+- Figure 1E: glutamate is spelled twice with 2 m (glutammate)- Figure 5A/B: "non targeting" is misspelled (non targeting)- fonts are non homogenous from one figure to another, as well as size of letterings- Figure 7G could be labelled with more self explanatory annotations.  
+
+<|ref|>text<|/ref|><|det|>[[93, 310, 884, 344]]<|/det|>
+Typos have been corrected and fonts have been homogenized. We have added the self- explanatory annotation of the patients' groups in Figure 7G.  
+
+<|ref|>text<|/ref|><|det|>[[94, 396, 415, 412]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[93, 430, 902, 465]]<|/det|>
+Unfortunately this reviewer does not find sufficient improvement in the fundamental stem cell experiments of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[92, 483, 875, 552]]<|/det|>
+The staining of MyHC seems very weak compared to other examples of stem cell- derived muscle (Choi et al, 2016; Martins et al, 2020; Bar- Nur et al 2018). Furthermore, the bright MyHC spots are outside the muscle fibers, and do not seem consistent with patterns or intensity seen in other iPSC or primary muscle cell types.  
+
+<|ref|>text<|/ref|><|det|>[[91, 552, 900, 763]]<|/det|>
+If the reviewer is referring to the intensity of the signal in the images, this cannot be directly compared in different publications in which different microscopes, parameters for the acquisition, kind of primary/secondary antibodies, etc. are used. If the comment is instead on the percentage of cells that are positive for MyHC, we had assessed in our previous publication that the original protocol yielded around \(77\%\) of MyHC- positive cells (Lenzi et al., 2016; DOI 10.1016/j.scr.2016.06.003). Prompted by the comment of the referee we have quantified the percentage of MyHC- positive cells with the improved protocol described here (Fig. S1D) and assessed that it is \(86.5 \pm 4.6\%\) at day 9 and \(90.0 \pm 2.5\%\) at day 13, which is in line with the literature (e.g. \(63.6\% \pm 9.68\%\) in Choi et al, 2016). This result has been included in the revised version of the paper (Supplementary Fig. S1 legend). For what concerns the MyHC signal outside the muscle fibers, this is due to antibody binding to cell debris which are present in long term co- cultures.  
+
+<|ref|>text<|/ref|><|det|>[[92, 780, 866, 815]]<|/det|>
+There is no demonstration of striations with sarcomeric alpha- actininin (as in Osaki et al 2018; Pereira et. al. 2021).  
+
+<|ref|>text<|/ref|><|det|>[[92, 816, 887, 886]]<|/det|>
+Demonstration of striations with sarcomeric \(\alpha\) - actininin in skeletal muscle monocultures and in co- cultures is shown in the new Supplementary Fig. S2 (panels A and B). The pattern is very similar to what reported in the literature, and in particular by Osaki et al. (2018) - Figures redacted due to lack of 3rd Party Rights
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 362, 896, 396]]<|/det|>
+Images in figure 1D do not convincingly demonstrate co- colocalization of synaptophysin with bungarotoxin.  
+
+<|ref|>text<|/ref|><|det|>[[92, 397, 878, 522]]<|/det|>
+The co- staining with \(\alpha\) - bungarotoxin ( \(\alpha\) - BTX; marking the postsynaptic side of the myoblast) and an antibody for the presynaptic protein synaptophysin (SYP; marking synaptic vesicles) is often used for characterization of NMJ formed by in vitro human iPSC- derived cultures. At a confocal microscope, these signals are not expected to be co- localized, being respectively post- and pre- synaptic, but located in proximity. The images in Figure 1D and new representative images shown in the new Supplementary Fig. S2C show indeed this situation, which is very similar to previous work by others,  
+
+<|ref|>text<|/ref|><|det|>[[140, 521, 700, 538]]<|/det|>
+Figures by others redacted due to lack of 3rd Party Rights
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 392, 870, 479]]<|/det|>
+The description of the contractions is not sufficient and would require video or dynamic collection of images to be convincing. Furthermore, the time course of glutamate effects, which could be ionotropic or metabotropic, are not sufficiently resolved. How do we know that the duration of glutamate effect doesn't taper in 5- 10 minutes, as there is no vehicle control for the addition of curare?  
+
+<|ref|>text<|/ref|><|det|>[[92, 480, 901, 567]]<|/det|>
+Video recording of contractions is now included as a new supplementary material (Supplementary Video S1). Such spontaneous contractions were absent in presence of curare (Supplementary Video S2). It should be noted that in this case we did not measure contractions induced by glutamate, but spontaneous ones, to avoid any possible issue due to attenuation of glutamate effects over time.  
+
+<|ref|>text<|/ref|><|det|>[[92, 585, 894, 687]]<|/det|>
+Unfortunately, give the concerns about the fundamental representation of the bungarotoxin staining and its relation to presynaptic NMJ components, this reviewer cannot interpret the results in the subsequent iPSC figures. In particular, the overlap of bungarotoxin on the muscle fibers is not clear in Figure 2 and in Figure 3 it seems clear that BTX puncta are not on the muscle? Thus it's not clear that the bungarotoxin puncta are directly connected to the muscle or putative NMJ.  
+
+<|ref|>text<|/ref|><|det|>[[92, 688, 892, 794]]<|/det|>
+We agree with the reviewer that in some figures part of the BTX signal is outside the muscle fibers, sometimes staining cell debris. However, we would like to clarify that for the quantification of the \(\alpha\) BTX- positive fibers of Figure 2 (and elsewhere in the paper for such kind of analysis) we have exclusively taken into consideration the specific BTX signal localized in bona fide MyHC- positive muscle fibers. This is shown in the images below, representing magnification of Figure 2 panels.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[90, 85, 760, 500]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[92, 536, 904, 678]]<|/det|>
+For what concerns Figure 3, while it is clear that the \(\alpha\) BTX signal is overlapped with the MyHC signal in the panels relative to "Diffuse puncta", "Dense" and "Cluster", the concern of the reviewer about the "Aligned puncta" panels is probably due to the fact that in these images the MyHC staining is weaker at the periphery of the fiber. Here below we show a magnification of the same merged image in which we have increased the brightness of the MyHC staining. In this image the green \(\alpha\) BTX spots are clearly inside the fiber. We prefer to keep the original image in the paper since the MyHC signal in the image below is oversaturated at the center of the fiber.  
+
+<|ref|>image<|/ref|><|det|>[[93, 692, 475, 925]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 118, 892, 156]]<|/det|>
+For the cell death experiments, one needs to understand more about the identity of the CC- 3 positive cells. For example, are these the muscle cells or not?  
+
+<|ref|>text<|/ref|><|det|>[[92, 156, 568, 172]]<|/det|>
+This point is addressed in the discussion as follows:  
+
+<|ref|>text<|/ref|><|det|>[[92, 172, 888, 310]]<|/det|>
+"In the co- cultures, we observed a significant increase on cell death in both neural and muscle populations upon MN- restricted FUS mutation or HuD overexpression. Both in vivo and in vitro, MNs are recognized for their ability to release trophic factors that support the survival and growth of nearby cells. The presence of MNs in the co- culture can thus enhance the survival of skeletal muscle cells and contribute to their overall health. We hypothesize that upon increased FUS<sup>P525L</sup> and FUS<sup>WT</sup>+HuD MNs death and/or failure to form mature NMJs, their support to muscle survival in vitro fails, thus explaining the increased apoptosis of muscle fibers."
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__d71fd36770c5d63442bd518c5d3e45031b33e8617cb1d616b4a8005bd8b6dbe1/images_list.json

@@ -0,0 +1,319 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Rev. Fig. 1 GO-term enrichment analysis of genes upregulated in \\(\\mathsf{CLDN6}^{\\mathsf{Low}}\\) EpiSCs displays significant overrepresentation of terms associated with MAPK/tyrosine kinase signalling cascades, cell adhesion, stem cell development, and DNA methylation.",
+    "footnote": [],
+    "bbox": [
+      [
+        130,
+        470,
+        830,
+        666
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Rev. Fig. 2 GO-term enrichment analysis of genes upregulated in CLDN6<sup>High</sup> EpiSCs does not show statistically significant term enrichment.",
+    "footnote": [],
+    "bbox": [
+      [
+        235,
+        214,
+        666,
+        377
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Rev. Fig. 3 ATAC-seq of Cdx2, Cdx1, Msgn1 and Tbx6 loci in EpiSCs not exposed to WNT agonists or antagonists (Epi) and EpiSCs cultured with WNT inhibitor XAV939 (Epi-X). The ATAC-seq data reveals similar accessibility patterns over the NMP and PSM (Presomitic Mesoderm) genes in Epi and Epi-X conditions. Chromatin accessibility increased at 6 hours post-WNT induction for the NMP genes Cdx1 and Cdx2, and at 12 hours for the early PSM genes Msgn1 and Tbx6. Accessible and open regions are indicated by the red arrows, while closed regions are with white arrows.",
+    "footnote": [],
+    "bbox": [
+      [
+        161,
+        100,
+        819,
+        760
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Rev. Fig. 5 ChlPseq and ATACseq coverage tracks showing absence of PBX1 recruitment over Hand1 and Tbx3 genes, even when the regulatory regions open upon differentiation towards PPS.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Rev. Fig. 6 a) WGBS (SeqMonk) visualization indicating differentially methylated regions (DMRs) across the Tbx3 locus in WT and Pbx1-KO ESCs and EpiSCs, showing hypermethylation in WT EpiSCs but not in ESCs and Pbx1-KO EpiSCs. The bar shows CpG methylation levels (%). b) Gene expression level (normalized FPKM data from RNAseq) of Tbx3 across ESCs (2i/LIF and serum/LIF), EpiSCs, PPS and ExM in WT and Pbx1-KO, shows no difference between WT and Pbx1-KO ESCs and no expression, but Tbx3 gets upregulated in Pbx1-KO PPS and ExM. c) WGBS (SeqMonk) visualization showing DMRs across Hand1 locus in WT and Pbx1-KO ESCs and EpiSCs, reveals hypermethylation in Pbx1-KO EpiSCs and hypomethylation in WT EpiSCs and across the ESCs. Bar shows CpG methylation levels (%). d) Analysis of normalized FPKM data from RNAseq reveals distinct expression patterns for Hand1 in different cell types and genotypes. Hand1 is not expressed in ESCs and EpiSCs, but gets specifically expressed in WT but not Pbx1-KO upon BMP-mediated differentiation towards PPS and ExM.",
+    "footnote": [],
+    "bbox": [
+      [
+        123,
+        106,
+        880,
+        425
+      ]
+    ],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_7.jpg",
+    "caption": "Rev. Fig. 7 FGF5 localization in the anterior epiblast as shown by the Fgf5 reporter generated by Khoa et al., PLoS One 2016. Panels are extracted from Khoa et al., 2016.",
+    "footnote": [],
+    "bbox": [
+      [
+        114,
+        533,
+        835,
+        850
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_8.jpg",
+    "caption": "Rev. Fig. 8 a) IF of HAND1 (green) and DAPI (blue) showing loss of HAND1 expression in \\(Pbx1 - KO\\) ExM cells. b) RT-qPCR mRNA analysis showing loss of Hand1 expression in \\(Pbx1 - KO\\) EpiSCs upon PPS and ExM differentiation. c) IF and qPCR d) analysis of HAND1 showing no difference in protein and mRNA levels in WT and \\(Pbx2 - KO\\) across ExM differentiation.",
+    "footnote": [],
+    "bbox": [
+      [
+        163,
+        201,
+        787,
+        655
+      ]
+    ],
+    "page_idx": 20
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_9.jpg",
+    "caption": "Rev. Fig. 9 Relative expression of Oct4 (pluripotency marker), Fgf5, Otx2 and Sox2 (neural markers), T-Bra, and Foxa2 (PS markers), revealing EpiSCs differentiation upon PD03 treatment.",
+    "footnote": [],
+    "bbox": [
+      [
+        125,
+        301,
+        808,
+        852
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_11.jpg",
+    "caption": "Rev. Fig. 11 Heatmap of RNAseq analysis, displaying relative expression of ACTIVIN/NODAL and FGF/ERK signalling components and target genes in CLDN6<sup>High</sup> and CLDN6<sup>Low</sup> EpiSCs. While CLDN6-sorted EpiSC populations exhibit distinct enrichment patterns in the FGF/ERK signaling pathway, there is no significant disparity in the ACTIVIN/NODAL signaling pathway between these populations.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 23
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_12.jpg",
+    "caption": "Rev. Fig. 12 Relative expression of Nanog revealing no difference in CLDN6Low and CLDN6High EpiSCs. Upon WNT-induced APSD differentiation Nanog is upregulated in CLDN6High APS.",
+    "footnote": [],
+    "bbox": [
+      [
+        215,
+        125,
+        707,
+        319
+      ]
+    ],
+    "page_idx": 25
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1. PBX1 is dispensable in ESCs, which rely on ERK inhibition. However, in EpiSCs where ERK is required, PBX1 plays a crucial role (modified from Supp Fig. 5f-g and Fig4b in the manuscript). A) Heatmap showing higher expression of FGF/ERK signalling components like Fgf4 and Fgf5 and Fgf15, ERK feedback inhibitors such as Spry2, Spry4 and Dusp6, and the ERK downstream TFs, Etv4 in Pbx1-KO EpiSCs, with downregulation of Jun and Elf4. B) In WT versus Pbx1-KO EpiSCs, significant overrepresentation of terms associated with the MAPK/ERK cascade was observed in the GO term enrichment analysis. C) RNAseq analysis comparing WT and Pbx1-KO ES cells cultured in 2i/LIF or serum/LIF conditions did not detect any differentially expressed genes.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 29
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2. Loss of Pbx1 results in reduced pERK activity without altering tERK and pAKT levels. (Modified from Fig 4c, and Supp Fig 5h-l in the manuscript). A) Western blot analysis confirms the absence of PBX1 and reduction of pERK1/2 in Pbx1-KO EpiSCs. Western blot analyses showing. B) unchanged levels of total ERK1/2 proteins and pAKT C), and reduced levels of pMEK D) in three Pbx1-KO EpiSC cell lines compared with WT.",
+    "footnote": [],
+    "bbox": [
+      [
+        142,
+        275,
+        856,
+        664
+      ]
+    ],
+    "page_idx": 30
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3. Treatment of PD03 in EpiSCs leads to differentiation and loss of epiblast state. Not adding bFGF (see \\*, as initially suggested by Reviewers 2 and 3, leads to upregulation of Fgf5 ligands with some cells spontaneously differentiation (increasing level of T-Bra and Foxa2).",
+    "footnote": [],
+    "bbox": [
+      [
+        172,
+        194,
+        806,
+        702
+      ]
+    ],
+    "page_idx": 30
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4. PBX1 is recruited to \\(Spry2\\) specifically in EpiSCs, and not in ESCs (Modified from Suppl Fig.5 k-l in the manuscript). A) ATACseq and ChIPseq data demonstrating specific recruitment of PBX1 on \\(Spry2\\) enhancer site that gets specifically accessible in EpiSCs, but is closed in ES cells (p2). B) EMSA showing affinity of PBX-PREP1 complex over \\(Spry2\\) enhancer site (p2) that is specifically accessible in EpiSCs.",
+    "footnote": [],
+    "bbox": [
+      [
+        293,
+        102,
+        765,
+        789
+      ]
+    ],
+    "page_idx": 34
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5. Treating EpiSCs with PD03 prompts differentiation towards neuroectoderm and ExM lineages (Modified from Supp Fig 1h in the manuscript).",
+    "footnote": [],
+    "bbox": [
+      [
+        225,
+        560,
+        722,
+        844
+      ]
+    ],
+    "page_idx": 35
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Figure 6. In Pbx1-KO EpiSCs, imprinted genes are dysregulated similarly to hpESCs treated with PD03 (ERK inhibitor), reinforcing our finding that PBX1 regulates DNA methylation through the ERK pathway.",
+    "footnote": [],
+    "bbox": [
+      [
+        123,
+        186,
+        380,
+        396
+      ]
+    ],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_7.jpg",
+    "caption": "Figure 7. Pbx1-KO EpiSCs are biased towards primitive erythroid fate, similar to PD03 treatment in the epiblast. A) PD03 treatment of EpiSCs results in the upregulation of primitive erythroid genes. B) In the red box, direct differentiation of Pbx1-KO towards ExM results in increased expression of primitive erythroid genes, Runx1, Tal1, Gata2 (Modified from Supp Fig.1h in the manuscript and new data).",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 37
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_8.jpg",
+    "caption": "Figure 8. \\(Pbx1 - KO\\) EpiSCs are biased towards neuroectoderm fate, similar to EpiSCs treated with PD03. A) PD03 treatment of EpiSCs results in the upregulation of neuroectodermal genes. B) In the green box, direct differentiation of \\(Pbx1 - KO\\) towards neuroectoderm results in increased expression of neural genes, Pax6, Tubb3, Nell2 (Modified from Supp Fig.1h in the manuscript and new data).",
+    "footnote": [],
+    "bbox": [
+      [
+        126,
+        100,
+        886,
+        288
+      ]
+    ],
+    "page_idx": 38
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_9.jpg",
+    "caption": "Figure 9. A) RT-qPCR analyses of DE markers (Cer1, Foxa2, Hhex, and Sox17) show reduced DE differentiation in \\(Pbx1 - KO\\) AME cells. B) Confocal maximum intensity projection images of E7.5 control and \\(Pbx1 / 2 - DKO\\) embryos probed with FOXA2 (green) and HAND1 (red) antibodies and counterstained with DAPI (grey) reveal reduced expression of HAND1 in ExM and FOXA2 in DE but not in the node (N). C) Visualization of WGBS (SeqMonk) indicating differentially methylated regions of the Foxa2 locus in WT and \\(Pbx1 - KO\\) EpiSCs shows hypomethylation in the \\(Pbx1 - KO\\) EpiSCs, while the WT EpiSCs contain methylated CpG regions, validating the previously reported paradoxical nature of DNA methylation regulation of Foxa2 expression (Bahar Halper et al., JCB 2014).",
+    "footnote": [],
+    "bbox": [
+      [
+        137,
+        550,
+        865,
+        771
+      ]
+    ],
+    "page_idx": 40
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_10.jpg",
+    "caption": "Figure 10. Effect of Pbx1-KO on EpiSC differentiation potential and ERK activity regulation during cardiac lineage segregation. A) Conventional differentiation protocols and WT cells show modest efficiency in generating Second Heart Field (SHF) progenitors. Pbx1-KO EpiSCs exhibit a deficiency in differentiation towards First Heart Field (FHF) cells, marked by low expression of FHF markers (Hand1, Foxf1, and Pitx1) and an enhanced differentiation towards SHF cardiac cells, characterized by the upregulation of SHF-specific markers including Tbx1, Isl1, Tbx3, Nkx2-5, Lhx2, and Gata4. B) Inhibition of ERK activity via PD03 fine-tunes cardiac lineage segregation. Specifically, PD03 treatment leads to the downregulation of FHF markers (Hand1, etc.) and the upregulation of SHF markers (Tbx1, etc.).",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        437,
+        876,
+        765
+      ]
+    ],
+    "page_idx": 40
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_11.jpg",
+    "caption": "Figure 11. ERK-Associated Marker Expression in Anterior and Distal Posterior Epiblasts. A) Heatmap illustrating specific gene expression in the anterior epiblast (dark yellow) and distal posterior epiblast (light yellow), based on FPKM values from E7.0-E7.5 epiblast GEO-sequencing data. Graphical Cornplot, highlighting the positional identity of anterior epiblast (dark yellow) and distal-posterior epiblast (light yellow). B) Differential expression of CLDN6<sup>High</sup> vs. CLDN6<sup>Low</sup> cells. CLDN6<sup>Low</sup> cells exhibit higher expression of FGF receptors (Fgfr3, Fgfr4) and specific ETS TFs (Ets1, Elf4, Etv1). CLDN6<sup>High</sup> cells are enriched with epithelial markers such as Cdh1, Cldn6, and Cldn7, along with Fgf ligands (Fgf4, Fgf5, etc.) and ETS TF Ets2 (Modified from Supp Fig.1f and Fig1. d in the manuscript).",
+    "footnote": [],
+    "bbox": [
+      [
+        130,
+        580,
+        895,
+        760
+      ]
+    ],
+    "page_idx": 40
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1. PI3K Inhibition Does Not Replicate ERK Effects in EpiSCs. WT EpiSCs (black bar) and PIK90-treated EpiSCs (blue bar) exhibited similar levels of epiblast (Oct4 and Fgf5) and lineage differentiation markers (Neuroectoderm Octx2 and Sox2; APSD Tbra and Foxa2). Treatment of high dose of PD03 (1 μM) results in spontaneous differentiation, marked by increased levels of Octx2, Sox2, T-Bra, and Foxa2. Thus, PI3K Inhibition Does Not Replicate ERK Effects in EpiSCs. Omitting bFGF (indicated by a star \\*), as initially suggested by Reviewers 2 and 3, leads to the upregulation of Fgf5 ligands, with some cells undergoing spontaneous differentiation, as shown by elevated levels of T-Bra and Foxa2. This result show that ERK inhibition feeds back on FGF signalling in the epiblast state.",
+    "footnote": [],
+    "bbox": [
+      [
+        175,
+        123,
+        738,
+        568
+      ]
+    ],
+    "page_idx": 41
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2. Loss of \\(Pbx1\\) results in reduced pERK activity without altering tERK and pAKT levels. (Modified from Fig 4c, and Supp Fig 5h-l in the manuscript). A) Western blot analysis confirms the absence of PBX1 and reduction of pERK1/2 in \\(Pbx1 - KO\\) EpiSCs. Western blot analyses showing, B) unchanged levels of total ERK1/2 proteins and pAKT C), and reduced levels of pMEK D) in three \\(Pbx1 - KO\\) EpiSC cell lines compared with WT.",
+    "footnote": [],
+    "bbox": [
+      [
+        160,
+        135,
+        803,
+        485
+      ]
+    ],
+    "page_idx": 41
+  }
+]

peer_reviews/supplementary_0_Peer Review file__daf022cfe2dadcc2063b0395d1f2d975d0e27498f5d78d4631ec409850e17903/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review file__daf022cfe2dadcc2063b0395d1f2d975d0e27498f5d78d4631ec409850e17903/supplementary_0_Peer Review file__daf022cfe2dadcc2063b0395d1f2d975d0e27498f5d78d4631ec409850e17903.mmd

@@ -0,0 +1,388 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+# Impact of Age-Related Changes in Buccal Epithelial Cells on Pediatric Epigenetic Biomarker Research  
+
+Corresponding Author: Dr Michael Kobor  
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+Version 0:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author)  
+
+Cheek swabs are often used as a source of DNA for studies of DNA methylation profiling. Because cellular heterogeneity within the tissue sampled can affect the DNA methylation data, changes in cellular composition of cheek swab samples with age could cause a confound in analysis of age- related epigenetic outcomes. Accordingly, the authors used existing Illumina HM450 and EPIC 850 data on cheek swab samples from 10 cohorts, representing children aged from 1 to 200 months, and estimated proportions of buccal epithelial cells (BEC) and various leukocyte subtypes to test for changes in cheek swab cellular composition with age. They conclude that "buccal epithelial cell proportions decrease reliably with age", and also that interindividual variation in the proportion of buccal epithelial cells increases with age, and that these findings have implications for studies utilizing cheek swab DNA for epigenetic research.  
+
+I have some concerns regarding the authors' interpretation of their results.  
+
+Examining the main data in the paper (Figure 1, a and b) it is apparent that the linear regression line is not a good fit to the data in either panel a or b. If the data are taken at face value, estimated BEC proportions decrease from birth to 50- 70 months of age, then go back up again by age 200 m. A reciprocal pattern is observed for estimated neutrophil proportions (Fig. 1b). How can the authors claim that "BEC proportions decrease reliably with age" when the majority of data from the oldest individuals (200m) are above the regression line?  
+
+Also, I am concerned that there are systematic differences in methylation data among the different cohorts analyzed, and that the authors have not adequately taken these into consideration. Most obviously, why would BEC proportion go down (and show more variation) at age 50 m, only to increase at 60 m, then go down again at 70 m? Most likely, much of this mysterious variation is due to systematic differences in the HM450 and EPIC 850 data generated at different facilities. In Figures 1a and 1b the authors should color code the data points according to which of the 10 cohorts the data were obtained from. This will clearly indicate to what extent their findings are confounded with technical (cohort- associated) biases, and suggest approaches for adjusting for these.  
+
+This problem is also evident in Fig. 3, in which the authors try to make the case that interindividual variation in BEC increases with age. By displaying cohort- specific data separately in panels a, b, and c, it is obvious that the greatest variation is between the SEED and GUSTO cohorts at age 48 m, rather than the age associations they are trying to highlight.  
+
+## Reviewer #2  
+
+(Remarks to the Author) The manuscript by Merrill et al. titled "Pediatric Buccal Epithelial Cell Proportions Decrease Reliably With Age: Considerations for PedBE in Pediatric Research" presents a comprehensive study on the dynamics of buccal epithelial cell (BEC) proportions across pediatric development, with a focus on their implications for epigenetic aging studies using cheek swab samples. The study analyzes DNA methylation profiles of 3,799 typically developing children ranging from 2 months to 20 years, drawn from ten independent cohorts.
+
+<--- Page Split --->
+
+Key Points of the study:  
+
+BEC Proportions: BEC proportions decline with chronological age and predicted epigenetic age in pediatric populations. The study highlights that correcting for BEC proportion when calculating epigenetic age acceleration (EAA) affects the association of PedBE EAA with certain conditions like OCD and cortisol slope but warns of the potential associations with oral inflammatory conditions that may be driven by socio- economic factors.  
+
+Variability and Influences: Increased variability in BEC proportions was noted with age. Factors influencing BEC proportions include oral health practices and visits to the dentist, but not daily tooth brushing. The study did not find significant associations with sex, mode of delivery, or breastfeeding status.  
+
+Methodological Approaches: EpiDISH, a reference- based bioinformatics method, was used to estimate the cell type proportions. Variance in BEC proportion was assessed through longitudinal analysis, showing a trend of increased variance with age.  
+
+Implications for Research: The findings suggest that pediatric cheek swab samples exhibit significant changes in cellular composition that could influence epigenetic age estimates. These changes need to be considered in epigenetic studies to avoid potential biases.  
+
+## Comments:  
+
+Lines 140- 153: Could you comment on the integration of EpiDiSH for 450K and EPIC? Were the results different when removing the nonoverlapping probes across the two arrays?  
+
+Figures 1A, Supple Fig 1: Could you add some information about the correlation of the results across the different methods? One area that seems quite divergent is for children 150- 200 months, but in general, the constrained projection seems to diverge compared to the CBS- RPC results for the most abundant epithelial cells. Is this a problem of scaling to one, or are there other differences that might be affecting the results?  
+
+Figure 1b: an interesting trend is observed during the period of change from deciduous to permanent teeth with a drastic increase in neutrophils. This point is further brought up with the longitudinal samples in Figure 4 (which also require revising the axes labels as some are missing). Do you have information about the teeth loss/replacement relative to the sample collection? The document points out that while BEC proportion declines with age, the consistency of this decline over time within the same individuals is not well established, potentially affecting the reliability of longitudinal studies using cheek swabs. That can strengthen the observation and explain some of the variability. Please expand on lines 207- 220 as needed. Supplementary Figure 3: I would try to understand this better. Why is this variation occurring? is it only due to scaling to one? what happens when you are restricted to the overlapping probes across both platforms? Are the missing probes in EPIC biasing some of the results?  
+
+Figures 5 and 6: Is it possible to show the relation between chronological age and PedBe Age in months? That can help with the interpretation of the figures. please correct the axis labels on 6b.  
+
+Lines 239- 255: I would bring attention to the fact that this is a small cohort (and, as you mention, a proof of concept analysis). Please include the n in the text, not only in table 2.  
+
+Lines 298- 307: I agree with your discussion, but I would add the fact that the peaks of neutrophils are by the time of teeth exchange so maybe there is blood and more debris during this period that are changing the direction of your findings. Try to add as much context about these physiological changes as possible.  
+
+Lines 430- 439: What method for the CP estimation was used constrained or not constrained to equality?  
+
+## Reviewer #3  
+
+(Remarks to the Author)  
+
+The increased between- person BEC variability combined with the lack of within- person temporal stability is a key finding of this study. That is, the correlations for BEC proportions for the same individuals over even a brief period of time were not significantly strong. What are the potential implications of this lack of within- person stability for conducting future longitudinal research with the same individuals as well as future cross- sectional or other research combining independent, age- heterogeneous samples to examine between- person change? Does a within- person or between- person approach provide unique advantages for epigenetic aging research?  
+
+Please report the beta weights and corresponding p- values for models where: 1) epigenetic age is regressed on chronological age, and 2) epigenetic age is regressed on both chronological age and BEC proportions. These results, particularly for model two, have implications for the uniqueness of each predictor in explaining variation in epigenetic age estimates.  
+
+The implications of the current set of results suggest the importance of including BEC proportions when generating estimates of epigenetic age and epigenetic age acceleration during child development. Are the translational implications then for relevant epigenetic clock developers to adjust the available code (e.g. GitHub for PedBE) or online calculators (e.g. Horvath's) so that BEC proportions are now included in the generation of epigenetic aging estimates?  
+
+Along these same lines, is it sufficient to adjust risk estimates of epigenetic aging on a particular outcome by including BEC proportions as an additional covariate/predictor in the statistical models? Or, must BEC proportions be included in the estimation of epigenetic aging?
+
+<--- Page Split --->
+
+Please provide a brief rationale for why diurnal cortisol slope and obsessive- compulsive disorder status were examined as outcomes important for epigenetic aging. Similarly, a rationale for examining dental visits at 48- months, as opposed to any other age, was not provided in the original manuscript.  
+
+On line 273, the authors report a significant association between chronological age and the residuals of the model where PedBE age acceleration was regressed only on chronological age. As reported in other places in the manuscript, I expected this association would also be near zero given PedBE age acceleration is determined by removing or accounting for the effects of chronological age. Could the authors provide some insight into this finding? Perhaps there is a non- linear relation here? Please also report the magnitude of the association observed here.  
+
+On line 160, the t- value reported is - 2.41 but the adjusted p- value is reported as greater than .05. In a typical t- distribution, a value greater than 1.96 in either direction would be statistically significant (less than .05). Could it be that the decimal is in the wrong place?  
+
+Please report means, standard deviations, and/or effect sizes in each of the figures.  
+
+Several different terms were used to refer to BEC dimensions (e.g. estimated, inferred, computed). Is it possible to use one of these terms throughout the manuscript?  
+
+Version 1:  
+
+Reviewer comments:  
+
+Reviewer #1  
+
+(Remarks to the Author) I am satisfied with the changes the authors have made in response to my concerns.  
+
+Reviewer #2  
+
+(Remarks to the Author)  
+
+The manuscript by Merrill et al. titled "Pediatric Buccal Epithelial Cell Proportions Decrease Reliably With Age From Infancy through Childhood: Considerations for PedBE in Pediatric Research" presents a comprehensive study on the dynamics of buccal epithelial cell (BEC) proportions across pediatric development, with a focus on their implications for epigenetic aging studies using cheek swab samples. The study analyzes DNA methylation profiles of 4,626 typically developing children ranging from 2 months to 20 years, drawn from ten independent cohorts.  
+
+the authors have responded to all my questions. i have no additional recommendations.  
+
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were
+
+<--- Page Split --->
+
+made.  
+
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source. The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+Cheek swabs are often used as a source of DNA for studies of DNA methylation profiling. Because cellular heterogeneity within the tissue sampled can affect the DNA methylation data, changes in cellular composition of cheek swab samples with age could cause a confound in analysis of age- related epigenetic outcomes. Accordingly, the authors used existing Illumina HM450 and EPIC 850 data on cheek swab samples from 10 cohorts, representing children aged from 1 to 200 months, and estimated proportions of buccal epithelial cells (BEC) and various leukocyte subtypes to test for changes in cheek swab cellular composition with age. They conclude that buccal epithelial cell proportions decrease reliably with age", and also that interindividual variation in the proportion of buccal epithelial cells increases with age, and that these findings have implications for studies utilizing cheek swab DNA for epigenetic research.  
+
+I have some concerns regarding the authors interpretation of their results.  
+
+Review 1 Comment 1: Examining the main data in the paper (Figure 1, a and b) it is apparent that the linear regression line is not a good fit to the data in either panel a or b. If the data are taken at face value, estimated BEC proportions decrease from birth to 50- 70 months of age, then go back up again by age 200 m. A reciprocal pattern is observed for estimated neutrophil proportions (Fig. 1b). How can the authors claim that BEC proportions decrease reliably with age" when the majority of data from the oldest individuals (200m) are above the regression line?  
+
+Authors Response: We appreciate Reviewer 1 s concern and astute insight into the data in bringing up an important point. We certainly agree that the oldest" pediatric individuals in our dataset did not conform well to the regression line. Given that we did not have a substantial representation of pediatric samples in the adolescence period, it was difficult to assess the association of estimated buccal epithelial proportions with age in the older" pediatric individuals using the initial cohorts from our original submission. To address this important issue, we have now included two new cohorts: BIBO and Cohort 12, which substantially filled the gap in the pre- adolescence to late adolescence period. While BIBO is a longitudinal cohort with samples at three time- points: 72 months, 120 months, 168 months, Cohort 12 is comprised of adolescent females between 138- 220 months. Our new assembled "full" cohort of 4626 samples from 12 independent cohorts with ages spanning across the pediatric developmental window has now enabled us to examine the relationship of estimated buccal epithelial proportion with age both at childhood and adolescence.  
+
+Using this more complete coverage across the pediatric age range, we conducted a spline regression analysis to determine if our model fit, ascertained by adjusted R- squared, significantly improved by allowing different slopes during childhood and adolescence - directly based on this observation and suggestion of Reviewer 1. By including a knot at 120 months/10 years, which represents the developmentally relevant age of transition from childhood to adolescence, our model fit was markedly improved from 0.17 (no knot, linear regression) to 0.25 (knot at 10 years, spline regression). Because the transition from childhood to adolescence is not at one age,
+
+<--- Page Split --->
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+per se, but defined as a developmental period, encompassing the ages around 10 as well, we examine the model fit by including the knot at 9 years, 11 years and 12 years independently. However, irrespective of when the knot was included (10 years/11 years/12 years), an adjusted R- squared of 0.25 was reported compared to 0.17 which was noted when one linear regression line was fitted across all the samples in the pediatric window. Upon using a cut- off of 10 years to split our pediatric population into i) infancy to childhood (0- 9 years \(< 120\) months) and ii) adolescence, including pre- through the end of adolescence (10- 18 years \(>120\) months), we observed an even stronger negative linear association between estimated BEC proportion and chronological age throughout infancy to end of childhood. However, this association was no longer observed in pediatric samples around adolescence.  
+
+This interesting set of findings which reaffirms our previous result of an association of estimated BEC proportion with age, but now highlights the significance of this association at a particular developmental period, specifically from the start of infancy to the end of childhood, but not in adolescence. As such, we now have significantly modified our second section of results Line 153- 194, made substantial changes to our Figure 1 and added Supplementary Figure 3 to accurately reflect these results. Accordingly, we have also adjusted our manuscript title to explicitly capture the developmental period when the association is consistently observed by specifying this association is from Infancy to Childhood" and removed the word reliably" in the title and throughout our manuscript.  
+
+We thank Reviewer 1 once again for asking these insightful questions that we believe have substantially improved the impact and accuracy of the manuscript, and we hope they have been appropriately addressed in our response and manuscript.  
+
+Figure 1: Significant association of estimated buccal epithelial cell (BEC) proportions with reported chronological age in the pediatric window. A) In the scatterplot, estimated BEC proportions estimated by the EpiDISH- RPC method was plotted on the y axis against chronological age in months on the x axis. Significant  
+
+![PLACEHOLDER_5_0]
+
+
+<--- Page Split --->
+
+decrease in estimated BEC proportion with reported chronological age was observed in pediatric buccal swabs. B) Pediatric samples from infancy to childhood, using a cut- off of \(< 120\) months (10 years), showed a stronger negative linear relationship between estimated BEC proportions and chronological age (reported in months). C) Pediatric samples in adolescence ( \(>120\) months/10 years) exhibited no association between estimated BEC proportion and chronological age (reported in months).  
+
+Supplementary Figure 3: Spline regression analysis by including a knot at age 10 to signify the transition from childhood to adolescence, showed improved model fit, ascertained by a higher adjusted R- squared of 0.25, compared to linear regression which yielded an adjusted R- squared of 0.17.  
+
+![PLACEHOLDER_6_0]
+  
+
+Reviewer 1 Comments 2 & 3: Also, I am concerned that there are systematic differences in methylation data among the different cohorts analyzed, and that the authors have not adequately taken these into consideration. Most obviously, why would BEC proportion go down (and show more variation) at age 50 m, only to increase at 60 m, then go down again at 70 m? Most likely, much of this mysterious variation is due to systematic differences in the HM450 and EPIC 850 data generated at different facilities. In Figures 1a and 1b the authors should color code the data points according to which of the 10 cohorts the data were obtained from. This will clearly indicate to what extent their findings are confounded with technical (cohort- associated) biases, and suggest approaches for adjusting for these. This problem is also evident in Fig. 3, in which the authors try to make the case that interindividual variation in BEC increases with age. By displaying cohort- specific data separately in panels a, b, and c, it is obvious that the greatest variation is between the SEED and GUSTO cohorts at age 48 m, rather than the age associations they are trying to highlight.  
+
+Authors Response: We agree with the reviewer that there is diversity, as expected amongst our 12 independent cohorts both in terms of the sample sizes and demographic characteristics including age which is reflected in Table 1. Importantly, the DNA methylation data which was used to estimate cell type proportions and epigenetic ages was measured using the same platform: Illumina microarrays, either with the Illumina Infinium HumanMethylation450 BeadChip (450K) or MethylationEPIC BeadChip (850K). Both 450K and 850K have identical probe type chemistry to capture DNA methylation intensities of the CpGs and except the one
+
+<--- Page Split --->
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+GEO cohort, all the cohort samples were run in the same facility (Kobor Lab). Further, samples across all the cohorts underwent the same preprocessing steps as outlined in the Methods. However, the majority of the cohorts are confounded by the chronological age at which buccal samples were collected and as expected, the individual cohorts did not have a widespread representation of all the ages across the pediatric window, except the four longitudinal cohorts (BEPAC, SEED, GUSTO and BIBO) which exhibited some level of spread in terms of age, though in their nicked range. As suggested by Reviewer 1, we have now also color coded the data points based on the cohorts to demonstrate that the negative association between estimated BEC proportions and age held true for the majority of the cohorts (Supplementary Figure 4) and was largely not influenced by differences in 450K or 850K array.  
+
+![PLACEHOLDER_7_0]
+  
+
+We appreciate the Reviewer's concern regarding differences in 450K versus 850K, some of which may be then reflected in the estimated buccal cell type proportions. Given that the EpiFibIC reference (716 CpGs) used by the EpiDISH R package is based on 450K array, inevitably there were missing CpGs (36 CpGs) when we used EpiDISH on our 850K cohort. To investigate this, we had used the strength of a small but matched subset of samples ( \(n = 24\) matched pairs) from the APrON FetalPro and APrON NeuroTox cohorts for which we had obtained DNA methylation measurements using 450K and 850K arrays. We would like to highlight our work reported earlier in the manuscript that the estimated buccal cell type proportions using the RPC method, the method we use to report all our findings in the manuscript, showed no significant differences based on the array. To specifically test the impact of the missing EpiFibIC CpGs from the 850K array, we now tested these differences in our matched dataset by calculating the cell type proportions using different sets of reference CpGs:  
+
+1. Using the EpiFibIC reference with 716 CpGs on 450K datasets and the same EpiFibIC reference excluding the 36 missing CpGs on the 850K datasets2. Restricting the EpiFibIC reference to the common/shared CpGs between the arrays and using this set of CpGs on both 450K/850K
+
+<--- Page Split --->
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+Estimated proportions of BEC by EpiDISH- RPC did not differ based on differing number of CpGs employed in the deconvolution EpiFibIC reference panel, for i) (t (23) = 1.1, \(p_{\mathrm{adj}} > 0.05\) ) and for ii) (t (23) = - .16, \(p_{\mathrm{adj}} > 0.05\) ). Overall, EpiDISH- RPC method consistently predicted higher BEC proportions compared to the other methods (Supplementary Figure 7). Further, CBS estimated BEC proportions showed no significant differences based on the array type, only when the cell type estimates were calculated by limiting the EpiFibIC reference to the shared/common CpGs between the arrays. In contrast, in the context of CP- based cell type algorithm, employing either equality or inequality constraint, we noted significant differences between 450K and 850K calculated BEC proportions, with a large Cohens s d effect size (>1), in both the scenarios of using all the CpGs in the EpiFibIC reference versus restricting the EpiFibIC reference to the common/shared CpGs between the arrays. However, the magnitude of difference between the arrays based on CP estimated BEC proportions was relatively larger when differing numbers of CpGs on the EpiFibIC reference were used to calculate the estimates. (Figure below: Supplementary figure 7). This is now included in Results Line 196- 223.  
+
+Supplementary Figure 7: Comparison of buccal epithelial cell (BEC) proportions estimated by EpiDISH_RPC, EpiDISH_CBS, EpiDISH_CP_equality constraint and EpiDISH_CP_inequality constraint methods across the two DNA methylation platforms based on A) shared/common CpGs (EpiFibIC reference) between 450K and 850K arrays and B) using all the EpiFibIC reference CpGs for the 450K samples and with 36 missing CpGs from the reference for the 850K samples.  
+
+For A), the mean BEC proportions with EpiDISH_RPC on the 450K array was 0.99 0.0 and on the 850K array it was 0.99 0.01. For EpiDISH_CBS, the mean BEC proportion on the 450K array was 0.98 0.01, while on the 850K array it was 0.99 0.01. For EpiDISH_CP with equality constraint, the mean BEC proportion on the 450K array was 0.82 0.01 and on the 850K array it was 0.8 0.01 and with Cohen s d = 1.85. For EpiDISH_CP with inequality constraint, the mean BEC proportion on the 450K array was 0.78 0.01 and on the 850K array it was 0.77 0.01 and with Cohen s d = 1.69.  
+
+For B), the mean BEC proportions with EpiDISH_RPC on the 450K array was 0.99 0.01 and on the 850K array it was 0.99 0.01. For EpiDISH_CBS, the mean BEC proportion on the 450K array was 0.98 0.01, while on the 850K array it was 0.99 0.01. With a moderate Cohen s d = 0.56. For EpiDISH_CP with equality constraint, the mean BEC proportion on the 450K array was 0.85 0.01 and on the 850K array it was 0.80 0.01 and with a large Cohen s d = 5.23. For EpiDISH_CP with inequality constraint, the mean BEC proportion on the 450K array was 0.78 0.01 and on the 850K array it was 0.77 0.01 and with Cohen s d = 1.69. \* Indicate significant adjusted p- value < 0.05
+
+<--- Page Split --->
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+![PLACEHOLDER_9_0]
+  
+
+Further, at cohort- specific level, we tested whether the differences in EpiFibIC reference CpGs impacted the estimations of the cell type proportions in samples ran across different arrays. To do this, we re- calculated the cell type proportions on the 450K cohorts (5/12) based only on the shared/common CpGs of the reference between 450K and 850K, creating a simulated 850K scenario within our 450K cohorts and then compared the cell type proportions to those obtained previously in the 450K cohorts using all the CpGs in the EpiFibIC reference. Across all the 450K cohorts we observed no significant differences in cell type estimates (see below are examples of two 450K cohorts: GECKO and PAWS), using both RPC and CBS methods, depending on whether we used the EpiFibIC reference with the complete set of CpGs or if the 36 missing CpGs on the 850K were selectively excluded from the reference. In the context of CP based cell type estimations we observed significant differences in both buccal epithelial proportions and fibroblasts proportions, irrespective of adopting an equality constraint or inequality constraint.  
+
+![PLACEHOLDER_9_1]
+  
+
+We also added sentences to the limitations section of the discussion Line 586- 597 to directly address this concern of systematic differences across the cohorts: While the cohorts were processed together, and several cohorts were characterized by the same laboratory, the combining of multiple cohorts (450K and 850K) rather than one, large, uniform collection likely introduced some technical variation at this level. However, in the
+
+<--- Page Split --->
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+context of DNA methylation based calculated cell types, we demonstrate that EpiDISH- RPC estimated BEC proportions are congruent between arrays, irrespective of whether all the CpGs in the EpiFibIC reference are used for BEC estimations or limiting the references to the shared CpGs between the arrays. This finding highlights EpiDISH- RPC as the robust algorithm for BEC estimations, specifically when collating proportions across the two platforms. Although four large longitudinal cohorts were available, the inclusion of more repeated samples with additional time points and using identical methods of collection and storage would have enhanced the robustness of the cell type proportion variability findings."  
+
+Reviewer #2 (Remarks to the Author):  
+
+The manuscript by Merrill et al. titled "Pediatric Buccal Epithelial Cell Proportions Decrease Reliably With Age: Considerations for PedBE in Pediatric Research" presents a comprehensive study on the dynamics of buccal epithelial cell (BEC) proportions across pediatric development, with a focus on their implications for epigenetic aging studies using cheek swab samples. The study analyzes DNA methylation profiles of 3,799 typically developing children ranging from 2 months to 20 years, drawn from ten independent cohorts.  
+
+Key Points of the study:  
+
+BEC Proportions: BEC proportions decline with chronological age and predicted epigenetic age in pediatric populations. The study highlights that correcting for BEC proportion when calculating epigenetic age acceleration (EAA) affects the association of PedBE EAA with certain conditions like OCD and cortisol slope but warns of the potential associations with oral inflammatory conditions that may be driven by socio- economic factors.  
+
+Variability and Influences: Increased variability in BEC proportions was noted with age. Factors influencing BEC proportions include oral health practices and visits to the dentist, but not daily tooth brushing. The study did not find significant associations with sex, mode of delivery, or breastfeeding status.  
+
+Methodological Approaches: EpiDISH, a reference- based bioinformatics method, was used to estimate the cell type proportions. Variance in BEC proportion was assessed through longitudinal analysis, showing a trend of increased variance with age.  
+
+Implications for Research: The findings suggest that pediatric cheek swab samples exhibit significant changes in cellular composition that could influence epigenetic age estimates. These changes need to be considered in epigenetic studies to avoid potential biases.  
+
+Review 2 Comments 1 and 4: Lines 140- 153: Could you comment on the integration of EpiDISH for 450K and EPIC? Were the results different when removing the nonoverlapping probes across the two arrays? For Supplementary Figure 3: I would try to understand this better. Why is this variation occurring? is it only due to scaling to one? What happens when you are restricted to the overlapping probes across both platforms? Are the missing probes in EPIC biasing some of the results?
+
+<--- Page Split --->
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+Authors Response: We thank Reviewer 2 for asking an excellent question about the integration of EpiDiSH for 450K and 850K profiled samples, which is important for the scientific community conducting DNA methylation based epigenetic research. Further, Reviewer 2 also highlights an important aspect regarding the concordance in findings when only the shared CpGs (EpiFibIC reference) between 450K and 850K are used in cell type estimations. We appreciate the reviewers curiosity to understand the source of variation noted in Supplementary Figure 3 (now Supplementary Figure 7).  
+
+Because both the comments are related to the figure and the associated findings, we are addressing this in the same response. Similar to our previous response to reviewer 1 comment 2, we recognize that the EpiFibIC reference (716 CpGs) used by the EpiDiSH R package is based on 450K array, so inevitably there were missing CpGs (36 CpGs) when we used EpiDiSH on our 850K cohorts. To demonstrate the influence of missing EpiDiSH CpGs for calculation of estimates on the 850K samples we used the strength of our matched matched subset of samples ( \(n = 24\) pairs) from APron FetalPro and APron NeuroTox cohorts for which we had DNA methylation measurements using both 450K and 850K. We had earlier shown that the buccal epithelial cell type proportions were comparable based on the RPC method of calculation, though differed based on CBS and CP (inequality constraint) methods. While in this case all the EpiFibIC reference CpGs were available for the 450K samples, there were 36 missing CpGs from the reference for the 850K samples. This may have contributed to some level of differences observed in the CBS and CP (inequality) estimations. For ease of navigation, we copy here the relevant portion included in our response to Reviewer 1:  
+
+To investigate this, we re- calculated our buccal cell type proportions for the matched dataset based on the shared/common CpGs (EpiFibIC reference) between 450K and 850K arrays. In addition, for CP based estimation we also compared between adopting an equality versus inequality constraint to explore its subsequent impact on the cell type estimates. We noted that irrespective of whether or not all the CpGs in the EpiFibIC reference were used to calculate the buccal cell type proportions, RPC estimates were congruent across both 450K and 850K measurements and overall predicted higher buccal cell type proportions compared to the other methods. Alternatively, in the context of CP based cell type proportion algorithm, employing either equality or inequality constraint, we noted significant differences between 450K and 850K calculated estimates, with a large Cohens s d effect sizes \((>1)\) , in both the scenarios of either using all the CpGs in the reference versus restricting the EpiFibIC reference to the common/shared CpGs between the arrays (Figure below). However, the magnitude of difference between the arrays based on CP estimated buccal cell type proportions were larger in the situation when differing numbers of CpGs on the EpiFibIC reference were used to calculate the estimates on 450K versus 850K samples. Further, CBS estimated buccal epithelial proportions showed no differences based on the array, only when the cell type estimates were calculated by limiting the EpiFibIC reference to the shared/common CpGs between the arrays. This work suggests the importance of using the common/shared CpGs in the EpiFibIC reference to calculate cell type estimates when comparing samples across the two platforms, specifically if a researcher chooses to use the CBS or CP algorithms. These findings are now in Results Line 196- 223 and included as a Supplementary Figure 7.
+
+<--- Page Split --->
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+Supplementary Figure 7: Comparison of buccal epithelial cell (BEC) proportions estimated by EpiDISH_RPC, EpiDISH_CBS, EpiDISH_CP_equality constraint and EpiDISH_CP_inequality constraint methods across the two DNA methylation platforms based on A) shared/common CpGs (EpiFibIC reference) between 450K and 850K arrays and B) using all the EpiFibIC reference CpGs for the 450K samples and with 36 missing CpGs from the reference for the 850K samples.  
+
+For A), the mean BEC proportions with EpiDISH_RPC on the 450K array was 0.99 0.0 and on the 850K array it was 0.99 0.01. For EpiDISH_CBS, the mean BEC proportion on the 450K array was 0.98 0.01, while on the 850K array it was 0.99 0.01. For EpiDISH_CP with equality constraint, the mean BEC proportion on the 450K array was 0.82 0.01 and on the 850K array it was 0.8 0.01 and with Cohen s d = 1.85. For EpiDISH_CP with inequality constraint, the mean BEC proportion on the 450K array was 0.78 0.01 and on the 850K array it was 0.77 0.01 and with Cohen s d = 1.69.  
+
+For B), the mean BEC proportions with EpiDISH_RPC on the 450K array was 0.99 0.01 and on the 850K array it was 0.99 0.01. For EpiDISH_CBS, the mean BEC proportion on the 450K array was 0.98 0.01, while on the 850K array it was 0.99 0.01. With a moderate Cohen s d = 0.56. For EpiDISH_CP with equality constraint, the mean BEC proportion on the 450K array was 0.85 0.01 and on the 850K array it was 0.80 0.01 and with a large Cohen s d = 5.23. For EpiDISH_CP with inequality constraint, the mean BEC proportion on the 450K array was 0.78 0.01 and on the 850K array it was 0.77 0.01 and with Cohen s d = 1.69.  
+
+![PLACEHOLDER_12_0]
+  
+
+Review 2 Comment 2: Figures 1A, Supple Fig 1: Could you add some information about the correlation of the results across the different methods? One area that seems quite divergent is for children 150- 200 months, but in general, the constrained projection seems to diverge compared to the CBS- RPC results for the most abundant epithelial cells. Is this a problem of scaling to one, or are there other differences that might be affecting the results?
+
+<--- Page Split --->
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+Authors Response: Thank you for inquiring on the correlation specifics across the different methods of cell type estimation. We had previously included those details in the methods section, Line 660 in our manuscript "Although the estimated BEC proportion were comparable across all approaches (Kendall s tau (4624) = 0.64 to 0.91, \(p < 2.2 \times 10^{- 16}\) ), both EpiDiSH- RPC and EpiDiSH- CBS consistently predicted higher BEC proportions across all cohorts, were strongly correlated (Kendall s \(\tau_{b}\) (4624) = 0.91, \(p < 2.2 \times 10^{- 16}\) ), and showed the same correlations when the cohorts were investigated independently". This pattern is consistently observed across all the datasets which is reflected in the new Supplementary Figure 1. Further, a detailed comparison of equality versus inequality constraints, implemented in the CP based estimation, has been discussed above as responses to Reviewer 2 Comment 1 and 4 and incorporated in the manuscript.  
+
+Review 2 Comment 3 & Comment 7: Figure 1b: an interesting trend is observed during the period of change from deciduous to permanent teeth with a drastic increase in neutrophils. This point is further brought up with the longitudinal samples in Figure 4 (which also require revising the axes labels as some are missing). Do you have information about the teeth loss/replacement relative to the sample collection? The document points out that while BEC proportion declines with age, the consistency of this decline over time within the same individuals is not well established, potentially affecting the reliability of longitudinal studies using cheek swabs. That can strengthen the observation and explain some of the variability. Please expand on lines 207- 220 as needed.; Lines 298- 307: I agree with your discussion, but I would add the fact that the peaks of neutrophils are by the time of teeth exchange so maybe there is blood and more debris during this period that are changing the direction of your findings. Try to add as much context about these physiological changes as possible.  
+
+Authors Response: We thank the Reviewer for this very insightful suggestion and while we wholeheartedly agree, we unfortunately do not have that data recorded in any of the cohorts, nor any other oral health information. We hope that perhaps our manuscript motivates future cohorts to include a more comprehensive data collection on oral health and oral cells in children. However, we added tooth loss and replacement" as a potential explanation for the observed increases in variability we observed, as this was an excellent point. We also included in our discussion section: There are many other potential explanations for this variability, including tooth development, tooth loss and eruption, oral hygiene and habits, use of orthodontic or dental apparatuses, diet and environmental exposures, and other oral health concerns 8,9,32- 34. These events may also differ across developmental stages, rather than attributable to only explanation, such as the importance of tooth loss and eruption in childhood versus oral hygiene later in the pediatric period, but may point to any event that would increase blood in the mouth transiently."  
+
+Review 2 Comment 5: Figures 5 and 6: Is it possible to show the relation between chronological age and PedBe Age in months? That can help with the interpretation of the figures. Please correct the axis labels on 6b.  
+
+Authors Response: Thank you for pointing out this inconsistency, we have confirmed all the reported and estimated ages are in months for the all figures and fixed our labels.
+
+<--- Page Split --->
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+Review 2 Comment 6: Lines 239- 255: I would bring attention to the fact that this is a small cohort (and, as you mention, a proof- of- concept analysis). Please include the n in the text, not only in table 2.  
+
+Authors Response: We agree and appreciate this excellent suggestion from Reviewer 2. We have now added to our results section: As a proof- of- concept, we leveraged a publicly available cohort (GSE147058), though small with 22 monozygotic twin pairs (n=44 children), that had uniquely measured daily salivary cortisol concurrently with buccal DNA methylation samples enabling us to examine the potential importance of accounting for estimated BEC proportion when predicting PedBE EAA in the association with a relevant biological variable." We also reiterated this in the discussion.  
+
+Review 2 Comment 8: Lines 430- 439: What method for the CP estimation was used constrained or not constrained to equality?  
+
+Authors Response: We appreciate this thoughtful question from Reviewer 2 and their methodological rigor. We used the default inequality" setting embedded in HEpiDiSH, which is an iterative hierarchical procedure of EpiDiSH, for CP- based estimations. We have now included this relevant information in our Methods. The same inequality" setting was implemented in Houseman et al (2012) for cell type estimations. However, in our matched dataset analysis where we tested differences in the estimations between the two array platforms, we also compared the two constraints methods of CP (inequality versus equality), which is explained in detail in our response to Reviewer 2 Comment 1 and 4 and this excellent question in part overlap with a similar to comment also raised by Reviewer 1.  
+
+Reviewer #3 (Remarks to the Author):  
+
+Reviewer 3 Comment 1: The increased between- person BEC variability combined with the lack of within- person temporal stability is a key finding of this study. That is, the correlations for BEC proportions for the same individuals over even a brief period of time were not significantly strong. What are the potential implications of this lack of within- person stability for conducting future longitudinal research with the same individuals as well as future cross- sectional or other research combining independent, age- heterogeneous samples to examine between- person change? Does a within- person or between- person approach provide unique advantages for epigenetic aging research?  
+
+Authors' Response: We thank Reviewer 3 for highlighting the importance of this work. We agree that we should be more explicit in our interpretive recommendations based on the findings. As such we have added the following to the discussion Line 541- 559: "These data thus strongly support and expand our previous recommendation when creating the PedBE clock to correct for estimated BEC proportion when calculating PedBE EAA. If this is not possible, or not preferred, then we recommend evaluating for associations and heteroskedasticity with estimated BEC proportions within pediatric samples and correct models accordingly. Due to statistical concerns with confounding and heteroskedasticity across the pediatric age range, specifically due the increased variability observed with increased age, we would also extend this sentiment to all epigenetic age investigations employing pediatric cheek swabs. We also caution researchers in future studies to
+
+<--- Page Split --->
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+not assume BEC proportion stability in pediatric cohorts over time, especially when conducting longitudinal research or when combining independent, age- heterogeneous samples."  
+
+Reviewer 3 Comment 2: Please report the beta weights and corresponding \(p\) - values for models where: 1) epigenetic age is regressed on chronological age, and 2) epigenetic age is regressed on both chronological age and BEC proportions. These results, particularly for model two, have implications for the uniqueness of each predictor in explaining variation in epigenetic age estimates.  
+
+Authors' Response: Thank you for these excellent recommendations. We agree that this will be helpful for interpretation and have included them as requested. Further, upon examining the beta weights from epigenetic age regressed on chronological age, estimate BEC proportion, and chronological age and estimated BEC proportion together, we discovered what was likely a suppression effect (large changes in the estimated BEC proportion model coefficient when a sole predictor versus when in a model with chronological age, but little change to chronological age's model coefficients). This is not an effect we thought to look for, and, in fact, is an often overlooked and underreported statistical effect \(^{1}\) . We thank Reviewer 3, as this would not come to our attention if not for this comment. To investigate the possibility of this further, we examined the significance of each predictor alone and in combination in a linear regression of PedBE age. It was here that we found that in childhood estimated BEC proportion was always found to have a significant association with PedBE age, but in adolescence, though there was no significant correlation between estimated BEC proportion alone with PedBE age, in the joint model with chronological age, estimated BEC proportion was a significant predictor of PedBE age. Thus, we followed the literature guidelines on exploring such a possible effect by examining the model fit (adjusted \(R^2\) ), structure coefficients, and conducting a commonality analysis \(^{2}\) . We found that including estimated BEC proportion with chronological age improved model fit, as measured by adjusted R2, in both childhood and adolescence models of PedBE age, though this was a very small improvement in childhood (.96 to .97) and a larger improvement in adolescence (.63 to .71). We also found through structure coefficients, which are a measure of the correlation between a predictor variable and the predicted values of the outcome commonly used to investigate multicollinearity and statistical suppression, that estimated BEC proportion was influential and negatively associated with PedBE age uniquely from chronological age in childhood, but had almost no direct influence on PedBE age when accounting for chronological age in adolescence. Further, we conducted a commonality analysis, which decomposes the total explained variance ( \(R^2\) ) into the unique and common shared effects of model predictors, allowing for a detailed understanding of how each contributes to the overall model based on Reviewer 3's suggestion of the importance of the uniqueness of each predictor to PedBE age. This analysis suggested a classical suppression effect in childhood, where the unique contribution of estimated BEC proportion was low, but the common shared contribution of chronological age and estimated BEC proportion together was high, explaining 1/3 of the explained variance. However, in adolescence, where estimated BEC proportion did not significantly associated with PedBE age alone, we observed estimated BEC proportion contributed little alone and, in fact, negatively to the common shared contribution.  
+
+Together, these statistical indices suggest a classical statistical suppression effect in childhood, which estimated BEC proportion largely contributes to the common shared variance rather than uniquely but is still a strong predictor of PedBE age. These alternatively suggest a negative statistical suppression in adolescence, where estimated BEC proportion is not directly or uniquely associated with PedBE age, but does still contributes
+
+<--- Page Split --->
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+meaningfully to the model by assisting chronological age to explain more variance, apparent by the improvement in adjusted \(\mathsf{R}^2\) . Our interpretation of these results are that estimated BEC proportion makes a statistically meaningful contribution to the ability of chronological age to explain the variance of PedBE age throughout the pediatric period, though differently in childhood and adolescence. Given the primary use of PedBE age, or any epigenetic age estimation, in the literature is to compare residuals from the regression of chronological age on PedBE age, the more accurate this estimation, then the more likely the resulting statistical analyses are to be valid and replicable.  
+
+We have included this research in the results, discussion, and methods section of the manuscript now, which read:  
+
+## "Developmental period specific statistical suppression effects of estimated BEC proportion were observed on the chronological age prediction of PedBE age  
+
+To investigate the contribution of estimated BEC proportion to PedBE age prediction by chronological age in childhood and adolescence, we examined models with and without the inclusion of estimated BEC proportion as a covariate. In childhood samples \((< 120\) months/10years), we fit the a regression model of chronological age and estimated BEC proportions on PedBE age, which was statistically significant, \(\mathsf{F}(2,640) = 54,350\) \(p< 2.2\) \(\times 10^{- 16}\) , adjusted \(\mathsf{R}^2 = 0.97\) . Next, removing the estimated BEC proportion from the regression model to leave chronological age as the sole predictor of PedBE age, the model was still statistically significant, \(\mathsf{F}(1,641) =\) \(91,110\) \(p< 2.2\times 10^{- 16}\) , though with a slightly lower adjusted \(\mathsf{R}^2\) of 0.96. Similarly, we also fit these two regression models on the adolescent samples \((>120\) months/10years). While the regression model of chronological age and estimated BEC proportions predicting PedBE age was statistically significant, \(\mathsf{F}(2,640) =\) \(813.4\) \(p< 2.2\times 10^{- 16}\) \(\mathsf{R}^2 = 0.71\) , removing the estimated BEC proportion from the model and leaving chronological age as the sole predictor of PedBE age, while still statistically significant, \(\mathsf{F}(1,641) = 1097\) \(p< 2.2\) \(\times 10^{- 16}\) yielded a more substantially lower adjusted \(\mathsf{R}^2\) of 0.63.  
+
+When examining the model \(\beta\) coefficients, as expected the coefficient for chronological age was approximately 1, and this held true both in the sole predictor models and the models with estimated BEC proportions as a covariate, as well as in both developmental periods. This model coefficient indicates that for every unit of chronological age, there was a unit increase for PedBE age across all models. However, for samples in the infancy to childhood period, the model \(\beta\) coefficient of estimated BEC proportion as the sole predictor of PedBE age was - 220.14, significantly more negative than when in the model with chronological age (Table 2). Similarly, for the adolescence period samples, the model \(\beta\) coefficient when BEC proportion was the sole predictor of PedBE age was 4.58, much smaller than when in the model with chronological age (Table 2). These substantial differences in model \(\beta\) coefficients for estimated BEC proportion indicated the possibility of suppressor effects in the more complex regression models \(^{1,2}\) . Therefore, we examined the model \(\beta\) coefficients, structure coefficients, and commonality analysis results with the inclusion of estimated BEC proportion for the chronological age prediction of PedBE age in childhood and adolescence (Table 2).
+
+<--- Page Split --->
+
+Tel 604.875.3803 michael.kobor@ubc.ca 117- 2194 Health Sciences Mall Vancouver, B.C. V6T 1Z3  
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+Table 2. Model indices of developmental period specific prediction of PedBe age by chronological age and estimated BEC proportion. Beta model regression coefficients, structural coefficients, representing the correlation between a predictor variable and the predictor- criterion values, and the unique, shared, and total contribution of each predictor from the commonality analysis are presented.  
+
+<table><tr><td>Predictor</td><td>Beta (β) coefficient</td><td>Structured coefficient</td><td>Unique contribution</td><td>Shared contribution</td><td>Total contribution</td></tr><tr><td></td><td></td><td></td><td></td><td>Commonality Analysis</td><td></td></tr><tr><td>Chronological age</td><td>1.1942</td><td>0.9974</td><td>0.6517</td><td>0.3170</td><td>0.9737</td></tr><tr><td>Estimated BEC proportion</td><td>-220.1476</td><td>-0.5750</td><td>0.0049</td><td></td><td></td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>Chronological age</td><td>0.9247</td><td>0.9378</td><td>0.7173</td><td>-0.0860</td><td>0.7177</td></tr><tr><td>Estimated BEC proportion</td><td>4.5898</td><td>0.0231</td><td>0.0864</td><td></td><td></td></tr></table>  
+
+## Relevant associated discussion section (new and integrated):  
+
+The increases in beta coefficients when estimated BEC proportion is in the model with chronological age to predict PedBE, as well as examination of the model fit \((R^2)\) , structural coefficients, and commonality analysis, suggested suppression effects with estimated BEC proportion across the pediatric age range. Specifically, we observed a classical suppression effect in childhood and a negative suppression effect in adolescence. In childhood, the high common variance measured by the commonality analysis implies that chronological age and BEC proportion explained overlapping components of PedBE age variance. This indicates a classical suppression effect, where estimated BEC proportion enhanced the predictive power of chronological age by accounting for variance that chronological age cannot explain alone. This results in only a small improvement in model fit, as chronological age alone is already predictive of PedBE age and estimated BEC proportion is correlated with chronological age. However, in adolescence, the low common variance coupled with a more substantial model fit improvement, suggests a negative suppression effect, which is notably statistically rare \(^{1 - }\) \(^{3}\) . In this context, we observed estimated BEC proportion was slightly suppressing irrelevant variance in chronological age, despite their lack of association in adolescence, which subsequently allowed the model to better explain the variance in PedBE age. Together, these indices suggested that including estimated BEC proportion helps clarify the prediction of PedBE age made by chronological age by eliminating noise. As such, though for differing reasons, we recommend researchers to account for estimated BEC proportion when examining EAA in children and adolescents. This association of epigenetic age with cell type in early life may contribute to the relative accuracy of the PedBE clock in pediatric cheek swabs in comparison to other epigenetic clocks \(^{4}\) . Therefore, the power of epigenetic age as a biomarker is likely related to some extent to the associations between cell type proportions and the trait of interest \(^{5}\) . The expected changes in cell type proportion heterogeneity with age may not be synonymous to, but informative of biological age prediction.
+
+<--- Page Split --->
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+Due to the stronger association between estimated BEC proportion and predicted epigenetic age than chronological age, it was not surprising to find a moderate correlation between EAA and BEC proportion during childhood. However, the addition of differences in estimated BEC proportion as a covariate when calculating EAA led to the disappearance of the correlation between EAA and estimated BEC proportion. These data thus strongly support and expand our previous recommendation when creating the PedBE clock \(^4\) to correct for estimated BEC proportion when calculating PedBE EAA. Given the primary use of PedBE age, or any epigenetic age estimation, to compare these residuals (PedBE EAA) from the regression of chronological age on PedBE age, incorporation of estimated BEC proportion, in calculating EAA can lead to robust and accurate epigenetic age estimates. This, in turn, will likely enhance the validity and replicability of the resulting statistical analyses. If this is not possible, or not preferred, then we recommend evaluating for associations and heteroskedasticity with estimated BEC proportions within pediatric samples and correct models accordingly. Due to statistical concerns with confounding and heteroskedasticity across the pediatric age range, specifically due the increased variability observed with increased age, we would also extend this sentiment to all epigenetic age investigations employing pediatric cheek swabs. We also caution researchers in future studies to not assume BEC proportion stability in pediatric cohorts over time, especially when conducting longitudinal research or when combining independent, age- heterogeneous samples.  
+
+## New Addition to the Method:  
+
+Assessment of estimated BEC proportion contribution to PedBE predictionTo further understand the association of estimated BEC proportion to the prediction of PedBE age by chronological age, we compared the differences in model \(\beta\) coefficients for estimated BEC proportion across the two developmental periods. Upon discovering large discrepancies in estimated BEC proportion model \(\beta\) coefficients when a covariate and when a sole predictor, we investigated the possibility of statistical suppression \(^{1 - 3}\) . This exploration included the effects of estimated BEC proportion as a covariate to chronological age in predicting PedBE age, specifically: a) employing model indices including model fit (adjusted \(\mathbb{R}^2\) ) and b) structure coefficients extracted from the regression models, which were fitted using the stats R package \(^6\) . Structure coefficients reflect the correlation between a predictor variable and the outcome predicted values and reflects the unique strength of the predictor as specified by the model \(^1\) . Additionally, we conducted a commonality analysis using the yhat R package \(^7\) to dissect the unique contribution of the independent predictors and the shared contribution of the predictors in explaining the overall variance of the model \(^2\) ."  
+
+Reviewer 3 Comment 3 and 4: The implications of the current set of results suggest the importance of including BEC proportions when generating estimates of epigenetic age and epigenetic age acceleration during child development. Are the translational implications then for relevant epigenetic clock developers to adjust the available code (e.g. GitHub for PedBE) or online calculators (e.g. Horvath's) so that BEC proportions are now included in the generation of epigenetic aging estimates? Along these same lines, is it sufficient to adjust risk estimates of epigenetic aging on a particular outcome by including BEC proportions as an additional covariate/predictor in the statistical models? Or, must BEC proportions be included in the estimation of epigenetic aging?  
+
+Authors' Response: This is an excellent point. Given our updated findings with the new adolescent cohorts, this illustrates how both the correlation and variability of estimated BEC proportion differs across the pediatric age
+
+<--- Page Split --->
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+range. As such, it would be helpful these calculators or code to be data- dependent for the needs of all researchers. For example, this work would not have been possible without the ability to examine models with and without BEC proportions. However, to address the importance of the reviewer's point, we did update our discussion with the following Line 531: "This association of epigenetic age with cell type in early life may contribute to the relative accuracy of the PedBE clock in pediatric cheek swabs in comparison to other epigenetic clocks, especially at the youngest ages 10. Therefore, the power of epigenetic age as a biomarker is likely related to some extent to the associations between cell type proportions and the trait of interest 45. The expected natural changes in cell type proportion heterogeneity with age may not be synonymous to, but be informative of, biological age prediction." We also agree that correcting for estimated BEC proportion in the model is also a viable methodological option. While it may be more difficult to interpret EAA differences without BEC independence, if the model including estimated BEC proportions is being presented, this would be equally useful. As such we included in a line in the discussion (included in our earlier response to Reviewer 3): "These data thus strongly support and expand our previous recommendation when creating the PedBE clock4 to correct for estimated BEC proportion when calculating PedBE EAA. If this is not possible, or not preferred, then we recommend evaluating for associations and heteroskedasticity with estimated BEC proportions within pediatric samples and correct models accordingly".  
+
+Review 3 Comment 5: Please provide a brief rationale for why diurnal cortisol slope and obsessive- compulsive disorder status were examined as outcomes important for epigenetic aging. Similarly, a rationale for examining dental visits at 48- months, as opposed to any other age, was not provided in the original manuscript.  
+
+Authors' Response: We appreciate this suggestion to make the usefulness of these comparisons more evident. Regarding diurnal cortisol slope and OCD status we now state in Line 567: "These comparisons are indicative of two areas of research in which PedBE EAA has been applied, i.e., developmental perspectives on mental health and stress responses 13,17,46- 48. " This is to highlight that these are the types of analyses likely for pediatric EAA to be applied in the field and can act as proof- of- concept exemplars (as was also emphasized in response to Reviewer 2's comment). Regarding dental visits at 48 months in GUSTO, while we were grateful to have these data, this (along with the also included tooth brushing behavior) was unfortunately the only available concurrent oral health variable available across any of the ages. To our knowledge, this was not measured at the other timepoints in GUSTO, unfortunately, nor in the other available cohorts. Ideally, as mentioned earlier, future research will measure oral health in detail along with oral cells and DNA methylation in pediatric samples to further investigate the potential causal mechanisms of these observed trends.  
+
+Review 3 Comment 6: On line 273, the authors report a significant association between chronological age and the residuals of the model where PedBE age acceleration was regressed only on chronological age. As reported in other places in the manuscript, I expected this association would also be near zero given PedBE age acceleration is determined by removing or accounting for the effects of chronological age. Could the authors provide some insight into this finding? Perhaps there is a non- linear relation here? Please also report the magnitude of the association observed here.  
+
+Authors' Response: We appreciate the reviewers' attention to detail and accuracy. Indeed, no significant association was reported between chronological age and PedBE age acceleration residuals (p=0.5, r=0.02). The
+
+<--- Page Split --->
+
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+p- value \(= 0.02\) which was earlier reported was actually the strength of association and the corresponding p- value is 0.54. We have now modified our results accordingly Line 392.  
+
+Review 3 Comment 7: On line 160, the t- value reported is - 2.41 but the adjusted p- value is reported as greater than .05. In a typical t- distribution, a value greater than 1.96 in either direction would be statistically significant (less than .05). Could it be that the decimal is in the wrong place?  
+
+Authors' Response: Agreed. Thank you for pointing that out, we very much appreciate it. The t- value that was earlier reported was incorrect, the correct t- value is - .16 which is now corrected in the manuscript  
+
+Review 3 Comment 8: Please report means, standard deviations, and/or effect sizes in each of the figures.  
+
+Authors' Response: We thank the reviewer for this suggestion and have included these in the graphs and figure legends (including supplementary figure) throughout the manuscript.  
+
+Review 3 Comment 9: Several different terms were used to refer to BEC dimensions (e.g. estimated, inferred, computed). Is it possible to use one of these terms throughout the manuscript?  
+
+Authors' Response: Thank you for this feedback. We have chosen the word "estimated" and changed all references throughout the manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__daf022cfe2dadcc2063b0395d1f2d975d0e27498f5d78d4631ec409850e17903/supplementary_0_Peer Review file__daf022cfe2dadcc2063b0395d1f2d975d0e27498f5d78d4631ec409850e17903_det.mmd

@@ -0,0 +1,526 @@
+<|ref|>title<|/ref|><|det|>[[73, 53, 295, 80]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[74, 96, 296, 118]]<|/det|>
+Peer Review File  
+
+<|ref|>title<|/ref|><|det|>[[74, 158, 855, 202]]<|/det|>
+# Impact of Age-Related Changes in Buccal Epithelial Cells on Pediatric Epigenetic Biomarker Research  
+
+<|ref|>text<|/ref|><|det|>[[74, 214, 425, 231]]<|/det|>
+Corresponding Author: Dr Michael Kobor  
+
+<|ref|>text<|/ref|><|det|>[[72, 298, 864, 313]]<|/det|>
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.  
+
+<|ref|>text<|/ref|><|det|>[[73, 352, 144, 365]]<|/det|>
+Version 0:  
+
+<|ref|>text<|/ref|><|det|>[[73, 378, 219, 391]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 404, 160, 417]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 430, 238, 443]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 443, 920, 548]]<|/det|>
+Cheek swabs are often used as a source of DNA for studies of DNA methylation profiling. Because cellular heterogeneity within the tissue sampled can affect the DNA methylation data, changes in cellular composition of cheek swab samples with age could cause a confound in analysis of age- related epigenetic outcomes. Accordingly, the authors used existing Illumina HM450 and EPIC 850 data on cheek swab samples from 10 cohorts, representing children aged from 1 to 200 months, and estimated proportions of buccal epithelial cells (BEC) and various leukocyte subtypes to test for changes in cheek swab cellular composition with age. They conclude that "buccal epithelial cell proportions decrease reliably with age", and also that interindividual variation in the proportion of buccal epithelial cells increases with age, and that these findings have implications for studies utilizing cheek swab DNA for epigenetic research.  
+
+<|ref|>text<|/ref|><|det|>[[72, 560, 575, 574]]<|/det|>
+I have some concerns regarding the authors' interpretation of their results.  
+
+<|ref|>text<|/ref|><|det|>[[72, 585, 905, 651]]<|/det|>
+Examining the main data in the paper (Figure 1, a and b) it is apparent that the linear regression line is not a good fit to the data in either panel a or b. If the data are taken at face value, estimated BEC proportions decrease from birth to 50- 70 months of age, then go back up again by age 200 m. A reciprocal pattern is observed for estimated neutrophil proportions (Fig. 1b). How can the authors claim that "BEC proportions decrease reliably with age" when the majority of data from the oldest individuals (200m) are above the regression line?  
+
+<|ref|>text<|/ref|><|det|>[[72, 662, 923, 755]]<|/det|>
+Also, I am concerned that there are systematic differences in methylation data among the different cohorts analyzed, and that the authors have not adequately taken these into consideration. Most obviously, why would BEC proportion go down (and show more variation) at age 50 m, only to increase at 60 m, then go down again at 70 m? Most likely, much of this mysterious variation is due to systematic differences in the HM450 and EPIC 850 data generated at different facilities. In Figures 1a and 1b the authors should color code the data points according to which of the 10 cohorts the data were obtained from. This will clearly indicate to what extent their findings are confounded with technical (cohort- associated) biases, and suggest approaches for adjusting for these.  
+
+<|ref|>text<|/ref|><|det|>[[72, 767, 923, 808]]<|/det|>
+This problem is also evident in Fig. 3, in which the authors try to make the case that interindividual variation in BEC increases with age. By displaying cohort- specific data separately in panels a, b, and c, it is obvious that the greatest variation is between the SEED and GUSTO cohorts at age 48 m, rather than the age associations they are trying to highlight.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 833, 161, 846]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[72, 859, 922, 937]]<|/det|>
+(Remarks to the Author) The manuscript by Merrill et al. titled "Pediatric Buccal Epithelial Cell Proportions Decrease Reliably With Age: Considerations for PedBE in Pediatric Research" presents a comprehensive study on the dynamics of buccal epithelial cell (BEC) proportions across pediatric development, with a focus on their implications for epigenetic aging studies using cheek swab samples. The study analyzes DNA methylation profiles of 3,799 typically developing children ranging from 2 months to 20 years, drawn from ten independent cohorts.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[73, 60, 235, 73]]<|/det|>
+Key Points of the study:  
+
+<|ref|>text<|/ref|><|det|>[[72, 85, 911, 137]]<|/det|>
+BEC Proportions: BEC proportions decline with chronological age and predicted epigenetic age in pediatric populations. The study highlights that correcting for BEC proportion when calculating epigenetic age acceleration (EAA) affects the association of PedBE EAA with certain conditions like OCD and cortisol slope but warns of the potential associations with oral inflammatory conditions that may be driven by socio- economic factors.  
+
+<|ref|>text<|/ref|><|det|>[[72, 137, 920, 176]]<|/det|>
+Variability and Influences: Increased variability in BEC proportions was noted with age. Factors influencing BEC proportions include oral health practices and visits to the dentist, but not daily tooth brushing. The study did not find significant associations with sex, mode of delivery, or breastfeeding status.  
+
+<|ref|>text<|/ref|><|det|>[[72, 176, 920, 216]]<|/det|>
+Methodological Approaches: EpiDISH, a reference- based bioinformatics method, was used to estimate the cell type proportions. Variance in BEC proportion was assessed through longitudinal analysis, showing a trend of increased variance with age.  
+
+<|ref|>text<|/ref|><|det|>[[72, 216, 910, 256]]<|/det|>
+Implications for Research: The findings suggest that pediatric cheek swab samples exhibit significant changes in cellular composition that could influence epigenetic age estimates. These changes need to be considered in epigenetic studies to avoid potential biases.  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 270, 153, 282]]<|/det|>
+## Comments:  
+
+<|ref|>text<|/ref|><|det|>[[72, 282, 884, 309]]<|/det|>
+Lines 140- 153: Could you comment on the integration of EpiDiSH for 450K and EPIC? Were the results different when removing the nonoverlapping probes across the two arrays?  
+
+<|ref|>text<|/ref|><|det|>[[72, 309, 920, 360]]<|/det|>
+Figures 1A, Supple Fig 1: Could you add some information about the correlation of the results across the different methods? One area that seems quite divergent is for children 150- 200 months, but in general, the constrained projection seems to diverge compared to the CBS- RPC results for the most abundant epithelial cells. Is this a problem of scaling to one, or are there other differences that might be affecting the results?  
+
+<|ref|>text<|/ref|><|det|>[[72, 360, 921, 477]]<|/det|>
+Figure 1b: an interesting trend is observed during the period of change from deciduous to permanent teeth with a drastic increase in neutrophils. This point is further brought up with the longitudinal samples in Figure 4 (which also require revising the axes labels as some are missing). Do you have information about the teeth loss/replacement relative to the sample collection? The document points out that while BEC proportion declines with age, the consistency of this decline over time within the same individuals is not well established, potentially affecting the reliability of longitudinal studies using cheek swabs. That can strengthen the observation and explain some of the variability. Please expand on lines 207- 220 as needed. Supplementary Figure 3: I would try to understand this better. Why is this variation occurring? is it only due to scaling to one? what happens when you are restricted to the overlapping probes across both platforms? Are the missing probes in EPIC biasing some of the results?  
+
+<|ref|>text<|/ref|><|det|>[[72, 477, 900, 504]]<|/det|>
+Figures 5 and 6: Is it possible to show the relation between chronological age and PedBe Age in months? That can help with the interpretation of the figures. please correct the axis labels on 6b.  
+
+<|ref|>text<|/ref|><|det|>[[72, 504, 920, 530]]<|/det|>
+Lines 239- 255: I would bring attention to the fact that this is a small cohort (and, as you mention, a proof of concept analysis). Please include the n in the text, not only in table 2.  
+
+<|ref|>text<|/ref|><|det|>[[72, 530, 920, 570]]<|/det|>
+Lines 298- 307: I agree with your discussion, but I would add the fact that the peaks of neutrophils are by the time of teeth exchange so maybe there is blood and more debris during this period that are changing the direction of your findings. Try to add as much context about these physiological changes as possible.  
+
+<|ref|>text<|/ref|><|det|>[[72, 570, 784, 583]]<|/det|>
+Lines 430- 439: What method for the CP estimation was used constrained or not constrained to equality?  
+
+<|ref|>sub_title<|/ref|><|det|>[[72, 633, 161, 645]]<|/det|>
+## Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[72, 660, 238, 672]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[72, 672, 920, 750]]<|/det|>
+The increased between- person BEC variability combined with the lack of within- person temporal stability is a key finding of this study. That is, the correlations for BEC proportions for the same individuals over even a brief period of time were not significantly strong. What are the potential implications of this lack of within- person stability for conducting future longitudinal research with the same individuals as well as future cross- sectional or other research combining independent, age- heterogeneous samples to examine between- person change? Does a within- person or between- person approach provide unique advantages for epigenetic aging research?  
+
+<|ref|>text<|/ref|><|det|>[[72, 762, 903, 815]]<|/det|>
+Please report the beta weights and corresponding p- values for models where: 1) epigenetic age is regressed on chronological age, and 2) epigenetic age is regressed on both chronological age and BEC proportions. These results, particularly for model two, have implications for the uniqueness of each predictor in explaining variation in epigenetic age estimates.  
+
+<|ref|>text<|/ref|><|det|>[[72, 827, 914, 880]]<|/det|>
+The implications of the current set of results suggest the importance of including BEC proportions when generating estimates of epigenetic age and epigenetic age acceleration during child development. Are the translational implications then for relevant epigenetic clock developers to adjust the available code (e.g. GitHub for PedBE) or online calculators (e.g. Horvath's) so that BEC proportions are now included in the generation of epigenetic aging estimates?  
+
+<|ref|>text<|/ref|><|det|>[[72, 892, 914, 932]]<|/det|>
+Along these same lines, is it sufficient to adjust risk estimates of epigenetic aging on a particular outcome by including BEC proportions as an additional covariate/predictor in the statistical models? Or, must BEC proportions be included in the estimation of epigenetic aging?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[73, 46, 911, 88]]<|/det|>
+Please provide a brief rationale for why diurnal cortisol slope and obsessive- compulsive disorder status were examined as outcomes important for epigenetic aging. Similarly, a rationale for examining dental visits at 48- months, as opposed to any other age, was not provided in the original manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[73, 99, 920, 166]]<|/det|>
+On line 273, the authors report a significant association between chronological age and the residuals of the model where PedBE age acceleration was regressed only on chronological age. As reported in other places in the manuscript, I expected this association would also be near zero given PedBE age acceleration is determined by removing or accounting for the effects of chronological age. Could the authors provide some insight into this finding? Perhaps there is a non- linear relation here? Please also report the magnitude of the association observed here.  
+
+<|ref|>text<|/ref|><|det|>[[73, 177, 914, 218]]<|/det|>
+On line 160, the t- value reported is - 2.41 but the adjusted p- value is reported as greater than .05. In a typical t- distribution, a value greater than 1.96 in either direction would be statistically significant (less than .05). Could it be that the decimal is in the wrong place?  
+
+<|ref|>text<|/ref|><|det|>[[73, 229, 640, 244]]<|/det|>
+Please report means, standard deviations, and/or effect sizes in each of the figures.  
+
+<|ref|>text<|/ref|><|det|>[[73, 254, 923, 283]]<|/det|>
+Several different terms were used to refer to BEC dimensions (e.g. estimated, inferred, computed). Is it possible to use one of these terms throughout the manuscript?  
+
+<|ref|>text<|/ref|><|det|>[[73, 295, 144, 308]]<|/det|>
+Version 1:  
+
+<|ref|>text<|/ref|><|det|>[[73, 321, 219, 334]]<|/det|>
+Reviewer comments:  
+
+<|ref|>text<|/ref|><|det|>[[73, 347, 161, 360]]<|/det|>
+Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[73, 373, 640, 400]]<|/det|>
+(Remarks to the Author) I am satisfied with the changes the authors have made in response to my concerns.  
+
+<|ref|>text<|/ref|><|det|>[[73, 412, 161, 425]]<|/det|>
+Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[73, 438, 238, 451]]<|/det|>
+(Remarks to the Author)  
+
+<|ref|>text<|/ref|><|det|>[[73, 450, 920, 515]]<|/det|>
+The manuscript by Merrill et al. titled "Pediatric Buccal Epithelial Cell Proportions Decrease Reliably With Age From Infancy through Childhood: Considerations for PedBE in Pediatric Research" presents a comprehensive study on the dynamics of buccal epithelial cell (BEC) proportions across pediatric development, with a focus on their implications for epigenetic aging studies using cheek swab samples. The study analyzes DNA methylation profiles of 4,626 typically developing children ranging from 2 months to 20 years, drawn from ten independent cohorts.  
+
+<|ref|>text<|/ref|><|det|>[[73, 527, 670, 541]]<|/det|>
+the authors have responded to all my questions. i have no additional recommendations.  
+
+<|ref|>text<|/ref|><|det|>[[73, 897, 915, 938]]<|/det|>
+Open Access This Peer Review File is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 48, 117, 60]]<|/det|>
+made.  
+
+<|ref|>text<|/ref|><|det|>[[72, 60, 910, 125]]<|/det|>
+In cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source. The images or other third party material in this Peer Review File are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.  
+
+<|ref|>text<|/ref|><|det|>[[72, 125, 618, 139]]<|/det|>
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 191, 366, 207]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[90, 219, 895, 391]]<|/det|>
+Cheek swabs are often used as a source of DNA for studies of DNA methylation profiling. Because cellular heterogeneity within the tissue sampled can affect the DNA methylation data, changes in cellular composition of cheek swab samples with age could cause a confound in analysis of age- related epigenetic outcomes. Accordingly, the authors used existing Illumina HM450 and EPIC 850 data on cheek swab samples from 10 cohorts, representing children aged from 1 to 200 months, and estimated proportions of buccal epithelial cells (BEC) and various leukocyte subtypes to test for changes in cheek swab cellular composition with age. They conclude that buccal epithelial cell proportions decrease reliably with age", and also that interindividual variation in the proportion of buccal epithelial cells increases with age, and that these findings have implications for studies utilizing cheek swab DNA for epigenetic research.  
+
+<|ref|>text<|/ref|><|det|>[[92, 404, 635, 420]]<|/det|>
+I have some concerns regarding the authors interpretation of their results.  
+
+<|ref|>text<|/ref|><|det|>[[90, 435, 904, 551]]<|/det|>
+Review 1 Comment 1: Examining the main data in the paper (Figure 1, a and b) it is apparent that the linear regression line is not a good fit to the data in either panel a or b. If the data are taken at face value, estimated BEC proportions decrease from birth to 50- 70 months of age, then go back up again by age 200 m. A reciprocal pattern is observed for estimated neutrophil proportions (Fig. 1b). How can the authors claim that BEC proportions decrease reliably with age" when the majority of data from the oldest individuals (200m) are above the regression line?  
+
+<|ref|>text<|/ref|><|det|>[[90, 565, 907, 783]]<|/det|>
+Authors Response: We appreciate Reviewer 1 s concern and astute insight into the data in bringing up an important point. We certainly agree that the oldest" pediatric individuals in our dataset did not conform well to the regression line. Given that we did not have a substantial representation of pediatric samples in the adolescence period, it was difficult to assess the association of estimated buccal epithelial proportions with age in the older" pediatric individuals using the initial cohorts from our original submission. To address this important issue, we have now included two new cohorts: BIBO and Cohort 12, which substantially filled the gap in the pre- adolescence to late adolescence period. While BIBO is a longitudinal cohort with samples at three time- points: 72 months, 120 months, 168 months, Cohort 12 is comprised of adolescent females between 138- 220 months. Our new assembled "full" cohort of 4626 samples from 12 independent cohorts with ages spanning across the pediatric developmental window has now enabled us to examine the relationship of estimated buccal epithelial proportion with age both at childhood and adolescence.  
+
+<|ref|>text<|/ref|><|det|>[[90, 794, 905, 904]]<|/det|>
+Using this more complete coverage across the pediatric age range, we conducted a spline regression analysis to determine if our model fit, ascertained by adjusted R- squared, significantly improved by allowing different slopes during childhood and adolescence - directly based on this observation and suggestion of Reviewer 1. By including a knot at 120 months/10 years, which represents the developmentally relevant age of transition from childhood to adolescence, our model fit was markedly improved from 0.17 (no knot, linear regression) to 0.25 (knot at 10 years, spline regression). Because the transition from childhood to adolescence is not at one age,
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 68, 616, 88]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>title<|/ref|><|det|>[[128, 100, 559, 140]]<|/det|>
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[90, 162, 905, 327]]<|/det|>
+per se, but defined as a developmental period, encompassing the ages around 10 as well, we examine the model fit by including the knot at 9 years, 11 years and 12 years independently. However, irrespective of when the knot was included (10 years/11 years/12 years), an adjusted R- squared of 0.25 was reported compared to 0.17 which was noted when one linear regression line was fitted across all the samples in the pediatric window. Upon using a cut- off of 10 years to split our pediatric population into i) infancy to childhood (0- 9 years \(< 120\) months) and ii) adolescence, including pre- through the end of adolescence (10- 18 years \(>120\) months), we observed an even stronger negative linear association between estimated BEC proportion and chronological age throughout infancy to end of childhood. However, this association was no longer observed in pediatric samples around adolescence.  
+
+<|ref|>text<|/ref|><|det|>[[90, 336, 900, 490]]<|/det|>
+This interesting set of findings which reaffirms our previous result of an association of estimated BEC proportion with age, but now highlights the significance of this association at a particular developmental period, specifically from the start of infancy to the end of childhood, but not in adolescence. As such, we now have significantly modified our second section of results Line 153- 194, made substantial changes to our Figure 1 and added Supplementary Figure 3 to accurately reflect these results. Accordingly, we have also adjusted our manuscript title to explicitly capture the developmental period when the association is consistently observed by specifying this association is from Infancy to Childhood" and removed the word reliably" in the title and throughout our manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[91, 500, 900, 555]]<|/det|>
+We thank Reviewer 1 once again for asking these insightful questions that we believe have substantially improved the impact and accuracy of the manuscript, and we hope they have been appropriately addressed in our response and manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[91, 565, 886, 620]]<|/det|>
+Figure 1: Significant association of estimated buccal epithelial cell (BEC) proportions with reported chronological age in the pediatric window. A) In the scatterplot, estimated BEC proportions estimated by the EpiDISH- RPC method was plotted on the y axis against chronological age in months on the x axis. Significant  
+
+<|ref|>image<|/ref|><|det|>[[145, 642, 860, 920]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 163, 904, 255]]<|/det|>
+decrease in estimated BEC proportion with reported chronological age was observed in pediatric buccal swabs. B) Pediatric samples from infancy to childhood, using a cut- off of \(< 120\) months (10 years), showed a stronger negative linear relationship between estimated BEC proportions and chronological age (reported in months). C) Pediatric samples in adolescence ( \(>120\) months/10 years) exhibited no association between estimated BEC proportion and chronological age (reported in months).  
+
+<|ref|>text<|/ref|><|det|>[[90, 264, 894, 319]]<|/det|>
+Supplementary Figure 3: Spline regression analysis by including a knot at age 10 to signify the transition from childhood to adolescence, showed improved model fit, ascertained by a higher adjusted R- squared of 0.25, compared to linear regression which yielded an adjusted R- squared of 0.17.  
+
+<|ref|>image<|/ref|><|det|>[[268, 336, 707, 567]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[90, 585, 904, 787]]<|/det|>
+Reviewer 1 Comments 2 & 3: Also, I am concerned that there are systematic differences in methylation data among the different cohorts analyzed, and that the authors have not adequately taken these into consideration. Most obviously, why would BEC proportion go down (and show more variation) at age 50 m, only to increase at 60 m, then go down again at 70 m? Most likely, much of this mysterious variation is due to systematic differences in the HM450 and EPIC 850 data generated at different facilities. In Figures 1a and 1b the authors should color code the data points according to which of the 10 cohorts the data were obtained from. This will clearly indicate to what extent their findings are confounded with technical (cohort- associated) biases, and suggest approaches for adjusting for these. This problem is also evident in Fig. 3, in which the authors try to make the case that interindividual variation in BEC increases with age. By displaying cohort- specific data separately in panels a, b, and c, it is obvious that the greatest variation is between the SEED and GUSTO cohorts at age 48 m, rather than the age associations they are trying to highlight.  
+
+<|ref|>text<|/ref|><|det|>[[90, 799, 896, 912]]<|/det|>
+Authors Response: We agree with the reviewer that there is diversity, as expected amongst our 12 independent cohorts both in terms of the sample sizes and demographic characteristics including age which is reflected in Table 1. Importantly, the DNA methylation data which was used to estimate cell type proportions and epigenetic ages was measured using the same platform: Illumina microarrays, either with the Illumina Infinium HumanMethylation450 BeadChip (450K) or MethylationEPIC BeadChip (850K). Both 450K and 850K have identical probe type chemistry to capture DNA methylation intensities of the CpGs and except the one
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 68, 616, 88]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>title<|/ref|><|det|>[[128, 100, 559, 140]]<|/det|>
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[90, 162, 905, 327]]<|/det|>
+GEO cohort, all the cohort samples were run in the same facility (Kobor Lab). Further, samples across all the cohorts underwent the same preprocessing steps as outlined in the Methods. However, the majority of the cohorts are confounded by the chronological age at which buccal samples were collected and as expected, the individual cohorts did not have a widespread representation of all the ages across the pediatric window, except the four longitudinal cohorts (BEPAC, SEED, GUSTO and BIBO) which exhibited some level of spread in terms of age, though in their nicked range. As suggested by Reviewer 1, we have now also color coded the data points based on the cohorts to demonstrate that the negative association between estimated BEC proportions and age held true for the majority of the cohorts (Supplementary Figure 4) and was largely not influenced by differences in 450K or 850K array.  
+
+<|ref|>image<|/ref|><|det|>[[88, 332, 660, 599]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[90, 611, 893, 817]]<|/det|>
+We appreciate the Reviewer's concern regarding differences in 450K versus 850K, some of which may be then reflected in the estimated buccal cell type proportions. Given that the EpiFibIC reference (716 CpGs) used by the EpiDISH R package is based on 450K array, inevitably there were missing CpGs (36 CpGs) when we used EpiDISH on our 850K cohort. To investigate this, we had used the strength of a small but matched subset of samples ( \(n = 24\) matched pairs) from the APrON FetalPro and APrON NeuroTox cohorts for which we had obtained DNA methylation measurements using 450K and 850K arrays. We would like to highlight our work reported earlier in the manuscript that the estimated buccal cell type proportions using the RPC method, the method we use to report all our findings in the manuscript, showed no significant differences based on the array. To specifically test the impact of the missing EpiFibIC CpGs from the 850K array, we now tested these differences in our matched dataset by calculating the cell type proportions using different sets of reference CpGs:  
+
+<|ref|>text<|/ref|><|det|>[[118, 828, 905, 910]]<|/det|>
+1. Using the EpiFibIC reference with 716 CpGs on 450K datasets and the same EpiFibIC reference excluding the 36 missing CpGs on the 850K datasets2. Restricting the EpiFibIC reference to the common/shared CpGs between the arrays and using this set of CpGs on both 450K/850K
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 68, 615, 88]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>title<|/ref|><|det|>[[128, 100, 559, 140]]<|/det|>
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[90, 161, 904, 389]]<|/det|>
+Estimated proportions of BEC by EpiDISH- RPC did not differ based on differing number of CpGs employed in the deconvolution EpiFibIC reference panel, for i) (t (23) = 1.1, \(p_{\mathrm{adj}} > 0.05\) ) and for ii) (t (23) = - .16, \(p_{\mathrm{adj}} > 0.05\) ). Overall, EpiDISH- RPC method consistently predicted higher BEC proportions compared to the other methods (Supplementary Figure 7). Further, CBS estimated BEC proportions showed no significant differences based on the array type, only when the cell type estimates were calculated by limiting the EpiFibIC reference to the shared/common CpGs between the arrays. In contrast, in the context of CP- based cell type algorithm, employing either equality or inequality constraint, we noted significant differences between 450K and 850K calculated BEC proportions, with a large Cohens s d effect size (>1), in both the scenarios of using all the CpGs in the EpiFibIC reference versus restricting the EpiFibIC reference to the common/shared CpGs between the arrays. However, the magnitude of difference between the arrays based on CP estimated BEC proportions was relatively larger when differing numbers of CpGs on the EpiFibIC reference were used to calculate the estimates. (Figure below: Supplementary figure 7). This is now included in Results Line 196- 223.  
+
+<|ref|>text<|/ref|><|det|>[[90, 398, 890, 490]]<|/det|>
+Supplementary Figure 7: Comparison of buccal epithelial cell (BEC) proportions estimated by EpiDISH_RPC, EpiDISH_CBS, EpiDISH_CP_equality constraint and EpiDISH_CP_inequality constraint methods across the two DNA methylation platforms based on A) shared/common CpGs (EpiFibIC reference) between 450K and 850K arrays and B) using all the EpiFibIC reference CpGs for the 450K samples and with 36 missing CpGs from the reference for the 850K samples.  
+
+<|ref|>text<|/ref|><|det|>[[89, 502, 904, 647]]<|/det|>
+For A), the mean BEC proportions with EpiDISH_RPC on the 450K array was 0.99 0.0 and on the 850K array it was 0.99 0.01. For EpiDISH_CBS, the mean BEC proportion on the 450K array was 0.98 0.01, while on the 850K array it was 0.99 0.01. For EpiDISH_CP with equality constraint, the mean BEC proportion on the 450K array was 0.82 0.01 and on the 850K array it was 0.8 0.01 and with Cohen s d = 1.85. For EpiDISH_CP with inequality constraint, the mean BEC proportion on the 450K array was 0.78 0.01 and on the 850K array it was 0.77 0.01 and with Cohen s d = 1.69.  
+
+<|ref|>text<|/ref|><|det|>[[89, 661, 904, 830]]<|/det|>
+For B), the mean BEC proportions with EpiDISH_RPC on the 450K array was 0.99 0.01 and on the 850K array it was 0.99 0.01. For EpiDISH_CBS, the mean BEC proportion on the 450K array was 0.98 0.01, while on the 850K array it was 0.99 0.01. With a moderate Cohen s d = 0.56. For EpiDISH_CP with equality constraint, the mean BEC proportion on the 450K array was 0.85 0.01 and on the 850K array it was 0.80 0.01 and with a large Cohen s d = 5.23. For EpiDISH_CP with inequality constraint, the mean BEC proportion on the 450K array was 0.78 0.01 and on the 850K array it was 0.77 0.01 and with Cohen s d = 1.69. \* Indicate significant adjusted p- value < 0.05
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 68, 615, 88]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>text<|/ref|><|det|>[[128, 99, 559, 140]]<|/det|>
+Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>image<|/ref|><|det|>[[93, 166, 864, 368]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[90, 394, 896, 595]]<|/det|>
+Further, at cohort- specific level, we tested whether the differences in EpiFibIC reference CpGs impacted the estimations of the cell type proportions in samples ran across different arrays. To do this, we re- calculated the cell type proportions on the 450K cohorts (5/12) based only on the shared/common CpGs of the reference between 450K and 850K, creating a simulated 850K scenario within our 450K cohorts and then compared the cell type proportions to those obtained previously in the 450K cohorts using all the CpGs in the EpiFibIC reference. Across all the 450K cohorts we observed no significant differences in cell type estimates (see below are examples of two 450K cohorts: GECKO and PAWS), using both RPC and CBS methods, depending on whether we used the EpiFibIC reference with the complete set of CpGs or if the 36 missing CpGs on the 850K were selectively excluded from the reference. In the context of CP based cell type estimations we observed significant differences in both buccal epithelial proportions and fibroblasts proportions, irrespective of adopting an equality constraint or inequality constraint.  
+
+<|ref|>image<|/ref|><|det|>[[90, 602, 860, 809]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[90, 825, 900, 904]]<|/det|>
+We also added sentences to the limitations section of the discussion Line 586- 597 to directly address this concern of systematic differences across the cohorts: While the cohorts were processed together, and several cohorts were characterized by the same laboratory, the combining of multiple cohorts (450K and 850K) rather than one, large, uniform collection likely introduced some technical variation at this level. However, in the
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 68, 615, 88]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>title<|/ref|><|det|>[[128, 100, 559, 140]]<|/det|>
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[90, 162, 907, 290]]<|/det|>
+context of DNA methylation based calculated cell types, we demonstrate that EpiDISH- RPC estimated BEC proportions are congruent between arrays, irrespective of whether all the CpGs in the EpiFibIC reference are used for BEC estimations or limiting the references to the shared CpGs between the arrays. This finding highlights EpiDISH- RPC as the robust algorithm for BEC estimations, specifically when collating proportions across the two platforms. Although four large longitudinal cohorts were available, the inclusion of more repeated samples with additional time points and using identical methods of collection and storage would have enhanced the robustness of the cell type proportion variability findings."  
+
+<|ref|>text<|/ref|><|det|>[[92, 330, 365, 347]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[91, 357, 893, 450]]<|/det|>
+The manuscript by Merrill et al. titled "Pediatric Buccal Epithelial Cell Proportions Decrease Reliably With Age: Considerations for PedBE in Pediatric Research" presents a comprehensive study on the dynamics of buccal epithelial cell (BEC) proportions across pediatric development, with a focus on their implications for epigenetic aging studies using cheek swab samples. The study analyzes DNA methylation profiles of 3,799 typically developing children ranging from 2 months to 20 years, drawn from ten independent cohorts.  
+
+<|ref|>text<|/ref|><|det|>[[92, 460, 263, 477]]<|/det|>
+Key Points of the study:  
+
+<|ref|>text<|/ref|><|det|>[[91, 487, 894, 579]]<|/det|>
+BEC Proportions: BEC proportions decline with chronological age and predicted epigenetic age in pediatric populations. The study highlights that correcting for BEC proportion when calculating epigenetic age acceleration (EAA) affects the association of PedBE EAA with certain conditions like OCD and cortisol slope but warns of the potential associations with oral inflammatory conditions that may be driven by socio- economic factors.  
+
+<|ref|>text<|/ref|><|det|>[[91, 589, 896, 644]]<|/det|>
+Variability and Influences: Increased variability in BEC proportions was noted with age. Factors influencing BEC proportions include oral health practices and visits to the dentist, but not daily tooth brushing. The study did not find significant associations with sex, mode of delivery, or breastfeeding status.  
+
+<|ref|>text<|/ref|><|det|>[[91, 654, 895, 709]]<|/det|>
+Methodological Approaches: EpiDISH, a reference- based bioinformatics method, was used to estimate the cell type proportions. Variance in BEC proportion was assessed through longitudinal analysis, showing a trend of increased variance with age.  
+
+<|ref|>text<|/ref|><|det|>[[91, 720, 903, 774]]<|/det|>
+Implications for Research: The findings suggest that pediatric cheek swab samples exhibit significant changes in cellular composition that could influence epigenetic age estimates. These changes need to be considered in epigenetic studies to avoid potential biases.  
+
+<|ref|>text<|/ref|><|det|>[[91, 784, 890, 875]]<|/det|>
+Review 2 Comments 1 and 4: Lines 140- 153: Could you comment on the integration of EpiDISH for 450K and EPIC? Were the results different when removing the nonoverlapping probes across the two arrays? For Supplementary Figure 3: I would try to understand this better. Why is this variation occurring? is it only due to scaling to one? What happens when you are restricted to the overlapping probes across both platforms? Are the missing probes in EPIC biasing some of the results?
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 68, 616, 88]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>title<|/ref|><|det|>[[128, 100, 559, 140]]<|/det|>
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[90, 165, 901, 285]]<|/det|>
+Authors Response: We thank Reviewer 2 for asking an excellent question about the integration of EpiDiSH for 450K and 850K profiled samples, which is important for the scientific community conducting DNA methylation based epigenetic research. Further, Reviewer 2 also highlights an important aspect regarding the concordance in findings when only the shared CpGs (EpiFibIC reference) between 450K and 850K are used in cell type estimations. We appreciate the reviewers curiosity to understand the source of variation noted in Supplementary Figure 3 (now Supplementary Figure 7).  
+
+<|ref|>text<|/ref|><|det|>[[90, 294, 904, 515]]<|/det|>
+Because both the comments are related to the figure and the associated findings, we are addressing this in the same response. Similar to our previous response to reviewer 1 comment 2, we recognize that the EpiFibIC reference (716 CpGs) used by the EpiDiSH R package is based on 450K array, so inevitably there were missing CpGs (36 CpGs) when we used EpiDiSH on our 850K cohorts. To demonstrate the influence of missing EpiDiSH CpGs for calculation of estimates on the 850K samples we used the strength of our matched matched subset of samples ( \(n = 24\) pairs) from APron FetalPro and APron NeuroTox cohorts for which we had DNA methylation measurements using both 450K and 850K. We had earlier shown that the buccal epithelial cell type proportions were comparable based on the RPC method of calculation, though differed based on CBS and CP (inequality constraint) methods. While in this case all the EpiFibIC reference CpGs were available for the 450K samples, there were 36 missing CpGs from the reference for the 850K samples. This may have contributed to some level of differences observed in the CBS and CP (inequality) estimations. For ease of navigation, we copy here the relevant portion included in our response to Reviewer 1:  
+
+<|ref|>text<|/ref|><|det|>[[90, 542, 908, 878]]<|/det|>
+To investigate this, we re- calculated our buccal cell type proportions for the matched dataset based on the shared/common CpGs (EpiFibIC reference) between 450K and 850K arrays. In addition, for CP based estimation we also compared between adopting an equality versus inequality constraint to explore its subsequent impact on the cell type estimates. We noted that irrespective of whether or not all the CpGs in the EpiFibIC reference were used to calculate the buccal cell type proportions, RPC estimates were congruent across both 450K and 850K measurements and overall predicted higher buccal cell type proportions compared to the other methods. Alternatively, in the context of CP based cell type proportion algorithm, employing either equality or inequality constraint, we noted significant differences between 450K and 850K calculated estimates, with a large Cohens s d effect sizes \((>1)\) , in both the scenarios of either using all the CpGs in the reference versus restricting the EpiFibIC reference to the common/shared CpGs between the arrays (Figure below). However, the magnitude of difference between the arrays based on CP estimated buccal cell type proportions were larger in the situation when differing numbers of CpGs on the EpiFibIC reference were used to calculate the estimates on 450K versus 850K samples. Further, CBS estimated buccal epithelial proportions showed no differences based on the array, only when the cell type estimates were calculated by limiting the EpiFibIC reference to the shared/common CpGs between the arrays. This work suggests the importance of using the common/shared CpGs in the EpiFibIC reference to calculate cell type estimates when comparing samples across the two platforms, specifically if a researcher chooses to use the CBS or CP algorithms. These findings are now in Results Line 196- 223 and included as a Supplementary Figure 7.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 70, 614, 89]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>title<|/ref|><|det|>[[128, 101, 559, 140]]<|/det|>
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[91, 162, 888, 254]]<|/det|>
+Supplementary Figure 7: Comparison of buccal epithelial cell (BEC) proportions estimated by EpiDISH_RPC, EpiDISH_CBS, EpiDISH_CP_equality constraint and EpiDISH_CP_inequality constraint methods across the two DNA methylation platforms based on A) shared/common CpGs (EpiFibIC reference) between 450K and 850K arrays and B) using all the EpiFibIC reference CpGs for the 450K samples and with 36 missing CpGs from the reference for the 850K samples.  
+
+<|ref|>text<|/ref|><|det|>[[90, 266, 904, 410]]<|/det|>
+For A), the mean BEC proportions with EpiDISH_RPC on the 450K array was 0.99 0.0 and on the 850K array it was 0.99 0.01. For EpiDISH_CBS, the mean BEC proportion on the 450K array was 0.98 0.01, while on the 850K array it was 0.99 0.01. For EpiDISH_CP with equality constraint, the mean BEC proportion on the 450K array was 0.82 0.01 and on the 850K array it was 0.8 0.01 and with Cohen s d = 1.85. For EpiDISH_CP with inequality constraint, the mean BEC proportion on the 450K array was 0.78 0.01 and on the 850K array it was 0.77 0.01 and with Cohen s d = 1.69.  
+
+<|ref|>text<|/ref|><|det|>[[90, 425, 904, 571]]<|/det|>
+For B), the mean BEC proportions with EpiDISH_RPC on the 450K array was 0.99 0.01 and on the 850K array it was 0.99 0.01. For EpiDISH_CBS, the mean BEC proportion on the 450K array was 0.98 0.01, while on the 850K array it was 0.99 0.01. With a moderate Cohen s d = 0.56. For EpiDISH_CP with equality constraint, the mean BEC proportion on the 450K array was 0.85 0.01 and on the 850K array it was 0.80 0.01 and with a large Cohen s d = 5.23. For EpiDISH_CP with inequality constraint, the mean BEC proportion on the 450K array was 0.78 0.01 and on the 850K array it was 0.77 0.01 and with Cohen s d = 1.69.  
+
+<|ref|>image<|/ref|><|det|>[[90, 588, 860, 790]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[90, 813, 904, 904]]<|/det|>
+Review 2 Comment 2: Figures 1A, Supple Fig 1: Could you add some information about the correlation of the results across the different methods? One area that seems quite divergent is for children 150- 200 months, but in general, the constrained projection seems to diverge compared to the CBS- RPC results for the most abundant epithelial cells. Is this a problem of scaling to one, or are there other differences that might be affecting the results?
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 68, 616, 88]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>title<|/ref|><|det|>[[128, 100, 559, 140]]<|/det|>
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[90, 165, 904, 348]]<|/det|>
+Authors Response: Thank you for inquiring on the correlation specifics across the different methods of cell type estimation. We had previously included those details in the methods section, Line 660 in our manuscript "Although the estimated BEC proportion were comparable across all approaches (Kendall s tau (4624) = 0.64 to 0.91, \(p < 2.2 \times 10^{- 16}\) ), both EpiDiSH- RPC and EpiDiSH- CBS consistently predicted higher BEC proportions across all cohorts, were strongly correlated (Kendall s \(\tau_{b}\) (4624) = 0.91, \(p < 2.2 \times 10^{- 16}\) ), and showed the same correlations when the cohorts were investigated independently". This pattern is consistently observed across all the datasets which is reflected in the new Supplementary Figure 1. Further, a detailed comparison of equality versus inequality constraints, implemented in the CP based estimation, has been discussed above as responses to Reviewer 2 Comment 1 and 4 and incorporated in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[90, 357, 904, 540]]<|/det|>
+Review 2 Comment 3 & Comment 7: Figure 1b: an interesting trend is observed during the period of change from deciduous to permanent teeth with a drastic increase in neutrophils. This point is further brought up with the longitudinal samples in Figure 4 (which also require revising the axes labels as some are missing). Do you have information about the teeth loss/replacement relative to the sample collection? The document points out that while BEC proportion declines with age, the consistency of this decline over time within the same individuals is not well established, potentially affecting the reliability of longitudinal studies using cheek swabs. That can strengthen the observation and explain some of the variability. Please expand on lines 207- 220 as needed.; Lines 298- 307: I agree with your discussion, but I would add the fact that the peaks of neutrophils are by the time of teeth exchange so maybe there is blood and more debris during this period that are changing the direction of your findings. Try to add as much context about these physiological changes as possible.  
+
+<|ref|>text<|/ref|><|det|>[[90, 553, 904, 771]]<|/det|>
+Authors Response: We thank the Reviewer for this very insightful suggestion and while we wholeheartedly agree, we unfortunately do not have that data recorded in any of the cohorts, nor any other oral health information. We hope that perhaps our manuscript motivates future cohorts to include a more comprehensive data collection on oral health and oral cells in children. However, we added tooth loss and replacement" as a potential explanation for the observed increases in variability we observed, as this was an excellent point. We also included in our discussion section: There are many other potential explanations for this variability, including tooth development, tooth loss and eruption, oral hygiene and habits, use of orthodontic or dental apparatuses, diet and environmental exposures, and other oral health concerns 8,9,32- 34. These events may also differ across developmental stages, rather than attributable to only explanation, such as the importance of tooth loss and eruption in childhood versus oral hygiene later in the pediatric period, but may point to any event that would increase blood in the mouth transiently."  
+
+<|ref|>text<|/ref|><|det|>[[90, 782, 896, 818]]<|/det|>
+Review 2 Comment 5: Figures 5 and 6: Is it possible to show the relation between chronological age and PedBe Age in months? That can help with the interpretation of the figures. Please correct the axis labels on 6b.  
+
+<|ref|>text<|/ref|><|det|>[[90, 832, 869, 871]]<|/det|>
+Authors Response: Thank you for pointing out this inconsistency, we have confirmed all the reported and estimated ages are in months for the all figures and fixed our labels.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 68, 615, 88]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>title<|/ref|><|det|>[[128, 100, 559, 140]]<|/det|>
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[92, 163, 892, 199]]<|/det|>
+Review 2 Comment 6: Lines 239- 255: I would bring attention to the fact that this is a small cohort (and, as you mention, a proof- of- concept analysis). Please include the n in the text, not only in table 2.  
+
+<|ref|>text<|/ref|><|det|>[[91, 212, 890, 333]]<|/det|>
+Authors Response: We agree and appreciate this excellent suggestion from Reviewer 2. We have now added to our results section: As a proof- of- concept, we leveraged a publicly available cohort (GSE147058), though small with 22 monozygotic twin pairs (n=44 children), that had uniquely measured daily salivary cortisol concurrently with buccal DNA methylation samples enabling us to examine the potential importance of accounting for estimated BEC proportion when predicting PedBE EAA in the association with a relevant biological variable." We also reiterated this in the discussion.  
+
+<|ref|>text<|/ref|><|det|>[[92, 342, 835, 379]]<|/det|>
+Review 2 Comment 8: Lines 430- 439: What method for the CP estimation was used constrained or not constrained to equality?  
+
+<|ref|>text<|/ref|><|det|>[[92, 392, 894, 555]]<|/det|>
+Authors Response: We appreciate this thoughtful question from Reviewer 2 and their methodological rigor. We used the default inequality" setting embedded in HEpiDiSH, which is an iterative hierarchical procedure of EpiDiSH, for CP- based estimations. We have now included this relevant information in our Methods. The same inequality" setting was implemented in Houseman et al (2012) for cell type estimations. However, in our matched dataset analysis where we tested differences in the estimations between the two array platforms, we also compared the two constraints methods of CP (inequality versus equality), which is explained in detail in our response to Reviewer 2 Comment 1 and 4 and this excellent question in part overlap with a similar to comment also raised by Reviewer 1.  
+
+<|ref|>text<|/ref|><|det|>[[93, 567, 366, 584]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[91, 594, 904, 723]]<|/det|>
+Reviewer 3 Comment 1: The increased between- person BEC variability combined with the lack of within- person temporal stability is a key finding of this study. That is, the correlations for BEC proportions for the same individuals over even a brief period of time were not significantly strong. What are the potential implications of this lack of within- person stability for conducting future longitudinal research with the same individuals as well as future cross- sectional or other research combining independent, age- heterogeneous samples to examine between- person change? Does a within- person or between- person approach provide unique advantages for epigenetic aging research?  
+
+<|ref|>text<|/ref|><|det|>[[91, 732, 905, 899]]<|/det|>
+Authors' Response: We thank Reviewer 3 for highlighting the importance of this work. We agree that we should be more explicit in our interpretive recommendations based on the findings. As such we have added the following to the discussion Line 541- 559: "These data thus strongly support and expand our previous recommendation when creating the PedBE clock to correct for estimated BEC proportion when calculating PedBE EAA. If this is not possible, or not preferred, then we recommend evaluating for associations and heteroskedasticity with estimated BEC proportions within pediatric samples and correct models accordingly. Due to statistical concerns with confounding and heteroskedasticity across the pediatric age range, specifically due the increased variability observed with increased age, we would also extend this sentiment to all epigenetic age investigations employing pediatric cheek swabs. We also caution researchers in future studies to
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 68, 615, 88]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>title<|/ref|><|det|>[[128, 100, 559, 140]]<|/det|>
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[92, 162, 875, 199]]<|/det|>
+not assume BEC proportion stability in pediatric cohorts over time, especially when conducting longitudinal research or when combining independent, age- heterogeneous samples."  
+
+<|ref|>text<|/ref|><|det|>[[92, 209, 899, 282]]<|/det|>
+Reviewer 3 Comment 2: Please report the beta weights and corresponding \(p\) - values for models where: 1) epigenetic age is regressed on chronological age, and 2) epigenetic age is regressed on both chronological age and BEC proportions. These results, particularly for model two, have implications for the uniqueness of each predictor in explaining variation in epigenetic age estimates.  
+
+<|ref|>text<|/ref|><|det|>[[90, 290, 904, 824]]<|/det|>
+Authors' Response: Thank you for these excellent recommendations. We agree that this will be helpful for interpretation and have included them as requested. Further, upon examining the beta weights from epigenetic age regressed on chronological age, estimate BEC proportion, and chronological age and estimated BEC proportion together, we discovered what was likely a suppression effect (large changes in the estimated BEC proportion model coefficient when a sole predictor versus when in a model with chronological age, but little change to chronological age's model coefficients). This is not an effect we thought to look for, and, in fact, is an often overlooked and underreported statistical effect \(^{1}\) . We thank Reviewer 3, as this would not come to our attention if not for this comment. To investigate the possibility of this further, we examined the significance of each predictor alone and in combination in a linear regression of PedBE age. It was here that we found that in childhood estimated BEC proportion was always found to have a significant association with PedBE age, but in adolescence, though there was no significant correlation between estimated BEC proportion alone with PedBE age, in the joint model with chronological age, estimated BEC proportion was a significant predictor of PedBE age. Thus, we followed the literature guidelines on exploring such a possible effect by examining the model fit (adjusted \(R^2\) ), structure coefficients, and conducting a commonality analysis \(^{2}\) . We found that including estimated BEC proportion with chronological age improved model fit, as measured by adjusted R2, in both childhood and adolescence models of PedBE age, though this was a very small improvement in childhood (.96 to .97) and a larger improvement in adolescence (.63 to .71). We also found through structure coefficients, which are a measure of the correlation between a predictor variable and the predicted values of the outcome commonly used to investigate multicollinearity and statistical suppression, that estimated BEC proportion was influential and negatively associated with PedBE age uniquely from chronological age in childhood, but had almost no direct influence on PedBE age when accounting for chronological age in adolescence. Further, we conducted a commonality analysis, which decomposes the total explained variance ( \(R^2\) ) into the unique and common shared effects of model predictors, allowing for a detailed understanding of how each contributes to the overall model based on Reviewer 3's suggestion of the importance of the uniqueness of each predictor to PedBE age. This analysis suggested a classical suppression effect in childhood, where the unique contribution of estimated BEC proportion was low, but the common shared contribution of chronological age and estimated BEC proportion together was high, explaining 1/3 of the explained variance. However, in adolescence, where estimated BEC proportion did not significantly associated with PedBE age alone, we observed estimated BEC proportion contributed little alone and, in fact, negatively to the common shared contribution.  
+
+<|ref|>text<|/ref|><|det|>[[92, 833, 902, 906]]<|/det|>
+Together, these statistical indices suggest a classical statistical suppression effect in childhood, which estimated BEC proportion largely contributes to the common shared variance rather than uniquely but is still a strong predictor of PedBE age. These alternatively suggest a negative statistical suppression in adolescence, where estimated BEC proportion is not directly or uniquely associated with PedBE age, but does still contributes
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 68, 616, 88]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>title<|/ref|><|det|>[[128, 100, 559, 140]]<|/det|>
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[91, 162, 904, 290]]<|/det|>
+meaningfully to the model by assisting chronological age to explain more variance, apparent by the improvement in adjusted \(\mathsf{R}^2\) . Our interpretation of these results are that estimated BEC proportion makes a statistically meaningful contribution to the ability of chronological age to explain the variance of PedBE age throughout the pediatric period, though differently in childhood and adolescence. Given the primary use of PedBE age, or any epigenetic age estimation, in the literature is to compare residuals from the regression of chronological age on PedBE age, the more accurate this estimation, then the more likely the resulting statistical analyses are to be valid and replicable.  
+
+<|ref|>text<|/ref|><|det|>[[92, 301, 890, 334]]<|/det|>
+We have included this research in the results, discussion, and methods section of the manuscript now, which read:  
+
+<|ref|>sub_title<|/ref|><|det|>[[92, 352, 900, 386]]<|/det|>
+## "Developmental period specific statistical suppression effects of estimated BEC proportion were observed on the chronological age prediction of PedBE age  
+
+<|ref|>text<|/ref|><|det|>[[90, 386, 904, 590]]<|/det|>
+To investigate the contribution of estimated BEC proportion to PedBE age prediction by chronological age in childhood and adolescence, we examined models with and without the inclusion of estimated BEC proportion as a covariate. In childhood samples \((< 120\) months/10years), we fit the a regression model of chronological age and estimated BEC proportions on PedBE age, which was statistically significant, \(\mathsf{F}(2,640) = 54,350\) \(p< 2.2\) \(\times 10^{- 16}\) , adjusted \(\mathsf{R}^2 = 0.97\) . Next, removing the estimated BEC proportion from the regression model to leave chronological age as the sole predictor of PedBE age, the model was still statistically significant, \(\mathsf{F}(1,641) =\) \(91,110\) \(p< 2.2\times 10^{- 16}\) , though with a slightly lower adjusted \(\mathsf{R}^2\) of 0.96. Similarly, we also fit these two regression models on the adolescent samples \((>120\) months/10years). While the regression model of chronological age and estimated BEC proportions predicting PedBE age was statistically significant, \(\mathsf{F}(2,640) =\) \(813.4\) \(p< 2.2\times 10^{- 16}\) \(\mathsf{R}^2 = 0.71\) , removing the estimated BEC proportion from the model and leaving chronological age as the sole predictor of PedBE age, while still statistically significant, \(\mathsf{F}(1,641) = 1097\) \(p< 2.2\) \(\times 10^{- 16}\) yielded a more substantially lower adjusted \(\mathsf{R}^2\) of 0.63.  
+
+<|ref|>text<|/ref|><|det|>[[90, 590, 904, 792]]<|/det|>
+When examining the model \(\beta\) coefficients, as expected the coefficient for chronological age was approximately 1, and this held true both in the sole predictor models and the models with estimated BEC proportions as a covariate, as well as in both developmental periods. This model coefficient indicates that for every unit of chronological age, there was a unit increase for PedBE age across all models. However, for samples in the infancy to childhood period, the model \(\beta\) coefficient of estimated BEC proportion as the sole predictor of PedBE age was - 220.14, significantly more negative than when in the model with chronological age (Table 2). Similarly, for the adolescence period samples, the model \(\beta\) coefficient when BEC proportion was the sole predictor of PedBE age was 4.58, much smaller than when in the model with chronological age (Table 2). These substantial differences in model \(\beta\) coefficients for estimated BEC proportion indicated the possibility of suppressor effects in the more complex regression models \(^{1,2}\) . Therefore, we examined the model \(\beta\) coefficients, structure coefficients, and commonality analysis results with the inclusion of estimated BEC proportion for the chronological age prediction of PedBE age in childhood and adolescence (Table 2).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[730, 40, 916, 101]]<|/det|>
+Tel 604.875.3803 michael.kobor@ubc.ca 117- 2194 Health Sciences Mall Vancouver, B.C. V6T 1Z3  
+
+<|ref|>title<|/ref|><|det|>[[130, 85, 614, 106]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>text<|/ref|><|det|>[[130, 101, 510, 140]]<|/det|>
+Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[92, 163, 895, 232]]<|/det|>
+Table 2. Model indices of developmental period specific prediction of PedBe age by chronological age and estimated BEC proportion. Beta model regression coefficients, structural coefficients, representing the correlation between a predictor variable and the predictor- criterion values, and the unique, shared, and total contribution of each predictor from the commonality analysis are presented.  
+
+<|ref|>table<|/ref|><|det|>[[88, 247, 976, 496]]<|/det|>
+
+<table><tr><td>Predictor</td><td>Beta (β) coefficient</td><td>Structured coefficient</td><td>Unique contribution</td><td>Shared contribution</td><td>Total contribution</td></tr><tr><td></td><td></td><td></td><td></td><td>Commonality Analysis</td><td></td></tr><tr><td>Chronological age</td><td>1.1942</td><td>0.9974</td><td>0.6517</td><td>0.3170</td><td>0.9737</td></tr><tr><td>Estimated BEC proportion</td><td>-220.1476</td><td>-0.5750</td><td>0.0049</td><td></td><td></td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>Chronological age</td><td>0.9247</td><td>0.9378</td><td>0.7173</td><td>-0.0860</td><td>0.7177</td></tr><tr><td>Estimated BEC proportion</td><td>4.5898</td><td>0.0231</td><td>0.0864</td><td></td><td></td></tr></table>  
+
+<|ref|>sub_title<|/ref|><|det|>[[92, 519, 548, 536]]<|/det|>
+## Relevant associated discussion section (new and integrated):  
+
+<|ref|>text<|/ref|><|det|>[[90, 536, 896, 893]]<|/det|>
+The increases in beta coefficients when estimated BEC proportion is in the model with chronological age to predict PedBE, as well as examination of the model fit \((R^2)\) , structural coefficients, and commonality analysis, suggested suppression effects with estimated BEC proportion across the pediatric age range. Specifically, we observed a classical suppression effect in childhood and a negative suppression effect in adolescence. In childhood, the high common variance measured by the commonality analysis implies that chronological age and BEC proportion explained overlapping components of PedBE age variance. This indicates a classical suppression effect, where estimated BEC proportion enhanced the predictive power of chronological age by accounting for variance that chronological age cannot explain alone. This results in only a small improvement in model fit, as chronological age alone is already predictive of PedBE age and estimated BEC proportion is correlated with chronological age. However, in adolescence, the low common variance coupled with a more substantial model fit improvement, suggests a negative suppression effect, which is notably statistically rare \(^{1 - }\) \(^{3}\) . In this context, we observed estimated BEC proportion was slightly suppressing irrelevant variance in chronological age, despite their lack of association in adolescence, which subsequently allowed the model to better explain the variance in PedBE age. Together, these indices suggested that including estimated BEC proportion helps clarify the prediction of PedBE age made by chronological age by eliminating noise. As such, though for differing reasons, we recommend researchers to account for estimated BEC proportion when examining EAA in children and adolescents. This association of epigenetic age with cell type in early life may contribute to the relative accuracy of the PedBE clock in pediatric cheek swabs in comparison to other epigenetic clocks \(^{4}\) . Therefore, the power of epigenetic age as a biomarker is likely related to some extent to the associations between cell type proportions and the trait of interest \(^{5}\) . The expected changes in cell type proportion heterogeneity with age may not be synonymous to, but informative of biological age prediction.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 68, 615, 88]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>title<|/ref|><|det|>[[128, 100, 559, 140]]<|/det|>
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[90, 162, 900, 435]]<|/det|>
+Due to the stronger association between estimated BEC proportion and predicted epigenetic age than chronological age, it was not surprising to find a moderate correlation between EAA and BEC proportion during childhood. However, the addition of differences in estimated BEC proportion as a covariate when calculating EAA led to the disappearance of the correlation between EAA and estimated BEC proportion. These data thus strongly support and expand our previous recommendation when creating the PedBE clock \(^4\) to correct for estimated BEC proportion when calculating PedBE EAA. Given the primary use of PedBE age, or any epigenetic age estimation, to compare these residuals (PedBE EAA) from the regression of chronological age on PedBE age, incorporation of estimated BEC proportion, in calculating EAA can lead to robust and accurate epigenetic age estimates. This, in turn, will likely enhance the validity and replicability of the resulting statistical analyses. If this is not possible, or not preferred, then we recommend evaluating for associations and heteroskedasticity with estimated BEC proportions within pediatric samples and correct models accordingly. Due to statistical concerns with confounding and heteroskedasticity across the pediatric age range, specifically due the increased variability observed with increased age, we would also extend this sentiment to all epigenetic age investigations employing pediatric cheek swabs. We also caution researchers in future studies to not assume BEC proportion stability in pediatric cohorts over time, especially when conducting longitudinal research or when combining independent, age- heterogeneous samples.  
+
+<|ref|>sub_title<|/ref|><|det|>[[92, 451, 314, 467]]<|/det|>
+## New Addition to the Method:  
+
+<|ref|>text<|/ref|><|det|>[[90, 468, 884, 688]]<|/det|>
+Assessment of estimated BEC proportion contribution to PedBE predictionTo further understand the association of estimated BEC proportion to the prediction of PedBE age by chronological age, we compared the differences in model \(\beta\) coefficients for estimated BEC proportion across the two developmental periods. Upon discovering large discrepancies in estimated BEC proportion model \(\beta\) coefficients when a covariate and when a sole predictor, we investigated the possibility of statistical suppression \(^{1 - 3}\) . This exploration included the effects of estimated BEC proportion as a covariate to chronological age in predicting PedBE age, specifically: a) employing model indices including model fit (adjusted \(\mathbb{R}^2\) ) and b) structure coefficients extracted from the regression models, which were fitted using the stats R package \(^6\) . Structure coefficients reflect the correlation between a predictor variable and the outcome predicted values and reflects the unique strength of the predictor as specified by the model \(^1\) . Additionally, we conducted a commonality analysis using the yhat R package \(^7\) to dissect the unique contribution of the independent predictors and the shared contribution of the predictors in explaining the overall variance of the model \(^2\) ."  
+
+<|ref|>text<|/ref|><|det|>[[91, 715, 900, 862]]<|/det|>
+Reviewer 3 Comment 3 and 4: The implications of the current set of results suggest the importance of including BEC proportions when generating estimates of epigenetic age and epigenetic age acceleration during child development. Are the translational implications then for relevant epigenetic clock developers to adjust the available code (e.g. GitHub for PedBE) or online calculators (e.g. Horvath's) so that BEC proportions are now included in the generation of epigenetic aging estimates? Along these same lines, is it sufficient to adjust risk estimates of epigenetic aging on a particular outcome by including BEC proportions as an additional covariate/predictor in the statistical models? Or, must BEC proportions be included in the estimation of epigenetic aging?  
+
+<|ref|>text<|/ref|><|det|>[[92, 873, 904, 908]]<|/det|>
+Authors' Response: This is an excellent point. Given our updated findings with the new adolescent cohorts, this illustrates how both the correlation and variability of estimated BEC proportion differs across the pediatric age
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 68, 615, 88]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>title<|/ref|><|det|>[[128, 100, 559, 140]]<|/det|>
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[90, 162, 903, 455]]<|/det|>
+range. As such, it would be helpful these calculators or code to be data- dependent for the needs of all researchers. For example, this work would not have been possible without the ability to examine models with and without BEC proportions. However, to address the importance of the reviewer's point, we did update our discussion with the following Line 531: "This association of epigenetic age with cell type in early life may contribute to the relative accuracy of the PedBE clock in pediatric cheek swabs in comparison to other epigenetic clocks, especially at the youngest ages 10. Therefore, the power of epigenetic age as a biomarker is likely related to some extent to the associations between cell type proportions and the trait of interest 45. The expected natural changes in cell type proportion heterogeneity with age may not be synonymous to, but be informative of, biological age prediction." We also agree that correcting for estimated BEC proportion in the model is also a viable methodological option. While it may be more difficult to interpret EAA differences without BEC independence, if the model including estimated BEC proportions is being presented, this would be equally useful. As such we included in a line in the discussion (included in our earlier response to Reviewer 3): "These data thus strongly support and expand our previous recommendation when creating the PedBE clock4 to correct for estimated BEC proportion when calculating PedBE EAA. If this is not possible, or not preferred, then we recommend evaluating for associations and heteroskedasticity with estimated BEC proportions within pediatric samples and correct models accordingly".  
+
+<|ref|>text<|/ref|><|det|>[[92, 464, 901, 519]]<|/det|>
+Review 3 Comment 5: Please provide a brief rationale for why diurnal cortisol slope and obsessive- compulsive disorder status were examined as outcomes important for epigenetic aging. Similarly, a rationale for examining dental visits at 48- months, as opposed to any other age, was not provided in the original manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[90, 529, 904, 731]]<|/det|>
+Authors' Response: We appreciate this suggestion to make the usefulness of these comparisons more evident. Regarding diurnal cortisol slope and OCD status we now state in Line 567: "These comparisons are indicative of two areas of research in which PedBE EAA has been applied, i.e., developmental perspectives on mental health and stress responses 13,17,46- 48. " This is to highlight that these are the types of analyses likely for pediatric EAA to be applied in the field and can act as proof- of- concept exemplars (as was also emphasized in response to Reviewer 2's comment). Regarding dental visits at 48 months in GUSTO, while we were grateful to have these data, this (along with the also included tooth brushing behavior) was unfortunately the only available concurrent oral health variable available across any of the ages. To our knowledge, this was not measured at the other timepoints in GUSTO, unfortunately, nor in the other available cohorts. Ideally, as mentioned earlier, future research will measure oral health in detail along with oral cells and DNA methylation in pediatric samples to further investigate the potential causal mechanisms of these observed trends.  
+
+<|ref|>text<|/ref|><|det|>[[91, 741, 901, 850]]<|/det|>
+Review 3 Comment 6: On line 273, the authors report a significant association between chronological age and the residuals of the model where PedBE age acceleration was regressed only on chronological age. As reported in other places in the manuscript, I expected this association would also be near zero given PedBE age acceleration is determined by removing or accounting for the effects of chronological age. Could the authors provide some insight into this finding? Perhaps there is a non- linear relation here? Please also report the magnitude of the association observed here.  
+
+<|ref|>text<|/ref|><|det|>[[92, 861, 900, 898]]<|/det|>
+Authors' Response: We appreciate the reviewers' attention to detail and accuracy. Indeed, no significant association was reported between chronological age and PedBE age acceleration residuals (p=0.5, r=0.02). The
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 68, 616, 88]]<|/det|>
+# THE UNIVERSITY OF BRITISH COLUMBIA  
+
+<|ref|>title<|/ref|><|det|>[[128, 99, 559, 140]]<|/det|>
+# Edwin S.H. Leong Centre for Healthy Aging Faculty of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[90, 162, 875, 199]]<|/det|>
+p- value \(= 0.02\) which was earlier reported was actually the strength of association and the corresponding p- value is 0.54. We have now modified our results accordingly Line 392.  
+
+<|ref|>text<|/ref|><|det|>[[91, 209, 904, 264]]<|/det|>
+Review 3 Comment 7: On line 160, the t- value reported is - 2.41 but the adjusted p- value is reported as greater than .05. In a typical t- distribution, a value greater than 1.96 in either direction would be statistically significant (less than .05). Could it be that the decimal is in the wrong place?  
+
+<|ref|>text<|/ref|><|det|>[[92, 273, 896, 311]]<|/det|>
+Authors' Response: Agreed. Thank you for pointing that out, we very much appreciate it. The t- value that was earlier reported was incorrect, the correct t- value is - .16 which is now corrected in the manuscript  
+
+<|ref|>text<|/ref|><|det|>[[92, 320, 866, 339]]<|/det|>
+Review 3 Comment 8: Please report means, standard deviations, and/or effect sizes in each of the figures.  
+
+<|ref|>text<|/ref|><|det|>[[91, 348, 901, 386]]<|/det|>
+Authors' Response: We thank the reviewer for this suggestion and have included these in the graphs and figure legends (including supplementary figure) throughout the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[91, 395, 884, 432]]<|/det|>
+Review 3 Comment 9: Several different terms were used to refer to BEC dimensions (e.g. estimated, inferred, computed). Is it possible to use one of these terms throughout the manuscript?  
+
+<|ref|>text<|/ref|><|det|>[[91, 441, 855, 479]]<|/det|>
+Authors' Response: Thank you for this feedback. We have chosen the word "estimated" and changed all references throughout the manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review file__deffadbb7826afdf2ab8ad33c642df1e1e47832002716b5ccbc5b5c2417b4676/images_list.json

@@ -0,0 +1 @@
+[]