Sử dụng phương pháp mô phỏng động lực học phân tử không cân bằng để nghiên cứu cơ chế mở rào cản máu não dưới tác động của siêu âm hội tụ

SỞ KHOA HỌC VÀ CƠNG NGHỆ TP HỒ CHÍ MINH VIỆN KHOA HỌC VÀ CƠNG NGHỆ TÍNH TỐN BÁO CÁO TỔNG KẾT Sử dụng phương pháp mô động lực học phân tử không cân để nghiên cứu chế mở rào cản máu não tác động siêu âm hội tụ Đơn vị thực hiện: PTN Khoa Học Sự Sống Chủ nhiệm nhiệm vụ: TSKH Nguyễn Hoàng Phương TP HỒ CHÍ MINH, THÁNG 6/2020 SỞ KHOA HỌC VÀ CƠNG NGHỆ TP HỒ CHÍ MINH VIỆN KHOA HỌC VÀ CƠNG NGHỆ TÍNH TỐN BÁO CÁO TỔNG KẾT Sử dụng phương pháp mô động lực học phân tử không cân để nghiên cứu chế mở rào cản máu não tác động siêu âm hội tụ Viện trưởng: Đơn vị thực hiện: PTN Khoa Học Sự Sống Chủ nhiệm nhiệm vụ: TSKH Nguyễn Hoàng Phương Nguyễn Kỳ Phùng Nguyễn Hồng Phương TP HỒ CHÍ MINH, THÁNG 6/2020 Sử dụng phương pháp mô động lực học phân tử không cân để nghiên cứu chế mở rào cản máu não tác động siêu âm hội tụ MỤC LỤC Trang MỞ ĐẦU ĐƠN VỊ THỰC HIỆN KẾT QUẢ NGHIÊN CỨU I Báo cáo khoa học II Tài liệu khoa học xuất 24 III Chương trình giáo dục đào tạo 26 IV Hội nghị, hội thảo 27 V File liệu 28 TÀI LIỆU THAM KHẢO 29 CÁC PHỤ LỤC 32 Phụ lục 1: Bài báo “Nonequilibrium atomistic molecular dynamics simulation of tubular nanomotor propelled by bubble propulsion “ Phụ lục 2: Bài báo “Interaction mechanism between the focused ultrasound and lipid membrane at the molecular level" Phụ lục 3: Bài báo “Molecular mechanism of ultrasound interaction with a blood brain barrier model.” Phụ lục 4: Bài báo “Molecular mechanism of ultrasound induced drug delivery pathways on liposomes.” Phụ lục 5: Các minh chứng đào tạo Viện Khoa học Cơng nghệ Tính tốn TP Hồ Chí Minh Page Sử dụng phương pháp mô động lực học phân tử không cân để nghiên cứu chế mở rào cản máu não tác động siêu âm hội tụ MỞ ĐẦU Hàng triệu người giới mắc loại rối loạn hệ thần kinh trung ương (CNS) thời điểm đời, ba người có người bị [1, 2] Theo dự đoán, khoảng tỷ người trải qua số loại rối loạn thần kinh trung ương vào năm 2030 không phát triển phương pháp điều trị [3] Trên toàn giới, 20 triệu người mắc bệnh Alzheimer (AD), bệnh Parkinson, bệnh đa xơ cứng bệnh xơ cứng teo bên Bởi lão hóa làm tăng đáng kể nguy mắc bệnh thần kinh trung ương, ngày có nhiều người mắc bệnh Ví dụ, có 4,7 triệu người từ 65 tuổi trở lên mắc chứng trí nhớ AD Hoa Kỳ vào năm 2010 tổng số người mắc chứng trí nhớ AD vào năm 2050 ước tính gấp ba, 13,8 triệu [4] Ngoài gánh nặng cảm xúc to lớn bệnh nhân người chăm sóc họ, cịn có vấn đề kinh tế Thật khơng may, có tiến lớn đạt năm gần để hiểu bệnh thần kinh trung ương, có phương pháp điều trị hiệu quả, khơng có phương pháp chữa trị hết bệnh Điều phần lớn thực tế hàng rào máu não (BBB) ngăn chặn đến 98% loại thuốc vào não để chữa bệnh Để khắc phục, số phương pháp phát triển phép thuốc xâm nhập vào não dễ dàng Phương pháp phẫu thuật phương pháp hiệu nhất, vùng não có bệnh chẩn đốn thuốc đưa vào cách tiêm trực tiếp Phương pháp gây nên thiệt hại mô khỏe mạnh vết kim rủi ro phẫu thuật gây Một cách tiếp cận khác sử dụng chế vận chuyển nội sinh thuốc sửa đổi tăng độ hấp thụ qua BBB thụ thể tế bào nội mô não Cách này, nhiên, loại thuốc khơng đạt nồng độ mong muốn để trị liệu bệnh [5] Phương pháp mở BBB cách tạm thời cách tiếp cận khác phép phân phối thuốc vào CNS Một kỹ thuật tiến hành tiêm chất hóa học rượu đường, dung môi thuốc giãn mạch để thu nhỏ tế bào nội mơ, kéo giãn BBB, thúc đẩy việc đưa thuốc vào não [6,7] Tuy nhiên, điều làm cho BBB bị mở rộng, tồn não bị tiếp xúc khơng với thuốc mà hợp chất khác từ hệ thống tuần hồn, khiến não bị độc Khoảng mười lăm năm trước, Hynynen đồng nghiệp phát triển phương pháp để mở BBB cách sử dụng siêu âm tập trung (FUS) kết hợp với tiêm bong bóng có kích thước micro vào tĩnh mạch [8] Kể từ đó, nhiều nghiên cứu động vật kỹ thuật hứa hẹn có tiềm ứng dụng cao gây nên tác động cục bộ, không xâm lấn đặc biệt an tồn Ví dụ, hình ảnh não kính hiển vi bình thường sau BBB mở [9], việc mở BBB lặp lặp lại nhiều tuần [10-12] Kỹ thuật cho phép kiểm soát biên độ gián đoạn BBB, kiểm sốt mức độ phân phối thuốc vào não, điều mà không làm với phương pháp khác Với lợi này, nhiều phịng thí nghiệm Mỹ, Châu Âu Châu Á nghiên cứu kỹ thuật để thiết lập thơng số tối ưu [13-16], định lượng tính thấm mô não bị phá vỡ [17,18], phân phối thuốc mới, đánh giá theo dõi đáp ứng với phương pháp điều trị [19-21] Mặc dù nghiên cứu cho kết tích cực, kỹ thuật chủ yếu thực động vật? Có lẽ, trở ngại làm hạn chế phát triển tương lai đưa vào ứng dụng lâm sàng phương pháp chế xác cấp độ phân tử việc mở BBB FUS gây chưa hiểu rõ ràng Một phần nguyên nhân phụ thuộc phức tạp việc mở BBB vào thơng số siêu âm bong bóng, việc sử dụng giao thức thử nghiệm khác nhau, đa dạng tiêu chí sử dụng để đánh giá việc mở, khó khăn để so sánh giải thích kết từ phịng thí nghiệm khác với độ tin cậy [22] Vì nói nay, kỹ thuật giai đoạn sơ khai Trong bối cảnh này, mục tiêu cốt lõi dự án thực mô động lực phân tử cách có hệ thống để hiểu cấp độ phân tử chế mở BBB siêu âm gây Viện Khoa học Cơng nghệ Tính tốn TP Hồ Chí Minh Page Sử dụng phương pháp mô động lực học phân tử không cân để nghiên cứu chế mở rào cản máu não tác động siêu âm hội tụ Lời cảm ơn đến ICST Đề tài tài trợ Viện Khoa học Cơng nghệ Tính tốn, Sở khoa học Cơng nghệ Hồ Chí Minh Viện Khoa học Cơng nghệ Tính tốn TP Hồ Chí Minh Page Sử dụng phương pháp mô động lực học phân tử không cân để nghiên cứu chế mở rào cản máu não tác động siêu âm hội tụ ĐƠN VỊ THỰC HIỆN Phịng thí nghiệm: Khoa Học Sự Sống Chủ nhiệm nhiệm vụ: Nguyễn Hoàng Phương Thành viên nhiệm vụ: Mai Xuân Lý Phạm Đăng Lân Phan Minh Trường Cơ quan phối hợp: Viện Khoa học Cơng nghệ Tính tốn TP Hồ Chí Minh Page Sử dụng phương pháp mô động lực học phân tử không cân để nghiên cứu chế mở rào cản máu não tác động siêu âm hội tụ KẾT QUẢ NGHIÊN CỨU I BÁO CÁO KHOA HỌC A Rào cản mạch máu não Hình Phác hoạ mơ hình rào cản máu não BBB Phần não máu ngăn cách lớp màng luminal abluminal Các tế bào nối với múi nối chặt chẽ, cho phân tử có kích thước nhỏ qua Các phân tử lớn vào não thông qua phương pháp chủ động (active), nhờ hạt vận chuyển (carriers) đầu nhận (receptor) Như trình bày Hình 1, rào cản mạch máu não (BBB) bao gồm tế bào nội mô kết nối với liên kết phức hợp bao gồm nút nối chặt chẽ (TJ: tight junction) nút nối adherens (AJ: adherens junction) Màng luminal abluminal ngăn cách máu với não, đóng vai trò rào cản thấm Một số phân tử vận chuyển qua BBB phương pháp thụ động chủ động Thông thường, phân tử phân cực hịa tan có khối lượng phân tử nhỏ 400 Da chuyển từ máu vào não qua TJ Các phân tử lớn đến não thông qua hệ thống vận chuyển nội sinh BBB, xảy chủ yếu thông qua hạt vận chuyển (carrier mediate) đầu nhận tế bào (receptor) [23,24] Chính rào cản mà có khoảng 5% khoảng 7000 thuốc biết vượt qua BBB để vào não để điều trị bệnh thối hóa thần kinh [25] B Nguyên tác mở rào cản mạch máu não siêu âm Nghiên cứu tính thấm BBB sau chiếu xạ siêu âm có từ năm 1950 Bakay đồng nghiệp phát siêu âm tập trung cường độ cao (HIFU) áp dụng cho não mèo làm cho BBB mở ra, gây tổn thương nhu mơ nghiêm trọng [26] Kể từ đó, số nhà nghiên cứu lặp lại thí nghiệm [27-29] Tuy nhiên, người ta xác định HIFU khơng phá vỡ BBB mà cịn gây tổn thương não nhiệt nóng tạo siêu âm [30] Do đó, sau gần nửa kỷ khơng phát triển kỹ thuật này, khoảng mười lăm năm trước, Hynynen đồng nghiệp làm cho nhiều người quan tâm lại đến Viện Khoa học Cơng nghệ Tính tốn TP Hồ Chí Minh Page Sử dụng phương pháp mô động lực học phân tử không cân để nghiên cứu chế mở rào cản máu não tác động siêu âm hội tụ Hình Nguyên tắc phương pháp sử dụng siêu âm bong bóng để đưa thuốc vào não Siêu âm (ultrasound) hội tụ qua hộp sọ vào vùng não quan tâm Bong bóng khí micrometer (microbubble, màu xanh dương) tiêm vào máu Khi bong bóng đến vùng siêu âm chúng bắt đầu co giãn Điều làm thành mạch máu mở cách tạm thời, cho phép phân tử trị liệu (therapeutic, màu xanh lục) di chuyển từ máu vào não phương pháp thú vị [8] Họ giải vấn đề liên quan đến cường độ cao HIFU cách kết hợp siêu âm tập trung (FUS) với bong bóng nhỏ có kích thước micrometer Nguyên lý hoạt động phương pháp minh hoạ Hình Trước hết, bong bóng khí có kích thước cỡ micrometer (có sẵn thị trường) bơm vào mạch máu Sau siêu âm hội tụ vào vị trí não, nơi mà cần đưa thuốc vào Dưới siêu âm, bong bóng co giãn làm thay đổi kích cỡ Nếu biên độ dao động nhỏ co lại giãn nở kích thước bong bóng gần đối xứng Hiện tượng này, gọi cavitation ổn định, tạo dòng chảy chất lỏng xung quanh bong bóng gây áp suất lên vật gần Ngược lại, dao động biên độ lớn dẫn tới co giãn nở bất đối xứng, sau nổ bong bóng, tượng gọi cavitation qn tính Điều tạo nhiệt độ, áp suất cao giải phóng sóng xung kích lan truyền tốc độ siêu âm[31] Trong hai trường hợp, bong bóng khí tạo lực học lên BBB làm cho BBB mở Kỹ thuật làm giảm công suất siêu âm cần thiết để mở BBB hai bậc độ lớn so với trường hợp khơng dùng bong bóng, tránh vấn đề nhiệt tăng lên siêu âm Kể từ đó, số lượng lớn nghiên cứu thực ống nghiệm động vật, tập trung vào khía cạnh khác Một khía cạnh quan trọng chế tương tác siêu âm tập trung cường độ thấp (LIFU) với bong bóng, ảnh hưởng bong bóng lên BBB Một khía cạnh khác ảnh hưởng thông số siêu âm biên độ áp suất, tần số, độ dài xung, tần số lặp lại xung, thời gian siêu âm thông số bong bóng liều lượng, đường kính loại vật liệu, lên chế mở BBB [32-35] Ngoài ra, nhiều cơng trình tập trung vào việc định lượng tính thấm mơ não bị phá vỡ [36-37], kiểm tra việc cung cấp thuốc mới, đánh giá theo dõi phản ứng phương pháp điều trị [38-41] Tuy nhiên, chế phân tử xác việc mở BBB LIFU chưa biết C Mục đích dự án Các chế phân tử việc mở BBB xạ HIFU LIFU phức tạp, có nhiều tương tác liên quan đến q trình này, bao gồm bốn tương tác bong bóng tế bào, bong bóng TJ, siêu âm tế bào, siêu âm TJ, kết hợp tương tác Một Viện Khoa học Cơng nghệ Tính tốn TP Hồ Chí Minh Page Sử dụng phương pháp mơ động lực học phân tử không cân để nghiên cứu chế mở rào cản máu não tác động siêu âm hội tụ hiểu biết chế mở BBB cấp độ phân tử đòi hỏi hiểu biết tất tương tác cách chi tiết Chiến lược phân chia vấn đề phức tạp thành nhiệm vụ đơn giản mà thực báo cáo báo cáo Nhiệm vụ nghiên cứu tương tác tạo bong bóng dao động với màng tế bào, nhiệm vụ thứ hai tìm hiểu tương tác siêu âm tập trung (khơng có bong bóng) với màng tế bào, nhiệm vụ thứ ba nghiên cứu tương tác siêu âm với mơ hình BBB, nhiệm vụ thứ tư nghiên cứu tác dụng phụ siêu âm lên protein albumin máu Chúng thực hai nghiên cứu bổ sung, có liên quan đến dự án Nghiên cứu phát triển nanomotor để đưa thuốc đến nơi mong muốn nghiên cứu thứ hai nghiên cứu chế phân tử đường vận chuyển thuốc siêu âm gây bề mặt hạt liposomes D Cơ chế phân tử tương tác siêu âm với mơ hình BBB (J Chem Phys Accepted for publication 2020) Hiện tại, cấu trúc phân tử đầy đủ BBB khơng có sẵn, điều làm cho việc mô động lực học phân tử việc mở BBB siêu âm khó khăn Gần đây, Nangia đồng nghiệp thu cấu trúc protein claudin-5, thành phần mối nối chặt chẽ, thông qua so sánh với cấu trúc thực nghiệm protein claudin-15 Các mô động lực phân tử họ monomers claudin-5 đặt vào bảy màng lipids với thành phần lipid khác dễ dàng tạo thành cis-dimers ổn định, sau tạo thành chuỗi liền kề cụm bậc cao [42] Các cis-dimers sử dụng để xây dựng giao diện trans, hình thành protein claudin-5 tế bào lân cận tương tác trực tiếp với để tạo thành tổ hợp nối chặt chẽ, cách sử dụng phương pháp lắp ghép kết hợp với tinh chỉnh đối xứng [43] Các đặc điểm mối nối, đường kính chiều dài tính tốn chi tiết [44] Những thơng tin sử dụng để xây dựng mơ hình BBB tối thiểu, cho phép thực mô động lực phân tử nhằm tìm hiểu tác động siêu âm lên màng tế bào mối nối chặt chẽ Cụ thể, muốn hiểu phần mơ hình BBB: màng tế bào mối nối chặt chẽ bị ảnh hưởng siêu âm Các tác động bong bóng việc xây dựng mơ hình BBB phức tạp thực dự án Xây dựng mơ hình BBB Như đề cập đây, BBB bao gồm tế bào kết nối với nút nối Mỗi múi nối bao gồm nhiều loại proteins khác nhau, Trong loại proteins loại protein có tên gọi claudin-5 thành phần cấu tạo nên múi nối Bởi hệ lớn để thực mơ với điều kiện máy tính nay, chúng tơi phải đơn giản hóa mơ hình BBB Mơ hình BBB chúng tơi bao gồm hai màng lipid đôi (lipid bilayer) kết nối với múi nối bao gồm proteins claudin-5 Ở mơ hình này, màng tế bào xây dựng lipids loại phosphatidylcholine (POPC) Cấu trúc protein claudin-5 có thơng qua mơ so sánh tương đồng (homology modelling) với cấu trúc thực nghiệm protein claudin-15 Protein claudin-5 bao gồm bốn bó xoắn helix nằm xuyên màng tế bào, hai vòng ngoại bào vịng nội bào Sau cấu trúc cặp claudin-5 protein (claudin-5 dimer) nằm màng tế bào lấy từ mô động lực học phân tử trước [42], hai protein claudin-5 (dimer) tương tác với thơng qua bó xoắn Sau đó, cặp hai màng tế bào có gắn hai dimers đặt gần nhau, mô động lực học cân thực để hai cặp dimers tương tác với nhau, tạo thành múi nối Mô hình trình bày Hình Mặc dù mơ hình BBB đơn giản chứa đựng thơng tin BBB [43,44] Trong bối Viện Khoa học Cơng nghệ Tính tốn TP Hồ Chí Minh Page Sử dụng phương pháp mô động lực học phân tử không cân để nghiên cứu chế mở rào cản máu não tác động siêu âm hội tụ cảnh chưa có mơ hình xây dựng, mơ hình cho phép lần thực mô tương tác siêu âm BBB Chúng sử dụng trường lực CHARMM36 [45] để mơ hình hóa protein lipid Tồn mơ hình đặt hộp nước bao gồm 180 000 phân tử nước Kích thước ban đầu hộp (𝐿𝑥 , 𝐿𝑦 , 𝐿𝑧 ) = (10,10,18) nm Xuất phát từ cấu trúc này, mô động lực học cân thực 100 ns sử dụng tập hợp NPT với áp suất P = bar nhiệt độ T = 300 K, sử dụng chương trình mơ GROMACS [46] Cấu trúc cuối sử dụng để làm cấu trúc ban đầu cho mô với siêu âm Hình Cấu trúc mơ hình BBB Mỗi cặp dimer claudin-5 proteins (màu cam hay màu xanh lục) gắn vào màng tế bào (màu xanh) đặt gần để tạo nên múi nối hai tế bào Toàn hệ đặt nước (màu đỏ) Phương pháp mô với siêu âm Trong mơ phỏng, ảnh hưởng sóng âm thanh, có dạng 𝑝(𝑡) = 𝐴𝑠𝑖𝑛(2𝜋𝜔𝑡) tính đến cách chia tỷ lệ tọa độ nguyên tử sau: 𝒓𝑖 = 𝜇𝒓𝑖 (𝑖 = ⋯ 𝑁) với hệ số: 𝛥𝑡 𝜇 = [1 − (𝑃(𝑡) − 𝑃0 − 𝐴𝑠𝑖𝑛(2𝜋𝜔𝑡))]1Τ3 , 𝜏𝑝 Trong 𝜔v A tần số biên độ siêu âm, N P (t) số lượng nguyên tử áp suất tức thời hệ Chiều dài hệ thống chia tỷ lệ 𝐿 = 𝜇𝐿 thể tích hệ trở thành 𝛥𝑡 𝑉 = (𝜇𝐿)3 = [1 − 𝜏 (𝑃(𝑡) − 𝑃0 − 𝐴𝑠𝑖𝑛(2𝜋𝜔𝑡))]𝑉0 𝑝 Chương trình mơ GROMACS [46] với chương trình siêu âm sử dụng cho tất mô Trong tất mô phỏng, siêu âm với tần số 𝜔 = 20 MHz sử dụng biên độ siêu âm thay đổi từ 100 đến 500 bar Áp suất tham chiếu, 𝑃0 = bar, số thời gian để nối hệ với bể áp suất 𝜏𝑝 = ps độ nén đẳng nhiệt 4.5 × 10−5 𝑏𝑎𝑟 −1 sử dụng Để đảm bảo thiệt hại cho màng nhiệt gây siêu âm, kết nối màng lipid nguyên tử nước vào bể nhiệt nhiệt độ 300 K cách sử dụng phương pháp Berendsen [47] với số kết nối với bể nhiệt 0.1 ps Các phương trình chuyển động tích hợp thuật tốn bước nhảy vọt với bước thời gian nhỏ fs Các tương tác tĩnh điện tính tốn phương pháp Ewald lưới hạt mức cắt 1.4 nm [48] Ngưỡng 1,4 nm sử dụng cho tương tác van der Waals Danh sách cặp không thay đổi cập nhật sau fs Dữ liệu lưu cho 25 ps cho phân tích Sự thay đổi cấu trúc màng lipid tác động siêu âm Viện Khoa học Cơng nghệ Tính tốn TP Hồ Chí Minh Page gether, after many decades, the induced-HIFU molecular mechanism of BBB opening is still unclear One of the reasons that prevents further development of the HIFU method is probably due to the fact that a safer method was latter developed by Hynynen and colleagues in 2001 In this method, a low intensity focused ultrasound (LIFU) is used in combination with microbubbles to open the BBB Microbubbles concentrate ultrasound energy to desirable areas, thus low intensity ultrasound is sufficient to open the BBB, and cell lesions could therefore be minimised[10] Since then, a large number of studies have been carried both in in vitro and in vivo, focusing on different aspects One of the important aspects is the fundamental interaction mechanisms of LIFU with microbubbles, and effects of bubble cavitation on the BBB[20] Another aspect is the influence of ultrasound parameters such as pressure amplitude, frequency, burst length, pulse repetition frequency, sonication duration, and parameters of microbubbles such as dose, diameter and type, on the BBB opening mechanism[21–24] Also, many works have focused on the quantification of the permeability of the disrupted brain tissues[25, 26], test the delivery of new drugs, evaluate and monitor the response of treatments[27–30] Nevertheless, the exact molecular mechanism of the LIFU induced BBB opening is still unknown Compared to experiments, theoretical studies of the ultrasound induced BBB opening are still at their infancy Several mathematical models have been developed, which are essentially composed of coupled partial differential equations and solved numerically[31–37] These studies provide valuable information on the effect of bubbles cavitation and ultrasound on the vessel wall However, relying on continuum mechanics, these models cannot capture detailed conformational changes of the BBB at the molecular level At the molecular level, a number of simulations employing coarse-grained and atomistic molecular models have been carried out to study the effects of the bubble inertial cavitation and shock waves on the cell membranes These studies have confirmed that inertial cavitation can induce the formation of membrane pores[38–47] Recently, a molecular dynamics (MD) simulation study based on a coarse-grained model explored the possibility of opening a simple tight junction due to shock wave induced bubble collapse[48] The molecular mechanisms of the BBB opening under HIFU and LIFU irradiations are very complex, because they results from interaction of ultrasound and bubble interaction with the cellular membranes and the tight junctions To simulate these interactions, in principle we must have a bubble model, an ultrasound simulation method and a molecular BBB model With this in mind, we have recently developed a bubble model and a focused ultrasound simulation method[49–51] This allows us to study the molecular mechanism of the interaction between ultrasound with lipid cell membrane, in the presence of bubble[51] or without bubble[52] For the molecular BBB model, Nangia and coworkers have recently obtained the structure of the claudin-5 protein via homology modelling of the experimental structure of the claudin-15 protein Their MD simulations have shown that claudin-5 monomers embedded in seven membranes with different lipid compositions readily form stable cisdimers, which subsequently form contiguous strands and high-order assemblies[53] These cis-dimers are used to construct the trans interfaces, which are formed when claudin-5 proteins on adjacent cells interact head-on to form the tight junction assembly, by using docking methods in combination with symmetric refinement[54] Fundamental pore features, such as diameter and length of the pore were characterized in details[55] As a first attempt to provide more insights into the ultrasound induced BBB opening molecular mechanism due to HIFU or LIFU, in this work, based on the above mentioned achievements, we carry out MD simulations aimed at understanding the impact of ultrasound alone on the cell membranes and tight junction using the minimal BBB model In particular, we wish to understand which part of the BBB model: membrane or tight junction resists more to the ultrasound The impacts of the bubble cavitation, and the construction of more sophisticated BBB models will be carried out in next studies II A METHODS The minimal BBB model The family of claudin proteins forms tight junctions in endothelial and epithelial cells[56] Among 27 known members of the claudin family, the claudin-5 is the key tight junction protein with an expression level that is much higher than other claudin proteins[57–59] Thus, we use only claudin-5 protein to construct the tight junction Our minimal BBB model consists of two lipid bilayer membranes connected together by a paracellular pore formed by two claudin-5 cis-dimers The molecular structure is shown in Fig.1 and briefly described here First, the structure of the claudin-5 monomer is obtained via homology modelling of the experimental structure of the claudin-15 protein It consists of four transmembrane helix bundles, two extracellular loops and an intracellular loop The structure of a claudin-5 cis-dimer in the phosphatidylcholine (POPC) membrane was obtained from previous MD simulations[53], where two claudin-5 monomers are mediated by the transmembrane helix contact Then, a pair of two cis-dimers is used to construct the paracellular pore by using the docking methods in combination with symmetric refinement This minimal model captures the fundamental interactions responsible for the BBB tight junction assembly[54, 55], and serves as a starting point for simulating effect of ultrasound on the BBB We use the CHARMM36 force field[60] to model the protein and lipid The whole model is solvated in a water box consisting of ∼ 180 000 waters The initial dimen- V [nm ] p [bar] 90 degree T [K] E [x 10 kJ/mol] P [bar] A Figure 1: (A) The molecular structure, viewed from two angles, of two claudin-5 cis-dimers (green and orange) forming a paracellular pore (B) The molecular structure of our minimal BBB model where cis-dimers are embed into two lipid membranes of two neighbouring cells The water is filled between and outside two lipid bilayers sions of the unit cell are (Lx , Ly , Lz ) = (10, 10, 18) nm Starting from this structure, an equilibrium MD simulation is carried out for 100 ns in the NPT ensemble with the pressure P = bar and temperature T = 300 K, employing the GROMACS simulation package[61] The last structure is used as initial structure for ultrasound simulations The ultrasound simulation method In the simulation, the effect of the sound wave, which has the form p(t) = A sin(2πωt), 315 310 305 300 A 20 40 60 80 A=100 bar A=200 bar 100 A=400 bar A=450 bar 20 40 60 80 100 20 40 60 80 100 120 B 120 C 120 D 20 40 60 80 100 120 E 20 40 60 80 100 120 Figure 2: Time evolution of various quantities, including the ultrasound (A), the system volume (B), the pressure (C), the potential energy (D) and the temperature (E) Shown are results obtained with different ultrasound intensities A = 100, 200, 400 and 450 bar The period of the ultrasound is τ = 20 ns (50 MHz) B B 100 50 -50 -100 1800 1750 1700 1650 1600 600 300 -300 -600 -1.26 -1.28 -1.3 -1.32 -1.34 320 The GROMACS simulation package[61] coupled to our code of ultrasound is used for all the simulations In all simulations, the ultrasound with frequency ω = 20 MHz is used, and the amplitude is varied from 100 to 500 bar The reference pressure, P0 = bar, the pressure coupling constant, τp = ps and an isothermal compressibility of 4.5 × 10−5 bar−1 are used To ensure that the damage to the membrane is not due to heat generated by work done by ultrasound, we couple both membrane and water to the heat bath at 300 K employing the Berendsen coupling method[62] with a temperature coupling constant of 0.1 ps The equations of motion are integrated using the leapfrog algorithm with a small time step of fs The electrostatic interactions are calculated using the particle mesh Ewald method and a cutoff of 1.4 nm [63] A cutoff of 1.4 nm is used for the van der Waals interactions The nonbonded pair lists are updated every fs The data is saved for every 25 ps for subsequent analyses III (1) is taken into account by scaling the coordinate of atoms as ri = µri (i = · · · N ) with i1/3 h ∆t , (2) (P (t) − P0 − A sin(2πωt)) µ = 1− τp where ω and A are the frequency and amplitude of the ultrasound, N and P (t) are the number of atoms and instantaneous pressure in the system, respectively The length of the system is scaled as L = µL, and the volume becomes V = (µL)3 = [1− ∆t τp (P (t)−P0 −A sin(2πωt))]V0 A RESULTS Response of the system to the ultrasound In all simulations, the ultrasound period τ = 20 ns (50 MHz) is used, and the amplitude is varied in the range A = 100 − 500 bar To obtain a first impression, Fig.2(A) shows, as an example, the time evolution of the ultrasound p(t) with amplitude A = 100 bar The negative and positive periods are called the rarefaction and compression phases, respectively For each ultrasound amplitude, the simulation is run for 120 ns, i.e the sys- B Response of the claudin-5 tight junction to the ultrasound To understand the effect of ultrasound on the claudin5 tight junction, we calculate the root-mean-square dis- 0.5 A RMSD [nm] 0.4 0.3 A=100 bar A=200 bar A=400 bar A=450 bar 0.2 0.1 0 20 40 60 80 100 120 20 B 15 Hbond tem is excited by five ultrasound pulses, and the response of the system to the ultrasound is shown in Fig Let us first present the results for the case A = 400 bar During the first rarefaction phase, i.e p(t) ≤ (t ≤ τ /2), the system is expanded with an increase in the volume from the initial value V ∼ 1650 nm3 to V ∼ 1700 nm3 at ns [Fig.2(B)] As a consequence, the pressure inside the system becomes negative, and decreases from the initial equilibrium value of bar to -400 bar [Fig.2(C)] As a result, atoms are pulled apart, thus the attractive interactions, including the Coulombic interaction, intramolecular lipid bending and torsion interactions, and long ranged van der Waals attractions become dominant Thus the potential energy of the system is increased from -1.30 ×106 to -1.29 ×106 kJ/mol [Fig.2(D)] Because the system is coupled to the heat bath, its temperature is always fluctuating around the reference value of 300 K [Fig.2(E)] Within the time interval τ /4 ≤ t ≤ τ /2, the ultrasound intensity is reduced, but still in the rarefaction phase, thus the system is relaxed, as indicated by the decrease in the volume, pressure and potential energy, and achieved the equilibrium state at t = τ /2 = 10 ns Next, the ultrasound is in the compression phase, i.e p(t) ≥ (τ /2 ≤ t ≤ τ ), the system is compressed as shown by the decrease and increase in the volume and pressure, respectively The atoms are pushed closely, thus the short ranged van der Waals attractions are dominated, therefore the potential energy of the system becomes more negative After t ≥ 3τ /4, the ultrasound intensity is reduced, but still in the compression phase, the system volume is increased, the pressure is reduced and the potential energy is increased At t = τ , the ultrasound is vanished and the system reaches the equilibrium state, finishing one ultrasound irradiation period The response of the system is repeated for the next four ultrasound periods As seen from Fig.2, the response of the system to the weaker ultrasounds with A = 100 or 200 bar, is similar to that of A = 400 bar, but less intense However, by increasing the ultrasound intensity to A = 450 bar, the system is able to oscillate with the ultrasound within the first period (t ≤ 20 ns), but then explosive during the second ultrasound period That is, at t = 25 ns, when the ultrasound is fully expanded, the system volume suddenly increases to a large value, the pressure is reduced immediately to zero, and the potential energy becomes very large because atoms are largely separated apart Although the system is coupled to the heat bath, as the explosion takes place very fast, the velocity of atoms are not rapidly rescaled and the temperature of the system is significantly increased Technically, the simulation is crashed 10 0 20 40 60 time [ns] 80 100 120 Figure 3: The time evolution of the RMSD of the claudin-5 tight junction with respect to the initial structure (A), and the total number of intermolecular Hbonds between two dimers (B) Shown are results obtained from simulations using ultrasound with period of 20 ns and various intensities placement (RMSD) of two cis-dimers (each anchored to each membrane) with respect to the initial structure for different ultrasound intensities As seen from Fig.3(A), the RMSDs undergo small increases of ∼ 0.25 nm compared to the initial structure, despite large variations in the ultrasound intensities from 100 to 450 bar Interestingly, although the system is explosive with A = 450 bar as shown above, but the RMSD does not change much, indicating that the overall structure of the tight junction is well-maintained Fig.3(B) shows the total number of the intermolecular hydrogen bonds (Hbond) between two dimers We observe that the number of Hbond fluctuates in time, but it does not follow the oscillation of the ultrasound pressure After 120 ns, two dimers are slightly separated, as indicated by a small reduction in the number of Hbond from to for A = 400 bar Even at 450 bar where the system is explosive, there is no large change in the number of Hbond, indicating that the tight unction is not disrupted Next, we wish to understand the impact of the ultrasound on the secondary structures of the claudin-5 proteins To this end, we calculate the population of the secondary structures of two dimers by using the STRIDE program[64] and the results are shown in Fig.4 As seen, the extracellular loops of claudin-5 proteins, which are mainly in the β structure, contribute initially to ≈ 15% of the secondary structure population The transmembrane helix bundles contribute to ≈ 65%, and ≈ 20% are attributed to the turn and coil structures of the linkers and of the intracellular loops We observe that the secondary structures hardly undergo any changes, irrespective large variations in the ultrasound intensities, even at 450 bar where the system is explosive All these results show that both 3D and 2D structures 60 60 β 50 40 30 30 20 20 10 10 0 20 40 60 80 100 120 70 A=100 bar A=200 bar A=400 bar A=450 bar 20 40 80 60 100 120 50 50 40 40 30 30 20 20 10 20 40 80 60 time [ns] 100 120 0 20 40 80 60 time [ns] 100 SCH sn-1 A=100 bar A=200 bar A=400 bar A=450 bar 0.1 0 20 40 60 80 100 120 0.2 SCH sn-2 0.1 0 20 40 60 80 100 40 120 time [ns] Figure 5: Time evolution of the carbon-hydrogen order parameters of the first sn−1 (upper) and second sn−2 lipid tails, respectively (lowver) These results are averaged over 15 and 17 order parameters of C-H vectors of the first and second lipid tails, respectively, and obtained from simulations using ultrasound with period of 20 ns and various intensities of individual claudin-5 proteins, and of the whole tight junction are largely maintained, and the tight junction is not disrupted by the ultrasound irradiation 60 t 80 100 120 B 0.9 0.6 0.3 20 40 60 80 100 120 C 0.2 0.18 0.16 20 40 60 time [ns] 80 100 120 Figure 6: Time evolution of the shortest intra-distances between two monolayers of the first (A) and second (B) cell membrane lipid bilayers The shortest inter-distance between two lipid bilayer cell membranes is shown in the (C) Shown are results obtained from simulations using ultrasound with period of 20 ns and various intensities C 0.2 20 1.2 120 Figure 4: The time evolution of the population of various secondary structures For each structure, shown is the total populations of four claudin-5 proteins pertaining to the junction Shown are results obtained from simulations using ultrasound with period of 20 ns and various intensities A 0.16 10 A=400 bar A=450 bar 0.18 coil 60 turn A=100 bar A=200 bar 0.2 1.5 70 60 helix 50 40 distance [nm] 70 distance [nm] 70 distance [nm] secondary structure content [%] secondary structure content [%] Response of the lipid bilayer membranes to ultrasound To investigate the response of the lipid bilayers to the ultrasound, we first calculate the carbon-hydrogen order parameters of the lipid tails SCH = h3 cos2 θ−1i/2, where θ is the angle between a C-H bond vector and the bilayer normal The angular brackets represent molecular and temporal ensemble averages For a POPC lipid, the number of C-H vectors in the first, sn−1 , and second, sn−2 , tails are 15 and 17, respectively To obtain an overall picture on the order of each tail, we simply take the average of the order parameters of all C-H vectors pertaining to that tail The results are shown in Figs.5 for two tails As seen, for the ultrasound intensities A = 100, 200 and 400 bar, the order parameters of both tails not undergo any oscillation with ultrasounds, indicating that lipids are quite ordered However, with A = 450 bar, the order parameters drop to zeros at t = 25 ns, a moment when the system is explosive, indicating significantly structural changes in the membranes To understand this, we calculate the shortest intra-distance between two monolayers of each bilayer, and the shortest inter-distance between two bilayer membranes The latter measures the gap of the tight junction between two cells The results are shown in Fig.6 As seen, two monolayers of individual lipid bilayer membranes are always separated by intradistances of ≈ 0.18 nm [Figs.6(A), (B)], and two lipid bilayer membranes of two adjunct cells are stay intact at ≈ 0.9 nm for ultrasounds with A = 100, 200 and 400 bar A visual inspection of a snapshot at 120 ns of the trajectory with A = 400 bar shown in Fig.7(A) confirms A C B D Figure 7: Snapshots of the system at 120 ns obtained by using ultrasound intensity of 400 bar (A), and at 24.5, 24.7 and 25 ns obtained by using ultrasound intensity of 450 bar (B, C, D respectively) The ultrasound period of 20 ns is used in all simulations While the bilayers are disrupted, the tight junction is well-maintained these results, that is the membranes and the claudin-5 tight junction structures are well-maintained, and similar to the initial structures shown in Fig.1(B) However, with A = 450 bar, we observe that the intra-distances between two monolayers of individual bilayers increase and become suddenly large at t = 25 ns when the system is explosive At the same time, the inter-distance between two bilayer membranes is reduced from 0.9 to 0.5 nm To understand this, we show in Fig.7 two snapshots just before collapsing at t = 24.5 and 24.7 ns, and one snapshot at the collapsed moment at 25 ns We observe that when the system is about explosive, the two monolayers of individual membranes are separated apart, and this pushes two lipid bilayer membranes of two cells to each others [Figs.7(B),(C)] When the system is fully explosive, two monolayers are highly separated, two cells get closer, and lipids become disorder This explains the reduction in the order parameters shown in Fig.5 Nevertheless, the claudin-5 tight junction is hardly affected, and still maintained by two monolayers of two adjunct cells [Fig.7(D)] IV DISCUSSION AND CONCLUDING REMARK In the HIFU experiments the direct interaction of ultrasound with the BBB is the cause of the BBB disruption For the LIFU experiments, it is widely believed that ultrasound induces the stable and/or inertial cavitation of bubble, which in turn exerts acoustic shear forces and/or shockwave on the BBB, leading to the BBB opening However, in principle, ultrasound also interacts directly with the cells and tight junction This results in complicated BBB opening mechanisms, especially at the molecular level Thus, our strategy is to split this complex problem into simpler tasks In this work, our primary aim is to understand how the lipid bilayer cell membrane and the tight junction respond to the ultrasound and whether the cell membrane and/or the claudin-5 junction are affected by ultrasound Before discussing the results, the limitations of our study should be pointed out First, our BBB model is very simple, composed of only two lipid bilayers, mimicking two portions of neighbouring cells The tight junction is composed of a paracellular pore formed by two cis-dimers of claudin-5 proteins Nevertheless, given the fact that currently there is no experimental molecular structure of the BBB, this computational model still can serve as a starting point for MD simulations, which can provide some insights into the BBB opening mechanism Second, the experimental frequency of the ultrasound is usually tuned between ω = 0.5−50 MHz, the ultrasound pressure is in the range 10 - 50 bar for LIFU experiments[65, 66], and 100 - 1000 bar for HIFU experiments[67] The ultrasound irradiation duration is usually in the second to minute timescales[65, 66] To be close to experiments, we use an ultrasound frequency of 50 MHz for our simula- Figure 8: (Upper) A structure of the BBB model where components are labeled as M1 and M2 for two monolayers of a cell membrane bilayer; W1 and W2 for waters between two bilayers and above the membrane M2, respectively; C1 and C2 for two claudin-5 dimers embed in two cell membrane lipid bilayers (Lower) Time evolution of the potential energy acting on M1 (black) and M2 (red) Waters are not shown for clarity tions To be computationally feasible with current computer technology, the simulation time can only be in the nanosecond scale This means that our system is only irradiated by several ultrasound cycles We use the typical HIFU intensity in the range 100-500 bar for our simulations, which are strong enough to allow us to obtain response of the BBB within the simulation timescales For low intensities used in LIFU experiments, the timescale of simulations would be much longer to obtain any significant responses of the BBB, and this is beyond our current capability Our simulations show that with weak ultrasound intensities A ≤ 400 bar, the structures of both lipid cell membranes and claudin-5 tight junction are well-maintained This implies that for the HIFU experiments, the direct interaction of ultrasound with BBB does not yield to the opening neither at the cells nor tight junction In the context of the LIFU experiments, where the ultrasound intensity is even weaker, this interaction can be safely excluded in the interpretation of the BBB opening mechanism, confirming that the interaction of the bubble cavitation with BBB is the main cause as hypothesised With stronger ultrasound intensities, say A = 450 bar, we observe that two monolayers within individual lipid bilayers are detached following the expansion of the simulation box during the ultrasound rarefaction phase, and this could result in the damage of the cell membrane In contrast, the tight junction between two cells is hardly affected This suggests that the BBB disruption likely takes place at the cell surface instead of at the cell junction To explain this, we calculate the interaction potential energy acting on two monolayers M1 and M2 of a bilayer shown in Fig.8 As seen, the monolayer M1 is stabilised by potential energy V(M1) which is contributed by four interactions V(M1) = V(M1-C1) + V(M1-W1) + V(M1C2) + V(M1-M2), and the monolayer M2 is stabilised by three interactions V(M2)=V(M2-C1)+ V(M2-W1) + V(M2-M1), where V(M1-C1) is the interaction potential energy between the monolayer M1 and the claudin cisdimers C1 embed in M1, V(M1-W1) is the interaction energy between M1 and waters between two cells, and so on Fig.8 shows the time evolution of V(M1) and V(M2) for an example case A = 400 bar As seen, these potential energies oscillate with the ultrasound, i.e, their values increase and decrease using the ultrasound rarefaction and compression phases, respectively However, on average, V(M1) is ∼ -2000 kJ/mol lower than V(M2), suggesting that the monolayer M1 is more stable than M2 Therefore, while the M1 layer is anchored by the tight junction, the M2 layer moves following the expansion of the system during the ultrasound rarefaction phase, resulting in the detachment of the layer M2 from the layer M1 In the real BBB, the tight junction is much more complicated, composed of many claudin-5 proteins and other proteins Of course, this will increase the tightness of the junction but also increase the interaction between cell membranes with transmembrane proteins of the tight junction The question is then whether the BBB is still disrupted at the cell membrane position? To this end, we construct an empirical BBB model whose tight junction is composed of four pairs of two cis-dimers, each pair is located on the corner of a square with edge lengths of 10 nm [Fig.9] This is obtained by translating a pair of two cis-dimers of the above model [Fig.1] in the membrane surface for 10 nm along the x and y-axis We then carry out simulations using the same parameters as used for the previous model Fig.9 shows snapshots after six ultrasound irradiation periods, i.e at 120 ns for A = 400 and 450 bar Consistent with the results of the single cis-dimer pair model, the structure of the four cis-dimer pairs model is well-maintained with A = 400 bar, but with A = 450 bar, two monolayers of individual bilayer membranes is detached, suggesting that the BBB is also damaged at the cell plasma membrane rather than at the tight junction Let us discuss how the separation of monolayers of the cell membrane results in the enhancement of the drug permeability The separation of two monolayers in a bilayer results in air compartments at the mid- Figure 9: Snapshots of the system whose tight junction is composed of four pairs of claudin-5 cis-dimers at 120 ns obtained by using ultrasound intensity of 400 bar (left), and at 25 ns obtained by using ultrasound intensity of 450 bar (right) The ultrasound period is 20 ns in all simulations Waters are not shown for clarity The separation of two monolayers is clearly seen from the snapshot at 450 bar dle of the lipid bilayer (Figs.7(B),(C), (D)) This phenomenon has been observed experimentally, in the context of drug delivery studies using liposomes That is, under LIFU irradiation, gas bubble nuclei may be formed in the hydrophobic region of the lipid bilayer of the liposomes These nuclei grow until they permeate the membrane, forming a transient pore through which the drug is released[68, 69] At the molecular level, the presence of the air compartments alternates the diffusive process of drugs across the membrane Indeed, a number of simulations have shown that the free energy profile for translocation across the lipid bilayer of drugs usually exhibits a high barrier at the middle of the two monolayers[70– 73] If two monolayers are separated, then this leaves a large space in the middle of the bilayer, resulting in high population of drugs in this area, i.e the free energy barrier is reduced, and therefore the drug permeability is increased We find that the molecular mechanism of the ultrasound induced permeability of BBB is similar to that observed in liposomes, but the origin of the air compartment formation may be different It is expected that considerable heterogeneity exists in the preparations of liposomes, and it is unlikely that all liposomes are unilamellar Thus, gas compartments are likely formed at the flexible areas of the membrane of liposomes For the BBB, because the outer layer of the cell membrane is anchored by the tight junction, but the inner layer is contracted during the rarefaction phase of the ultrasound, therefore this separation results in air compartments inside the cell membrane bilayer [Fig.8] As reviewed earlier in this paper, under HIFU irradiation, several experiments observed the BBB disruption within the damaged areas[16, 17], but other results showed the BBB disruption without damage of the brain parenchyma[18, 19] In both cases, it was unclear whether the BBB disruption is due the cell damage or opening of the tight junction[11] Our simulations could suggest that the BBB may be disrupted due to the damaged of the cells, but the tight junction is unlikely happened A large body of literature confirms that LIFU-induced BBB opening involves at least four mechanisms: transcythosis, transendothelial opening, i.e fenestration and pore formation, tight junction opening at low ultrasound intensities, and passage through damaged cell at high ultrasound intensities[20, 74] Our simulation study suggests that with low ultrasound intensities used in LIFU experiments, the impact of the direct interactions of ultrasound with cells and with tight junctions can be negligible Thus, the interaction between bubble and the BBB plays an essential role Indeed, it has been suggested that the acoustic microstreaming generated by stable bubble cavitation stimulates the transcythosis, transendothelial opening, and the inertial cavitation may induce cell membrane pore formation The tight junction opening may due to the volume expansion of microbubbles in the vessel, which in turn exerts direct mechanical forces on the BBB However, at high ultrasound intensity, the direct interaction of ultrasound with the BBB becomes significant, leading to the cell damage as observed by experiments[20, 74], and confirmed by our simulations In conclusions, we have performed a molecular dynamics simulation study to understand the molecular impacts of HIFU and LIFU on the BBB opening The study shows that at low ultrasound intensities (A ≤ 400 bar), the structures of the cell membranes and tight junction are well-maintained, implying that the direct interaction of ultrasound with the BBB is not responsible for the experimentally observed BBB opening At high intensities, the rarefaction of ultrasound pulls two monolayers of individual cell membrane lipid bilayers apart, creating air compartments inside the bilayer This reduces the free energy barrier for translocation across the lipid bilayer of drugs, thus enhances the drug permeability At very high intensities, two monolayers are largely separated, resulting cell damage, implying that the BBB is primarily disrupted at the damaged areas as observed by experiment The MD simulations of the full system composed of ultrasound, bubbles and BBB will be carried out to understand impacts of bubble cavitation on the BBB model Also, more detailed BBB models will be constructed These works may open the door of drugs delivery for Alzheimer’s diseases[75, 76] Acknowledgments This work has been supported by the Department of Science and Technology at Ho Chi Minh City, Vietnam (grant grant 10/2018/HDKHCNTT), CNRS, the World Bank and the Vietnam Ministry of Science and Technology FIRST project (grant number 13/FIRST/1.a/VNU1), the National Science Foundation (NSF, grant SI2-SEE-1534941), the National Institute of Health (NIH-R01GM118508) and the CINES center for providing computer facilities (project A0010707721) DATA AVAILABILITY The data that support the findings of this study are available from the corresponding author upon reasonable request [1] R Daneman and A Prat, Cold Spring Harb Perspect Biol 7, a020412 (2015) [2] D Janigro, Epilepsia 53, 26 (2012) [3] N Vykhodtseva, N McDannold, and K Hynynen, Ultrasonics 48, 279 (2008) [4] W M Pardridge, Discov Med 6, 139 (2006) [5] R E Gross, R L Watts, R A Hauser, and et al., Lacet Neurol 10, 509 (2011) [6] J W J Marks, J L Ostrem, L Verhagen, and et al, Lacet Neurol 7, 400 (2008) [7] W M Pardridge and R J Boado, Methods Enzymol 503, 269 (2012) [8] K Matsukado, M Sugita, and K L Black, Brain Res 792, 10 (1998) [9] S I Rapoport, Proc Natl Acad Sci USA 97, 5696–5698 (2000) [10] K Hynynen, N McDannold, N Vykhodtseva, and F A Jolesz, Radiology 220, 640 (2001) [11] A H Mesiwala, L Farrell, H J Wenzel, D L Silbergeld, L A Crum, H R Winn, and P D Mourad, Ultrasound Med Biol 28, 389 (2002) [12] B.Baseri, J J Choi, Y Tung, and E E Konofagou, Ultrasound Med Biol 36, 1445 (2010) [13] N McDannol, C D Arvanitis, N Vykhodtseva, and M S Livingstone, Cancer Res 72, 3652 (2012) [14] C Horodyckid, M Canney, A Vignot, R Boisgard, A Drier, G Huberfeld, C Francois, A Prigent, M Santin, C.Adam, et al., J Neurosurg 10, (2016) [15] J W Barnard, W J Fry, F J Fry, and R F Krumins, J Comp Neurol 103, 459 (1995) [16] L Bakay, J H T Ballantine, T F Hueter, and et al, Arch NeurPsych 76, 457 (1956) [17] J T Patrick, M N Nolting, S Goss, K Dines, J Clendenon, M Rea, and R Heimburger, Adv Exp Med Biol 267, 369 (1990) [18] J H Ballantine, E Bell, and J Manlapaz, J Neurosurg 17, 858 (1960) [19] N Vykhodtseva, K Hynynen, and C Damianou, Ultrasound Med Biol 21, 967 (1995) [20] N Sheikov, N McDannold, N Vykhodtseva, F Jolesz, and K Hynynen, Ultrasound Med Biol 30, 979 (2004) [21] J J Choi, S A Small, and E E Konofagou, IEEE Ultrasonics Symposium pp 540–543 (2006) [22] K Hynynen, N McDannold, N A Sheikov, F A Jolesz, and N Vykhodtseva, Neuroimage 24, 12 (2005) [23] M A O’Reilly, Y Huang, and K Hynynen, Phys Med Biol 55, 5251 (2010) [24] N Sheikov, N McDannold, S Sharma, and K Hynynen, Ultrasound Med Biol 34, 1093 (2008) [25] F Vlachos, Y Tung, and E E Konofagou, Phys Med Biol 55, 5451 (2010) [26] F Vlachos, Y Tung, and E E Konofagou, Mag Reson Med 66, 821 (2011) [27] P Chen, H Liu, M Y Hua, H Yang, C Y Huang, P C Chu, L Lyu, I C Tseng, L Feng, H C Tsai, et al., Neuro-Oncology 12, 1050 (2010) [28] H L Liu, M Hua, P Chen, P Chu, C Pan, H Yang, C Huang, J Wang, T Yen, and K Wei, Radiology 255, 415 (2010) [29] J Jordao, C A Ayala-Grosso, K Markham, Y Huang, R Chopra, J McLaurin, K Hynynen, and I Aubert, PloS One 5, e10549 (2010) [30] S B Raymond, L H Treat, J D Dewey, N J McDannold, K Hynynen, and B J Bacskai, PloS One 3, e2175 (2008) [31] T Nhan, A Burgess, L Lothar, and K Hynynen, Phys Med Biol 59, 5987 (2014) [32] W Wiedemair, Z Tukovic, H Jasak, D Poulokakos, and V Kurtcuoglu, Phys Med Biol 57, 1019 (2012) [33] N Hosseinkhah, D E Goertz, and K Hynynen, IEEE Trans Biomed Eng 62, 1293 (2015) [34] N Hosseinkhah and K Hynynen, Phys Med Biol 57, 785 (2012) [35] N Hosseinkhah, H Chen, T Matula, P N Burns, and K Hynynen, J Acoust Soc Am 134, 1875 (2013) [36] S Qin and K W Ferrara, Phys Med Biol 51, 5065 (2006) [37] W Wiedemair, Z Tukovic, H Jasak, D Poulikakos, and V Kurtcuiglu, Biomech Model Mechaobiol pp DOI:10.1007/s102337–016–0840–z (2016) [38] G C Ganzenmuller, S Hiermaier, and M O Steinhauser, Soft Matter 7, 4307 (2011) [39] A Choubey, M Vedadi, K Nomura, R K Kalia, A Nakano, and P Vashishta, Appl Phys Lett 98, 023701 (2011) [40] D Schanz, B Metten, T Kurz, and W Lauterborn, New J Phys 14, 113019 (2012) [41] A S K Nomura, R K Kalia, A Nakano, and P Vashishta, Phys Rev Lett 111, 184503 (2013) [42] K Koshiyama, T Kodama, T Yano, and S Fujikawa, Biophys J 91, 2198 (2006) [43] K P Santo and M L Berkowitz, J Phys Chem B 119, 10 8879 (2014) [44] K P Santo and M L Berkowitz, J Chem Phys 140, 054906 (2014) [45] H Fu, J Comer, W Cai, and C Chipot, J Phys Chem Lett 5, 413 (2015) [46] U Adhikari, A Goliaei, and M L Berkowitz, J Phys Chem B 119, 6225 (2015) [47] V Agrawal, P Peralta, Y Li, and J Oswald, J Chem Phys 145, J Chem Phys 145, 104903 (2016) [48] A Goliaei, U Adhikari, and M L Berkowitz, ACS Chem Neurosci 6, 1296 (2015) [49] M H Viet, P Derreumaux, and P H Nguyen, J Chem Phys 145, 174113 (2016) [50] M H Viet, M Li, P Derreumaux, and P H Nguyen, J Chem Phys 148, 094505 (2018) [51] M H Viet, M T Phan, M Li, P Derreumaux, W Junmei, N T Van-Oanh, P Derreumaux, and P H Nguyen, J Phys Chem B 123, 71 (2019) [52] M H Viet, M Li, J Wang, P Derreumaux, and P H Nguyen, J Chem Phys 150, 215101 (2019) [53] F J Irudayanathan, J P Trasatti, P Karande, and S Nangia, J Phys Chem B 120, 77 (2016) [54] F J Irudayanathan, N Wang, X Wang, and S Nangia, Ann N Y Acad Sci 1405, 131–146 (2017) [55] N Rajagopal, F J Irudayanathan, and S Nangia, Int J Mol Sci 20, 5583 (2019) [56] J Piontek, L Winkler, H Wolburg, S L Muller, N Zuleger, C Piehl, B Wiesner, G Krause, and I E Blasig, The FASEB Journal 22, 146 (2008) [57] T Nitta, M Hata, S Gotoh, Y Seo, H Sasaki, N Hashimoto, M Furuse, and S Tsukita, J Cell Biol 161, 653 (2003) [58] J Rossa, D Lorenz, M Ringling, A Veshnyakova, and J Piontek, Ann N Y Acad Sci 1257, 59 (2012) [59] J Cording, J B J N Kading, C Bellmann, C Tscheik, J K Westphal, S Milatz, D Gunzel, H Wolburg, J Piontek, and et al., J Cell Sci 126, 554 (2013) [60] A D M Jr., D Bashford, M Bellott, R L Dunbrack, [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] J D Evanseck, M J Field, S Fischer, J Gao, H Guo, S Ha, et al., J Phys Chem B 102, 3586 (1998) E Lindahl, B Hess, and D van der Spoel, J Mol Mod 7, 306 (2001) H J C Berendsen, J P M Postma, W F van Gunsteren, A Dinola, and J R Haak, J Chem Phys 81, 3684 (1984) T Darden, D York, and L Pedersen, J Chem Phys 98, 10089 (1993) D Frishman and P Argos, Proteins 23, 566 (1995) M Aryal, C D Arvanitis, P M Alexander, and N McDannold, Adv Drug Deliv Rev 72, 94 (2014) K T Cheng, K C Wei, and H L Liu, Front Pharmacol 7, 86 (2019) M S Canney, M R Bailey, and L A Crum, J Acoust Soc Am 124, 2406–2420 (2008) S L Huang and R C MacDonald, Biochim Biophys Acta 1665, 134 (2004) A Schroeder, J Kost, and Y Barenholz, Chem Phys Lipids 162, (2009) W Shinoda, Biochimica et Biophysica ActaBiomembranes 1858, 2254 (2016) M Orsi, W E Sanderson, and J W Essex, J Phys Chem B 113, 12019 (2009) T X Xiang and B D Anderson, J; Chem Phys 110, 1807 (1999) Y Wang, J Cohen, W F Boron, K Schulten, and E Tajkhorshid, J Struct Biol 157, 534 (2007) N Sheikov, N McDannold, F Jolesz, Y Z Zhang, K Tam, and K Hynynen, Ultrasound Med Biol 32, 1399 (2006) J Nasica-Labouze, P H Nguyen, O Berthoumieu, F Sterpone, N V Buchete, S Cote, A D Simone, A Doig, P Faller, A Garcia, et al., Chem Rev 115, 3518 (2015) A J Doig and P Derreumaux, Curr Opin Struct Biol 30, 50 (2015) Molecular mechanism of ultrasound induced drug delivery pathways on liposomes Man Hoang Viet1 , Mai Suan Li2,3 , Philippe Derreumaux4 , Junmei Wang1 , Phuong H Nguyen4∗ , Department of Pharmaceutical Sciences, University of Pittsburgh, USA Institute of Physics, Polish Academy of Sciences, Al Lotnikow 32/46, 02-668 Warsaw, Poland Institute for Computational Science and Technology, SBI Building, Quang Trung Software City, Tan Chanh Hiep Ward, District 12 , Ho Chi Minh City, Vietnam CNRS, Universite de Paris, UPR9080, Laboratoire de Biochimie Th´eorique, Paris, France Institut de Biologie Physico-Chimique Fondation Edmond de Rothschlid, PSL Research University, Paris, France (Dated: June 3, 2020) The use of ultrasound in combination with liposomes is a promising approach to improve drug delivery To achieve an optimal drug release rate, it is important to understand the molecular mechanism by which ultrasound opens the pathways for drugs to release from the liposome To this end, we carry out nonequilibrium molecular dynamics simulations for three large liposomes with diameters of ∼ 80 nm under ultrasound with different intensities The results show that the drug release pathway from the DOPC liposome may be through the low lipid packing areas on the surface induced by stretched bilayer, while for the DPPC and DOPC liposomes, the pathways are through pores and/or large damaged sections on the liposome surface The molecular origins of the difference between two types of pathway is explained based the competition between the interaction energy of lipids within monolayers and between monolayers Our results could support experimental findings which show that the rate of release is sensitive to the lipid composition, but disagree with theoretical modelling showing that the rate is dominated by the bilayer thickness Introduction In medicine, an administered drug must penetrate many obstacles in the living system before reaching the desired targets However, most of drugs are rapidly cleaned from the blood stream, lacked of targeting and difficult to cross cell membranes, rendering the drug delivery ineffective Fortunately, these problems have been solved by the use of nanoparticles (NP), which play a role as drug carriers[1–5] The basic idea is that NP carry drugs to desired targets, and then drugs are released from NP by stimuli[6–13] This way offers several advantages, including improves drug stability, bioavailability[14], highly site specific delivery[15], reduced toxicity[16] and drug release ”on demand” without perturbing surrounding cells Currently, liposomes are widely used as carriers because of their high biocompatibility and stability[2, 3, 17, 18] Ultrasound is an efficient stimulus because it can induce locally and invasively drug release via the thermal and/or cavitation effects[19–23] Therefore, the use of liposomes in combination with ultrasound should provide an excellent method for drug delivery Understanding the molecular pathways by which ultrasound enhances the permeability of liposomes is therefore of fundamental interest, and essential for safe and effective implementation of this method To this end, various ultrasound experiments have been carried out to measure the rate and identify pathways of released drugs from liposomes[24– 29] Various mathematical and computational models have also been developed to fit and elucidate experimental release rates[24, 28, 30–35] All together, these studies have suggested that ultrasound stimulates drug release through (i) destruction of large parts of the bilayer, (ii) ∗ Email: nguyen@ibpc.fr formation of pores on the bilayer, and (iii) stretch, i.e enhanced diffusion of the bilayer To support these findings, this work presents for the first time a large scale molecular dynamics simulation study of the ultrasound induced pathways in liposomes Methods In this work, we study three liposomes, each is purely composed of 1,2dioleoyl-sn-glycero-3-phosphocholine (DOPC), 1,2dipalmitoylphosphatidylcholine (DPPC) and phosphatidylcholine (POPC) lipids A liposome is constructed by distributing uniformly lipids on the surface of a sphere having diameter of 80 nm Given the value of the area per lipid of ∼ 0.46 nm2 at 290 K[36], this results in ∼ 63000 lipids for each liposome Each liposome is solvated in a water box consisting of 16.331.931 waters The initial dimensions of the unit cell are (Lx , Ly , Lz ) = (129, 129, 129) nm, leaving a water layer with a thickness of 25 nm between the liposome surface and box edge The coarse-grained MARTINI 2.2 force field[37, 38] is employed to describe the membrane and water As an example, the initial structure of the DPPC liposome is shown in Fig.1(A) An equilibrium MD simulation is carried out for 500 ns for each system in the NPT ensemble with the pressure P = bar and temperature T = 300 K The last structure is used as initial structure for the ultrasound simulations The GROMACS simulation package[39] coupled to our code of ultrasound is used for all the simulations The ultrasound of the form p(t) = A sin(2πωt) is taken into account by scaling the coordinate of atoms as ri = µri (i = · · · N ) with h i1/3 ∆t µ= 1− , (1) (P (t) − P0 − A sin(2πωt)) τp where ω and A are the frequency and amplitude of the ultrasound, N and P (t) are the number of atoms and Figure 1: (A) Initial structure of the DPPC liposome whose inner and outer monolayers are coloured as red and green, respectively The selected snapshots at various moments of the DOPC (B), DPPC (C) and POPC (D) Shown are results obtained from simulations using the ultrasound frequency of 20 MHz and different intensities instantaneous pressure in the system, respectively In all simulations, the frequency ω = 20 MHz is used, and the amplitude is varied from 100 to 600 bar The reference pressure, P0 = bar, the isotropic pressure coupling constant, τp = ps and an isothermal compressibility of 4.5 × 10−5 bar−1 are used To ensure that the damage to the membrane is not due to heat generated by work done by ultrasound, we couple both liposome and water to the heat bath at 300 K employing the Berendsen coupling method[40] with a temperature coupling constant of 0.1 ps The equations of motion are integrated using the leapfrog algorithm with a small time step of fs The electrostatic interactions are calculated using the particle mesh Ewald method and a cutoff of 1.4 nm [41] A cutoff of 1.4 nm is used for the van der Waals interactions The nonbonded pair lists are updated every fs The data is saved for every 25 ps for subsequent analyses Results and discussion During the ultrasound rarefaction and compression phases, the system is expanded and compressed, respectively, and this induces changes in the system pressure We perform simulations with different ultrasound intensities, and find that all three systems are stable up to 500 bar, as indicated by harmonic oscillation in their pressures shown in Fig.2 To find the thresholds at which pathways are formed, we increase progressively the intensity and find that each liposome responds differ- 25 50 100 75 DPPC inner-outer -1.20 ×106 -1.41 ×106 -1.38 ×106 Table I: The total potential energy (in kJ/mol) of the inner, outer monolayers and between two layers liposomes Shown are results obtained from the average of conformations of 100 ns equilibrium MD trajectories at 300 K 500 bar 550 bar 100 75 POPC 500 bar 580 bar 25 50 time [ns] 75 100 Figure 2: Time evolution of the pressure in three liposome systems The pressures of all systems are stably increased and decreased under ultrasound intensity of 500 bar (black lines) At higher ultrasound intensities, the pressures drop to zero at maximum ultrasound amplitudes (red lines), indicating that the systems are explosive A frequency of 20 MHz is used in all simulations ently to the ultrasound At intensities A = 640, 550 and 580 bar, the pressures in the DOPC, DPPC and POPC systems, respectively, are initially negatively increased, but around t = τ /2 = 12.5 ns, when the ultrasound is maximum, the pressures are suddenly decreased to zero, indicating that the systems undergo very large expansions [Fig.2] To reveal the structural changes, Fig.3 shows the time evolution of the inner and outer radii Pn of each liposome A radius is calculated as R(t) = n1 i=1 ri (t), where ri (t) is the distance between the centre of the liposome to the center of mass of the i-th lipid, and n is the total number of lipids pertaining to the inner or outer monolayer At A = 500 bar, all three liposomes are stably expanded and compressed following the decrease and increase of the system pressures as shown in Fig.2 We note that for the DOPC liposome, the outer radius is increased more than the inner radius, thus leaving an empty space in between two monolayers In contrast, for the DPPC and POPC liposomes, the increase or decrease in the outer radius and inner radii are similar, and no empty space is created in between two monolayers Overall, we observe that the expansion amplitude is larger than the compression amplitude This is due to the presence of water inside the liposomes that makes the compression is difficult At intensity A = 640 bar, both inner and outer radii of the DOPC liposome increase rapidly from the initial values ∼ 41 nm to ∼ 65 nm at t ∼ 12 ns [Fig.3] As seen from Fig.1, at t = 12.25 ns, the spherical structure of the inner monolayer is still maintained with a radius ∼ 67 nm, but the outer monolayer is largely stretched with a radius ∼ 200 nm [Fig.3], creating a large air compartment between two monolayers At t = 12.5 ns, the outer layer is largely damaged, the inner layer is, however, remained its spherical shape though very stretched [Fig.1] For the DPPC and POPC liposomes, at intensities A = 550 and 580 bar, respectively, the increase or decrease in the inner radius and outer radius are similar [Fig.3] This means that the distance between two monolayers is always maintained, and no air compartment is formed between two monolayers A visualisation of snapshots, shown in Fig.1, reveals that sizes of the DPPC and POPC liposomes not change much, only pores are formed on the surface The large increases in the radii shown in Fig.3 are due to the contribution from the expelled lipids from the pores, which are far from the liposomes 46 45 44 43 42 41 40 39 46 45 44 43 42 41 40 39 46 45 44 43 42 41 40 39 200 500 bar 640 bar 150 DOPC 50 100 10 15 20 25 30 35 40 45 50 500 bar 50 12.1 200 12.2 12.3 12.4 550 bar 150 100 10 15 20 25 30 35 40 45 50 500 bar 50 11.2 200 11.3 11.4 11.5 11.6 11.7 580 bar 150 100 10 15 20 25 30 35 40 45 50 time [ns] 50 12.6 12.7 12.8 time [ns] 12.9 DPPC 25 11.8 POPC radius [nm] 600 400 200 -200 -400 -600 500 bar 640 bar radius [nm] pressure [bar] 600 400 200 -200 -400 -600 Liposome inner outer DOPC -4.84 ×106 -5.34 ×106 DPPC -4.57 ×106 -5.07 ×106 POPC -4.59 ×106 -5.11 ×106 DOPC radius [nm] pressure [bar] 600 400 200 -200 -400 -600 pressure [bar] 13 Figure 3: Time evolution of the inner (black lines) and outer (red lines) radii of three liposomes All liposomes are stable at ultrasound intensity of 500 bar (left panels), but undergo large changes at higher intensities (right panels) A frequency of 20 MHz is used for all simulations Results and discussion Our simulation results show that the ultrasound induced drug release pathways depend on the lipid composition of the liposome For the DOPC liposome, the pathway is through the diffusion due to the low lipid packing density induced by stretched bilayer In addition, the air compartment between two monolayers also enhances the diffusive process of drugs across the membrane We explain this based on the fact that a number of MD simulations have shown that the free energy profile for translocation across the lipid bi- layer of drugs usually exhibits a high barrier at the middle of the two monolayers[42–45] If two monolayers are separated apart, then the population of drugs between two monolayers is increased, i.e the free energy barrier is reduced, and therefore the drug permeability is increased For the DPPC and DOPC liposomes, the pathways are through pores and/or large damaged sections on the liposome surface To understand the molecular origin of the pathway formation, we first calculate the thickness of the bilayers, and the results of three liposomes are similar, ∼ nm This suggests that the bilayer thickness should not be the dominant parameter that determines the pathways as suggested by theory[33] Next, we calculate the potential energy of the inner and outer layers, and the innerouter interaction energy for each liposome in an equilibrium state; i.e without ultrasound, and the results averaged over structures of 100 ns trajectory are listed in Tab.I As seen, the inter-layer energy of the DOPC liposome is ∼ × 105 kJ/mol weaker than that of the DPPC and POPC, and this explains why two monolayers of the DOPC liposome tend to be separated apart, but those of the DPPC and POPC are remained close under ultrasound [Figs.1,3] In contrast, the potential energy of individual layers of the DOPC liposome is ∼ × 105 kJ/mol stronger than that of the DPPC and POPC counterparts, meaning that the interaction between lipids inside DOPC monolayers is stronger than that in the DPPC and POPC counterparts This explains that although two layers of the DOPC are separated apart but their spherical shape is largely maintained, while lipids tend to be disrupted and expelled from the surface of the DPPC and POPC liposomes, forming transient or permanent pores on the surface [Fig.1] It is known that the DOPC is in the liquid-disordered phase while the POPC and DPPC are in the liquidordered phase Small et al showed that as membrane phase changes toward liquid-ordered, the membrane becomes increasingly resistant to destruction such that the rate of diffusion decreases[26] However, in another study, the authors found, quite surprisingly, that there was no consistent correlation between the rate of release and the liquid phase types as expected, i.e the DPPC liposome did not display release rates that were consistently lower than that of the DOPC liposome[28] The authors also showed that the difference in membrane thickness between phases does not account for the observed difference in rates However, the theoretical modelling of Dan showed that the rate of release is insensitive to bilayer composition, but sensitive to the bilayer thickness[33] Our simulation shows that the type of the pathway depends on the lipid composition of the liposome The rate of release through the surface pores de- [1] D Kohane, Biotechnol Bioeng 96, 203 (2007) pend on the relative pore surface area, the pore size and distribution[35], while the diffusion through the stretched surface could depend on the disrupted lipid packing of the hydrophobic lipid tails, and also on the air compartments inside the bilayer Taken together, the optimisation of the drug release rate is quite challenge In this context, our simulation may provide complementary information to experiment and mathematical modelling, with the final aim is to help manufactures asses the specific role of each pathway to design efficient drug-delivery systems that meet user requirements The present study has some limitations of which we are fully aware First, to be close to experiments, the diameter of our liposomes, ∼ 80 nm, is similar to that of experimental liposomes (∼ 100 nm) To our best knowledge, this is the first MD simulation study of such big liposomes However, current computer technology only allows simulation in the nanosecond timescale which is much shorter than the second timescale in experiments Thus, we have to use high intensity and fast frequency ultrasound in order to obtain reasonable ultrasound effects on reasonable timescales However, since the formation of pathways is essentially determined by the molecular interaction between particles as shown above, we expect that the pathways seen in the simulation would also be the pathways obtained in experiment But, we note that the rate of release may vary between simulation and experiment, and proving this is beyond the scope of this paper Second, to verify whether the release pathways depend on the used coarse-grained MARTINI force field, we carry out a simulation using the all-atom CHARMM36 force field[46] for a DOPC liposome To be feasible, the all-atom liposome is smaller with a diameter of 20 nm We use an ultrasound of the same intensity and frequency as we used with the coarse-grained liposome Interestingly, the snapshots shown in Fig.1 indicate that the ultrasound induced pathway is through the stretched membrane, well-agreed with that obtained by the MARTINI force field Moreover, the result appears to be insensitive to the size of the liposome Author’s contributions V H Man and M S Li contributed equally to this work Acknowledgments This work has been supported by the Department of Science and Technology at Ho Chi Minh City, Vietnam (grant grant 10/2018/HDKHCNTT), CNRS, the National Science Foundation (NSF, grant SI2-SEE-1534941), the National Institute of Health (NIH-R01GM118508) and the CINES center for providing computer facilities (project A0010707721) DATA AVAILABILITY The data that support the findings of this study are available from the corresponding author upon reasonable request [2] F Wang, Y Shi, L Lu, L Liu, Y Cai, H.Zheng, X Liu, F Yan, C Zou, C.Sun, et al., PloS One 7, e52925 (2012) [3] J Chen, H Jiang, Y Wu, Y Li, and Y Gao, Drug Des Devel Ther 9, 2265 (2015) [4] S Parveen, R Misra, and S Sahoo, Nanomed Nanotechnol Biol Med 8, 147 (2012) [5] C Martinelli, C Pucci, and G Ciofani, APL Bioeng 3, 011502 (2019) [6] E Blanco, H Shen, and M Ferrari, Nat Biotechnol 33, 941 (2015) [7] Y Obata, S Tajima, and S Takeoka, J Controlled Release 142, 267 (2010) [8] I F Tannock and D Rotin, Cancer Res 49, 4373 (1989) [9] Y C Wang, F Wang, T M Sun, and J Wang, Bioconjugate Chem 22, 1939 (2011) [10] D Needham, G Anyarambhatla, G Kong, and M W Dewhirst, Cancer Res 60, 1197 (2000) [11] B M Dicheva and G A Koning, Expert Opin Drug Delivery 11, 83 (2014) [12] N Fomina, C McFearin, M Sermsakdi, O Edigin, and A Almutairi, J Am Chem Soc 132, 9540 (2010) [13] A Schroeder, R Honen, K Turjeman, A Gabizon, J Kost, and Y Barenholz, J Controlled Release 137, 63 (2009) [14] D M Benival and P V Devarajan, Int J Pharm 423, 554 (2012) [15] M Alibolandi, K Abnous, F Sadeghi, H Hosseinkhani, M Ramezani, and F Hadizadeh, Int J Pharm 500, 162 (2016) [16] J A Olson, J Schwartz, D Hahka, N Nguyen, T Bunch, G M Jensen, and J P Adler-Moore, Med Mycol 53, 107 (2015) [17] S R D’Mello, C N Cruz, M Chen, M Kapoor, S L Lee, and K M Tyner, Nat Nanotechnol 12, 523 (2017) [18] (????) [19] V Frenkel, Adv Drug Deliv Rev 60, 1193 (2008) [20] W G Pitt, G Husseini, and B Staples, Expert Opin Drug Deliv 1, 37 (2004) [21] S Dromi, V Frenkel, A Luk, B Traughber, M Angstadt, M Bur, J Poff, J Xie, S Libutti, K Li, et al., Clin Cancer Res 13, 2722 (2007) [22] S Rapoport, Expert Opin Investing Drugs 10, 1809 (2001) [23] Z Gao, H Fain, and N Rapoport, J Control Release 102, (2005) [24] A Schroeder, Y Avnir, S Weisman, Y Najajreh, A Gabizon, Y Talmon, J Kost, and Y Barenholz, Langmuir 23, 4019 (2007) [25] A Schroeder, J Kost, and Y Barenholz, Phys Chem Liq 162, (2009) [26] E F Small, M C Willy, P A Lewin, and S P Wrenn, Colloids Surf A 390, 40 (2011) [27] S W Wrenn, S M Dicker, E F Small, N R Dan, M Mleczko, G Schmitz, and P A Lewin, Theranostics 2, 1140 (2012) [28] E F Small, N R Dan, and S P Wrenn, Langmuir 28, 14364 (2012) [29] T J Evjen, S Hupfeld, S Barnert, S Fossheim, R Schubert, and M Brandl, J Pharm Biomed Anal 78-79, 118 (2013) [30] G Enden and A Schroeder, Ann of Biomed Eng 37, 2640 (2009) [31] L Simon and J Ospina, Physica A 451, 366 (2016) [32] J Siepmann and F Siepmann, Int J Pharm 364, 328 (2008) [33] S P Wrenn, E Small, and N Dan, Biochimica et Biophysica Acta 1828, 1192 (2013) [34] N Dan, Langmuir 30, 13809 (2014) [35] N Dan, Colloids and Surfaces B: Biointerfaces 126, 80 (2015) [36] Y Wang, P Gkeka, J E Fuch, K R Liedl, and Z Cournia, Biochimica et Biophysica Acta (BBA) - Biomembranes 1858, 2846 (2016) [37] S J Marrink, J Risselada, S Yefimov, D P Tieleman, and A H de Vries, J Phys Chem B 111, 7812 (2007) [38] S J Marrink and D P Tieleman, Chem Soc Rev 42, 6801 (2013) [39] E Lindahl, B Hess, and D van der Spoel, J Mol Mod 7, 306 (2001) [40] H J C Berendsen, J P M Postma, W F van Gunsteren, A Dinola, and J R Haak, J Chem Phys 81, 3684 (1984) [41] T Darden, D York, and L Pedersen, J Chem Phys 98, 10089 (1993) [42] W Shinoda, Biochimica et Biophysica ActaBiomembranes 1858, 2254 (2016) [43] M Orsi, W E Sanderson, and J W Essex, J Phys Chem B 113, 12019 (2009) [44] T X Xiang and B D Anderson, J; Chem Phys 110, 1807 (1999) [45] Y Wang, J Cohen, W F Boron, K Schulten, and E Tajkhorshid, J Struct Biol 157, 534 (2007) [46] A D M Jr., D Bashford, M Bellott, R L Dunbrack, J D Evanseck, M J Field, S Fischer, J Gao, H Guo, S Ha, et al., J Phys Chem B 102, 3586 (1998)