同一个吡嗪酰胺骨架，换一个靶点：人源 ProRS 抑制剂的虚拟筛选、热迁移与共晶
下划线 = 中文容易“变味”或丢信息的词句：红线带编号 = 15 条注释卡（悬停看、点击跳转）。这一页讲的事一句话能说完：把一个本来为抗结核准备的化合物库，按人源脯氨酰-tRNA 合成酶的结构重新筛一遍，筛出五个会结合的分子，再把它们贴着 ATP 口袋的样子拍下来。注意它做到哪一步就停了——这一点比结论本身更值得学。英文原文附于文末
作者：Luping Pang、Stephen D. Weeks、Martin Juhás、Sergei V. Strelkov、Jan Zitko、Arthur Van Aerschot
出处：International Journal of Molecular Sciences 2021;22(15):7793
DOI：10.3390/ijms22157793 PMID：34360555
单位：以鲁汶大学（KU Leuven）药物设计与发现实验室为主，同位素与化合物库部分来自合作组
译自英文原题：Towards Novel 3-Aminopyrazinamide-Based Prolyl-tRNA Synthetase Inhibitors: In Silico Modelling, Thermal Shift Assay and Structural Studies（Toward 新的 3-氨基吡嗪酰胺类脯氨酰-tRNA 合成酶抑制剂：计算建模、热迁移实验与结构研究）
摘要（全译）
人源胞质脯氨酰-tRNA 合成酶（HcProRS）催化脯氨酰-tRNAPro 的生成，在蛋白质合成中起重要作用。已有研究表明，抑制 HcProRS 的活性，对纤维化、自身免疫疾病与癌症的治疗可能有益。不久前，有研究者用高通量筛选（high-throughput screening，HTS）找到了一批强效的吡嗪酰胺类抑制剂，但没有报道后续的深入改造。吡嗪酰胺核心是一个生物活性骨架，在许多经过临床验证的药物里都能找到它，也被做过各式各样的修饰。因此，我们把一套虚拟筛选流程用在自建的小分子库上（库里成员都含吡嗪酰胺），以寻找可能的新型 HcProRS 抑制剂。我们识别出一系列 3-苄氨基吡嗪-2-甲酰胺衍生物作为阳性苗头化合物。其中五个由热迁移实验（thermal shift assay，TSA）确认；在 1 mM 脯氨酸（proline，Pro）与 3.45 µM 酶浓度的条件下，最好的化合物 3b 与 3c 的 EC50 分别为 3.77 与 7.34 µM。HcProRS 与这些化合物以及 Pro 的共晶结构证实了最初的对接结果，并且显示 Pro 如何促进这些与 ATP 竞争的配体结合。把 3b 建模进其他的人源 class II 氨酰-tRNA 合成酶（aaRS）后，提示这些酶 ATP 结合位点上的细微差异，很可能就是它对 HcProRS 具有潜在选择性结合的原因。总的来说，本研究从我们的抗结核内部化合物库中成功识别出新的 HcProRS 结合物，展示了把老候选药物重定位到新用途（例如治疗 HcProRS 相关疾病）的机会。
摘要里其实有四个动作，值得一个一个数出来：① 别人已经找到一个强效分子（2a），但没人接着往下做；② 我们手里正好有一个化学上很像的库；③ 先用计算筛，再用实验验，最后用晶体结构看；④ 顺手把“为什么只打这个酶”算了一遍。这是一篇“跟做 + 扩展”的文章，不是“发现新机制”的文章——读它的时候，把期待放在方法链上，不要放在“找到了新药”上。
引言：aaRS 为什么是好靶点
氨酰-tRNA 合成酶（aaRS）是一类存在于所有细胞中的必需酶。它们负责以依赖 ATP 的方式，把氨基酸连接到与之对应的 tRNA 上；这些被“充好电”的 tRNA 随后在核糖体上用于蛋白质合成。到目前为止，人类细胞中已知有 36 种 aaRS，由不同的基因编码。其中 16 种位于细胞质，17 种被转运并专一性地定位在线粒体，剩下三种（GlnRS、LysRS 与 GlyRS）呈双重定位。许多人类疾病与 aaRS 功能异常相关，例如某些癌症中 aaRS 的过表达与催化活性增强，以及遗传病中出现的具有临床意义的 aaRS 突变体。除了在蛋白质合成中的重要角色，aaRS 还被证明具有某些非经典功能，参与多种生理与病理过程，包括翻译后修饰、自身免疫疾病、血管生成、纤维化与神经病变。因此，aaRS 是开发针对多种人类疾病治疗药物的有吸引力的靶点。
特别值得一提的是，谷氨酰-脯氨酰-tRNA 合成酶（glutamyl-prolyl-tRNA synthetase，EPRS）是一个双功能酶，在细胞质中同时催化谷氨酰-tRNA 与脯氨酰-tRNA 的生成。人源 ProRS（HcProRS）的活性位于这个蛋白的 C 端区域，抑制该酶已被认为是对付 HcProRS 相关疾病的一条有前景的途径。卤夫酮（halofuginone）是一个经过验证的 ProRS 抑制剂，它以依赖 ATP 的方式，同时占据脯氨酸（Pro）结合位点与同源 tRNA 的 3′ 端结合位点。卤夫酮已显示出优异的抗疟、抗纤维化与抗癌活性。然而，由于它在物种之间的选择性很差，进一步的临床应用受到严重限制。最近，Adachi 等人基于生物发光酶活测定建立的高通量筛选（HTS）体系，识别出一批吡嗪酰胺类 HcProRS 抑制剂（化合物 2a–c）。其中最好的化合物 2a，在 Pro 存在时对 HcProRS 表现出强效结合亲和力，Kd 为 0.76 nM。生化数据加上晶体结构进一步证明，这个化合物是一个 ATP 竞争性抑制剂，且以依赖 Pro 的方式起作用。
吡嗪酰胺（pyrazinamide，PZA）是一个经过临床验证的一线抗结核药，用于活动性结核分枝杆菌感染的临床治疗，也被当作一个有生物活性的化学骨架，用于各种化学修饰。Zitko 的课题组曾报道过一大批吡嗪酰胺衍生物，属于开发新型抗菌药（尤其是抗分枝杆菌药）的努力。内部化合物库里有一系列 3-取代-N-苄基吡嗪-2-甲酰胺衍生物，它们与已确认的 HcProRS 抑制剂 2a 在结构上高度相似。这促使我们系统地考察这些化合物作为 HcProRS 抑制剂的潜力。于是我们用当时可得的、结合了化合物 2a 的 HcProRS 结构（PDB ID：5VAD）作为起始模型，对本组的小分子库做分子对接虚拟筛选。阳性苗头随后由热迁移实验（TSA）与共晶结构确认。这些结果让我们识别出五个先导化合物，它们对将来基于结构设计更强效的 HcProRS 抑制剂会有用。
这里插一句，把“靶点”这个词说清楚。做药物的人说某蛋白是“好靶点”，意思其实是三件事同时成立：它在一个你愿意干预的病理过程里（这里是纤维化、自身免疫、肿瘤）；它有一个能被小分子占据、而且占据之后酶就干不了活的口袋；最后一件事最容易被忽略——你打它的时候，不能顺手把别的几十个同族蛋白一起打了。人类有 36 个 aaRS，它们的催化逻辑几乎一样，口袋又长得很像，所以“选择性”在这个家族里不是加分项，是入场券。卤夫酮之所以卡在临床上，正是因为它的选择性。
而卤夫酮的位点恰好是家族里最保守的那一块：氨基酸结合位点。Pro 结合口袋在物种之间几乎不能动，动了酶就不认识底物了——这种地方做抑制剂，几乎注定会误伤。反过来，ATP 结合位点虽然也保守，但它保守的是“怎么把 ATP 捏住”这件事的几何关系，口袋边缘和通道是可以有差异的。这篇文章押的就是这一注：打 ATP 位点，用同族之间的细微差别换选择性。这个判断是全文的骨架，后面所有实验都在为它提供证据（或者至少在试探它）。
方法：虚拟筛选 + 热迁移 + 共晶
这篇的方法是一条三段式的流水线，每一段解决的问题不一样，证据强度也不一样。先说清楚这个区别，后面的结果你才不会被“证实了”三个字带走。
第一段，计算（in silico）。用已经解出的 HcProRS:Pro:2a 三元复合物（PDB ID：5VAD）当受体，把库里的 2,3-二取代吡嗪（共 97 个化合物）逐个对接到 ATP 口袋里。对接是有打分的，但打分是经验函数，不是自由能；所以作者还做了一件事——把每个对接构象与 2a 在晶体里的真实构象叠合，算RMSD。分数高、RMSD 又小的构象，才算“像是照着 2a 的样子坐进去的”。这一步的产物是一份待验证名单，不是结论。
第二段，结合（TSA）。原理很朴素：蛋白受热会变性（展开、聚集），变性过程中会暴露疏水核心，荧光染料结合上去，信号上升；信号上升最快的那一点就是熔解温度 Tm。如果有小分子结合上去、把蛋白“捏紧”了一点，Tm 就会往右移，这个差值叫 ∆Tm。所以 TSA 测的是“这个分子有没有让蛋白变结实”，它是一份结合证据，不是抑制证据。具体条件：0.2 mg/mL HcProRS（相当于 3.45 µM）、1× 热迁移染料、50 mM HEPES pH 7.0、150 mM KCl、10%（v/v）乙二醇，20 µL 每孔、三复孔，从 4 °C 升到 95 °C、0.05 °C/s，最后用 Boltzmann S 形曲线拟合求 Tm。每个化合物先在 100 µM 单浓度下预筛，加与不加 1 mM Pro 各做一遍——这个“±Pro”的设计是全文的关键，后面会解释为什么。
第三段，构象（共晶）。把蛋白与化合物（以及 Pro）一起养晶体，用 X 射线衍射解结构。这一步给的不是“结合了没有”，而是“结合在哪里、以什么姿势、和谁形成了什么相互作用”。它也是唯一能直接反驳对接预测的一步——计算说是这样坐的，晶体说不是，那就不是。
三段的分工可以这么记：计算负责把 97 个变成几个，TSA 负责把“几个”变成“确实结合的几个”，晶体负责说清“怎么结合”。任何一段单独拿出来都不够，三段的组合才构成这篇的论证。你把这条链记住，以后读任何一篇“虚拟筛选 + 实验验证”的文章都能套。
结果：3-苄氨基吡嗪-2-甲酰胺系列
4.1 对接：97 个化合物里，谁排到了前面
以 HcProRS:Pro:2a 三元复合物（5VAD）为起始模型，我们对内部库里 97 个 2,3-二取代吡嗪做了基于分子对接的虚拟筛选。结果指向的最高分衍生物，是 3-苄氨基-N-苄基吡嗪-2-甲酰胺（通式 3）——这批化合物最初是作为抗分枝杆菌化合物发表的。
其中一些衍生物的对接构象与已确认的抑制剂 2a 完全同一种结合模式，除了一个氢键：2a 的 C3-氨基位点与羰基氧之间有一个氢键，而我们的结构里没有。以 2a 的晶体构象为参照，对共有原子子集算 RMSD：苄环 2 位有取代的衍生物最小到 0.42 Å，苄环 4 位有取代的（R = 4-CF3、4-OCH3、4-Cl、4-CH3、3,4-二氯）最大到 0.86 Å。这些 4-取代衍生物的构象不仅打分更差，而且因为吡嗪 C2 位上的甲酰胺基团被扭歪了，连甲酰胺 NH 与 Thr1164 主链之间的那个氢键也丢了。
作者还注意到一件事：无论是已确认的抑制剂 2a（晶体结构），还是通式 3 的化合物，C2-甲酰胺上的取代基与酶之间只有非特异的疏水相互作用。换句话说，这个位置不是靠“精确匹配”拿分的。于是他们又去看库里另一批通式 4 的衍生物——C2-甲酰胺上没有取代基的版本。结果不出所料：打分普遍更差（少了重原子，本来就不该指望），而且结合模式更灵活，出现了好几种坐法。多数 4 位取代的 4 系化合物甚至伸不到该去的地方。
小结这一步：计算认为通式 3 与 4 都能以 2a 的方式结合 HcProRS，并且偏好苄环 2 位取代——打分与 RMSD 两个指标都指向这一边；反过来，4 位取代基本被否掉，只有位阻很小的 4-F 是例外。
4.2 热迁移：17 个化合物里，5 个让酶变结实了
从计算里挑出来的化合物（通式 3 与 4，包括计算预测为活性和非活性的），加上几个相关片段（吡嗪酸 POA、吡嗪酰胺 PZA、3-氨基吡嗪酸 3-NH2-POA），一起拿去测重组 HcProRS 的荧光热迁移。化合物结合通常会稳定蛋白，表现为热变性过程中熔解温度 Tm 升高。∆Tm 就是“有化合物”与“无化合物”两次 Tm 之差。
先看没有 Pro 的那一组：17 个受试化合物里，只有 4h 与 4j 两个轻微抬高了 Tm，分别 0.9 °C 与 2 °C；其余要么不变，要么还略微降低。这个结果本身已经很重要——它说明在缺 Pro 的时候，这些化合物基本贴不上去。
再看加了 1 mM Pro 的那一组：Pro 自己就让 HcProRS 的 Tm 升高了 4 °C（这是对照，所有 ∆Tm 都要按它归一）。归一之后，17 个里有 5 个显著提高了 HcProRS 的热稳定性，∆Tm 落在 2 到 7 °C 之间——提示这些化合物很可能是以依赖 Pro 的方式结合 HcProRS 的。第二轮做了剂量-反应，求出这 5 个苗头的 Tm 基准 EC50：活性范围从通式 3 骨架化合物的低微摩尔（3.77 µM）一直到 4 系衍生物的 91.11 µM。
作为参照，已报道的抑制剂 2a（100 µM）在无 Pro 时把 Tm 抬高 7 °C（EC50 = 30.35 ± 3.63 µM），在有 Pro 时抬高 13.15 °C（EC50 = 3.74 ± 0.67 µM）。这与它此前报道的更好结合亲和力与抑制活性是一致的。
把完整数据集放在一起看，还有两条信息：单环化合物 PZA、POA、3-NH2-POA 对 Tm 完全没有影响——也就是说，光有那个“骨架”不行，得是完整的分子；以及，在 3-苄氨基衍生物（通式 4）里，邻位（ortho）修饰比间位、对位更容易被接受。EC50 上，4h（R = 2-Cl）与 4j（R = 2-CF3）比 4d（R = 2-CH3）好 6 倍，提示吸电子性质可能对结合有利。再把 C2-甲酰胺也修饰上第二个邻甲基苄基（3c）后，EC50 相对 4d 又降了 12.4 倍——说明 C3-氨基与 C2-甲酰胺两个位置同时改是更划算的；同样地，双取代的 3b（R = 2-Cl）比单取代的 4h 低了 3.6 倍。
这一段是全篇最像“药物化学”的地方：同样一个环，改哪个位置、加什么基团，读数就动。但也别被“12.4 倍”“6 倍”这类数字带着跑——它们全是 TSA 里的 EC50，不是酶活 IC50，更不是细胞里的效应。同一篇文章里，2a 的结合 Kd 是 0.76 nM、酶活 IC50 是 12 nM、TSA 的 EC50 是 3.74 µM，三个数字差着好几个数量级，说的却是同一个分子。这三个数不是一个东西，见注 6。
4.3 共晶：Pro 先把 Arg1152 挪开，位子才空出来
为了说清这些苗头到底怎么结合，作者解了五个共晶结构：HcProRS 分别与对应的化合物以及 Pro 形成复合物，分辨率 2.2–2.7 Å。与已报道的无配体 HcProRS 结构（PDB ID：4K86）相比，这些结构的不对称单位里有两个蛋白分子，对应生理上具有活性的同源二聚体。多数情况下，电子密度图显示两个活性位点里各结合了一个化合物分子；但 4d 只在一个链里能被明确搭建出来，另一条链里密度很弱——这与它在 TSA 里 EC50 偏高（结合偏弱）的结果一致。
这五个三元复合物确认：这些 3-氨基吡嗪酰胺类化合物与已验证的抑制剂 2a 共享相似的结合模式，都占据 ATP 结合口袋，而 Pro 则待在氨基酸结合位点里。
两条链叠合后高度相似，全体重原子的 RMSD 为 0.373 ± 0.103 Å。因此后面的结构描述只聚焦一个单体——由于 4d 在 A 链里密度弱，作者选了 B 链。
把结合了 Pro 的结构与无配体的 HcProRS（4K86）对比，能看到一件很关键的事：Pro 的结合诱导 Arg1152 的侧链转开。Arg1152 是 motif-2 上一个在所有 class II aaRS 中都保守的残基；它从 ATP 结合位点里退出去，与 Pro 的羧基形成盐桥。位置让出来了，吡嗪酰胺类化合物才有地方坐进 ATP 口袋。这就是为什么这些化合物只在 Pro 存在时才结合的分子解释——TSA 里那个“±Pro”的设计，最后在这里被兑现了。
至于“五个化合物姿势是否一致”，叠合结果显示它们采取相似的构象。最好的化合物 3b 的细节，下一节单独讲。
4.4 3b 在口袋里到底摸到了谁
五个配体结合结构叠合起来看，五个化合物采取相似的构象。细看最好的化合物 3b，会发现它主要靠在两个 class II 的特征区上：motif-2（残基 1142–1176）与 motif-3（残基 1269–1287）与它形成了多处相互作用。
吡嗪环被 Phe1167 通过 π-π 堆叠、被 Arg1278 通过阳离子-π 相互作用“夹”在中间；吡嗪环的 N4 与 motif-3 里 Thr1276 的侧链羟基形成一个直接氢键；甲酰胺基团的氮原子 N9（氢键供体）与吡嗪酰胺核心的 N1（氢键受体），分别与 motif-2 里 Thr1164 的主链氧和主链氮形成两个氢键。此外，C3-氨基上的邻氯苄基伸进了 ATP 的核糖结合位点（与结合腺苷的 HcProRS 结构 4K87 相比），而 C2-甲酰胺上挂的邻氯苄基落在一个由 motif-2 环区残基围成的小腔里，靠范德华相互作用稳定。
注意最后这一句里的“尽管是非特异的”：C2-甲酰胺上的取代基并没有形成什么漂亮的氢键网络，它就是靠范德华力多贴了一点。但“非特异”不等于“不重要”——3b、3c 的 TSA EC50 比 4h、4d 低，说明这点额外的贴合确实换来了亲和力。
4.5 为什么偏偏喜欢邻位：一个被取代掉的结构水
对接与 TSA 都指向同一个偏好：C3-氨基上接邻位取代的苄环，比间位、对位好。结构上怎么解释？
C3-氨基苄环的落点是一段 β 链的主链，疏水性很强。更关键的是位置：它紧挨着保守 motif-3 α 螺旋的 N 端，而 α 螺旋本身有固有偶极，这使该区域的静电势整体偏正；在接纳邻位取代基的那个小腔里，正电势尤其明显。
再来一个对照。把抑制剂 2a（5VAD）与结合腺苷的结构（4K87）放在一起比，会发现那个小腔里原本坐着一个结构化水分子，它桥接着配体（2a 的甲酰胺氧、腺苷核糖的 2′-OH 与 3′-OH）与 Thr1276、Gly1238。而在我们这批配体结合的结构里，看不到这个水分子对应的电子密度——邻位取代基（2-Cl、2-CF3、2-CH3）似乎把它顶掉了。
把“正电势的小腔”和“被顶掉的水”两件事合起来，作者的推理是：邻位放一个带负电性的取代基，应该比中性或缺电子的更好。数据对得上：2-Cl 与 2-CF3 取代的 4h、4j，结合强度比 2-CH3 的 4d 好 7 倍；3b（2-Cl）的 EC50 比 3c（2-CH3）低 2 倍。
再看 3b 与前辈 2a 的关系。把 3b 复合物叠到 2a 复合物上——按蛋白骨架对齐（383 个残基上 Cα 的 RMSD 为 0.362 Å），或按吡嗪酰胺核心的原子对齐（RMSD 0.111 Å）——两个化合物的构象非常接近。也就是说，3b 几乎复制了 2a 所有重要的相互作用，除了那几处由水分子介导的氢键。
4.6 分子动力学：10 ns，看苄环会不会自己翻
为了解释“邻位取代为什么更稳”，作者跑了 10 ns 的分子动力学模拟。对象是三个 2-取代化合物 4d（2-CH3）、4h（2-Cl）、4j（2-CF3），外加一个无取代的 4a（R = H）作为对照。4a 的初始模型是从 4h 的复合物“炼”出来的——把 2-Cl 简单换成氢，属计算上的替换，不动别的。
判断配体姿势稳不稳，用了三个读数：配体各个重原子的均方根涨落（RMSF）；配体与受体之间那三个氢键的成键原子间距离随时间的变化；以及苄环平面相对初始状态的角度。
结果很干脆：2-取代的 4j（2-CF3）与无取代的 4a，两者的吡嗪酰胺部分都稳定；但 4j 的邻位取代基让苄环稳定在原来的构象里——碳原子 C12–C17 的 RMSF 很小，苄环平面的相对角度也几乎不动。反过来，无取代的 4a，苄环被预测要灵活得多，角度在 −135° 到 +135° 之间来回摆。另外两个 2-取代化合物 4d、4h 也观察到类似的苄环稳定化。
这项模拟支持了前面的假设：苄环 2 位取代把配体“钉”在 HcProRS 的 ATP 结合位点里；在 MD 稳定下来的构象（与晶体构象一致）中，邻位取代基正好落在 2a 结构里那个水分子所在的位点。计算与 TSA 合起来指向同一件事——邻位取代有利。还有一条旁证：所有成功与 HcProRS 共结晶的化合物，都带这样的邻位取代基。
4.7 选择性：把 3b 塞进另外九个 class II 酶
20 种标准 aaRS 按催化核心的两种完全不同的折叠分成两类，这两种折叠原本是为了特异识别与结合共同的底物 ATP 而演化出来的。class I 的催化结构域采取 Rossmann 折叠，class II 则组织成六股 β 折叠。尽管 ATP 在每个 aaRS 类别内部的结合模式是严格保守的，但作为 ATP 竞争性抑制剂，cladosporin 被报道专一性地打 LysRS 而不是其他 class II aaRS；化合物 2a 也被报道对 HcProRS 的选择性比对另一个 class II 成员 ThrRS 高 2500 倍。既然我们的化合物与 2a 共享相似的结合机制，就值得用已有的人源 class II tRNA 合成酶结构来考察一下潜在的选择性。
以 3b 为例，作者把它建模进其余九个人源 class II aaRS 结构的 ATP 结合位点：先把已有含腺嘌呤配体的结构与 HcProRS:Aze-SA 复合物（Aze-SA 即氮杂环丁烷-2-羧酸偶联 5′-氨磺酰腺苷，是一个脯氨酰-腺苷酸类似物，PDB ID：5V58）按腺嘌呤基团叠合，再把 HcProRS:Pro:3b 与 HcProRS:Aze-SA 叠合。为了把“家族选择性”讲清楚，配体被切成三块：C3 取代基、吡嗪酰胺环、C2-甲酰胺部分。
在 HcProRS 里，接纳 C3-氨基上 2-取代苄基的小亚口袋由 Trp1169、Gly1274 与 Gly1238 构成。去看其他 class II aaRS，这三个位置的残基变化很大。在与 ProRS 同属 IIa 亚类的 SerRS、ThrRS、HisRS、GlyRS 中，对应 Trp1169 的位置换成了极性残基（分别是 Lys323、Gln493、Gln173、Met294），于是与配体 C3-氨基苄环之间的疏水相互作用就没了；motif-3 β 链上的 Gly1274 被更大的氨基酸替换（Thr429、Ala625、Ser383、Ser524），很可能与苄环发生位阻冲突；Gly1238 也被带侧链的残基替换（Leu392、Cys591、Val357、Ile404）——这些侧链虽然不直接接触化合物，但它们会把主链“固定住”，限制羰基朝向 motif-3 β 链并与之平行，结果可能与化合物的邻位取代基撞上。这三处替换，让“结合”在其他 class II 酶里变得不利。
吡嗪酰胺那一块则坐在 motif-3 α 螺旋 N 端的顶上。结构比较显示，相当于 Arg1278 与 Phe1167 的堆叠残基高度保守，相当于 Thr1164（提供主链相互作用）的残基在其他 class II aaRS 里也总是存在——这意味着吡嗪酰胺核心可能有一个“通用”的结合方式。但有一处明显不同：motif-3 α 螺旋 N 端的长度。HcProRS 在这里是三个残基（Thr1276、Thr1277、Arg1278），而其他所有 class II tRNA 合成酶都是四个。这种不同的结构排布可能影响上游的 β 链，而这个区域的独特残基也可能为吡嗪环提供一个带位阻的专属结合位点。
C2-甲酰胺上的 2-取代苄基伸向 motif-2 环区，只形成非特异的范德华相互作用；而这个环区的序列长度与活动性都非常多样，很难预测这些化合物结合时它会怎么排布。不过有一个例子：在 SerRS 中，motif-2 环的对侧坐着 Arg192 的大侧链（在 HcProRS 里对应的是一个甘氨酸），会对 3b 的结合造成类似的位阻排斥——提示我们确认的这些结合物不太可能结合 SerRS。
总之，作者在这里给出的判断是：ATP 结合位点上的细微差异，看来对设计有选择性的 ATP 竞争性 aaRS 抑制剂是重要的。注意这句话的措辞——是“看来”，是建模得到的提示，不是测出来的选择性。这一节没有做任何一个其他酶的结合实验。
讨论：从抗结核化合物库到人源 ProRS
aaRS 催化氨酰-tRNA 的生成，一直是被当作抗菌药靶点来开发的。但 HcProRS 不一样：它位于双功能酶 EPRS 的 C 端，被认定为抗癌、抗纤维化与抗自身免疫反应药物开发的一个有前景的靶点。到现在为止，已报道的 HcProRS 抑制剂有几个：卤夫酮专一性地同时打氨基酸位点与 tRNA 3′ 端结合位点；化合物 2a 与 T-3833261 则占据 HcProRS 的 ATP 结合位点。
这里有一个容易被忽略的临床逻辑：卤夫酮对 HcProRS 的抑制活性可以被升高浓度的 Pro 逆转，而 Pro 在纤维化组织里是积累的。这直接削弱了卤夫酮的抗纤维化活性——你打的位点，正好被病理状态下的底物浓度保护着。所以后两个化合物（ATP 竞争性的）有可能绕开这个问题。但也因此有了新的麻烦：class II aaRS 共享非常相似的 ATP 结合机制，要想选择性抑制 HcProRS 而不影响其他 class II 成员，是药物开发的必要条件。
回到本文的工作。化合物 2a 用 ATP/PPi 交换实验测得的 HcProRS 抑制活性 IC50 是 12 nM，并对 class II 的 ThrRS 显示出潜在选择性，但没有后续研究。我们用结合了 2a 的结构作起始模型，对自建的合成小分子库做虚拟筛选，识别出两种通式的取代 3-氨基吡嗪酰胺（3、4），它们被预测以与 2a 相同的结合模式、以高亲和力结合 HcProRS。挑出的化合物做了热迁移实验：五个化合物在 Pro 存在时明显提高了 HcProRS 的热稳定性，说明它们与 2a 一样，是依赖 Pro 的结合物。这一点随后被共晶结构证明：Pro 的结合诱导 Arg1152 侧链退出 ATP 结合位点，腾出空间容纳吡嗪酰胺衍生物。吡嗪酰胺核心模拟 ATP 的腺嘌呤，C3 取代基则通过与活性位点的疏水相互作用替代 ATP 的核糖。C2-甲酰胺上的修饰伸向 motif-2 环区，改善了结合亲和力——3b、3c（R = 2-Cl；2-CH3）的 EC50 比 C2 未取代的 4h、4d 更低，就是证据。
对 HcProRS 的高类别选择性（而不是同时打其他 class II 成员）是 aaRS 抑制剂用于治疗的根本。尽管 ATP 在 class II aaRS 中的结合姿势非常相似，不同 class II 酶活性位点之间的细微差异，可能正是可设计的 selectivity 来源。把 3b 建模进其他所有 class II aaRS 结构后可以看到：围绕 C3 位邻取代苄基、与配体形成疏水相互作用的三个残基，在其他 class II 酶里被替换成了极性或更大的残基，因此这些酶无法与 C3 取代部分形成稳定相互作用。此外，HcProRS 中 motif-3 α 螺旋 N 端的特定排布（比其他 class II 成员短一个残基）决定了吡嗪环的精确位置，这也可能贡献选择性。
| 数字 | 测的是什么 | 条件 | 出处 |
|---|---|---|---|
| Kd 0.76 nM（2a） | 结合亲和力（热力学） | 有 Pro | Adachi 等，本文引用 |
| IC50 12 nM（2a） | 酶活抑制（ATP/PPi 交换） | 体外酶活体系 | 本文讨论，引用前人 |
| TSA EC50 3.74 ± 0.67 µM（2a） | 热稳定性迁移的半最大浓度 | 100 µM 参照、加 Pro | 本文 TSA |
| TSA EC50 3.77 µM（3b）/ 7.34 µM（3c） | 同上 | 1 mM Pro、3.45 µM 酶 | 本文摘要与 TSA |
| EC50 范围 3.77–91.11 µM（五个苗头） | 同上 | 加 Pro | 本文 TSA |
这张表想让你带走一个习惯：看到“活性”两个字，先问是哪种活性。同一个 2a，结合是纳摩尔级，抑制酶活是纳摩尔级，而它在 TSA 里的读出是微摩尔级——不是谁错了，是三个实验问的问题不同。你以后在组会或综述里对比化合物，务必把单位与实验类型一起抄下来，别只抄数字。
最后是本文的结论，以及它没有做的事。结论：把虚拟筛选与生物物理、晶体学研究结合起来，成功识别出五个新的吡嗪酰胺类 HcProRS 配体，它们以依赖 Pro 的方式特异性地与 ATP 底物竞争结合。这项研究不仅指向老化合物重定位到新用途的可能，也提供了通过基于结构的设计进一步提高结合亲和力与潜在选择性的空间，尤其是 C3 取代基那一侧。这之后应该接着做抑制活性测定与全细胞实验。
最后这句话是全篇最诚实的一句，也是你读这篇文章时最该划出来的地方：到这里为止，作者证明的是“结合”，不是“抑制”。TSA 的 EC50 是 1–90 µM 量级，没有酶活 IC50，没有细胞表型，没有毒性数据。“苗头化合物”这个词的准确含义就是“值得往下做的起点”，它离“候选药物”还隔着好几道门。写综述的时候如果把这类结果写成“发现了 HcProRS 抑制剂”，就过头了。
材料与方法（择要）
蛋白怎么来的。编码人源 ProRS（HcProRS）第 1001–1512 位残基的 DNA 序列（UniProt 登录号 P07814），用 PCR 从 HEK293T 细胞构建的 cDNA 库里扩增；扩增产物克隆进自建的 pETRUK 载体——它是 pETHSUL 的衍生载体，能在 N 端带上 SUMO 标签，在 E. coli Rosetta 2 (DE3) pLysS 里表达。用自动诱导培养基培养后收菌、超声裂解（缓冲液 A：25 mM Hepes pH 8、200 mM NaCl、5 mM β-巯基乙醇，补 1 mM MgCl2 与 100 U Cryonase 核酸酶），18,000×g 离心 30 min，上清过 5 mL Hitrap SP 柱；收集 SUMO 融合的 HcProRS 组分，用 SUMO 水解酶切掉标签，混合液在含 10 mM Tris pH 7、10%（w/v）甘油的缓冲液里透析过夜去盐，再上一次 SP 柱把标签与水解酶去掉；流穿液里的 HcProRS 继续用阴离子交换与凝胶过滤纯化。纯化后的 HcProRS 浓缩到 45 mg/mL（终缓冲液：10 mM Tris pH 7、100 mM NaCl、2.5 mM β-巯基乙醇），−80 °C 保存。
晶体怎么长的。取 30 mg/mL 蛋白（同样缓冲液），加 10 mM L-脯氨酸与 12%（v/v）DMSO，在有或无 2 mM 化合物的条件下冰上孵育 1 h，然后用 Microbatch 法点晶体：1 µL 预混液与 1 µL 池液（0.25–0.4 M SrCl2、15–20%（w/v）PEG3350、100 mM HEPES pH 7.5）混合，液滴上覆石蜡油，在 Terasaki 微批板里生长。合适的晶体直接从板上挑出、液氮速冻。
结构怎么解的。所有衍射数据在同步辐射装置上从冷冻晶体收集，用 autoPROC 处理；初始模型用已报道的 HcProRS:Pro:2a 结构（5VAD）做分子置换（Phenix 套件里的 PHASER）；随后在 Phenix.refine 里精修、在 COOT 里手工搭建，结构图用 PyMol 2.0.4 画。六个结构的 PDB 登录号：7OSY、7OSZ、7OT0、7OT1、7OT2、7OT3。
TSA 怎么做的。0.2 mg/mL HcProRS（3.45 µM）与不同浓度的化合物在有或无 1 mM L-Pro 的条件下孵育，加 1× 热迁移染料，缓冲液为 50 mM HEPES pH 7.0、150 mM KCl、10%（v/v）乙二醇；20 µL 分装到 96 孔 PCR 板、三复孔，离心去气泡后在 Applied Biosystems Protein Thermal Shift 仪器上测，4 °C 到 95 °C 梯度升温、0.05 °C/s，荧光由 Protein Thermal Shift 软件读取；用 Boltzmann S 形拟合算 Tm。
计算怎么做的。对接与 MD 的输入都在 MOE 2020.09（Chemical Computing Group）里准备，对接在 MOE 里跑；MD 用 NAMD2（GIT20190909），分析用 VMD 1.9.4a51。
给只做湿实验的人一句提醒：这一节看着像“流水账”，其实每一句都在回答“这个结论的边界在哪里”。比如 SUMO 标签是为了让融合蛋白别乱折叠、又能被切干净；SrCl2 是这批晶体特有的沉淀剂；Microbatch 加石蜡油是为了在极小体积里慢慢长。你在实验室复现别人的实验，卡住的地方常常不是“原理”，而是这类没人写进正文的细节。
这篇在轨迹里的位置
先把时间摆正：这篇是 2021 年，庞露苹在鲁汶大学读博与做博后期间的工作（她 2020 年 9 月博士毕业，2020.09–2021.08 在鲁汶做博后，2021 年 10 月入职郑州大学医学科学院）。所以它不是“回国后的第一篇文章”，而是鲁汶那几年两条线里“做抑制剂”那条线的收尾。
为什么说是收尾？因为这条线在鲁汶其实是两条腿走路：一条腿是“把酶怎么工作看清楚”，另一条腿是“看清楚了就拿它当靶点设计分子”。这一页属于第二条腿。它的起点是别人（Adachi 等）找到的 2a——一个纳摩尔级的 ATP 竞争性抑制剂，但没人接着做；它的做法是拿自己实验室里现成的、化学上高度相似的抗结核化合物库去对接、去验、去拍结构；它的产出是五个微摩尔级的结合物，加一套能解释“为什么邻位取代好”的结构依据。
有意思的是，同一个酶家族的另一条腿，发表在本页的下一篇（02 页，2022 年 Communications Biology）：用淋病奈瑟菌的 LeuRS，把氨酰化第一步的每个酶学状态都拍下来，发现一个肽平面翻转充当了两步催化之间的开关。按发表年份，01 在 02 之前；按逻辑，02 在 01 之前——先知道酶怎么动，才知道该往哪儿设计分子。这个顺序被打乱，不是作者搞错了，而是结构生物学的常态：一个实验室同时推几条线，谁先投出去、谁先被接收，跟逻辑顺序没关系。你以后看任何一位 PI 的论文列表，都要有这个心理准备。
再往后看这条轨迹怎么走。2022 年还有一篇 IJMS，做的是 MshC 抑制的结构基础——靶点从人源酶换成了细菌酶；2023 年 STTT 是一篇 CLK 激酶的综述，属于“药物化学那一侧”的整理；2026 年 JBC 把 MshC 放回真正的病原体（脓肿分枝杆菌）里当靶点；2026 年 ACS Omega 又回到吡嗪酰胺衍生物，这次是抗结核。注意最后这一篇——这一页从抗结核库里借了分子去打人源靶点，六年后她自己回到了抗结核本身。化学骨架在两边流动，问题从“人的病”换成了“细菌的病”。
一句话概括这条线：先把酶怎么工作看清楚（机制），再把它当靶点设计分子（结构 + 药物化学），最后推到真正的病原体和候选化合物上。这一页站在第二格的地基上，而且是“借来的地基”——用的是别人的先导化合物、自己的化合物库、和自己刚学会的结构解析手艺。
给刚进组的你
你现在的日常是跑 WB，可能觉得“结构生物学”离你很远。其实不远，只是顺序反了：你做 WB 用的抗体，认的是一段肽或一个构象；你跑胶用的蛋白，是从细胞里纯化出来的；你看到的条带弱，可能是蛋白降解、可能是聚集、也可能是表达量本来就低。这一页讲的每一件事，追问下去都会落到“这个蛋白现在是什么状态”上——而这正是结构生物学家每天在解决的问题。
如果只从这一页拿走三样东西，我建议是这三样。
第一，一个概念：选择性来自细微差异。class II aaRS 的 ATP 口袋，长得几乎一样——ATP 是它们共同的底物，能不一样吗？但就是三个残基的替换（Trp1169 → 极性残基、Gly1274 → 大残基、Gly1238 → 带侧链残基），加上螺旋 N 端少一个残基，让同一个分子在别的酶里坐不舒服。邻位、间位、对位这三个词，在这里不是化学课上的名词，而是“这个分子能不能只打一个靶点”的分水岭。
第二，一个手法：TSA 是性价比最高的结合筛选。它不需要标记、不需要固定、不需要特殊的仪器（很多组里的 qPCR 仪就能做），一个下午能筛几十个化合物。但它只回答“结合没有”，不回答“有没有抑制”。所以标准做法永远是：TSA 先筛，酶活实验再定量，细胞实验最后说话。你哪天要帮师兄筛化合物，先问清楚：这个读数是 Tm 还是 IC50？
第三，一个可以问的问题：“这个化合物是抑制，还是只是结合？”这一页的结论停在结合。这不是缺陷，是分寸。你在组会或文献汇报里，如果能把“结合”和“抑制”分开说，导师会立刻知道你读懂了——因为大部分本科生的汇报会把两者混成一句“这个化合物有效”。
还有一句实际的话。她 2021 年 10 月入职郑州大学医学科学院，本行是结构生物学与药物化学，做的是肿瘤与代谢性疾病相关蛋白的结构、机制，以及基于结构的小分子设计；教学上她也在基础医学院《医学细胞遗传基础》整合课程的团队里，属于后备力量。这意味着什么？意味着这个专区里的六篇，不是六篇“跟脑科学无关的闲文”，而是一条完整的方法链示范：从蛋白怎么工作，到分子怎么设计，到病原体怎么验证。你要练的迁移能力，就是在别人的领域里认出这条链，然后把它搬到你自己的课题上。
▲ 本页翻译难点（对照学习用）
附录 · 英文原文
Towards Novel 3-Aminopyrazinamide-Based Prolyl-tRNA Synthetase Inhibitors: In Silico Modelling, Thermal Shift Assay and Structural Studies
Abstract
Human cytosolic prolyl-tRNA synthetase (HcProRS) catalyses the formation of the prolyl-tRNAPro, playing an important role in protein synthesis. Inhibition of HcProRS activity has been shown to have potential benefits in the treatment of fibrosis, autoimmune diseases and cancer. Recently, potent pyrazinamide-based inhibitors were identified by a high-throughput screening (HTS) method, but no further elaboration was reported. The pyrazinamide core is a bioactive fragment found in numerous clinically validated drugs and has been subjected to various modifications. Therefore, we applied a virtual screening protocol to our in-house library of pyrazinamide-containing small molecules, searching for potential novel HcProRS inhibitors. We identified a series of 3-benzylaminopyrazine-2-carboxamide derivatives as positive hits. Five of them were confirmed by a thermal shift assay (TSA) with the best compounds 3b and 3c showing EC50 values of 3.77 and 7.34 µM, respectively, in the presence of 1 mM of proline (Pro) and 3.45 µM enzyme concentration. Co-crystal structures of HcProRS in complex with these compounds and Pro confirmed the initial docking studies and show how the Pro facilitates binding of the ligands that compete with ATP substrate. Modelling 3b into other human class II aminoacyl-tRNA synthetases (aaRSs) indicated that the subtle differences in the ATP binding site of these enzymes likely contribute to its potential selective binding of HcProRS. Taken together, this study successfully identified novel HcProRS binders from our anti-tuberculosis in-house compound library, displaying opportunities for repurposing old drug candidates for new applications such as therapeutics in HcProRS-related diseases.
1. Introduction
Aminoacyl-tRNA synthetases (aaRSs) are a class of essential enzymes found in all cells. They are responsible for catalysing the ligation of an amino acid to its cognate tRNA in an ATP-dependent manner [[1]]. The charged tRNAs are subsequently used for protein synthesis at the ribosome. To date, there are 36 known aaRSs encoded by distinct genes present in human cells. Among them, 16 are found in the cytoplasm, 17 are transported and exclusively localized in the mitochondria and the remaining three (GlnRS, LysRS and GlyRS) show dual localization [[2]]. Many human diseases are associated with aaRS dysfunction, such as the overexpression and the enhancement of aaRS catalytic activity in some cancers [[3]] and the emergence of clinically relevant aaRS mutants in genetic diseases [[4],[5],[6]]. In addition to their important roles in protein synthesis, aaRSs have also been shown to have certain non-canonical functions [[7]] being involved in various physiological and pathological processes, including post-translational modifications, autoimmune diseases, angiogenesis, fibrosis and neuropathy [[4]]. Therefore, aaRSs are attractive targets in the development of therapeutic agents against multiple human diseases [[8],[9]].
Notably, glutamyl-prolyl-tRNA synthetase (EPRS) is a dual function enzyme that catalyses the formation of charged glutamyl-tRNA and prolyl-tRNA in the cytoplasm. The human ProRS (HcProRS) activity is located at the C-terminal region of this protein and inhibition of this enzyme has been recognized as a promising approach for treatment of HcProRS-related diseases. Halofuginone (), a validated ProRS inhibitor, occupies both the proline (Pro) binding site and the 3′-end of cognate tRNA binding site in an ATP-dependent manner [[10]]. Halofuginone has shown to have excellent anti-malaria, anti-fibrosis and anti-cancer activities [[10],[11],[12],[13],[14]]. However, due to its poor selectivity between species, further clinical application has been severely hampered [[15]]. Recently, Adachi et al. identified a number of pyrazinamide-based HcProRS inhibitors (, compounds 2a–c) via a high-throughput screening (HTS) system based on a bioluminescent enzymatic assay. Among them, the best compound 2a demonstrated potent binding affinity with a Kd of 0.76 nM for HcProRS in the presence of Pro [[16]]. Biochemical data complemented with a crystal structure further proved that this compound is an ATP-competitive inhibitor that works in a Pro-dependent fashion.
Pyrazinamide (PZA) is a clinically validated first-line antitubercular drug for the clinical treatment of active Mycobacterium tuberculosis infections and has also been considered as a bioactive chemical scaffold used for various chemical modifications [[17],[18],[19],[20],[21],[22]]. Zitko’s group reported a large series of pyrazinamide derivatives as part of efforts for the development of new antimicrobials, especially as antimycobacterial drugs [[17],[18]]. The in-house library contains a series of 3-substituted-N-benzylpyrazine-2-carboxamide derivatives, which share high structural similarity with confirmed HcProRS inhibitor 2a. This motivated us to systematically investigate the potential of these compounds as HcProRS inhibitors. We therefore performed molecular docking to virtually screen our small molecule library using the available compound 2a-bound HcProRS structure (PDB ID: 5VAD) as a starting model. Positive hits were further confirmed by thermal shift assay (TSA) and co-crystal structures. These results enabled us to identify five lead compounds which would be useful for future structure-based drug design of more potent HcProRS inhibitors.
2. Results
2.1. Docking Compound Library with HcProRS Ligand-Bound Structure
Using the ternary complex of HcProRS:Pro:2a (PDB ID: 5VAD) as a starting model, we performed in silico screening based on molecular docking of 2,3-disubstituted pyrazines from our in-house library (97 compounds in total; for the experimental procedure, see). The results indicated that the highest scoring derivatives were 3-benzylamino-N-benzylpyrazine-2-carboxamides (, general formula 3), originally published as antimycobacterial compounds [[18]]. Some of our derivatives showed the same binding mode compared to the confirmed inhibitor 2a, exerting all H-bond interactions except for the H-bond to the carbonyl oxygen at C3-amine position, which is missing in our structures (and). The RMSD calculated for the shared subset of atoms (bolded in), taking the crystallographic pose of 2a as a reference, ranged from 0.42 Å (minimum value) for derivatives with a substitution in position 2 of the benzyl ring up to 0.86 Å (maximum value) for derivatives (3) with substituents in position 4 of the benzyl ring (R = 4-CF3, 4-OCH3, 4-Cl, 4-CH3, 3,4-diCl). The poses of these 4-substituted derivatives demonstrated worse docking scores and, due to the distortion of the carboxamide moiety attached at the C2 position of pyrazine, also lacked the H-bond between the NH of the carboxamide group and the backbone of Thr1164. Both in the confirmed inhibitor 2a (crystallographic structure) and in our compounds of general structure 3, the substituent on the C2-carboxamide group exerts only non-specific hydrophobic interactions to the enzyme. For this reason, we also focused on our in-house derivatives of general structure 4 with an unsubstituted C2-carboxamide group [[17]]. For these, we generally observed worse docking scores (as expected because of the lower number of heavy atoms due to the lack of the substituent on the C2-carboxamide) and more flexibility in terms of different binding modes. In most cases, compounds 4 with substitution in position 4 of the benzyl ring were not able to reach the binding mode of the confirmed inhibitor.
To sum up, the initial in silico screening suggested that both compounds of general structure 3 and 4 are capable of binding to HcProRS in the same mode as the confirmed inhibitor 2a. The docking also showed preference for derivatives with substitution in position 2 of the benzyl ring as indicated by both docking scores and RMSD values. On the contrary, 4-substituted derivatives were discouraged by molecular docking, except for sterically small R = 4-F. The docking scores and RMSD values of selected derivatives are present in, while full results are present in the.
2.2. Confirmation of the Hits from Virtual Screening by a Thermal Shift Assay
Selected compounds from in silico screening (general structures 3 and 4, both active and inactive as predicted by docking) along with relevant fragments (pyrazinoic acid, POA; pyrazinamide, PZA; and 3-aminopyrazinoic acid, 3-NH2-POA) were evaluated against recombinant HcProRS using a fluorescence-based thermal shift assay. Compound binding usually helps to stabilize the protein due to protein-ligand interactions, resulting in an increase of the melting temperature (Tm) of the target protein during the thermal denaturation process [[23]]. Each compound was pre-tested at 100 µM concentration against 3.45 µM HcProRS in the absence and presence of 1 mM Pro. The ∆Tm was calculated based on the difference between the Tm values of HcProRS with and without compound. In the absence of Pro, only two of the 17 tested compounds, 4h and 4j, slightly increased the Tm by 0.9 and 2 °C, respectively, while the remaining compounds did not change or slightly decreased the Tm (and). In the presence of the natural substrate Pro, the Tm of HcProRS increased by 4 °C. Normalized by this control, the presence of five of the 17 compounds, including 4h (R = 2-Cl) and 4j (R = 2-CF3), significantly improved the thermal stability of HcProRS resulting with ∆Tm values in a range of 2 to 7 °C suggesting these compounds are likely binding to HcProRS in a Pro-dependent manner (). In a second-tier characterization, the Tm-based EC50 values of these five hits were determined. These compounds demonstrated an activity range varying between lower micromolar (3.77 µM) for the compounds based on scaffold 3 up to 91.11 µM for derivatives of the series 4 compounds (and). Using the previously reported inhibitor 2a (100 µM) as the reference, we observed a significant increase of the Tm of HcProRS by 7 °C (EC50 = 30.35 ± 3.63 µM) and 13.15 °C (EC50 = 3.74 ± 0.67 µM) in the absence or presence of Pro, respectively (and). This is in good agreement with the prior reported improved binding affinity and inhibitory activity of 2a in the presence of Pro when compared to the corresponding values obtained without Pro [[16]]. Therefore, taking the limitation of TSA measurements into account, these results suggested that our best compound 3b and the confirmed inhibitor 2a behave similarly in binding the enzyme.
Comparison of the full datasets show that the single ring compounds PZA, POA and 3-NH2-POA have no effect on Tm. In 3-benzylamino derivatives of general structure 4, modification at the ortho-position appeared to be better accepted than at meta- and para-positions. The EC50 values of 4h (R = 2-Cl) and 4j (R = 2-CF3) proved 6-fold better than R = 2-CH3 in compound 4d, indicating that the electron withdrawing properties might be beneficial for binding. Further modification of the C2-carboxamido moiety with a second ortho-methylbenzyl group (3c) decreased the EC50 12.4-fold with respect to 4d, suggesting the modifications of both C3-amino and C2-carboxamido groups of the 3-aminopyrazinamide scaffold are preferable for binding. Similar phenomenon was also observed for the 3.6-fold lower EC50 of double substituted compound 3b (R = 2-Cl) when compared with the single substituted 4h.
2.3. Structural Studies of the Binding Mechanism of the Compounds with HcProRS
To further clarify the binding mechanism of the identified hits, we determined five co-crystal structures of HcProRS in complex with the corresponding compound and Pro at a resolution range of 2.2–2.7 Å (and). Compared with the reported ligand-free HcProRS structure (PDB ID: 4K86), our structures have two macromolecules, corresponding to the biologically active homodimer, in the asymmetric unit (a). In most cases, the calculated electron density map showed a molecule of the respective compound bound in both active sites of the dimeric HcProRS. However, compound 4d can only be unambiguously built in one chain while showing weak density in the other (b and). This is consistent with the observed higher EC50 for this compound in TSA measurements compared with the other four identified hits (). These ternary complexes confirmed that these 3-aminopyrazinamide-based compounds share a similar binding pattern with the validated inhibitor 2a, occupying the ATP binding pockets with Pro found in the amino acid binding site (b).
Superimposition of two chains in the HcProRS complex shows a high degree of similarity as evidenced by the overall mean all-atom RMSD value of 0.373 ± 0.103 Å. Therefore, for the later structural description, we only focus on one monomer, in this case chain B due to the weak electron density of 4d in chain A (). Comparison of Pro-bound and apo-form HcProRS (PDB ID: 4K86) structures shows that the binding of Pro induces the movement of the side chain of Arg1152, a motif-2 invariant residue found in all class II aaRSs [[25]], away from the ATP binding site to form a salt bridge between the carboxylic group of Pro and guanidino moiety of Arg1152 (a). This subsequently provides space to accommodate the binding of pyrazinamide-based compounds in the ATP binding pocket which gives a good rationale why these compounds only show binding in the presence of Pro as seen in the TSA measurements ().
Superposition of the ligand-bound structures showed that the five compounds adopt a similar conformation (b). Detailed analysis of the best compound 3b highlights that the class II signature motif-2 (residues 1142–1176) and motif-3 (residues 1269–1287) of HcProRS form multiple interactions with this compound. The pyrazine ring is sandwiched by Phe1167 via π-π stacking interaction and Arg1278 via cation-π interaction. The N4 of pyrazine ring makes one direct H-bond with the side chain hydroxyl group of Thr1276 in motif-3. The nitrogen atom N9 of the carboxamide moiety (H-bond donor) and N1 of the pyrazinamide core (H-bond acceptor) establish two H-bonds with the backbone oxygen and nitrogen of Thr1164 in motif-2, respectively (c). Moreover, the ortho-chlorobenzyl moiety at C3-amine extends into the ribose binding site of ATP compared with the adenosine-bound HcProRS structure (d, PDB ID: 4K87). The ortho-chlorobenzyl moiety attached to the C2-carboxamide is located in a cavity surrounded by residues of the motif-2 loop region and is stabilized via van der Waals interactions (d). Despite being non-specific, the interactions of the substituent at C2-carboxamide clearly enhance the binding affinity which can be evidenced by the decreased TSA EC50 values of compounds 3b and 3c compared to 4h and 4d.
As indicated by docking and TSA studies, there is a clear preference for binding of compounds having an ortho-substituted benzyl ring at C3-amine instead of meta- or para-substitution. The binding region of the C3-amine benzyl ring is a β-strand backbone with hydrophobic characteristics. In addition, due to its position at the N-terminus of conserved motif-3 α-helix, the inherent dipole of the α-helix makes the electrostatic potential of this zone of HcProRS overall positive. The positive electrostatic potential is especially profound in a small cavity accepting the ortho-substituent (e). Comparisons of inhibitor 2a (PDB ID: 5VAD) and adenosine-bound (PDB ID: 4K87) structures of HcProRS revealed a structured water molecule is placed in this cavity, bridging the interactions between the ligands (carboxamidic oxygen of 2a and 2′-OH and 3′-OH of ribose of adenosine) and Thr1276 and Gly1238 (e,f). However, in our ligand-bound structures, no electron density of equivalent water is observed, and the ortho-substituents (2-Cl, 2-CF3 and 2-CH3) of our compounds seem to replace this structured water. Combined, the positive electrostatic potential of this binding pocket indicated that an electronegative substituent at ortho-position of the C3-amine benzyl ring is likely better than neutral or electron-deficient substituents. This is evidenced by the 7-fold better binding potency of 2-Cl and 2-CF3 substituted compound 4h and 4j compared with 2-CH3 substituted 4d and 2-fold lower EC50 of 3b (2-Cl) relative to 3c (2-CH3) (). Superposition of compound 3b-bound structure onto the previously reported inhibitor 2a (PDB ID: 5VAD) bound HcProRS complex either by aligning the protein structure (RMSDCα of 0.362 Å over 383 residues) or by superposing the atoms of the pyrazinamide core (RMSD of 0.111 Å), shows that both compounds have very similar conformations. Therefore, 3b mimics almost all the important interactions of compound 2a with the exception of the water-mediated H-bonds formed by carbonyl oxygen in the C3-substituted part of 2a (the carbonyl oxygen atom is missing in our compounds) (f). This experimentally determined binding mode of 3b is also fully consistent with our predictions based on docking ().
2.4. Molecular Dynamics (MD) Simulations
In an attempt to explain the preference for binding compounds with ortho-substitution on the benzyl ring as suggested by TSA and crystallographic data discussed above, we ran 10 ns MD simulations of the respective ternary complexes. Simulations were performed for 2-substituted compounds 4d (R = 2-CH3), 4h (R = 2-Cl), 4j (R = 2-CF3), and compared to the results for unsubstituted 4a (R = H). The initial model for the complex of 4a was created from the complex of 4h by simple alchemical transformation of the ligand (replacement of the 2-Cl substituent with a hydrogen). The stability of the ligand’s pose in time was judged by calculating the root mean square fluctuation (RMSF) values for individual heavy atoms of the ligand. Additionally, the stability in terms of the position of the pyrazinamide core was assessed by monitoring the distances of the H-bond forming atoms of the ligand and the receptor, covering all three above-described H-bonds (to carboxamidic hydrogen and to pyrazine N1 and N4). We also observed the conformational rotation of the benzyl plane that was expressed by relative angle to the original state. As shown in, the comparison between 2-substituted 4j (R = 2-CF3) and unsubstituted 4a demonstrated that both ligands were stable in their pyrazinamide part. However, the ortho substituent in 4j led to the stabilization of the benzyl ring in the original conformation, indicated by small RMSF values for carbon atoms C12-C17 and by negligible changes in the relative conformation angle of the benzyl plane. In contrast, the benzyl ring of the unsubstituted derivative 4a is predicted to be much more flexible, with the angle oscillating back and forth approximately in the range of −135° to +135°. For full results, see. The similar stabilization of the benzyl ring was also detected for the other studied 2-substituted compounds 4d and 4h ().
The results of MD simulations confirm our hypothesis that 2-substitution of the benzyl ring stabilizes the ligand in the ATP-binding site of HcProRS. In the MD-stabilized conformation, which is identical with the crystallographic conformation, the ortho substituent is located in a water binding site as seen in the 2a-bound structure. Our in silico experiments and the results of the thermal shift assay jointly suggest the ortho-substitution to be favourable. This is further confirmed by the fact that all compounds which we succeeded to co-crystallize with HcProRS bear such an ortho substituent.
2.5. Potential Selectivity of Title Compounds towards HcProRS over Other Class II aaRS Members
The 20 standard aaRSs are split into two classes based on two completely different folds of the catalytic cores that originated for the specific recognition and binding of the shared substrate ATP [[27],[28],[29]]. The catalytic domain of class I adopts a Rossmann fold while that of class II is organized as a six-stranded β-sheet. Although the binding mode of ATP is strictly conserved in each aaRS class, as ATP-competitive aaRS inhibitors, cladosporin has been reported to specifically target LysRS rather than other class II aaRSs [[30]] and compound 2a reportedly also demonstrated 2500-fold selectivity for HcProRS versus another class II member ThrRS. Since our compounds share a similar binding mechanism with compound 2a, we were encouraged to examine the potential selectivity of our compounds using the available structures of human class II tRNA synthetases.
Taking compound 3b as an example, we modelled 3b in the ATP binding site of all other nine human class II aaRS structures by superimposing the adenine group of available adenine-containing ligand bound structures with HcProRS:Aze-SA (azetidine-2-carboxylic acid (Aze) coupled to 5′-sulfamoyladenosine (SA) is a prolyl-adenylate analogue; PDB ID: 5V58) complex followed by superimposing HcProRS:Pro:3b and HcProRS:Aze-SA structures [[31]]. To thoroughly examine the molecular mechanisms defining the potential family selectivity of 2a and its derivative 3b, we split the ligand molecule into three fragments: the C3 substituent, pyrazinamide ring and C2-carboxamide moiety (). The former two parts are found in all five confirmed hits and 2a. In HcProRS, the subpocket binding the 2-substituted benzyl at C3-amine is constituted by Trp1169, Gly1274 and Gly1238. Examination of other class II aaRSs showed these equivalent residues are highly variable. In SerRS, ThrRS, HisRS and GlyRS, belonging to the same subclass (IIa) with ProRS, the corresponding Trp1169 in HcProRS is substituted with polar residues (such as Lys323, Gln493, Gln173 and Met294, respectively) leading to the loss of hydrophobic interaction with the C3-amine benzyl ring of our ligands (a). In addition, Gly1274 in motif-3 β-strand is replaced with larger amino acids (Thr429, Ala625, Ser383 and Ser524) that would likely cause a steric clash with the benzyl ring. The Gly1238 is also replaced with residues with side chains (Leu392, Cys591, Val357 and Ile404). Despite these side chains do not make direct contacts with the compound, their presence will immobilize backbone which restricts the carbonyl group pointing to and in parallel with the motif-3 β-strand resulting in potential sterically clash with the ortho-substituent of compound (a and). Therefore, these residue replacements likely lead to unfavourable binding. For other class II members (AspRS, AsnRS, LysRS, AlaRS and PheRS), two out of three residues are different from those in HcProRS, and the residue equivalent to Gly1238 was replaced by isoleucine or valine. Similarly, both could also result in potential steric repulsion (a).
The pyrazinamide moiety sits on the top of the N-terminus of motif-3 α-helix. Structural comparison showed that the stacking residues equivalent to Arg1278 and Phe1167 are highly conserved, and that the residue corresponding to Thr1164 (making backbone interactions) is always present in other class II aaRSs, which may provide a general binding for the pyrazinamide core. However, an obvious distinction was observed in the length of the N-terminus of motif-3 α-helix, where three residues (Thr1276, Thr1277 and Arg1278) are present in HcProRS that are replaced by four residues in all other class II tRNA synthetases (b,c). This different structural rearrangement may affect the upstream β-strand and the distinct residues in this region may also provide a steric exclusive binding site for the pyrazinamide ring. The 2-substituted benzyl moiety at C2-carboxamide extends towards the motif-2 loop region and only makes non-specific van der Waals interactions. However, both sequence length and mobility of this loop region are highly diverse. It is difficult to predict how this region will be arranged when these compounds would bind. However, in SerRS, due to the presence of the large side chain of Arg192 sitting in the opposite side of motif-2 loop (equivalent to a glycine in HcProRS (a)), a similar steric repulsion is observed for the binding of 3b, suggesting our confirmed binders are unlikely to bind to SerRS. In summary, we generally note subtle differences in the ATP binding site which appear to be important for designing potential selective ATP competitive inhibitors of aaRSs.
3. Discussion
Aminoacyl-tRNA synthetases (aaRSs), catalysing the formation of aminoacyl-tRNA, have been explored as targets for the development of antimicrobials [[9],[32],[33],[34],[35]]. However, HcProRS, located at the C-terminal region of a dual function EPRS, is likewise recognized as a promising target for anti-cancer [[12],[13],[14]], anti-fibrosis [[11],[36]] and anti-autoimmune response [[37]] drug development. Up to date, several HcProRS inhibitors were reported. Halofuginone specifically targets both the amino acid site and 3′-end of tRNA binding sites [[10]] while compounds 2a [[16]] and T-3833261 [[36]] occupy the ATP binding site of HcProRS. Since the HcProRS inhibitory activity of halofuginone can be reverted by increased concentration of Pro [[38]] and Pro is accumulated in fibrotic tissues [[39]], the anti-fibrosis activity of halofuginone is hampered. Therefore, the latter two compounds, as ATP-competitive inhibitors, may overcome this issue. However, since class II aaRSs share a very similar ATP binding mechanism, selective inhibition of HcProRS without affecting other class II members is necessary for drug development.
Compound 2a demonstrated potent HcProRS inhibitory activity with an IC50 value of 12 nM using ATP/PPi exchange assay in vitro and potential selectivity towards HcProRS over the class II ThrRS, without further follow-up studies. Virtual screening of our in-house library of synthetic small molecules using compound 2a-bound structure as the initial model, identified two general structures of substituted 3-aminopyrazinamide (3, 4) [[17],[18]], which were predicted to bind to HcProRS with the same binding mode as 2a and with high affinity. Selected compounds were subjected to a thermal shift assay. Five compounds showed clearly increased thermal stability of HcProRS in the presence of Pro, indicating these compounds are Pro-dependent binders of HcProRS similar to compound 2a. This was further proven by the co-crystal structures of HcProRS in complex with a respective compound and Pro. Pro binding induced the movement of the side chain of Arg1152 out of the ATP binding site, providing additional space to accommodate the binding of pyrazinamide derivatives. The pyrazinamide core mimics the adenine of ATP while the C3-substituent replaces the ribose of ATP via hydrophobic interactions with the active site of HcProRS. The modifications at the C2-carboxamide extending toward the motif-2 loop region of HcProRS improved the binding affinity which can be evidenced by the decreased EC50 values for 3b and 3c (R = 2-Cl; 2-CH3) compared to those of 4h and 4d (R = 2-Cl; 2-CH3) having unsubstituted carboxamide at C2.
High class selectivity against HcProRS rather than simultaneous targeting of other class II members is fundamental for therapeutic applications of aaRS inhibitors. Despite ATP having a very similar binding pose in class II aaRSs, the subtle differences between active sites in different class II aaRS enzymes may contribute to designable selectivity. Modelling compound 3b in all other class II aaRS structures pointed out that three residues surrounding the ortho-substituted benzyl moiety at C3 of the pyrazinamide core and making hydrophobic interactions with the ligand, were replaced with either polar or larger residues in other class II enzymes. Therefore, these latter enzymes cannot form stable interactions with the C3-substituted moiety. In addition, the specific arrangement of the N-terminus of motif-3 α-helix in HcProRS (being one residue shorter than that in other class II members) is responsible for the exact positioning of the pyrazinamide ring, which may also contribute to its selectivity.
In conclusion, a combination of virtual screening with biophysical and crystallographic studies successfully identified five new pyrazinamide-based HcProRS ligands that specifically compete with the binding of the ATP substrate in a Pro-dependent manner. This research not only applies to the potential repurposing of old compounds for new applications, but also provides the possibility to further improve the binding affinity and potential selectivity of HcProRS inhibitors by structure-based drug design particularly at the C3 substituent. This should be followed by inhibitory activity measurements and whole cell assays. Taken together, our study leads to the successful exploration and broadening of the chemical space around the confirmed HcProRS inhibitor reported by Adachi. These may provide new drug candidates for the clinical treatment of ProRS-associated diseases.
4. Materials and Methods
4.1. Protein Preparation
The DNA sequence encoding human ProRS (HcProRS) encompassing residues 1001–1512 (UniProt accession ID: P07814) was amplified by PCR from a cDNA library produced from HEK293T cells. The amplified gene was cloned into the in-house pETRUK vector, a derivative of pETHSUL [[40]] that yields a fusion protein with a SUMO tag at the N-termini for expression in E. coli Rosetta 2 (DE3) pLysS cells. After culture by using auto-induction media [[41]], cells were harvested and lysed by sonication in cation exchange buffer A containing 25 mM Hepes pH 8, 200 mM NaCl and 5 mM β-mercaptoethanol supplemented with 1 mM MgCl2 and 100 U Cryonase (Takara). The lysis was cleared by centrifugation at 18,000× g for 30 min and the resulting supernatant was applied to a 5 mL Hitrap SP column (Cytiva, Marlborough, MA, USA). The fractions corresponding to the SUMO-fused HcProRS were collected followed by SUMO hydrolase treatment to remove the SUMO tag. This combined mixture was dialyzed in buffer containing 10 mM Tris pH 7, glycerol 10% (w/v) overnight to remove the salt which was then loaded onto a 5 mL Hitrap SP column (Cytiva, Marlborough, MA, USA) to remove the SUMO tag and SUMO hydrolase. The flow through containing HcProRS was further purified by anion exchange chromatography and gel filtration. Purified HcProRS was concentrated to 45 mg/mL in the final buffer (10 mM Tris pH 7, 100 mM NaCl and 2.5 mM β-mercaptoethanol) and stored at −80 °C.
4.2. Crystallization of HcProRS Complexes
For crystallization of HcProRS complexes, the purified protein at 30 mg/mL in 10 mM Tris pH 7, 100 mM NaCl, 2.5 mM β-mercaptoethanol was incubated with 10 mM L-proline and 12% (v/v) DMSO in the absence or presence of 2 mM compound on ice for 1 h before setting up crystallization using the Microbatch method. Crystals were grown in a Terasaki Microbatch plate by mixing 1 µL the pre-mix with 1 µL of reservoir solution containing 0.25–0.4 M SrCl2, 15–20% (w/v) PEG3350 and 100 mM HEPES pH 7.5. The drops were covered with paraffin oil. Suitable crystals were mounted and flash frozen in liquid nitrogen directly from the plate.
4.3. Data Collection and Structure Determination
All X-ray diffraction datasets were collected from cryo-cooled crystals at Synchrotron radiation facility and processed by using autoPROC software package [[42]]. The initial model of structure was determined by molecular replacement with reported HcProRS:Pro:2a structure (PDB ID: 5VAD) using the program PHASER from the Phenix program suite [[26]]. Subsequent structure refinement was conducted in Phenix.refine and the model was manually built in COOT [[43]]. Data collection and refinement statistics are summarized in, and all structural figures were generated using PyMol (http://www.pymol.org, version 2.0.4). All the crystal structures were deposited in PDB and the related accession codes are as follows: 7OSY, 7OSZ, 7OT0, 7OT1, 7OT2 and 7OT3.
4.4. Thermal Shift Assay
A total of 0.2 mg/mL HcProRS (corresponding to 3.45 µM) was incubated with various concentrations of different compounds in the absence or presence of 1 mM L-Pro with 1x thermal shift dye (Life Technologies, California, USA) in reaction buffer containing 50 mM HEPES pH 7.0, 150 mM KCl and 10% (v/v) ethylene glycol. Aliquots (20 µL) were transferred to a 96-well PCR plate in triplicate. After centrifugation to remove air bubbles, the plate was measured in Applied Biosystems® Protein Thermal Shift machine and subjected to a thermal gradient from 4 °C to 95 °C with the increasing rate of 0.05 °C/s. The fluorescence was detected by Protein Thermal Shift™ Software (Life Technologies, California, USA). Data analysis was performed by Boltzmann sigmoidal fitting to calculate melting temperature (Tm).
4.5. In Silico Studies
Inputs for molecular docking and MD were prepared in Molecular Operating Environment (MOE), 2020.09 (Chemical Computing Group ULC, Montreal, QC, Canada). The docking was run in MOE. MD simulations were run in NAMD2 [[44]] (version GIT20190909) and analysed in VMD version (1.9.4a51) [[45]]. For experimental details, see the.
Acknowledgements
AcknowledgmentsThe authors thank the China Scholarship Council for providing a scholarship to Luping Pang. We further thank Kalyan Das for providing the equipment for TSA measurements and to V.S.K. Pallabothula for the synthesis of compound 2a. Computational resources were supplied by the project “e-Infrastruktura CZ” (e-INFRA LM2018140) provided within the program Projects of Large Research, Development and Innovations Infrastructures.