Cen­tral ner­vous sys­tem (CNS) tox­i­c­i­ty ac­counts for 25% of safe­ty fail­ures across drug dis­cov­ery and de­vel­op­ment, yet it is rarely de­tect­ed dur­ing stan­dard GLP tox­i­col­o­gy stud­ies¹. Seizure risk re­mains one of the most dif­fi­cult CNS ad­verse ef­fects to pre­dict be­fore hu­man ex­po­sure be­cause 2D cell-based as­says and an­i­mal mod­els can­not ad­e­quate­ly mea­sure hu­man neur­al ac­tiv­i­ty.

Seizures arise from in­sta­bil­i­ty in neur­al net­works, dri­ven by dis­rup­tions in the bal­ance be­tween ex­ci­ta­to­ry and in­hibito­ry sig­nal­ing. These ef­fects emerge at the lev­el of in­ter­con­nect­ed sys­tems, not iso­lat­ed cells. As a re­sult, re­duc­tion­ist in vit­ro mod­els and an­i­mal mod­els of­ten gen­er­ate sig­nals that fail to align with clin­i­cal out­comes.

The 28bio En­durance Study ap­proach­es this chal­lenge dif­fer­ent­ly by ask­ing whether func­tion­al neur­al ac­tiv­i­ty in CNS-3D Brain Organoids can serve as a di­rect pre­dic­tor of seizure risk.

The study is a ret­ro­spec­tive, non-in­ter­ven­tion­al eval­u­a­tion of 66 small mol­e­cules with doc­u­ment­ed hu­man clin­i­cal out­comes from 120,551 pa­tients and known pos­i­tive con­trols. By an­chor­ing the analy­sis in known hu­man out­comes, the study is de­signed to as­sess pre­dic­tive per­for­mance in a clin­i­cal­ly rel­e­vant con­text. The drug set spans 30 seizure-in­duc­ing and 36 clin­i­cal­ly safe drugs with di­verse mech­a­nisms of ac­tion, in­clud­ing ion chan­nel mod­u­la­tors, neu­ro­trans­mit­ter path­way agents, and com­pounds with para­dox­i­cal ef­fects, re­flect­ing the com­plex­i­ty of re­al-world drug de­vel­op­ment.

Rather than re­ly­ing on in­di­rect mark­ers, the ap­proach mea­sures drug-in­duced changes in func­tion­al neur­al ac­tiv­i­ty us­ing hu­man iP­SC-de­rived cor­ti­cal organoids. These pat­terns are trans­lat­ed in­to quan­ti­ta­tive fea­tures de­scrib­ing os­cil­la­to­ry be­hav­ior, burst struc­ture, and net­work syn­chro­niza­tion. To com­ple­ment the func­tion­al da­ta, mol­e­c­u­lar fea­tures de­rived from chem­i­cal struc­ture are in­cor­po­rat­ed us­ing es­tab­lished fin­ger­print­ing meth­ods. To­geth­er, these in­puts form the ba­sis of an AI neu­ro­tox­i­col­o­gy pre­dic­tive mod­el trained to clas­si­fy seizure risk, by learn­ing how drugs per­turb neur­al sys­tems and how those per­tur­ba­tions re­late to clin­i­cal out­comes.

Re­sults from the En­durance Study demon­strate CNS-3D Brain Organoids pre­dict clin­i­cal seizure li­a­bil­i­ty with 83% sen­si­tiv­i­ty and 89% speci­fici­ty. This per­for­mance in­di­cates strong dis­crim­i­na­tion be­tween seizure-as­so­ci­at­ed and clin­i­cal­ly safe drugs, while main­tain­ing a bal­ance be­tween de­tect­ing risk and avoid­ing un­nec­es­sary ex­clu­sion. In prac­tice, this al­lows teams to de­pri­or­i­tize com­pounds with a high like­li­hood of seizure li­a­bil­i­ty while re­duc­ing the risk of elim­i­nat­ing clin­i­cal­ly vi­able can­di­dates.

Be­yond bi­na­ry clas­si­fi­ca­tion, the mod­el pro­duces con­tin­u­ous scores that strat­i­fy drugs by risk lev­el. These scores tend to align with known mech­a­nisms and with the clin­i­cal preva­lence of seizure events, sug­gest­ing that the sys­tem cap­tures bi­o­log­i­cal­ly mean­ing­ful gra­di­ents rather than ar­bi­trary thresh­olds.

Pub­lished com­par­isons in­di­cate that CNS-3D Brain Organoids demon­strate 13x high­er pre­dic­tive per­for­mance than an­i­mal mod­els and out­per­form 2D cell-based as­says.

Seizure li­a­bil­i­ty is of­ten treat­ed as an emer­gent risk that can­not be ful­ly pre­dict­ed be­cause com­pounds with sim­i­lar pro­files can di­verge sig­nif­i­cant­ly in clin­i­cal out­comes, re­flect­ing the com­plex­i­ty of neur­al sys­tems.

The 28bio ap­proach ad­dress­es this lim­i­ta­tion. By in­cor­po­rat­ing func­tion­al hu­man da­ta, mol­e­c­u­lar fea­tures, and AI mod­el­ing, re­searchers can as­sess drug can­di­dates ear­li­er in de­vel­op­ment, de­pri­or­i­tiz­ing com­pounds that desta­bi­lize neur­al net­works ahead of in vi­vo stud­ies and mov­ing for­ward safe com­pounds with greater con­fi­dence.

For a field de­fined by late-stage sur­pris­es, the abil­i­ty to mea­sure dys­func­tion be­fore it man­i­fests clin­i­cal­ly rep­re­sents a mean­ing­ful ad­vance­ment.

28bio is a neu­rotech­nol­o­gy com­pa­ny en­gi­neer­ing hu­man brains at-scale ex­hibit­ing mem­o­ry, learn­ing, and cog­ni­tive func­tions. Its Nex­on™ plat­form in­te­grates tis­sue en­gi­neer­ing, neur­al in­ter­fac­ing, and AI to re­verse to­day’s neu­ro­log­i­cal health cri­sis by im­prov­ing the abil­i­ty to pre­dict which ther­a­pies will work in hu­mans. 28bio is com­mit­ted to ad­vanc­ing eth­i­cal stan­dards in the de­vel­op­ment of brain organoid tech­nol­o­gy and en­gi­neered hu­man cog­ni­tion. Fol­low us on LinkedIn.

Ref­er­ences:

¹Walk­er, et al. Drug dis­cov­ery and de­vel­op­ment: Bio­mark­ers of neu­ro­tox­i­c­i­ty and neu­rode­gen­er­a­tion. Ex­per­i­men­tal bi­ol­o­gy and Med­i­cine 2018; 243: 1037-1045.

Rev­o­lu­tion Med­i­cines re­port­ed Mon­day that its ex­per­i­men­tal RAS in­hibitor called darax­on­ra­sib suc­ceed­ed in a reg­is­tra­tional tri­al for pan­cre­at­ic can­cer.

And now, the com­pa­ny plans to ask the FDA to ap­prove the ther­a­py un­der the agency’s Com­mis­sion­er’s Na­tion­al Pri­or­i­ty Vouch­er pro­gram that’s meant to speed up re­view times for a small num­ber of se­lect­ed drugs.

In a Phase 3 tri­al, darax­on­ra­sib led pa­tients with metasta­t­ic pan­cre­at­ic duc­tal ade­no­car­ci­no­ma to sur­vive for a me­di­an 13.2 months com­pared to 6.7 months with chemother­a­py, the com­pa­ny said. That means darax­on­ra­sib cut the risk of death by 60% ver­sus chemother­a­py.

“Break­ing the one-year bar­ri­er in pan­cre­at­ic can­cer is an enor­mous achieve­ment,” Rev­o­lu­tion Med­i­cines CEO Mark Gold­smith told End­points News.

Al­lo­gene Ther­a­peu­tics’ CAR-T ther­a­py re­moved all re­main­ing de­tectable lym­phoma cells in just over half of treat­ed pa­tients, ac­cord­ing to an ear­ly look at the Phase 2 da­ta.

The Bay Area biotech is seek­ing to ap­ply "off-the-shelf" cell ther­a­py in a dif­fer­ent way — and how its t