What makes a good neoanti­gen?

For all the promis­es of the bold new ap­proach to can­cer vac­cines and ther­a­pies — ze­ro­ing in on spe­cif­ic mu­tat­ed anti­gens ex­pressed on­ly by tu­mors — com­pa­nies and aca­d­e­mics have lit­tle way of know­ing how good they are at pre­dict­ing which neoanti­gens rep­re­sent the best tar­gets. There’s no stan­dard or base­line for play­ers to stack them­selves against ri­vals in the nascent field, and by the time they find out, it could be too late.

No sin­gle group could re­al­ly build that bench­mark. Pre­cious pro­pri­etary in­for­ma­tion is at stake, not to men­tion tremen­dous re­sources re­quired.

Dan­ny Wells

But four years ago, just as the con­cept was tak­ing off, it struck the Park­er In­sti­tute for Can­cer Im­munother­a­py as the ex­act kind of prob­lem its col­lab­o­ra­tive mod­el was built to solve.

“We’re like Switzer­land of neoanti­gens,” Dan­ny Wells, the prin­ci­pal da­ta sci­en­tist at PI­CI, told End­points News.

So work­ing with the Can­cer Re­search In­sti­tute and a non­prof­it named Sage Bionet­works, it brought to­geth­er over 40 bio­phar­ma com­pa­nies and aca­d­e­m­ic labs, gave them the same melanoma and non-small cell lung can­cer tis­sue, and asked each team to sub­mit its most promis­ing neoanti­gen pre­dic­tions. PI­CI re­searchers then went in­to the lab and cross-com­pared the pre­dic­tions, check­ing whether the neoanti­gens were in­deed rec­og­nized by T cells. The re­sult is a base­line dataset that the ini­tia­tive — named TES­LA, short for Tu­mor Neoanti­gen Se­lec­tion Al­liance — is mak­ing pub­lic to the sci­en­tif­ic world to­day.

To the re­searchers’ sur­prise, the dif­fer­ences be­tween the pre­dic­tion al­go­rithms were “tremen­dous,” said Wells, a co-se­nior au­thor in the pa­per pub­lished in Cell. No team man­aged to iden­ti­fy every neoanti­gen or even a large ma­jor­i­ty of them: “The over­lap be­tween pre­dic­tions, no mat­ter how we sliced it, was re­al­ly low.”

It high­lights the need for new knowl­edge in the field, he added, and PI­CI be­lieves TES­LA has yield­ed some in­sights.

Na­dine De­fra­noux

A set of five dis­tinct fea­tures, it turned out, could pre­dict good neoanti­gens with high ac­cu­ra­cy and speci­fici­ty when in­te­grat­ed in­to a mod­el. They are bind­ing affin­i­ty, tu­mor abun­dance and bind­ing sta­bil­i­ty, which has to do with how the neoanti­gens are pre­sent­ed; as well as agre­topic­i­ty and for­eign­ness, which re­lates to their recog­ni­tion by im­mune cells.

“These are all fea­tures that had been talked about, but I think we were sur­prised that just by in­te­grat­ing them to­geth­er in­to this sin­gle mod­el that it works so well,” Wells said.

When par­tic­i­pat­ing teams reap­plied these char­ac­ter­is­tics in­to their al­go­rithms, PI­CI re­port­ed, the pre­dic­tions im­proved. A da­ta mod­el em­pha­siz­ing all five fea­tures came out of a test against an­oth­er set of can­cer sam­ples ac­cu­rate­ly pre­dict­ing 75% of ef­fec­tive neoanti­gen tar­gets and fil­ter­ing out 98% of in­ef­fec­tive ones. De­pend­ing on the ther­a­peu­tic strat­e­gy drug de­vel­op­ers may cal­i­brate their al­go­rithms dif­fer­ent­ly, Wells said. But the hope he and co-se­nior au­thor Na­dine De­fra­noux have is that they can pro­vide a com­mon base­line both for those al­ready in the field and oth­ers look­ing to jump in.

Lisa But­ter­field

“This re­search has the po­ten­tial to im­prove drug mak­ers’ and re­searchers’ math­e­mat­i­cal al­go­rithms,” Lisa But­ter­field, vice pres­i­dent of re­search and de­vel­op­ment at PI­CI, said in a state­ment. “It can pri­or­i­tize anti­gens most like­ly to be present on each pa­tient’s can­cer and most vis­i­ble to the im­mune sys­tem while de­pri­or­i­tiz­ing the ones that aren’t.”

Robert Schreiber

While TES­LA won’t be mon­i­tor­ing the space lon­gi­tu­di­nal­ly to see how com­pa­nies com­pare over time, Wells does en­vi­sion new evo­lu­tions for the coali­tion. Many ques­tions still need to be an­swered; for in­stance, while they fo­cused on class 1 pre­dic­tion, or how CD8+ T cells see the tu­mor, co-se­nior au­thor Robert Schreiber’s re­search has sug­gest­ed that class 2, or how CD4+ T cells see the tu­mor, are just as im­por­tant.

“Un­til now, neoanti­gen pre­dic­tion has been a black box,” said Schreiber, a pro­fes­sor at the Wash­ing­ton Uni­ver­si­ty School of Med­i­cine in St. Louis. TES­LA has be­gun shed­ding light on it, and it in­tends to con­tin­ue.

The FDA’s Arthritis Advisory Committee on Thursday voted 10 for and 8 against the approval of ChemoCentryx’s $CCXI investigational drug avacopan as a treatment for adults with a rare and serious disease known as anti-neutrophil cytoplasmic autoantibody (ANCA)-vasculitis.

The vote on whether the FDA should approve the drug was preceded by a split vote of 9 to 9 on whether the efficacy data support approval, and 10 to 8 that the safety profile of avacopan is adequate enough to support approval.

As if Biogen’s aducanumab isn’t controversial enough, the researchers at drug pricing watchdog ICER have drawn up the contours of a new debate: If the therapy does get approved for Alzheimer’s by June, what price should it command?

Their answer: At most $8,290 per year — and perhaps as little as $2,560.

Even at the top of the range, the proposed price is a fraction of the $50,000 that Wall Street has reportedly come to expect (although RBC analyst Brian Abrahams puts the consensus figure at $11.5K). With critics, including experts on the FDA’s advisory committee, making their fierce opposition to aducanumab’s approval loud and clear, the pricing pressure adds one extra wrinkle Biogen CEO Michel Vounatsos doesn’t need as he orders full-steam preparation for a launch.