Grouping of Per- and Polyfluoroalkyl Substances (PFAS) for Human Health Risk Assessment

Expert 1's comment above correctly asks if production and use predict exposure. They do not. A low volume chemicals (or a low concentration contaminate of a product) with high releases may be found more often than a high volume site-limited intermediate. But the complete absence of production can be used to exclude a chemical.

To be clear creating a Cumulative Assessment Group (CAG) is not a function of persistence, it is defined by having a common mechanism of action consisting of a common AOP Network and preferably the same molecular initiation event. As Expert 5 stated "The toxicological outcome, the adverse effect is the optimal means to conduct a grouping activity for mixtures or cumulative risk assessment. It is agnostic to physical chemical properties, environmental fate and transport, biological longevity and chemical structural attributes." Persistence could be used in a definition of PFAS as a regulatory decision but not as a basis for a CAG.

I agree with the comment from Expert 3.

I and experts 4 and 8 argued that the inclusion of compounds like Prozac or Fipronil are not a problem because such compounds need to be evaluated, and if necessary, regulated under the SDWA whether they are considered PFAS or not. This is correct. However, what I did not consider is the pressure to consider all PFAS or all substances in a subgroup of PFAS in a Combined Assessment Group (CAG). I could see an agency being pressured to include Prozac or Fipronil in a PFAS CAG because the agency could not "prove" that the chemical did not belong in the CAG. As a result, it may not be optimal to consider these compounds as PFAS but rather to assess them separately.

The updated draft risk assessment EPA just released on GenX has some interesting thoughts about PPAR alpha and liver toxicity. The updated RfD for GenX is based on liver toxicity and a constellation of effects that may or may not be mediated via PPAR alpha. A key characteristic described for PPAR alpha-mediated hepatocarcinogenicity in rodents includes certain types of pathology (i.e., apoptosis) and this pathways is thought not to be operable in humans. Pathological findings in livers from mice exposed to GenX showed apoptosis and necrosis as well as some other pathologies, which suggests that not all effects in the liver are PPAR alpha-mediated. Again, this raises a potential question about what subtype of toxicity should drive RPFs or PODs or other values.

Very cool Expert 5! I didn't know about this approach. It seems very reasonable.

Maybe we need "NHANES Biopsy Biomonitoring" to determine if we can detect PFAS in tissues that we can't detect in blood. This is meant to be a bit irreverent but we also have to remember that what is measured in blood isn't always the best proxy for what is inside people's bodies. We have very little human tissue-level information on PFAS...and actually, very little experimental animal tissue data except for liver levels.

Regarding the comment from Expert 1, we also have to consider the byproducts associated with production. Some are produced only after release into the environment so are weird "byproducts of environmental transformation of byproducts." I'm sure a chemist could explain that much more eloquently.

The list of 150 PFAS is supposed to be somewhat representative - from the EPA - "Substances were selected based on a prioritization scheme that considered EPA Agency priorities, exposure/occurrence considerations, availability of animal or in vitro toxicity data, and ability to procure in non-gaseous form and solubilize samples in DMSO." This highlights that even if we want to generate data on a set of PFAS that are truly representative, we may be limited by the availability of purified product for experimentation.

In states that have MCLs on specific PFAS, additivity is the prevailing assumption when PFAS are grouped together.

It's interesting that there seems to be an almost even split between the workability potential of the problem formulation. I think the problem formulation is workable if it is specific to a single PFAS or a very defined subgroup of PFAS. If only PFAS were subgrouped in the environment/living organisms.

I am not sure if we speak the same language:
TOF = total organic fluorine
TOF = time of flight mass spectrometer ( I think this is useful for screening a large number of PFAS)

In response to Expert 3, our limit of quantification for PFBS in human milk is 0.006 ng/g.
We could not quantify PFBS in the general population at this LOQ - maybe an argument to exclude PFBS from a priority list? or leave it "special cases" like the firefighters?

My response on data gaps was more than a little bit snarky. For this I apologize to the group. But The point I was trying to make is that the wording in the draft problem formulation did not appear to think through the technical issues required in a statement. As a result, it was hard to actually answer the questions.

I agree with the above but the outcomes of the assessment - iterative approach - must be linked or followed by a risk management step; otherwise the assessment is pretty academic.
There may be intervention options such as eliminating an emitting present source or apply end-of-pipe measure to reduce/eliminate the exposure.
Also need to distinguish between short-time high exposures (of high levels? and very toxic compounds) vs. chronic exposures to low levels of (less toxic) compounds.

Expert 4's comment on apportionment is correct. But in the past apportionment has been been done based on policy in the absence of data. So this might not be a problem.

Good analytical data seem to be the basis but they need good interpretation; the pure numbers (with the units) are not sufficient and a value can be seen in many different ways.

In environmental chemistry, there are ambiguous uses of certain terms:
- half-life as half-life of the chemical under consideration or as the time to reduce the environmental concentration by 50%.
- Persistence is not a defined parameter for a chemical and as such does not have a unit, the proxy needs to be defined and the matrix (in air, water, biota, sediment, ...) with the conditions (natural or enforced)

I agree with the above (around Expert 10) but 150 PFAS may be too many; suggest to reduce to "representative" for the different groups of PFAS and a reasonable number of toxicological parameters (how many? In the same range as the number of chemicals?.

In response to expert 8's comment:
The text is unclear, but I took the statement “the assumed physical-chemical properties of some PFAS such as is necessary to make basic assumptions about the potential for a given PFAS to be present in the general population’s drinking water supply” to imply that phys-chem properties could be used to exclude certain PFAS (polymers, etc). Not imply that all soluble PFAS must be included. I agree that the latter point is correct. Sufficient solubility is not proof of the presence of a substance.

"...how should “certainty” or “variance” related to differences in study concentrations and durations be factored into extrapolation of data.. ?" Nobody touched on the concentration aspect as addressed within the Human Relevance Framework. One of its three points, on quantitative differences, is loosely stated as, "Are the concentrations studied comparable to exposures likely in the human?" Certainty will be increased in doing so.
Boobis et al. 2006. IPCS framework for analyzing the relevance of a cancer mode of action for humans Crit Rev Toxicol 36:781-792.

Yes. It will be interesting to see exactly what and to learn how the Whole Chemical Approach will be implemented under TSCA.

There are actually at least a couple of slants in this question. Quite a good question. Relative to PPAR alpha, and to liver weight, I feel that the jury is still out. The development of RPF values for these endpoints might be feasible, but the uncertainty in whether they represent, respectively, an MOA or an outcome related to toxicity remain a matter of discourse. Given differences in expression and binding affinities of the receptor between animals and humans, it seems that a focus exclusively on PPAR alpha binding is a incomplete enough assessment as to be accompanied by insufficient certainty as to be applicable. So, the issue of "robust" emerges. So, the caveats emerge related to the relationship of the events observed (in vivo or not) to the observed toxicities or adverse health effects. This represents a significantly valuable target for research. And, of course, when toxicities relate to MOAs that are different, RPF values will not be developed, because using the similarity or difference (independence) of MOAs is grounds for assigning components to teh same, or to different mixtures subgroups.

When considering multiple Modes and multiple potential (or real) target organs, the Target Organ Toxicity Dose (as proposed by Mumtaz and colleagues) has merit. Under TTD, not only is the "single" critical effect considered, but secondary effects (those that occur at higher doses/concentrations) are also considered. The TTD approach would be an appropriate construct under which to accommodate multiple target organs/systems. Within The TTD approach, consideration of MOA (within tissues/systems) could also be accommodated, though I cannot identify an example of that, presently.

If we "can" know the concentrations of PFAS for which we do have toxicity/risk estimates, then the EOF/TOF content of those PFAS should be subtracted from the actual EOF/TOF value, and some estimate of the remaining "unidentified" EOF/TOF content could then be addressed. A comparison of the risk of the "known" fraction of the EOF/TOF to the remaining unidentified portion could be informative in a number of ways. As before, sure such an approach could be conducted, but the elephant is the level of confidence that could be placed in the results. Several of the panel have previously commented on the structure and verbiage communicating the queries, and this one might have been communicated a little differently. On the other hand, it/these queries may have been very deliberately phrased in the manner communicated.

Seems like we've had a considerably valuable discourse on this one, seemingly a fairly good general agreement. I appreciate the attention to hotspots, and limited (small) segments of the population. I have read nothing to disagree with, and haven't much left to offer. Good discussion.

The Assessment Group (Cumulative Assessment Group) will include all chemicals included in the assessment. So, "Exposure" is a must-do - might be estimated from production/use, but may also vary geographically, given possible geospatial distribution of point sources. No exposure, no risk. Forming Subgroups requires some repeatable criteria established for further segregation of chemicals, be it on the basis of this or that criteria/criterion. There will be limitations with each step, depending mostly on lack of appropriate data which would have reduced uncertainty. An iterative approach to the endeavor may also be worthwhile.

The comments of experts 9 and 10 reflect my comments on the data availability of those PFAS that have been measured in the public drinking water facilities (UCMR3). There are sufficient data for those specific PFAS . It gets more difficult if other PFAS are included as the data will be lacking for the most part.

There are fairly extensive toxicology data and pk data for the straight chain perfluoroalkyls such as PFBS, PFHxS, PFOS, PFBA, PFOA etc and there has been a great deal of work in developing PBPK models. There are very limited data on mixtures. If the PFAS world is extended beyond these then the data are very limited. Some studies would have been conducted for the EPA under the new chemicals program for TSCA but these are largely confidential studies.

I appreciate the useful comments to the question I initially posted. It reinforces my view that this is the way forward, although there will be many caveats to the way this information can be used to support mixtures risk assessment.

While most comments recognise the importance of understanding MOA/AOP, Expert 9 has raised the important issue of how to assign this information in a grouping exercise, where individual PFAS could be assigned to multiple MOA/AOP pathways. Expert 11 also makes some very pertinent points about whether the MIE is common, or whether a converging AOP network can be identified.

While the ‘yes’ votes slightly outnumber the ‘no’ votes for utilising TOF as a screening tool, I stick with my earlier vote. Experts 2 & 10 have raised an important additional point about whether TOF or EOF is the more appropriate measure for screening.

Overall, a useful set of comments on the utility, and limitations, of the NHANES dataset. The point about NHANES data missing sub-groups in highly exposed occupational settings is a good one.

I like the comment from Expert 11, that seems similar to my initial response. It recognises, as do most of the comments, that analytical data on various types of samples provides the best information on potential exposures, but that it could be supplemented by information from other sources.

Once again, Expert 10 seems to have proposed the best way forward - a research-focussed approach to developing these in silico “tools” so that they may be more suited to the task of characterising the key toxicological properties of PFAS that will inform ‘grouping’ for risk assessment.

Appreciate the comment from Expert 10. A pragmatic approach assuming additivity may work even where there is a lack of information, or consensus, on a possible common critical effect or mode of action.

There seems to be general agreement that limiting the scope to the development of drinking water guidelines makes the process more manageable, although I would still prefer a broader objective. We need to return to the key issue, of what is the most scientifically defensible way of grouping PFAS for conducting a ‘mixtures analysis’ type of risk assessment. I agree with Expert 10, that such a solution will need to be pragmatic and cost-effective, even if not 100% scientifically sound.

I agree that Experts 9 & 10 have outlined the key data gaps, and I also like the "decision tree" approach suggested by Expert 7. It would be useful to know how such a decision tree approach could be fleshed out.

{Further editorial note - I found it frustrating that comments typed into the box would disappear if they were not 'saved' before moving to the next question. That prevents any return for later editing if further thoughts occur as a result of reading more comments}

Interesting array of responses. I assumed "health" was a throw-away word, and focused on human exposures. Analytical measurements are the gold standard, like personal biomonitoring. I had not considered production and use data as a "rule-out" tool, but that offers an appealing way to de-prioritize certain chemicals, but for those living near the hypothetical only production facility in the US won't feel well-represented. TSCA exposure analyses, job analyses and non-user analyses can be pretty solid estimates of exposure, though they sometimes prompt contemporary exposure analyses, often using updated methods and models. The use of "best" seems to ignore practicality, and may have prompted some us to differently consider "likelihood of success" in exposure evaluation. Analytic sensitivity relative to environmental contamination levels seems to be a limitation to success in detecting contaminant levels.

Expert 3 identified an issue important to consider. Persistence applies to chemicals presently in the environment, and does not completely address the issue of continued release. Persistence seems to be a less than optimal means of conducting a grouping exercise. And, persistence has yet to be precisely defined, as alluded to by several experts.

The responses from Experts 10 and 4 are appealing, inasmuch as I opted for choice 1: generate data to parameterize in silico models. I envisioned a governmental or cooperative effort to generate sufficient in vitro data from a list of selected PFAS chemicals, as discussed by Expert 10. As approaches move forward, like this week's announcement from EPA seems to indicate, we may begin to see significant movement in this area.

I am curious if the experts who answered yes to this question have any specific ideas on what the RPFs would be based on. For the fairly well studied PFAAs the toxicity profile is quite different as one moves from C4 to C8. This is true for both the sulfonates and the carboxylic acids. The only common feature between the C4, C6 and C8 is activation of PPARa and subsequent effects on the liver. Many scientists view the early liver weight changes as adaptive, and the liver tumors observed in the rodent C8 studies as not relevant to humans. The developmental effects observed with PFOS and PFOA are by different MOAs and are not observed in the rodent studies of C4 and C6. Would the RPFs be based on half-life and what toxicity endpoint?

There is much agreement on this use of TOF methods as screening approaches. I still think that they cannot be strictly used for risk assessment, but might be a cost-effective way of flagging hot spots for further scrunity.

Some experts point our that NHANES covers the general population, but may miss occupational exposure and contamination hot spots. Seems like a good justification for NOT excluding PFBS in these cases.

Agree with expert 1's conclusion on the various responses to this question.

This question caused huge confusion. It was unclear if the question referred to drinking water guidelines as discussed in earlier questions or not.

Expert 11's answer was clever and I didn't think of that. However, I think the answers of other experts (e.g. 5, 6 and 9) were more in line with what SciPinion wanted. They covered the important data gaps.

The question formulation did appear to cause some confusion and surprise. The PF seemed fine to me as a general approach for (sub)grouping PFAS for the specific purposes of setting drinking water guidelines. I agree with the other experts that many details (regarding mixture exposure and toxicity) remain unresolved. In my opinion, we will never be able to group all PFAS for risk assessment (there are thousands of PFAS and many will not be present in drinking water) and we will need pragmatic and cost-effective solutions for (sub)grouping, even if they are not 100% scientifically sound.

There seems to be large agreement among the panel. If we are unsure which PFAS chemicals comprise the exposure, then it follows that we cannot conduct an exposure assessment, which would serve as the basis of risk characterization. In the case that we have analytical detection capability for specific PFAS chemicals, but the drinking water concentrations are below the limits of detection, then there are statistical means available, and fuzzy logic approaches that may be used to estimate exposures. I can't really get my head to accept a machination of point 3, which I will state as: "Exposure estimates for some chemicals are lacking [but] this limitation is not likely to substantially affect the risk assessment." That may not be a fair thing to do, but that's how I see it.

It sounds like among ourselves, there is a recognition that we have a less than optimal level of knowledge, and that, for some, those data gaps are considered an insurmountable obstacle to completing a risk assessment. Or at least a risk assessment that would be accompanied by an acceptable level of confidence. It may be fruitful for us to explore what level of data "completeness" would be sufficient to serve as the basis for a reliable risk assessment.

Yeah, what Expert 4 said. The question asks for a straight-forward interpretation (which is unnecessary) of (US-centric) EPA mixtures guidance that says unless there are data that support independent (different) modes of action among mixtures constituents, dose additivity is the default additivity approach (meaning, response addition is not to be applied). Since we don't know the MOA, we can't assess interactions (results not predicted by dose additive models). And, sure, we can perform a quantitative risk assessment, but there will be uncertainties, and these uncertainties may rise to the point that they are unsurmountable - that the level of confidence in the QRA is not high enough to accept/apply the results.

It seems that the definition has different implications for regulation versus mixtures risk assessment. It is recognized that regulations may drive a mixtures risk assessment, so the two applications of the definition are not mutually exclusive. Agree with Expert 1, noting that we likely now and likely in the near future will not have enough data to characterize or to identify, even, the multiple possible structures of PFAS that may be present in drinking water. Regardless of the definition, Risk = Hazard X Exposure, and if the component is not identified in the drinking water, then we have *some* level of certainty that it does not pose a risk. So, understanding what chemicals comprise drinking water contamination will inform an optimal definition - with caveats.

Experts 9, 10 and 6 have pretty much captured what I think are the technical data (information) gaps. The response to these comments from Expert 7 (immediately above) rings true with me. We must develop a process that is clearly described, internally consistent, and - hopefully - workable. There will always be data gaps, unless every "mixture of concern" has adequate human dose response data from a chronic exposure.

RE: Expert 3's response, above, sure, I can't argue with that. The suggestion of an iterative approach to problem formulation or refinement could be based on a careful review of the data available and the confidence warranted. Given the state of knowledge available "now", my concern over a too-prescriptive approach is that it might be overly limiting, and reduce the chances of success. My concern is that we somehow let "the perfect" stand in the way of "the good", so to speak.

Interesting responses. Expert 9 raises a general, at least perhaps broadly accepted point of frustration ... in that there are limitations. Some of these limitations may be deal breakers. As a risk assessor, it isn't that the guidelines or standard approaches can't be modified to fit the issue of a "now" risk assessment of "all" or "most" PFAS in water. Rather, I see the most limiting issue is that lack of information on what the contaminants are and what kind of a hazard they present - let alone dose response information. At the end of round 3, we responded to a rather interesting query regarding total F, or the like. An approach like that, posed as a screening approach might be tenable. However, when we have most of a hazard characterization and some dose response for 3, 4, or 5 constituents, that makes estimating potency expressed as "total F" quite uncertain. We have recognized that grouping based on toxicity or mode of action represent the means accompanied by the highest level of confidence, but there are other approaches. The issue becomes, how will we address the intra-subgroup estimates of relative potency? And, for subgroups, will there be a component with a hazard characterization sufficiently developed, and is there enough dose response data upon which we can benchmark a dose response. So - sure, we can group most any way, that isn't an issue, the issue becomes how to estimate risk for the subgroups.

total agreement!!!!!

If drinking water is the media that we are supposed to focus on, then there are data for concentrations in public water facilities for 6 PFAS. This, of course, does not address potential hot spots. If the problem formulation statement goes beyond these 6 PFAS, then we do not have exposure data for drinking water.

NHANES data is useful for determining which which chemicals have widespread exposure. This is useful for determining which chemicals should have MCLs developed. The NHANES data is not helpful in determining hotspots - but hotspots are a local problem and should not be considered necessarily for national standards unless there is also widespread exposure.

It seems to me that there is general agreement that the problem formulation statement defines the media (drinking water) but needs further work to define the relevant PFAS. Then the assessor can determine whether the PFAS should be grouped together, and if so, how.

My overall feeling (with regards to this entire evluation, not only PK) is that the industry crafted a group of chemicals that are difficult to group when using traditional toxicological/risk assessment concepts (regardless of whether this is intentional). Measuring or estimating toxicity and toxicokinetics for all these fluorinated compounds is unrealistic, and unsustainable as new chemicals will replace the ones that are phased out. It seems unavoidable that we'll have to come up with a way to group these chemicals in the face of substantial uncertainty, and consequently, we probably should err on the side of caution and use a broad definition/grouping that may be too conservative, but won't minimize risks. Such an approch would also send a strong message to the industry: it would no longer be an option to outpace regulators by synthesizing new fluorinated chemicals.

The general feeling seems to be that understanding the MOA/AOP is the gold standard, but feasibility is very low.

In response to Expert 2, the EPA mentioned in the Strategic Roadmap (https://www.epa.gov/system/files/documents/2021-10/pfas-roadmap_final-508.pdf) that they will draft a method for total adsorbable fluorine this Fall. So we can expect some documentation to support harmonization in the very near future.

Do we have any data suggesting that presence in drinking water is associated with production and use? It seems to me that we are likely to exclude chemicals which may be present in substantial amounts in contaminated areas, even at low production volumes, if disposed of inappropriately.

I wholeheartedly agree with experts 3, 8 and 9: This should not be the government's responsibility. There seems to be a consensus that in silico methods are interesting, but there are concerns regarding the training set of chemicals not covering the entire PFAS chemical space. A combination of in silico and in vitro tests is likely the most promising approach, with results from in vitro tests allowing to update the predictive models and validation sets as they become available.

I agree with Expert 4 above. The MCL or MCLs for the PFAS mixture or mixtures will not be as robust as the ones for PFOS and PFOA. However, we're at a point where evaluating one PFAS at the time is no longer acceptable on a scientific basis (data supports additivity for some combinations). We basically HAVE to come up with one or multiple MCLs for PFAS groups, even though these values will be tainted with uncertainties.

If a more restrictive definition is used, we must make sure that we are not excluding chemicals which are present in significant amount in drinking water and have a mode/mechanism of action similar to that of main PFAS. This step requires information we do not have at the moment, and probably won't have in the near future. Say there's a site with telomer contamination (which are not included in some of the definitions above) where telomers contribute to 50% of the total organofluorine load in drinking water (some of which activate genes in a way that is similar to legacy PFAS (Reardon et al. 2021)), it seems like excluding these chemicals could lead to an underestimation of risk. Using the OECD definition may be conservative, but it makes sure we're not overlooking some emerging chemistries.

Having read the other experts' comments, I now agree that MOA/Adverse Outcome pathway is the "gold standard" However, my original comment still stands: exposure information is at least as important, regardless of whether there are shared MOAs or not.

Whatever problem addressed or approach taken, the process (risk assessment, setting MCL) must be consistent. We cannot state that the PFAS have different characteristics and not agree on grouping and then use one critical effect or one parameter (bioaccumulation, persistence, etc.) for all compounds or other groups or mixtures.
It does not make sense to ask for many groups and then use one approach in the risk assessment, neither the way backwards to have one critical effect and then assign all PFAS to this effect or parameter.
From the answers and the debate, I conclude that there are deficiencies, gaps, uncertainties, etc. at all ends with the chemical identity, the physical chemical parameters, which impact fate and transport, we do not know enough about the sources and the source strength and we lack knowledge about mode of action, endpoints, etc.
We must be more honest to state the facts and the gaps in each step of the risk assessment - then state how to close the gaps and do so and move forward towards the final risk assessment or MCL (or whatever is the target). We cannot accumulate the gaps (and uncertainties until finish line), have to make assumptions, agree on these and move forward. Otherwise, the process of the risk assessment is not understood.

Maybe it is time to develop one (or more) decision trees to elucidate the risk assessment approach.

My original comment shares Expert 7's concern about exposures via media other than drinking water. However, I think these important data needs can be filled, to a degree, with analytical information. Again, the deveil's in the details: "to a degree" depends on how much time and money you're willing to spend getting data vs making assumptions. Also important that analytical data not only reduces uncertainty, but also allows an understanding of variability.

I cannot agree with "Exposure estimates for some exposures are lacking..." because of the end of that category's description: "...this limitation is not likely to substantially affect the risk assessment." I think the limitations discussed by the panel are likely to substantially affect any risk assessment.

Speaking to those that suggest no QRA can be conducted: I think a QRA can be conducted, however, the uncertainty analysis must be clear about the confidence in the risk estimates. A QRA can always be done, provided one is willing to make enough assumptions. That is not the same as saying a ROBUST QRA can always be completed.

Expert 4 has a very pragmatic answer to this question.

I strongly concur with the opinion presented by Expert 10.

I read "can't" a lot in the opinions of my colleagues, as if risk assessment is a static process with a proscribed set of agreed upon rules. Isn't risk assessment a dynamic process with varied guidelines that also is infused with scientific judgement across all levels. If the current approach is insufficient for a class of chemicals, doesn't it raise the question about the appropriateness of the approach?

I would like to see comments from the analytical experts on expert 7's comment that the LOD is high for PFAS in NHANES data.

I appreciate the comments of experts 9 and 1.

A clarification: essentially I agree with experts 10 and 8. By "systematic testing" in the question I inferred testing in animals for short periods through lifetime. No single government is likely to undertake that job. But there are some systematic testing programs in place that rely on high throughput tests.

I appreciate the comment from expert 4 regarding accuracy and precision.

Of note, a news report I read on 10 19 21 indicates that US EPA is planning to move forward with regulatory determination under SDWA for PFAS. Presumably there will be a quantitative risk assessment produced for at least PFOS and PFOA.

I concur with experts 1 and 4.

I concur with Expert 11.

Re expert 5 comments. I suggest that the more loosely the problem formulation is constructed, the less likely that the risk assessment will provide information needed for choices among risk management options.

One can start with a "loose" draft and refine after some review and comment.

How precursor PFAS are evaluated in risk assessment can be tricky. They (i.e., precurosors) may not be found in human specimens as is, but they form the source for metabolites that accumulate in tissues. Thus, the toxicokinetics of many precursor PFAS needs to be assessed.

Although there were 3 yes's, it is clear that 'scope' definition and 'subgrouping' of PFAS need to be accomplished for any risk assessment purpose. As I indicated in my comment above, for most subgroups of PFAS, we do not have adequate toxicology data and mixture toxicity information. Lumping PFAS as a single group for risk assessment is daunting and prone to large uncertainty. First step would be develop subgroups of PFAS based on some common features.

Having considered the information provided by the other Experts in round 3, I can see why there is some support for the US TSCA definition as a practical matter, for example for consistency. However, even taking that into account, I continue to think that the OECD definition is the most appropriate, even if imperfect.

... identifies some of the data gaps on the exposure side. I continue to think it's important to know how drinking water exposures compare to exposures via other media (e.g. food), since this is important for apportionment.

[EDITORIAL NOTE FOR SCIPI: It would be great if we could edit our comments. I accidently hit the return key and could not edit my response. Thank you.]

Experts 9 and 10 have detailed the most important data gaps, particularly in the dose-response/toxicity portion of the risk assessment process. Expert 8

I agree with scope definition and subgroups of PFAS for risk assessment. Risk assessment should be broader in scope and substances than a MCL, which has to be enforced and labs must be capable to control; thus, needs robustness. Further, the issue of (all) PFAS should be put into perspective since PFAS are not the only contaminants in drinking water and in my view, not the most important ones. Therefore, I doubt that all PFAS with a certain structural group have to be included.

I also did not take this question as being limited to PFAS in drinking water. I took the question as referring to the need to classify PFAS in drinking water as one of the examples (e.g. = for example) of how the classification would be used.

I support the notion of Expert 10: de-prioritization of certain substances/groups based on combinations of properties that make exposure unlikely. However, note that this again depends on the purpose/scope of the RA. In Expert 10's example (not emitted to water and high Henry's Law constant), the substance might be de-prioritized in a drinking water risk assessment, but might still be a priority for assessment of exposure via inhalation. Bottom line: it's critical that the problem formulation be designed to meet the purpose of the risk assessment.

One priority that hasn't really been mentioned is that we obtain more complete phys/chem and tox information on new PFAS rather than letting the information gap continue to widen as new PFAS and their analogues enter commerce.

Given the current state of knowledge, most PFAS risk assessments will be driven by assumptions made to address uncertainty in the face of extremely limited data. In the face of uncertainty, risk assessments to make assumptions that are more likely to overestimate risk than to underestimate risk ("erring on the side of caution"). Since these assumptions are often multiplicative, they can lead to overestimates (by an unknown amount!) of both potency and exposure, and therefore risk. Against this backdrop, I don't think it is possible to complete a "robust" RA for groups of PFAS at this time. However, for specific PFAS for which solid exposure and toxicity (and related) data are available, robust assessments may be possible.

I'm not sure I agree with Expert 6 that the panel is divided. Seven of us selected relative potency and/or dose additivity. Of those seven, three selected both of these options. I also note that dose additivity and relative potency are often used together. Dose additivity is an assumption about how mixtures interact. Relative potency is about which substances contribute most to the toxicity of a mixture.

Having read the thoughts of the other experts, I have re-evaluated my original answer. I the caveats I outlined there amount to a screening step that would be a part of problem formulation. This leads me to concur with expert 5 that PFAS should not be "excluded" from quantitative risk assessments. However, QRAs should use the problem formulation step to focus on the most relevant individual PFAS or groups (which may vary given the purpose and scope of the risk assessment).

I appreciated Expert 8's comment that feasibility of a particular approach can change over time, for example as new data becomes available. I also agreed with several experts that feasibility of any approach depends on the context of the risk assessment. In my mind feasibility is not just about whether the data is available. It is also about whether a useful result can be produced within an appropriate time frame and budget, and can inform risk management strategies if necessary.

Whether or not the question was well-designed, it has led to some good discussion. After reading everyone's thoughts, I conclude that the most appropriate grouping approach will vary by purpose, and by practical considerations (budget, time constraints, availabiillity of suitable data). I agree with reviewer 10 that any of these approaches could be best, depending on the situation.

I agree with many of the comments, but I think many are focused on the assessment of risk, rather than the management of risk. When managing risk from PFAS, it makes sense to group them based on whether and how they will be managed. Note that decisions about risk management are informed by risk assessment, but other factors, for example economics, play into risk management decisions at every level. I like Experts' 3 and 5 suggestion of using a screening tool like total extractable fluoride. Although it is not specific to any substance nor a good predictor of risk on its own, monitoring total extractable fluoride could be a way of evaluating the success (or not) of risk management measures, for example regulations intended to reduce entry of PFAS to the environment to the environment.

Regarding expert 10's comment re limiting scope of risk assessments to relevant (my word) substances, I agree. However, this particular question is about how to define PFAS rather than about how to prioritize them.

Very interesting comments from everyone. At the end of the day, I still think the OECD definition, in spite of various imperfections, is the best available, and represents a consensus of a broad range of experts. It's okay to focus on subsets of PFAS for specific applications /situations, but we should avoid excluding substances too quickly at the screening stage of any assessment, so a broad definition is preferable.

Interesting. I did not interpret this question as focusing on drinking water criteria. That actually would have made it much easier for me to answer! I would have limited the deffinition to those PFAS that have actually been measured in drinking water. If this was the focus for al questions, then the answers would also have been much easier. However, the background document did not focus on drinking water so that is why I did not interpret the question as having that narrow a focus.

Problem formulation problem formulation problem formulation. In the absence of a clear goal it is impossible to judge which information will most add value.

Enjoyed reviews by Experts 9, 8 and 5. Agree wholeheartedly with Expert 7 comment above.

Neglected to add in my above clarification that dose additivity would also be used.

Hi, all. A clarification to Expert 7. It is most likely that PFAS are NOT equitoxic. What I was trying to convey is that for a screen, one could assume that PFAS are all as toxic as the reference chemical (best studied or most toxic). This will result in a conservative or overestimate of toxic potency. One then applies exposure assumptions germane to the exposure scenario identified in the Problem Formulation (there is is again). This Risk 21-like approach can lead to prioritization of the need to do a more refined risk assessment. For, example if with all the conservative hazard assumptions and the appropriate exposure assumptions, the risk is in the "green" area, then either dismiss or give a low priority. For example see Douglas C Wolf, Ammie Bachman, Gordon Barrett, Cheryl Bellin, Jay I Goodman, Elke Jensen, Angelo Moretto, Tami McMullin, Timothy P Pastoor, Rita Schoeny, Brian Slezak, Korinna Wend, Michelle R Embry. Illustrative case using the RISK21 roadmap and matrix: prioritization for evaluation of chemicals found in drinking water. Critical Reviews in Toxicology, 1-11, 2015.

Properties may be the most likely data available, but we be careful that we collect relevant properties given the unusual behaviour of PFAS, e.g. the octanol-water partition coefficient (Kow) is not relevant for fluorinated surfactant soprtion and bioaccumulation behavior. Also because of the unusual behaviour of PFAS, property estimation software often performs poorly.

One should consider the lifecycle of fluoropolymers and not just their use phase. Most of the problems with fluoropolymers relate to their manufacture and there are also likely to be issues with end of life (lack of recycling, incineration, release of persistent microplastics, etc.). I agree with expert 9 that this has been discussed at length in the Lohmann et al. paper. There are many important uses of fluoropolymers in society so total phase out is unrealistic.However, fluoropolymers can be manufactured in a more responsible way (e.g. use of non-hazardous non-fluorinated processing aids as Arkema now do for fine power PVDF) and even end of life issues can be solved, with effort and incentive.

It seems that several experts agree that the question was not well formulated. Different approaches can be used depending on the purpose of the grouping exercise.

I thought that this question was relatively well formulated compared to later questions. It was clear that the purpose of grouping was for example to set drinking water criteria. I'm not sure therefore how the OECD definition can be applied here when it covers thousands of substances for which we have little or no information on exposure and effects. In practice, we have to limit the definition to a few PFAS. However, this does not preclude grouping a larger number of PFAS for example in a restriction proposal (as is ongoing in the EU).

A paper just came out on extractable organic fluorine in human whole blood in Sweden (https://pubs.acs.org/doi/10.1021/acs.est.1c04031). This paper clearly shows that unidentified organofluorines (those that were not included in the 63 PFAS they could measure) contributed to a large portion of the extractable organic fluorine. That was especially true in females ages 18-44: 70% of their extractable organic fluorine was unidentified. This is the strata in the population for which many (if not most) exposure guidelines are developed because points of departure are often based on developmental toxicology studies. Basically, this study shows that we do not have a good understanding of overall exposure to organofluorines... that would support a broader definition of PFAS to make sure we encompass all PFAS that are present in biological samples.

In terms of hazard, I agree there may be more sophisticated ways to define PFAS, and maybe certain groups like CFC and fluorinated pesticides would be excluded if they do not share a common mode of action. It just seems like there is a clear knowledge gap regarding the mode of action of most organofluorines, and I for one can't predict what the next synthesized chemicals will be like.

I agree that problem formulation would help constrain this discussion. In the meantime, I would err on the side of caution, namely because of unidentified organofluorines that are detected in biological samples, and stick with a broader definition.

Expert 7 may be a bit too pessimistic. We certainly can define a reasonable list of possible metabolites for a compound with a defined structure. Since many compounds have common metabolites the universe of metabolites for a complex mixture of PFAS may still be tractable.

I would not support approaches that make the assumption that if we have no ability to measure a compound then we can assume that it is not there. Absence of exposure monitoring data is not proof of the absence of exposure. However, data on the lack of creation of a substance (either intentional or unintentional) is evidence for the absence of exposure, risk, and need for management. This has implications for arguments that monitoring data miss most of the PFAS compounds. The missing compounds will not include chemicals that are not intentionally or unintentionally created.

I would not support excluding chemicals from the class of PFAS because they are not intentionally or unintentionally created. Such chemicals could be used in commerce in the future. I would support using the structure of a chemical that is associated with toxicological and exposure properties as the basis for defining what is PFAS. This is based on my belief that the category of PFAS should be defined in a way that facilitates the protection from human and ecological effects.

In response to Expert 9’s 09/20/2021 11:09 comment/question "Is it better to move forward now with sub-groups of PFAS with known hazard and exposure characteristics only to have to do this again years down the road when more data become available on additional PFAS or spend money now to move toward managing PFAS as a class or as very large subgroups to protect populations".

The answer to this question is “only if there is evidence that using a chemical that is a member of the PFAS class will results in a high likelihood of risk”. I think for chemicals that are minor variations of PFAS and PFOA or are precursors of PFOA, the answer could well be "yes there is excellent evidence". All such chemicals should only be used when exposures and releases are controlled and testing to set safe levels should be required. It is also reasonable to require proof that PFAS that are less closely related (shorter chain length or those with CH2 groups) should he required to prove that they do not behave like PFAS and PFOA or do not have other issues. But, for some chemicals included in some definitions of PFAS (e.g., with no polar groups, that are high molecular weight, or only have a single CF3 group) it gets harder to justify.

I disagree with the exclusion of fluoropolymers from any sub-groupings. Lohmann et al. (2020) provides many counter points to arguments put forward in the Henry et al. (2018) publication, especially as related to production and disposal of fluoropolymers.

The Cousins et al (2020) paper addresses many of the concerns raised by the experts here and could provide some further discussion points that may enhance the depth of the debate.

One approach to grouping that could work for PFAS with commercial applications (and would have to include all precursors, byproducts, and final degradation products that are PFAS) would be PFAS that are essential versus PFAS that are not essential (a definition to essential uses can be found in Cousins, 2019), but like other approaches or a class approach, this would be a huge departure from "traditional" risk assessment approaches.

One point I've seen echoed throughout responses is that a very narrow definition of PFAS may not encompass PFAS for which we currently lack standards and therefore environmental monitoring and/or biomonitoring data => lack of exposure data. Another point is that a very broad definition of PFAS becomes impractical from a management standpoint. Both viewpoints may lead to stasis in action and continue to leave highly exposed communities without any options for reducing ALL PFAS (known and unknown, measured and unmeasured) in their drinking water. Is it better to move forward now with sub-groups of PFAS with known hazard and exposure characteristics only to have to do this again years down the road when more data become available on additional PFAS or spend money now to move toward managing PFAS as a class or as very large subgroups to protect populations (whoops, I may have hinted at the precautionary principle here)?

Please allow the following question (also in connection with production/use): Is there any regulation that handles degradation products or metabolites?

This step would come at the very end and needs lot of work and consensus before.
Values would need specification: threshold value is different from intervention or action value. Any value must be enforceable (and must have a high accuracy in the measurement).

Agree. Should be accompanied by a ranking according to confidence (certainty by Expert 5).

Formulation of defined objectives and scope with clear outputs would be first step. Methods to achieve these to be specified.

The comments bring together the difficulties. Needs a roadmap and results/conclusions from earlier Questions before consideration of addressing or resolving this Question.
"Robust" does not sound reasonable at this stage.

I agree with Expert 6; whole mixture approach is impracticable.
Cannot agree with Expert 3 that all PFAS are equitoxic. Also cannot agree that such approach should be taken due to lack of data.
Same applies for the chemical lab: there is not ONE method to analyse all PFAS. The screening approaches are not mature at all and at research stage.

Approach should be realistic and not be driven by achieving a large number of chemicals and scenarios handled.
Chemical structure alone would not be sufficient for risk assessment.
We also have to think about functionality of the molecule (PFAS have different purposes and persistence can be a desired criterion) and chemical structural groups therein.

I agree on Reviewer 1 first statement that "Given the number of past- and current-use PFAS and the lack of toxicological data on most of them, it seems like using chemical descriptors is the most realistic approach". However, I believe that in the view of the discussions before as to "PFAS" grouping together will only confuse and dilute the process. I prefer concrete examples.
I do not agree that production and use data are easy to access. There are historic data available but they do not match and newer "inventories" use past production numbers and processes. Further, amounts may only be available for regulated chemicals.
Finally, PFAS precursors are blamed for most of the occurrence (in monitoring exercises) and they cannot be covered by production/use amounts.
For any grouping, the source of information needs to be defined and accessible for verification.

I agree with expert 4 that sectoral approach is necessary.
Further it shall be noted that some of the parameters above govern exposure, others effect. Both must be combined and weighted.
A database containing the values for all of the above grouping parameters should form the basis and then being weighted as is typically done for criteria of high confidence (measured/proven in real life situations by many experiments/different groups/ etc ) down to QSAR- or otherwise derived data.

Differentiation between hazard and risk is important and should be iterative. For hazard, a screening process like using EOF or similar can be made at a very early stage but not necessarily by regulatory bodies. Methods are not validated or harmonized and are purely chemical analytical.
Context for risk assessment would be important: comparing C4-C9 PFAS with metals (lead, mercury) or brominated flame retardants or PCB or dioxins, they would form an own group because of fate and transport and exposure (and many other parameters); however, within the PFAS, they may behave very different from each other. Thus, definition of the chemical is necessary together with the scope.
On a regulatory basis, there is no single authority that would be able to regulate all exposure pathways and the endpoint.

I agree that a consensus cannot be found or made when there is a lack of objectives and of applicants. Also, the question should be asked "why is an all encompassing definition of PFAS needed"? Many/most of the chemicals under the long lists of PFAS have not been proven to exist.
Different groups have different needs and definitions. The OECD definition since named as such would address OECD member states only and would have difffulties and bias for others.
As a chemist, I cannot agree on certain terms in the OECD 2021 document because "convenience" from earlier uses in the OECD Global PFC Group has been maintained intentionally. In addition, there is inconsistent use of terms: "PFAS = ..alkyl" substances but perfluoro"octane"sulfonic acid (wrong use of "alkyl" and "alkane"), "per" means "fully" fluorinated but "a CF2 group" is sufficient. Further, I think, the OECD 2021 has too many exemptions or needs explanations.
What concerns nomenclature, I would recommend to refer to IUPAC.
For other disciplines, including toxicology or regulation, such errors or inconsistencies may/do not matter; therefore, a clear scope is necessary and then, an agreed definition for the substances concerned has to be defined in the relevant document.

These comments highlight some of the difficulties to be encountered in moving from a screening level to a more robust risk assessment. They are informative and identify some useful avenues for exploration (e.g. MoA. AOP, PBPK differences) that could eventually provide better information for risk assessment, and especially to better inform possible grouping of PFAS.

The Panel seems most divided on this question and that probably reflects the difficulties in applying a common approach to a screening level risk assessment where there are significant data gaps in regard to toxicological mechanisms and /or common targets for adverse effects. I agree with those reviewers that have concluded that a 'whole mixture' is impractical.

Most reviewers have noted the difficulty of grouping PFAS for risk assessment/risk management purposes where the available information on toxicological properties, relative potencies and environmental behaviour may by lacking. The inclusion of exposure into the grouping process (reviewer 1) is more novel, and while it equates more to the potential for health risk, the initial approach to grouping is, to my mind, probably better based on 'hazard' potential (i.e. toxicological properties). The exposure component will be formally incorporated into any risk assessments that ensue. Some reviewers have raised the quite pertinent question about whether some PFAS allocated to any group or sub-group can actually be measured by current analytical techniques.

There seems to be a consensus that broad definitions do not work if they catch up chemicals with any number of C-F bonds (e.g. the chemicals with a single CF3 moiety, as pointed out by reviewer 11). To my mind, that precludes using the broader OECD 2021 definition. The definition by Buck et al 2011 seems to include the perfluoro- an polyfluoro- chemicals of most concern and puts greater focus on the perfluorinated alkyl compounds similar to those for which more toxicological and environmental fate knowledge exists. While my knowledge of chemicals classification is more limited, it seems to me that the simple Buck et al definition should also capture some precursors and 'replacement' PFAS

I would suggest that an in silico structure-activity relationship model be developed for most widely/commercially used PFASs (probably thousands of them) and development of a potency factor (or toxic equivalency factor) be attempted. This may be aid in screening level risk assessment (as an interim measure).

I still believe that there should several subgroups within the current class of PFAS defined in Buck et al. for the purpose of risk/hazards of PFAS. I do not support that idea that PFAS can be measured as total fluorine or extractable organic fluorine. That is a very crude approach, not only from analytical point of view (current EOF/TOF methods are not standardized/harmonized) but also from toxicological point of view (all PFASs do not have same mechanisms of toxicity); furthermore, inorganic fluorine which is present at much higher levels than those of PFASs can confound/affect the accuracy of the risk/hazard assessment.

It is my understanding that the term PFAS as coined by Buck et al. 2011, was to distinguish perfluorochemical 'surfactants' from chlorofluorocarbons (CFCs)/ozone depleting perfluorocarbons. Earlier the term PFC (perfluorochemicals) was widely used by the scientific community but that term was very broad and could include perfluorocarbons (CFCs). While coining a chemical name based on chemical structure is appropriate (talking as a chemist), coining based on the purpose/utility (surfactant/plasticizer/flame retardant) or usage can lead to confusion. Now we know that perfluorochemical class compasses thousands of chemicals. Therefore, for risk assessment or management purposes, it is better to define what chemicals are in question, within the class of PFAS and then describe a scope of the document with chemical names that are under evaluation.

In response to reviewer 3, I would argue that true additivity, where many small doses can add to cause an effect, is limited to chemicals that operate on a common MIE. But chemicals with multiple MIEs on a common AOP network can at high concentrations also follow a dose additive model, cause synergy, or cause antagonism. But these interactions will only occur when both chemicals occur at concentrations that trigger their respective MIEs. It is the subsequent events on the AOP that interact and not the chemicals. If neither chemical (or if only one of the chemicals) triggers an MIE then there are no separate effects in the AOP network to interact and the mixture will not follow dose additivity. This can be thought of a threshold in dose additivity, that there are doses of such chemicals below which the chemicals do not add.

Now MIEs can be triggered at concentrations below the doses that cause apical effects, so in practice it is hard to say when a predicted concentration is not large enough to trigger an MEI. But it does suggest that if you have a mixture of a PFAS where one compound predominates and 20 other PFAS compounds are at much lower low doses, then even if all compounds caused a common endpoint in practice they may not add.

I agree with expert 3 and 5. If total organic fluoride or total extractable fluoride was used as a low tier screen for samples, then a single measurement could be used for all PFAS with the assurance that no PFAS would be missed. The problem with such a screen is that if there is no higher tier assessment approach available, then measurements that fail the screen would trigger regulation. Depending on how the screening assessment was designed, this could be a substantial cause of over regulation. Thus, such a screen should be proposed with guidance on what to do if the finding comes back as positive.

In several of the comments above, the issue of data gaps appears (can we detect all PFAS in a sample, can we characterize individual's combined exposures to all PFAS, do we know the toxicity and dose response of every PFAS compound). It would be good to take this on as a specific aspect of decision making for PFAS in a systematic way. How many and what are the exposure implications of the unknown PFASs seen in biomonitoring samples or in relevant exposure media. What are the likely ranges of toxicity for the compounds? How much risk is being missed. Which sources of exposure have the biggest data.

Up to now I have only seen this issue invoked on a piecemeal basis as a justification for a testing approach. I am concerned that without giving this question proper attention we may be making inappropriate decisions.

As noted in the earlier comments in this debate, the overall definition of PFAS is dependent on the purpose for establishing the class i.e., a problem formulation. (I think this point was made or implied in the original comments from experts 3, 4, 8, 9, 11 (me).) Without a clear description of the problem formulation, the trend has been to make the definition increasingly broad over time. That is without knowing purpose of the class you can’t offer a justification as to why a chemical should be excluded.

For example, I am not comfortable is including the many pesticides that have one or more -CF3 moieties as PFAS. They do not behave like PFOA or PFOS, they are currently well regulated, and in many cases their structures contain very biologically active moieties that drive the pesticides toxicity not the -CF3.

The answers to questions 1 and 2 are linked. The broader the definition of PFAS the more need for characterizing and managing risks posed by PFAS in terms of subgroups. As OECD acknowledges for its broad definition, many findings about PFAS only apply to portions of PFAS and not the entire class.

I agree with Expert 5's comment above, but I wonder if we can be sure that a non-bioavailable compound will not break down into a metabolite that is bioavailable. If there's a way to ensure a chemical, e.g., polymer, will not be broken down by UV, bacteria, enzymes, etc., then it may be possible to exclude it. Not being a chemist, I don't know if there are criteria that would allow such an evaluation solely based on chemical structure.

Each of the approaches will have its own limitations of feasibility, its intrinsic value and applicability and its own level of confidence; and these attributes may vary internally from one compound to another. Thus, a broad approach, taking advantage of each approach's strengths might be considered. I doubt that there will be a single "best" approach, a situation based on the high likelihood of differential data of varying levels of certainty for given chemicals. Yes, all will be useful.

Problem formulation would be an important first step. With that taken, then its limits on data evaluation may help identify (and prioritize) data gaps.

Keeping in mind that risk is a function of both hazard and exposure, I agree that exclusion may be considered for those parent chemicals that are non-bioavailable, and/or do not cross biological membranes ... to a point. Some consideration might be given to their environmental degradation products. Kudos to reviewer 7, "conditions that do not occur do not need to be assessed (since often they disturb the assessment)"- this would be an important point to include in the problem formulation.

I relate best to answers provided by experts 1,4 and 6 - in that the data most likely to be available and undisputed may be chemical properties. Grouping to discriminate among chemicals might be best accomplished on the basis of factors that differ among groups, and so the use of persistence, as common to many, may induce complications, especially if potencies differ among chemicals with similar half-life values. Production and use data might be used to prioritize (but not necessarily group) chemicals for risk evaluation

Regarding scientific merit as focussed here on risk, I find the ratings reassuring in that toxicity parameters seemed to receive generally high marks. It must be noted, however, and may be addressed in further discussions that such data may not be available. Thus, some hierarchy of grouping strategies might be sought.

Managing the risk assumes that the risk has been assessed and the risk rises to the level that management is required. A screening risk assessment of on some worst-case scenario, and perhaps based on total fluorine might be feasible. However, such an approach might also be fraught with damning uncertainty. It is interesting to consider, and the management of risks could be accomplished on a class by class or group by group basis. given divergent differences in half-life and potency, I can imagine that management as a single class/group may exclude from use some fairly innocuous compounds whose economic importance may be fairly high.

Once the purpose of the definition is established, the definition should be carefully crafted to fit the purpose. Because the emphasis is on risk assessment, some sentiment may be given to including only those chemicals with sufficient data to enable a risk assessment unencumbered by crippling uncertainty. However, developing such a strict definition (based on ability to assess risk presently) will preclude including other chemicals for which data may be forthcoming. Respectfully, I can't agree that commercial significance or biological half-life should be included in the definition - the name of which implies structural characteristics.

Looks like good agreement that all of these would be useful!

Key data gaps depends on the specific group/subgroup being assessed and the purpose of the assessment.

Agree with the comment above from reviewer 3.

I think that the feasibility of each of these depends on the purpose and scope of the assessment. Risk assessment is an iterative process and some mey be initially feasible and others may become feasible over time. What is lacking, however, from this list as well as from all the questions, is exposure. Production is often used as a surrogate, but for some PFAS compounds there are actual measures of human exposure.

I agree with reviewer 3 comment. I also had difficulty answering all of the questions as the background document was not a real problem formulation statement. Without a clear purpose for an assessment, and no conceptual model, etc., it is very difficult to answer the questions as there is no context for them.

As mentioned by some of the reviewers, risk is only present when there's both hazard and exposure. Consequently, grouping strategies should ideally account for both hazard and exposure.

In terms of hazard, it's unrealistic to think we will have tox data on all existing PFAS and be able to keep track of new PFAS coming to the market every year. When facing uncertainty regarding MOA and potency of PFAS, it may be preferable to assume similar MOA and dose additivity. Yes, this may be a little on the conservative side, but the approach is simple and sustainable (doesn’t require tox data for each new PFAS).

With regards to exposure, data from national surveys could be used to assess internal exposure to a certain number of PFAS, but there’s still a lot we don’t know about the total PFAS load because many PFAS are not measured. Where total extractable fluorine and targeted analyses were performed on the same serum samples, targeted PFAS explained only a portion of the total organic fluorine. Data from NHANES and ATSDR contaminated communities show that the composition of PFAS in serum changes over time, and geographically. Again, we are facing uncertainty regarding exposure.

Although I agree that the most robust approach would be to group PFAS based on MOA and/or PK, it seems like this approach would exclude a large portion of PFAS, and potentially underestimate health risks.

I don't think that a whole mixture approach will be suitable. That is, there would be entirely too many PFAS "whole mixtures" to test for toxicity of any sort, including high throughput screenings.

And once again in the absence of a problem formulation, I am hard pressed to say what a whole mixture is.

Agree that production and use data are likely to be most available. Appreciate comments by reviewers that for screening could use these data with default assumptions for conservative estimate.

I did not agree with reviewer 1's comment that grouping all the PFAS together is the way to proceed. But when I got to discussion of screening approaches -- maybe that is the way to go. In my original comments on screening I suggested considering all PFAS to be equitoxic, assume additivity, and compare estimated exposure to a health benchmark.

I appreciate reviewer 4's comment that the definition of PFAS ought not rely on the availability of toxicological data. But that may enter into iteration of subgroups.

I found reviewer 2's comment interesting, but I am not sure I agree.

Reviewer 8's comments appear to me to describe development of a conceptual model in my parlance.

I interpreted several responses to this question as echoing my quandary with all the questions. That is, the document is not a problem formulation in that it does not specify the reason for doing the assessment, provides no conceptual model, sets no limits. It is hard to define a group when there is no indication as to how the grouping will be used.

I appreciated reviewer 11's comments on PFOA and PFOS and inclusion or not in groups

I agree with reviewer 10 that 1.3 is a poorly designed question.

I found reviewer 11's comments to be particularly useful and comprehensive. I may differ, however, whether additivity should be limited only to those PFAS with the same MIE. I agree that interconnecting AOP tend to be complex (connections or branches at other than the first key event), but I am uncertain as to how that affects the likelihood of interaction among components of a mixture. Are not most interactions noted at high exposures?

None of these approaches is bad. But I appreciate the comments that several of these criteria for grouping are linked.

Just want to stress that risk assessment is an iterative process. As the assessment proceeds, it is very likely that the subgroups will change both in composition and in rationale. All this needs to be documented.

Reviewer 4's comment made sense to me

As noted by several reviewers, one large group will likely not be practical. Rather iterations of subgroups can be defined as the assessment proceeds -- based on both need and data availability.

A few of the reviewers stated that all PFAS should be grouped together and managed together. I am curious how one would propose to actually combine all PFAS and what regulatory action and what agency could possible have purview over all uses and exposures?

It sounds like several of us agree that the definition shoould be determined by the risk assessment question being addressed - i.e. we need a definitive problem formulation statement.

It may be helpful to test these definitions against certain compounds that may not fit the more stringent criteria. For example, we have to make sure the definition includes chemicals like perfluoro-4-ethylcyclohexanesulfonate (PFECHS), a compound which has been found in human serum (Miaz et al. 2020 - https://pubmed.ncbi.nlm.nih.gov/32182307/) and contains a cyclic structure. Some definitions mentionned above may be too stringent for chemicals like PFECHS, which I believe should be included when assessing the risks of PFAS.